This report is made possible by the generous support of the American people through the United States Agency 
for International Development (USAID) under the terms of contract no. 7200AA18C00057, which supports the 
Research Technical Assistance Center (RTAC). This report was produced by Duke University under the RTAC 
contract. The contents of this report are the sole responsibility of RTAC and NORC at the University of Chicago, 
and do not necessarily reflect the views of USAID or the United States Government.
Evaluating Climate-Related 
Migration Forecasting Models
Kerilyn Schewel, Sarah W. Dickerson, B. Madson, Gabriela Nagle Alverio
ADD IMAGE HERE.
BE SURE TO REPLACE PHOTO WITH ONE THAT IS RELEVANT AND CONTEXTUALLY 
APPRIOPRIATE.
MOVE OR REMOVE ENTIRELY AS NEEDED.
WHERE CAN YOU FIND PHOTOS? CHECK OUT OUR PHOTO RESOURCES
(See Branding and Communications Quick Reference and FAQ document).
DON’T FORGET TO ADD PHOTO CREDITS
Report | December 2022
R E S E A R C H T E C H N I C A L A S S I S T A N C E C E N T E R
Evaluating Climate-Related Migration Forecasting Models 1
Research Technical Assistance Center
The Research Technical Assistance Center is a network of academic researchers generating timely 
research for USAID to promote evidence-based policies and programs. The project is led by NORC at 
the University of Chicago in partnership with Arizona State University, Centro de Investigación de la 
Universidad del Pacifico (Lima, Peru), Davis Management Group, the Duke Center for International 
Development, Forum One, the Institute of International Education, the Notre Dame Initiative for Global 
Development, Population Reference Bureau, the Resilient Africa Network at Makerere University 
(Kampala, Uganda), the United Negro College Fund, the University of Chicago, and the University of 
Illinois at Chicago.
Cover photo credit: USAID/SERVIR
Suggested Citation
Schewel, Kerilyn et. al. 2022. Evaluating Climate-Related Migration Forecasting Models. Research 
Technical Assistance Center: Washington, DC.
Report | December 2022 2
Table of Contents
Introduction ........................................................................................................................................................................5
Background..........................................................................................................................................................................6
Categories of Climate-Related Migration................................................................................................................6
Population and Migration Modeling ..........................................................................................................................8
Climate-Related Migration Forecasting Model Types ..........................................................................................8
State of the Field..............................................................................................................................................................11
Model Inputs.................................................................................................................................................................12
Model Outputs.............................................................................................................................................................13
Nine Models......................................................................................................................................................................14
Model 1. The Groundswell Reports.......................................................................................................................15
Model 2. The Great Climate Migration Model ....................................................................................................16
Model 3. The Universal Model.................................................................................................................................17
Model 4. The Systems Dynamics Model................................................................................................................18
Model 5. The Dynamic Model..................................................................................................................................18
Model 6. The Inequality Model ................................................................................................................................19
Model 7. The Global Model......................................................................................................................................20
Model 8. The Small Island Model.............................................................................................................................21
Model 9. The Statistical Extrapolation Model ......................................................................................................22
Summary of the Nine Models ..................................................................................................................................25
Areas of Future Research..............................................................................................................................................28
Policy Implications...........................................................................................................................................................29
Exposure Modeling .....................................................................................................................................................30
Foresight Research .....................................................................................................................................................30
Improved Data Analytics...........................................................................................................................................31
Key Takeaways.................................................................................................................................................................32
References.........................................................................................................................................................................34
Appendix A.1: Background on Search Strategy........................................................................................................38
Appendix A.2: Background on Expert Interviews....................................................................................................41
Appendix A.3: Details of the Nine Models................................................................................................................43
Evaluating Climate-Related Migration Forecasting Models 3
List of Tables
Table 1. Overview of Key Characteristics of Climate Migration Forecasting Models Reviewed.................11
Table 2. Description of nine climate-related migration forecasting models......................................................24
List of Figures
Figure 1. Conceptual Framework for the Drivers of Migration.............................................................................7
Figure 2. Literature Search Process ............................................................................................................................39
Report | December 2022 4
List of Acronyms
ABM Agent Based Model 
GDP Gross Domestic Product
IPCC Intergovernmental Panel on Climate Change
OECD Ogranisation for Economic Co-operation and Development
RCP Representative Concentration Pathways
SPEI Standardized Precipitation Evapotranspiration Index
SSP Shared Socioeconomic Pathways
USAID Unites States Agency for International Development
Evaluating Climate-Related Migration Forecasting Models 5
Introduction
Climate change will have significant impacts on all aspects of human society, including population 
movements. In some cases, populations will be displaced by natural disasters and sudden-onset climate 
events. In other cases, climate change will slowly reshape the economic, social, and political realities of a 
place, which will influence how and where people migrate. Planning for the wide spectrum of future 
climate-related mobility is a key challenge facing development planners and policy makers. 
Human migration brings opportunities and challenges to both sending and receiving societies. Migration is 
a key adaptation mechanism vulnerable households can use to cope with climate change, and whether 
receiving societies experience benefits or strain from population growth will depend on key investments—
in housing, jobs, infrastructure, and social services. To best plan these investments requires an 
understanding of how future population movements will be affected by climate change. 
This report reviews a number of prevailing and promising modeling approaches for forecasting the nature, 
magnitude, and direction of climate-related migration over the next 30 years. We pay particular attention 
to how well models are likely to forecast migration across geographic contexts, for different population 
groups (including women and marginalized groups), the degree to which models integrate other 
developmental or conflict-related drivers of migration, and whether models capture the potential for 
trapped populations. Our findings are based on a systematic literature review (see Appendix A.1 for 
details) and benefit from the insight and expertise of eight climate migration experts and modelers (see 
Appendix A.2 for details). 
Our report finds that the field of climate-related migration forecasting is still in its infancy. Modeling 
experts caution that at this stage of model development, numerical projections to 2050 should be seen as 
notional at best. Modeling human behavior, including migration, is fraught with uncertainty, and adding the 
dimension of climate change only compounds that uncertainty. For this reason, a scenarios-based approach 
is preferable to an approach based on single narratives of future trends. Attention should be given to 
variation across the full spectrum of future scenarios, and policies oriented towards encouraging best-case 
outcomes. 
Exposure mapping remains a useful tool to identify at-risk populations, and the United States Agency for 
International Development (USAID) should support a multi-pronged strategy to enhance climate resilience 
in vulnerable regions, including facilitating migration as an adaptive strategy, investing in in-situ adaptation, 
and enhancing the capacity of local and regional urban centers to accommodate and benefit from in￾migration. To further strengthen foresight capacities, we recommend convening foresight exercises that 
bring modelers together with climate scientists, migration scholars, development practitioners, and other 
stakeholders to provide a more comprehensive picture of potential future trends for specific countries or 
regions. 
Climate-related migration modeling is significantly hampered by limited data on past and present migration. 
To improve the frequency and accuracy of census data collection, investment in statistical bureaus of 
developing countries is crucial. Finding innovative ways to capture flow-data is also important to capture 
short-term mobility, displacement, and irregular migration trends. Continued investment in individual and 
household level surveys that include questions on migration aspirations, plans, ability, and non-climate￾related drivers of migration will strengthen capacity to distinguish voluntary and involuntary forms of 
climate-related migration and immobility, and to anticipate divergent outcomes for various social groups 
and marginalized populations. 
Report | December 2022 6
Background
For more than thirty years, research published by scientists and reports in the news media have warned 
that climate change will cause mass migration and displacement on a global scale. A predominant early 
assumption was that climate change and migration have a linear, cause-and-effect relationship, in which 
climate induced drought, rising sea levels, and natural disasters result in the movement of affected 
populations. Approaches to forecasting migration initially focused on exposure mapping (also called hazard 
mapping): identifying areas threatened by climate change and assuming the vast majority of residents of 
affected areas would be forced to leave. This led to catastrophic projections of climate migrations and 
environmental refugees (Brown 2008; McLeman 2014). Fortunately, these early projections have failed to 
become reality. They did not adequately account for how climate-related factors interact with non-climate 
related drivers of migration, the potential for in-situ adaptation, and instances in which climate change 
impacts may suppress mobility, particularly in low-income countries. Their shortcomings spurred more 
nuanced investigations of how the impacts of climate change intersect with existing mobility systems and 
development conditions to affect the nature, volume, direction, and composition of migration flows. 
Attempts to forecast climate-related migration have grown in number and sophistication in recent years, 
particularly since the early 2010s. Before reviewing these models, this section discusses the 
conceptualization of climate-related migration, the origins of climate-related migration models, and the 
most common types of models used today to forecast climate-related migration.
Categories of Climate-Related Migration
The International Organization for Migration, a United Nations agency, defines climate migration as “the 
movement of a person or groups of persons who, predominantly for reasons of sudden or progressive 
change in the environment due to climate change, are obliged to leave their habitual place of residence, 
or choose to do so, either temporarily or permanently, within a State or across an international border” 
(International Organization for Migration 2019). This definition is broad; it encompasses many different 
kinds of climate-related migration, spanning the spectrum of forced to voluntary, internal and international, 
temporary and permanent. This breadth poses a challenge to research, forecasting, and policy-making 
related to migration and climate change, because the kinds of migration being studied can vary considerably 
and require significantly different policy responses. 
Brown and McLeman (2013) suggest climate-related migration may be categorized according to the nature 
of the climatic stimulus (sudden-onset climate events versus gradual changes in prevailing conditions) and 
the nature of the migration response (distress migration versus adaptive or amenity-seeking migration). 
Sudden-onset climatic events, such as floods and storms, are often associated with distress migration or 
displacement, in which large numbers of households abandon their place of residence at short notice. 
Alternatively, households can become trapped in place by sudden-onset events (e.g., floods that shut down 
roads). Climate-induced displacement often takes place over relatively short distances, and return 
migration tends to be common. The link between a sudden-onset climate event and distress migration is 
more direct. But where, how, and whether people move in response to that event is shaped by pre￾existing migration systems, the resources and networks of affected households, government or 
humanitarian interventions, and the broader development context. 
Slow-onset changes include increasing temperature, irregular rainfall patterns, sea-level rise, ocean 
acidification, soil salinization, loss of biodiversity, and desertification (de Sherbinin 2020). These slow-onset 
changes interact with migration outcomes indirectly and often in a non-linear fashion. Other political, 
Evaluating Climate-Related Migration Forecasting Models 7
economic, cultural, or conflict-related drivers of migration or immobility may be stressed by climate 
change, and these social phenomena mediate climate impacts and the responses of both individuals and 
households. For this reason, slow-onset climate changes contribute to migration but are often not the 
proximate cause of migration. This is why, for example, many migrants in drought-affected areas will 
attribute their primary reason for migrating to economic rather than climate-related factors (United States 
Agency for International Development Honduras Brief on Climate Change, Food Security and Migration). 
Some describe slow-onset climate impacts as threat multipliers—for example, drought can reduce crop 
yields and thus household incomes, or it can exacerbate conflict in water-scarce regions (McLeman 2014; 
Sofuoğlu and Ay 2020). In the context of slow-onset climate change, “thresholds” or “tipping points”
become important; many populations will attempt to adapt in place despite significant livelihood stress, 
until in-situ adaptation fails and/or a tipping point is reached in which a significantly higher share of 
households choose migration as their primary adaptation strategy (McLeman 2018). 
Figure 1. Conceptual Framework for the Drivers of Migration
Source: Foresight (2011, page 33)
Figure 1 presents a theoretical framework from the influential Foresight: Migration and Global Environmental 
Change study conducted by government of the United Kingdom in 2011. It illustrates how climate changes 
intersect with environmental, political, economic, social, and demographic factors at the macro- and 
community-level as well as micro-level variables related to personal and household characteristics and 
meso-level intervening obstacles and facilitators. This figure does not illustrate how the decision to migrate 
or stay is then affected by individual and household capabilities to migrate. Those who aspire to migrate 
but lack the ability to do so may become trapped or “involuntarily immobile” (Carling 2002). The potential 
for climate change to trap poor populations in place is as important a humanitarian concern as climate￾related distress migration. 
Because climate change acts indirectly on pre-existing migration systems, it is difficult to disentangle the 
relative impact of climate-related variables from other migration drivers. This is especially challenging for 
forecasting models. To date, forecasting has tended to operate under the assumption that populations are 
relatively fixed unless uprooted by a climate event. In reality, populations are constantly moving internally 
and internationally. They move in seasonal, temporary, or permanent ways for work, education, adventure, 
security, or family, and across stages of life (Cundill et al. 2021; Van Praag 2021). Population movements 
Report | December 2022 8
within and from a country even have a patterned relationship with levels of economic development. For 
example, more people move to towns and cities as economies industrialize, and in many countries, a 
growing percentage of the population migrates internationally as countries move from low- to middle￾income status–a phenomenon referred to as a country’s “mobility transition” (Zelinsky 1971; de Haas 
2010; Clemens 2020; Schewel and Asmamaw 2021). This suggests that modeling should not focus 
exclusively on identifying climate migrants, but rather on clarifying how climate change will reshape or 
constrain existing mobility systems. 
Population and Migration Modeling
Current approaches to forecasting climate-related migration have their roots in a longer history of 
forecasting population growth and distribution, which later evolved into more focused migration 
forecasting. Population forecasting is meant to predict future population distributions across rural and 
urban places based on factors like fertility, mortality, and migration. Population forecasting has been 
utilized by statistical agencies and development planners for many decades (Shryock and Siegel 1980). 
Over time, population forecasting tools have advanced, allowing researchers to incorporate a greater 
array of assumptions and to project more specific forecasts (e.g., by age) (Wiśniowski et al. 2015). 
Migration modelers build on the sub-component of migration within population forecasting through a 
variety of methods. Often modelers utilize past migration data to predict future migration trends, 
incorporating additional variables, such as economic or demographic factors, into their models (Disney et 
al. 2015). All models have a certain degree of uncertainty baked into them, as no model can predict future 
shocks or gamechangers like wars, pandemics, or major technological breakthroughs. The lack of high 
quality longitudinal historic migration data in most places makes it difficult to know precisely how people 
have moved in the past, and thus how they might move in the future. Further, migration modelers still 
struggle to capture the nuanced relationship between interacting drivers of migration in different 
socioeconomic contexts (Lutz and Goldstein 2004). Models tend to focus on economic or demographic 
variables, but social, political, and cultural factors also play a role in determining who migrates, where they 
go, and the degree of choice in the migration process. However, we often lack reliable data to capture 
these non-economic drivers of migration and immobility. 
Climate-related migration modeling fits into this context, as researchers incorporate climate-related 
indicators into population and migration models of all kinds. Climate-related variables are not yet 
commonly included in more general migration modeling, but a sub-field of climate-related migration 
modeling has emerged to address this gap. A distinct challenge to climate-related migration forecasting is 
the added uncertainty about what major tipping points may impact climate change in the future, along with 
the ability of humans to adapt to climate change via breakthroughs in technology or coordination. 
Climate-Related Migration Forecasting Model Types
Researchers utilize a variety of models from different disciplines to forecast climate-related migration. 
Most climate-related migration forecasting modeling to date has been experimental, focusing on refining 
new methods. Model types include: 
• exposure models that overlay climate-related hazards on a population distribution map to 
identify at-risk populations; 
• agent-based models that simulate the actions and interactions of individual agents; 
Evaluating Climate-Related Migration Forecasting Models 9
• gravity models that use population size, distance, and other variables to project future 
population distributions;
• radiation models that use population size and distance to model the flows of people 
between places;
• statistical extrapolation models or discrete history event models that model historical 
climate-migration interactions to project future trends;
• systems dynamics models that simulate the non-linear behavior of complex systems 
using complex econometric models and utility functions; 
• computable general equilibrium models that use large, economic models to assess 
potential policy effects on real-world economic problems; 
• integrated assessment models that integrate human systems and natural systems into 
one modeling framework to support informed policy making;
• machine learning models that have the potential to identify thresholds or tipping points 
in migration systems. 
The above models differ significantly in assumptions made, data inputs required, and the nature of results. 
Each model type has advantages, disadvantages, and a preferred scope of application. For example, agent￾based models (ABMs) are bottom-up, data-intensive models that tend to be better at exploring nuanced 
questions and mechanisms in circumscribed geographic settings such as villages, cities, and other sub￾national settings. Gravity models are top-down models that tend to be better at forecasting spatial patterns 
over larger geographic areas, such as entire nations or regions.
Our review finds that ABMs are the most common model type used in climate-related migration 
forecasting. ABMs model the behavior of autonomous agents to explore how individual decision processes 
lead to changes at the population level. ABMs tend to require rich data inputs to calibrate the model. 
When this data exists, they are well-suited to explore causal and feedback mechanisms, migration 
motivations, and variations in migration or staying behavior based on individual and household 
characteristics. Typically, ABMs use data obtained through household survey (ideally longitudinal) 
research. One limitation to ABMs is the limited spatial precision of model outputs. ABMs tend to estimate 
volumes, or relative increases or decreases in migration flows from a particular area. They do not typically 
model source or destination areas. Like other models, ABMs are more accurate at predicting short-term 
changes in the volume and composition of migration flows. They are less reliable over the long-term.
Gravity models are another popular but fundamentally different alternative. The name comes from 
Newton's law of gravity, which states that any two bodies attract one another with a force proportional 
to the product of their masses and inversely proportional to the square of the distance between them. In 
gravity models, population size is indicative of relative “attractiveness,” but the force of attraction decays 
with distance. From this basic interplay of population size and distance, additional inputs can be added to 
explore their effects on future population distributions, including climate variables. 
Gravity models do not directly model migration. Instead, climate-related migration is assumed to be the 
primary driver of deviations between population distributions in model runs that include climate impacts 
and the development-only (the “no climate”) models that include non-climate related drivers (Rigaud et 
al. 2018). This relatively straightforward approach makes them an attractive choice for migration modelers. 
However, using population as a proxy for migration can also be problematic, because population is also a 
function of changing administrative boundaries, fertility, and mortality. We know that fertility and mortality 
Report | December 2022 10
rates are also affected by climate change, yet in many gravity models, fertility and mortality rates are held 
constant. 
Strengths of the gravity model include the ability to reproduce past shifts in population distribution, 
providing some assurance that if population trends continue to behave as in the past, these models can be 
trusted to forecast future trends. Gravity models are also relatively flexible in terms of data inputs and 
can generate results over broad geographic scales. These models also have limitations. First, much data 
used in global or regional models rely on sparsely collected census data over 10 year increments, which 
means the models may not accurately represent mobility patterns over short periods of time, short 
distances, and from places where census data is rarely collected. Census data typically focuses on 
interprovincial moves, though research suggests that in many places climate-related migration occurs 
mostly over short distances. Thus, as one expert shared, gravity models can overrepresent long-distance 
rather than short-distance movers, which may reflect different segments of society. Further, these models 
cannot tell us anything about migration motivations or the degree to which migration is voluntary or 
forced. Finally, they do not forecast migration well at a micro level. For this reason, they are not typically 
utilized for forecasting migration from small island states or other small geographic regions. 
Another approach to forecasting climate-related migration draws on discrete history event modeling,
which uses historical data to evaluate how sudden- or slow-onset climate impacts affect internal or 
international migration patterns. Those historical relationships are then used to project the probability of 
future migration under different climate scenarios. Discrete history event modeling has the advantage of 
basing model assumptions in real-world experiences—as one expert put it, “ground-truthing” 
relationships between climate change and migration—rather than extrapolating migration trends from 
population projections as gravity models do. These studies need to carefully control for other 
determinants of migration. When done well, they have been fundamental to enhancing our theoretical 
understanding of the climate-migration-development nexus (see, for example, Henry, Schoumaker, and 
Beauchemin 2004; Feng, Krueger, and Oppenheimer 2010; Nawrotzki et al. 2012; Gray and Mueller 2012; 
Gray and Wise 2016). Not all studies that use this approach attempt to forecast migration; nevertheless, 
discrete history event modeling contributes to the evidence base and theoretical assumptions that inform 
the development of other forecasting models, like ABMs. In this report, we refer to those that do forecast 
migration as statistical extrapolation models. 
Although ABMs, gravity models, and statistical extrapolation approaches are some of the better-known 
model types, modelers continue to experiment with many others. Radiation models, for example, have 
been used to forecast migration flows between places with very few data inputs. As do gravity models, 
they rely on basic inputs related to population size, distance, and climate. The same limitations associated 
with using population as a proxy for migration discussed above for gravity models also apply to radiation 
models. Systems dynamics models offer the opportunity to explore complex systems dynamics and the 
potential impact of different policy scenarios, but they tend to lack spatial specificity. Computable general 
equilibrium and integrated assessment models share a similar focus on exploring the effects of policies on 
societal outcomes, but have only recently been applied to forecast climate-related migration. There is 
significant interest in machine learning models, but the large data-inputs required to use machine-learning 
significantly limits their application in data-scarce contexts. 
Evaluating Climate-Related Migration Forecasting Models 11
State of the Field
To assess the state of climate-related migration forecasting models, we conducted a systematic literature 
search of climate-related migration forecasting models. Our search initially yielded a total of 30 models, 
which we coded with regard to a variety of factors including type, model prediction, geography, intervening 
variables, and data sources. Based on USAID’s strategic interests, we removed models that focused solely 
on developed countries such as the United States and Australia. We focused on models that provide 
projections roughly over the next thirty years. We removed those that provided predictions for the year 
2100 and beyond, as we judged that date to be too distant to inform USAID programming. Finally, based 
on our expert review, we added one statistical extrapolation model to ensure we had an example of this 
approach to forecasting. The final analysis included 20 models (see Appendix A.2). The following reviews 
the wide range of climate-related migration forecasting model types and key features of these twenty 
models. 
Forecasting models covered multiple scales of migration (e.g., sub-national, national, regional, and global). 
Some forecast migration at one scale, while others combine scales and are counted twice in the 
summary statistics presented in Table 1. For example, 65 percent of the models forecasted trends at the 
national level, 30 percent at the global level, 15 percent at the regional level, and 15 percent at the 
community level. Of those models that focused on specific countries and fit the inclusion criteria, four 
focused on Bangladesh, two on Brazil, and two on Thailand. One model focused on Burkina Faso, 
Central America and Mexico (together), Mexico (alone), Kiribati, Maldives, and Nigeria, respectively. 
Table 1. Overview of Key Characteristics of Climate Migration Forecasting Models Reviewed
Geographic 
Coverage
Count Model Count Climate 
Hazard
Count Type of 
Migration
Count
National 8 Agent-Based 
Model
7 Precipitation 11 Internal 15
National, 
Regional, 
Global
1 Gravity Model 2 Temperature 9 International 12
National, 
Sub-national
4 Computable 
General 
Equilibrium 
Model
2 Sea level rise 9 Involuntary 3
Regional 2 Radiation 
Model
2 Storms 2 Voluntary 2
Global 5 Economic 
Model
3 Drought 3 Permanent 4
Other models* 4 Temporary/ 
Circular/ 
Seasonal
2
*Including Integrated Multi-Regional Applied General Equilibrium, Spatial Equilibrium, Statistical Extrapolation, and Systems Dynamics models.
18 models did not state whether future migration would be voluntary/involuntary and 16 models did not explicitly consider permanent versus 
temporary migration. 
Report | December 2022 12
Of the models we reviewed, 35 percent employed the agent-based model. Fifteen percent used an 
economic model (e.g., 2SLS, probit). Ten percent employed the gravity model, the computable general 
equilibrium model, and the radiation model. Integrated multi-regional, applied general equilibrium, spatial 
equilibrium, statistical extrapolation, and system dynamics models were each utilized once. Notably, not 
all models predicted migration for the same time horizon. Forty percent predicted to 2050, while others 
predicted to 2040, 2045, 2055, 2060, or 2080. 
Model Inputs
The most common climate-related data inputs included in the forecasting models we analyzed were 
related to precipitation (e.g., average rainfall, monthly and yearly) (55 percent), temperature (e.g., average 
temperature) (45 percent), sea level rise (45 percent), droughts (15 percent), and storms (10 percent). 
The pairing of precipitation and temperature hazards was most common across all models, with 30 percent 
of the papers analyzing these together. Some models also included crop yields, water availability, sunshine 
hours, elevation, days of extreme heat, erosion, and flooding. Forecasts that take a scenarios approach 
most often use representative concentration pathways (RCPs) to project future climate scenarios. A RCP 
is a greenhouse gas concentration trajectory adopted for climate modeling and research by the 
Intergovernmental Panel on Climate Change (IPCC). 
The most common development-related data inputs in the forecasting models we assessed are the 
shared socioeconomic pathways (SSPs). SSPs are scenarios of societal change developed by members of 
multiple research communities which include indicators of population, economic growth, education, 
urbanization, and the rate of technological development. SSPs describe scenarios for how the world might 
evolve in the absence of the implementation of additional climate policies. SSPs and RCPs are the most 
common aggregate indicators used to forecast potential future scenarios of climate change and 
development trajectories. 
Some models include additional or independent inputs related to population distribution and the 
economic, political, or social context. Additional demographic inputs include gender, education level, 
age, rural or urban location, marital status, and number of children. Many of these variables are more 
easily incorporated into models that focus on micro-level household dynamics, such as ABMs. Some 
models also include population, fertility, or mortality projections. 
Economic inputs—including Gross Domestic Product (GDP), household income, and occupation—are 
included in well over half the models we analyzed. Fewer models consider other economic factors, such 
as wage-differences between regions, percent of the population working in agriculture, employment data, 
household assets, level of education, local amenities, and local wage rates. Many models do not specify 
economic drivers directly, instead utilizing SSPs as proxies. 
Political and social factors are not included in models as frequently as economic inputs. Of the models, 
only 25 percent included political inputs and 30 percent included social inputs. Political inputs can include 
government stability, and the freedoms, rights, and liberties enjoyed by citizens. In some cases, political 
inputs measure the political feasibility of climate adaptation through, for example, policies or global 
cooperation. Notably, only one model we assessed considered conflict. Social inputs tend to capture social 
networks, or the connections an individual or household has to others at origin or at a potential 
destination. Data can come from survey research that asks about this directly or can be implied by data 
on remittances or the size of the diaspora. Some papers utilized pre-existing datasets, such as those from 
the World Bank or United Nations. 
Evaluating Climate-Related Migration Forecasting Models 13
Model Outputs
Models vary in the type of migration projected. Of the models we reviewed, the majority (75 percent) 
forecast internal migration trends, and just over half assess international migration (60 percent). Some 
models explore both internal and international migration. Most models do not state explicitly whether 
they are forecasting temporary or permanent migration. Short-term or temporary migration generally 
refers to migration lasting between three and 12 months, and long-term or permanent migration refers 
to a change of residence for one year or more. Only 20 percent of the models explicitly focus on 
permanent migration, while 10 percent investigate temporary, seasonal, or circular migration.1 Only one 
model considers cascading migration, which refers to the impacts that in-migration may have on out￾migration from the same location (De Lellis 2021). One model directly forecasts immobility (Smirnov et 
al. 2022). Gravity models have indirectly estimated trapped populations, suggesting many people will 
become trapped in a closed border scenario, but the models are not designed to directly forecast 
involuntary immobility (Jones 2020; Rigaud et al. 2018; Clement et al. 2021).
Some models forecast climate-related migration indirectly. For example, in gravity models, climate-related 
migration is assumed to constitute the difference in projected population distributions between scenarios 
with and without varying degrees of climate change. Other models, such as agent-based models, estimate 
relative changes in total out-migration, in-migration, or return migration. However, their results tend to 
be aspatial, meaning they do not indicate the trajectories migrants follow. Some models present numerical 
estimates for future climate-related migrants, while others present their findings in terms of percentage 
increases or decreases in migration. 
Many models communicate their findings in terms of, a range of possible future trends based on different 
climate scenarios—for example, baseline, low emissions, or high emissions scenarios. Various 
development scenarios can also be paired with these climate scenarios. Experts consulted for this study 
who forecast using a scenarios-based approach encourage readers to consider the full range of possible 
outcomes, rather than choosing one most-probable outcome. Finally, some models may pair estimated 
increases in migration to urban areas with data on the food, housing, and job demands the migration would 
generate (Davis et al. 2018). 
1 As one expert stressed, the lack of a consistent definition of permanent migration among the studies makes 
comparing results challenging. This is a function of the data available to measure migration. Censuses usually 
include questions only about previous residences in the last year, 5 years, and 10 years. How social scientists treat 
a one-year move relative to a five-year move is at their discretion. The meaning of “temporary” migration can vary 
significantly among different societies. One expert shared that in one region where she does fieldwork, it is 
common to visit relatives for a month or more, which some might categorize as a temporary move while others 
would not. 
Report | December 2022 14
Box 1. When are populations trapped?
Trapped populations—those who aspire to move but lack the ability to do so—are receiving 
increased attention in climate-migration research, but have not yet been given significant 
attention in forecasting (exceptions include Benveniste et al. (2022) and Smirnov et al. (2022)). 
According to experts interviewed, historical data can give some estimate of populations “left 
behind,” which are then used to forecast future immobility. In gravity models, migration and 
trapped populations are modeled indirectly; the share of the population that does not move 
when exposed to climate stress is characterized as “trapped.”
These estimates of trapped populations, however, are not based on data that tells us whether 
populations actually want to move. A household that chooses to stay behind and adapt in place 
may not see themselves as trapped. A rich body of empirical and qualitative research shows 
that many people do not want to move in environmentally stressed areas, even if better and 
more secure livelihoods could be obtained elsewhere (Zickgraf 2021; Czaika and Reinprecht 
2022; Schewel 2020). Consider, for example, Indigenous populations in the Pacific Islands, 
where rising sea-levels and coastal degradation threaten local livelihoods, but where many 
prefer to remain on their ancestral homelands for cultural and spiritual reasons, including a 
deep connection to land and place-based identity, knowledge, and culture (Farbotko and 
McMichael 2019). This voluntary immobility in climate-stressed contexts has important policy 
implications. If populations are characterized as trapped, policy responses will naturally focus 
on facilitating migration and relocation programming. If populations are characterized as 
voluntarily immobile, more investments may be made to support in-situ adaptation (Farbotko 
et al. 2020; Schewel 2021).
Forecasting models cannot yet discern where and for whom future immobility will be voluntary 
or involuntary. A more accurate and appropriate term to describe this immobility is “resource 
constrained immobility,” used by Benveniste et al. (2022) in their study of decreases in 
international migration under scenarios of climate change. Particularly for poorer and more 
vulnerable populations, climate change can reduce household resources and thus diminish the 
capacity to move and to invest in in-situ adaptation. ABMs are one model type well suited to 
discern between voluntary and involuntary immobility under conditions of climate stress; 
models and baseline assumptions can be calibrated using data from surveys that ask directly 
about migration aspirations and ability (Carling and Schewel 2018).
Nine Models
To illustrate more clearly the various approaches, strengths, and weaknesses of different climate-related 
migration forecasting models, we conducted an in-depth analysis of nine models purposely selected for 
variation across world regions, geographic scope (global, regional, or country-level forecasts), source 
(academic or gray literature), migration type (internal, international, and trapped populations), and model 
types (gravity, ABM, radiation, etc.). Key features of the models are summarized in Table 5, and a more 
detailed overview of each model and its data inputs/outputs is provided in Appendix A.3. The main 
approaches, contributions, findings, and limitations of each model are reviewed below. Models 1 and 2 are 
the most data- and time- intensive to produce. The first Groundswell report involved six European and 
Evaluating Climate-Related Migration Forecasting Models 15
American institutions and took nearly two years to complete. It took a supercomputer four days to 
estimate results for just one scenario of Model 2. Both were oriented towards a broader audience and 
published as a World Bank report (Model 1) or long-form magazine article (Model 2). Models 3 to 9 are 
published in peer-reviewed journals and produced by smaller teams of academic experts. 
Model 1. The Groundswell Reports
Model 1, the Groundswell reports, provide the first global picture of the potential scale of internal climate￾related migration across six world regions: Sub-Saharan Africa, South Asia, and Latin America (Part I) and 
East Asia and the Pacific, North Africa, and Eastern Europe and Central Asia (Part II). Within each region, 
a country-level case study is included: Ethiopia, Bangladesh, Mexico, Vietnam, Morocco, and the Kyrgyz 
Republic. The Groundswell reports apply a scenarios-based gravity model to forecast future population 
distributions and areas that are likely to see greater in- and out-migration under different climate and 
development scenarios. The model applies demographic, socioeconomic, and climate impact data at a 14-
square kilometer grid cell level using the Gridded World Population dataset to model likely shifts in 
population within countries. To address the uncertainties of analyzing migration over the next 30 years, 
the report considers three potential climate and development scenarios (using RCP and SSP scenarios as 
inputs): a pessimistic reference scenario, a more inclusive development scenario, and a more climate￾friendly scenario. Climate impacts considered by the model include water scarcity, declining crop yields, 
and sea level rise. 
Part I of the model projects that without concrete climate and development action, more than 143 million 
people—about 2.8 percent of the population of Sub-Saharan Africa, South Asia, and Latin America—may 
move within their own countries due to the slow-onset impacts of climate change. Part II, which 
additionally models East Asia and the Pacific, North Africa, and Eastern Europe and Central Asia—provides 
a global estimate of up to 216 million internal climate migrants by 2050 across all six regions. Projections 
tend to be higher for pessimistic scenarios versus more optimistic climate and development scenarios. 
However, there are interesting regional differences. In East Africa, for example, there is a larger share of 
climate-related migrants relative to the general population under more inclusive development scenarios 
as compared to the pessimistic scenario. This is largely explained by development-driven migration; more 
people tend to migrate as they gain access to higher education, incomes, and infrastructure (de Haas 
2010). Estimates of climate migrants as a percentage of a region’s total population generally fluctuates 
between zero and three percent of a region’s total population, with North Africa being an exception 
where climate-related internal migration could reach as much as six percent (13 million, or half of all 
internal migrants) in 2050 under the pessimistic scenario. 
The Groundswell reports are arguably the most ambitious climate-related migration forecasting model to 
date. Their findings reveal important insights into how slow-onset climate change impacts, population 
dynamics, and development contexts might shape future mobility trends. The model is notable for its 
flexible data inputs, scalability, and application across world regions. Importantly, Groundswell projections 
include relatively granular maps depicting which locations are likely to see more or less in- or out￾migration, while many other model types lack this spatial information. 
However, there are also limitations to the Groundswell model. Because it uses population as a proxy for 
migration, the model is subject to the same limitations reviewed above for gravity models generally. 
Further, the Groundswell model does not attempt to forecast international migration, planned relocation, 
involuntary immobility, or cascading effects. It omits many political and economic factors (like access to 
Report | December 2022 16
land, resources, jobs, conflict, or shocks) that will certainly affect future migration trends. It cannot capture 
migration over distances of less than 14 kilometers, and thus cannot be applied to smaller geographic 
areas such as small island states. The model does not include short-term climate variations or sudden￾onset events. Like other forecasting models, it does not incorporate the impact of future adaptation efforts 
(e.g., improved crop varieties, irrigation, water conservation agriculture, or coastal defenses) into its 
projections.
Model 2. The Great Climate Migration Model
Model 2, the Great Climate Migration Model, provides an adapted and extended version of the 
Groundswell scenarios-based gravity model. ProPublica, in a collaboration with The New York Times 
Magazine funded by the Pulitzer Center, hired Bryan Jones, a geographer who had worked on the 
Groundswell report, to create a version focused on how climate change might lead to population shifts in 
Central America and Mexico, including how people may move within this region and to the United States. 
This model goes beyond the Groundswell model by forecasting international migration and highlights the 
potential for trapped populations. 
The modeling approach consists of two modules loosely coupled over time, one focused on internal 
migration and the second on international migration. The internal migration module is similar to the 
Groundswell model, but adds additional climate hazards like flood risk, fresh water availability, and 
extreme heat days. It adds additional non-climate related variables including the age-sex structure of the 
population. The model uses the same Gridded World Population dataset used by Groundswell, but rather 
than uniformly distributing a population across a cell, it modifies the dataset to distribute populations in 
areas that show more buildings and infrastructure. The international migration module uses existing 
bilateral flow data to train and project the model as a function of sectoral impacts (crops, water, and 
productivity), political instability and corruption, global income levels (GDP per capita), and existing 
diaspora to estimate potential changes in origin-destination flows under five alternative futures scenarios 
based on different RCP and SSP combinations. 
The Groundswell report projected an average of 1.4 to 2.1 million internal climate migrants in Central 
America and Mexico by 2050, depending on the scenario. To give a sense of scale, climate migration as a 
share of other internal migration ranges from 8.5 to 12.6 percent across scenarios. Model 2 does not 
present numerical estimates of internal migration that can be compared to Groundswell figures, but it 
does estimate between 680,000 to more than one million international migrants to the United States 
depending on government responses to climate change. The model considers the effects of more open or 
closed borders. It finds that closed borders reduce economic growth and urbanization in Central America 
and deepen poverty and hunger in rural areas, though projections for the location and scale of potential 
trapped populations are not reported. 
The Great Climate Migration Model improves notably on the Groundswell model by including 
international migration, a wider array of environmental, political and population variables, and modeling 
policy effects related to closed or open borders. However, Jones (2020) helpfully notes several areas of 
uncertainty related to the socioeconomic and climactic dimensions of the model. Some inputs, such as age 
structure, sex ratio, built-up land, groundwater, and political stability remain constant in future 
projections—an assumption that is almost certainly incorrect because projections for these factors do 
not exist at one kilometer resolution and attempting to model them would add further uncertainty to the 
model. Similar to the Groundswell model, SSPs also have assumptions baked into them regarding future 
Evaluating Climate-Related Migration Forecasting Models 17
age-specific fertility and mortality, education, wealth, and international mobility, each of which would have 
significant impacts on future population outcomes but are also subject to error. 
Other core limitations concern limited historical data on migration flows to calibrate the international 
migration model and the linear nature of climate-migration interactions in the model. For example, if a 
five percent decrease in water availability led to a two percent decline in the population of a given region 
in the past, the model assumes a future 10 percent decrease in water availability would lead to a four 
percent decrease in population. In reality, it is more likely that human response to climate stimuli will vary 
as a function of the intensity of those stimuli; as water scarcity worsens, the percent of the population 
that decides to leave is likely to rise in a non-linear fashion (Jones 2020; McLeman 2014). Gravity models 
are not yet able to capture these non-linear effects. 
Finally, although the article discusses the potential for trapped populations, particularly under scenarios of 
closed borders, Jones is clear that the model was not designed to estimate them. Trapped populations are 
assumed to be those who do not move when exposed to environmental hazards, but the model has no 
way of determining who chooses to adapt in place and who is unable to move due to resource scarcity, 
disability, health problems, or family responsibilities. Because this approach models aggregate trends as 
opposed to household decisions, it is unable to explicitly model a decision not to migrate due to lack of 
resources. 
Model 3. The Universal Model
Model 3, the Universal Model (Davis et al. 2018) applies a diffusion-based model of human mobility in 
combination with population, geographic, and climatic data to estimate the sources, destinations, and flux 
of potential migrants as driven by sea level rise in Bangladesh in the years 2050 and 2100. By linking the 
sources of migrants displaced by sea level rise with their likely destinations, the model purports to offer 
an effective approach for predicting climate-driven migrant flows, especially in data-limited settings. The 
authors describe the model as universal because it uses few data inputs and is parameter-free. 
The baseline model results showed good agreement with available division-level internal migration from 
the 2011 Bangladesh census, meaning it successfully replicated internal migration in Bangladesh using 
information on population distribution and distance. However, this constitutes only a one-year projection 
(using 2010 data to forecast 2011 trends). By mid-century, the model estimates that nearly 900,000 people 
are likely to migrate as a result of direct inundation from mean sea level rise alone, and Dhaka will be the 
top destination for migration. In large part because of the generally high population density across 
Bangladesh, however, the authors find most migrants will choose destinations close to their homes. The 
authors also analyze the additional jobs, housing, and food that would be required to support these 
migrants at their expected destinations. 
The model is distinct for how few data inputs are required. It builds on a radiation model published by 
Simini et al. (2011), which estimated internal and commuting mobility trends in the United States based 
on population distribution and distance estimates alone. By adding data inputs on elevation and projected 
sea level rise, Davis et al. (2018) present a very streamlined and simple approach to forecasting 
displacement from sea level rise. The model was expanded by De Lellis et al. (2021), who added a single 
parameter to the model on baseline migration rates, as well as a “resilience index.” These additions led 
to different predictions, however, namely a more country-wide distribution of migrants and a predicted 
outflow from Dhaka, raising concerns that radiation model outputs are too heavily dependent on the 
inputs and parameters modelers decide to include. Other models focused on sea level rise and migration 
Report | December 2022 18
in Bangladesh come to still different conclusions. Chen and Mueller (2018), for example, use a statistical 
extrapolation approach and find that inundation has negligible effects on internal migration in Bangladesh; 
gradual increases in soil salinity have more direct and important effects on internal and international 
migration trends. This suggests that, although the authors see the limited data inputs required for the 
universal model as a strength, it may also be a limitation. 
Model 4. The Systems Dynamics Model
Model 4, the Systems Dynamics Model (Naugle et al. 2022), couples migration decision making and 
behavior with the interacting dynamics of economy, labor, population, violence, governance, water, food, 
and disease. The model is applied to a test case of migration within and beyond Mali. The model is notable 
for the wide range of factors incorporated beyond climate and population variables, particularly political, 
economic, health, and conflict-related factors, and its experimentation with how several different policy 
interventions might affect migration outcomes. The model outputs include the fraction of the Malian 
population choosing to live in each simulated region. Potential locations are rural Mali, urban Mali, 
neighboring countries (Burkina Faso, Cote d’Ivoire, Gabon, Gambia, Ghana, Guinea, Mauritania, Niger, 
Senegal), the United States, and ‘the rest of the world’ (as one category).
The model generally finds that as temperatures increase, economic factors make migration from Mali to 
other locations more attractive, and the population tends to move out of both urban and rural areas of 
Mali toward neighboring countries, the United States, and the rest of the world. The model examines the 
impact of various policy options on migration outcomes and finds that providing contraception (reducing 
birth rates) reduces migration by limiting pressures on the economy, resources, food availability, and water 
availability. Increasing the effectiveness of governance in the model improves the economic situation by 
increasing the internal gross regional product, and ultimately reduces migration. Increasing infrastructure 
and services provided by the Malian government was relatively ineffective at reducing migration pressures.
The focus of this model is on exploring feedback effects and policy interventions, which is an understudied 
area in climate-related migration modeling. The model does not identify geographic hotspots of out- or 
in-migration. This model is best interpreted as a simulation exercise, based on some unrealistic 
assumptions. For example, in the base case simulation, temperatures remain stable throughout the time 
horizon and gross regional product tracks World Bank (2017) projections. The authors do not take a 
position on the validity of the base case projection; rather, they emphasize the opportunity to explore the 
differences between this and climate change scenarios to understand the causes of variations in the results. 
Again, this highlights that the model is best seen as an exploration of systems dynamics and potential policy 
effects, rather than offering concrete numerical projections of future climate-related migration. 
Model 5. The Dynamic Model
Model 5, the Dynamic Model (Entwisle et al. 2020), uses an ABM focused on land use, social networks, 
and household dynamics to examine how extreme floods and droughts affect migration from 41 rural 
villages in Northeast Thailand where rice cultivation is common. The ABM explicitly models the dynamic 
and interactive pathways through which climate-migration relationships might operate, including out and 
return migration, for each village. This model pays attention to variables that existing migration research 
confirms are fundamental to migration systems but are often left out of forecasting models: namely social 
networks, life-course dynamics, and return migration. This is one of the more sophisticated models to 
explore how climate change acts on already established migration processes that, as the authors describe, 
Evaluating Climate-Related Migration Forecasting Models 19
“are part and parcel of the life course, embedded in dynamic social networks, and incorporated in larger 
interactive systems where out-migration and return migration are integrally connected” (Entwisle et al. 
2020, 1469). 
The ABM is grounded in longitudinal survey, qualitative, spatial, social, and environmental data. Like Models 
1 and 2, the model does not assume a direct “climate effect” on migration; focusing rather on how changes 
in precipitation affect crop yields and thus livelihoods and household assets. It incorporates potential 
feedback into the model, through, for example, social networks and remittances. Interestingly, the results 
find minimal to no climate-related effects on out-migration. One potential reason for this is that out￾migration is already a normal part of social systems in this area, and these continue for other non-climate 
related reasons. However, the model finds that increased floods and droughts lead to a notable decline in 
return migration. Over time, the implication is that rural populations will decline. People will continue to 
leave as they always have, but many will no longer choose to return.
One strength of the ABM approach is the ability to run experiments. Running the model with and without 
social networks yields important differences in outcomes. In every scenario, without the facilitating effects 
of social networks, climate impacts decrease out-migration. This may be because social networks 
significantly lower the costs of moving, by sending money and helping prospective migrants (usually 
relatives) find housing and work. Without social networks, migration is a more individual decision and 
becomes costlier. The model also finds return migration stays constant without social networks. A likely 
explanation is that social networks are the main mechanism through which prospective return migrants 
learn about the hardships of floods or drought. Without social ties, decisions to return home are less 
affected by these climate impacts. 
This model has several strengths. It is built on rich data inputs, including decades of panel survey data for 
this one region. Further, it looks at the full spectrum of migration, both out- and return-migration, and 
incorporates different propensities for migration across the life-course. Although the authors do not 
explore this much in their paper, the model is better suited than others to assess heterogeneity within 
and between villages and provide micro-level evaluations of how potential climate-related stresses might 
disproportionately impact marginalized or vulnerable populations. 
The authors expressly state they “do not seek to replicate reality or predict any future,” rather they seek 
to predict the theories embedded within the model. A related limitation is that the model predicts 
increases and decreases in rates of out- or return-migration and yields no information on the spatial 
trajectories of migration flows. The data demands of this model also make it challenging to scale or apply 
in other areas that lack reliable, longitudinal data. Finally, this ABM model, like many other ABMs and 
computational models, rely on assumptions about what the underlying utility function looks like—that is, 
what people prioritize when they consider the costs and benefits of migration. In reality, we know utility 
functions can differ across populations, based on education, socioeconomic status, gender, culture, or 
personal disposition. 
Model 6. The Inequality Model
Model 6, the Inequality Model (Burzyński et al. 2022), investigates the long-term implications of climate 
change on global migration and inequality. The model combines a discrete choice model with random 
utility and a gravity-based approach. It assumes an ambitious global scope, directly focuses on the impact 
of rising inequality on migration outcomes, and incorporates a range of climate variables (including 
changing temperatures, sea levels, and the frequency and intensity of natural disasters). The model analyzes 
Report | December 2022 20
the impact of these climate changes on productivity and utility in a dynamic general equilibrium framework. 
It considers slow-onset changes in temperature and sea level rise as well as sudden-onset natural disasters 
and extreme events, and it predicts global migration patterns at a 5 kilometer x 5 kilometer pixel level. 
The model finds that climate change strongly intensifies global inequality and poverty, reinforces 
urbanization, and boosts migration from low- to high-latitude areas. However, it finds that only a small 
fraction of people suffering from the negative effects of climate change will manage to move beyond their 
homelands. Those who do move across borders tend to be more educated (especially those leaving Africa) 
compared to those who stay in their home countries. Median projections suggest that climate change will 
induce a permanent relocation of 62 million working-age individuals over the course of the 21st century. 
Massive international flows only occur under extremely pessimistic climate scenarios combined with highly 
permissive migration policies. The authors note that climate-related migration will be relatively small 
compared to migration driven by rising educational attainment and population growth differentials. 
Strengths of this global model include a detailed spatial resolution of climatic and economic factors. It 
incorporates individuals with multiple characteristics, distinguished by age category, education levels, and 
origin. The model is also well suited for extensions of policy experiments. Limitations of the model include 
a focus only on economic, voluntary migrants. In the model, choices to stay or to relocate are made by 
adults between 30 and 60 years old. This choice seems unusual, considering that migration propensities 
are highest for adults between 18 and 35 years. The authors also make stylized assumptions about 
migration and migration decision-making that do not necessarily reflect real-world dynamics. For example, 
migration decision-making is modeled under conditions of full information, which is rarely the case. 
Migration decisions consider wage rates, migration costs, and the congestion externally related to the 
total population. This leaves out other consequential factors in migration decision-making, like the 
influence of social networks (see Model 6), migration ability, or place attachment. 
Model 7. The Global Model 
Model 7, the Global Model (Smirnov et al. 2022), applies insights from climate science and computational 
modeling to generate a satisfactory agent-based model that forecasts relative increases in internal and 
international migration and immobility in response to drought over the 21st century. ABMs are often used 
to capture micro-level migration decision-making at the household level at smaller geographic scales. This 
ABM forecasts global and national-level trends for drought-induced migration and immobility. Smirnov et 
al. (2022) use 16 climate models in conjunction with high-resolution geospatial population data to consider 
different policy scenarios. Although this model forecasts trends beyond our 2050 time horizon, we include 
it here because of its direct focus on forecasting immobility. No other model projecting to 2050 directly 
modeled climate-related immobility. 
Model outputs are presented in relative terms: as the percentage change in total migration between 
different scenarios. The model’s simulations suggest that a potential for drought-induced migration 
increases by approximately 200 percent under the current international policy scenario (corresponding 
to the current Paris Agreement targets). In contrast, total migration increases by almost 500 percent, 
should current international cooperation fail and unrestricted greenhouse gas emissions prevail. The 
authors are clear, however, that this model does not offer a representation of actual future realities. “In 
an effort to convey this point explicitly, we frame our key findings in relative, as opposed to absolute, 
terms.” (Smirnov et all 2022, 3) Rather, the model should be taken as a tool to compare how the number 
Evaluating Climate-Related Migration Forecasting Models 21
of drought-induced migrants “might increase or decrease, in relative terms, between different climate and 
policy scenarios and for different countries.” 
This model is distinct for directly forecasting immobility. Immobile persons are defined as those exposed 
to extreme drought but unable to migrate due to the absence of suitable destinations in the final step of 
the model’s algorithm. Immobility is projected to increase by about 17 percent by the end of the 21st 
century in the baseline scenario, which holds drought conditions based on the 2008-2017 decade constant 
throughout the 21st century, in contrast to 175.7 percent in the low emissions scenario and 568.2 percent 
in the high emissions scenario. The implication is that more severe droughts could significantly entrap 
populations. Interestingly, both migration and immobility increase under the high emissions scenario; 
greater immobility, the authors state, does not imply less overall migration. 
The main sources of uncertainty in this model concern climate and population projections, the 
standardized precipitation evapotranspiration (drought) index (SPEI), and agent-based model parameters. 
The authors do not aim to explore migration motivations, causal mechanisms, or feedback dynamics. The 
model, by the authors’ own admission, is based on coarse assumptions about human behavior and 
migration. “Humans are essentially automata following very basic rules… [They] do not have age, gender, 
resources, social capital, diaspora networks, or other characteristics, all of which undoubtedly influence 
migration behavior” (Smirnov et all 2022, 3). Another important limitation is that the model assumes all 
populations have equal exposure and vulnerability to drought within each grid cell of the world map, which 
does not reflect real-world heterogeneity in who is most affected by drought and who has greater or 
lesser capabilities to migrate or adapt in place. Finally, the model claims to forecast involuntary immobility, 
but does not distinguish between potential voluntary and involuntary outcomes. The model does not 
directly include factors that affect real-world involuntary immobility, like resource-scarcity, conflict, 
border restrictions, or sociocultural factors. The authors suggest future iterations of the model should 
include social networks, border restrictions, and other institutional constraints on international migration. 
Model 8. The Small Island Model
Model 8, the Small Island Model (Speelman et al. 2021), uses an agent-based model to examine migration 
flows in the Maldives under various climate, political, and social scenarios to 2050. Small island states face 
unique climate-related challenges, including sea level rise, lack of land for managed retreat, and limited 
international migration pathways. Small island states pose distinct challenges to migration modelers, as 
they are often too small for gravity models and other approaches that rely on gridded population data. 
For this reason, Speelman et al. (2021) use an ABM that simulates migration from 1985 to 2014 and then 
forecasts a range of potential demographic futures to 2050. They develop a conceptual model for migration 
that highlights key push and pull factors and combine it with three factors that influence migration 
intentions: attitudes, subjective norms, and perceived behavioral control (Ajzen 2011). They apply the 
model across six possible future scenarios, each characterized by a combination of high/low population 
growth, little or strong government intervention, open or closed borders, and high or low emissions. 
Using census data as the primary population data input, Speelman et al. show that the Maldives has been 
characterized by an active migration system over recent decades, with many islands having declining 
populations and a large urbanization trend around the capital Malé. They find that overall, across all future 
scenarios, these trends will continue in the decades to come. The growth of Malé continues while many 
Report | December 2022 22
other islands see population decline.2 Revealing the importance of fertility trends, the model projects a 
decline in migration in the early years of the projections, due to a sharp decline in birth rates between 
2000 and 2006. It rises again between 2023 and 2030 due to higher birth rates from 2006 to 2014. Overall, 
the findings suggest “migration in the Maldives has a strong inertia, and radical change to the environmental 
and/or socio-economic drivers would be needed for existing trends to change” (Speelman et al. 2021, 
283).
The authors caution that the scenarios considered provide examples of future scenarios, not predictions. 
“Their purpose is simply to propose contrasting ways in which political and economic factors would 
combine to influence migration” (Speelman et al. 2021, 290). Limitations of the model highlighted by the 
authors include the need for more specific formulation of economic, social, political, and environmental 
change, linked to the need for more and better data. The model also excludes foreign residents, which is 
understandable given the focus on native populations. But this remains a consequential population group 
(∼64,000 residents), and their in-migration can affect inter-island migration dynamics. Finally, the model 
does not integrate future shocks, adaptation, or rapid changes in demographic processes, like the short￾lived but sharp decline in fertility rates in the early 2000s, which may occur again. 
Model 9. The Statistical Extrapolation Model
Model 9, the Statistical Extrapolation Model (Chen and Mueller 2019), approaches model climate￾migration interactions using historical data; the relationships uncovered are then used to forecast future 
trends. This model is used to forecast climate-induced international migration from Bangladesh. It is 
distinct for modeling several climate-related factors, including remote-sensing measures of flooding and 
rainfall and in situ measures of monsoon onset, temperature, radiation, and soil salinity. It uses nationally 
representative migration data to evaluate which locations and population groups are more likely to migrate 
across the border to other South-East Asian countries, with a particular focus on international migration 
to India. 
Chen and Mueller find that climate variables vary in their relationships to cross-border migration. Short￾term, adverse weather events are associated with decreased international migration, while increases in 
soil salinity increase cross-border migration. Soil salinity has a stronger effect on migration from poorer 
households. The model outputs do not provide an estimate of climate-related migrants by a particular
date; rather, projections are based on the degree of future climate change. For example, the model 
predicts a total of 17,874 more migrants moving to India in response to a one standard deviation increase 
in soil salinity, with out-migration concentrating in the southwestern coastal regions of the country.3 The 
model predicts a total of 5,754 fewer migrants moving to India in response to an increase in flooding of 
one standard deviation. The authors do not attempt to forecast when those changes in soil salinity or 
inundation will occur. 
2 To give a sense of the scale of change, the population of the capital of Maldives, Malé, grew from 67,939 in 2000 
to 109,635 in 2014. In 2050, its population is projected to be between 169,819 and 217,976, depending on the 
scenario considered. The smallest island, Kandoodhoo, grew from 2,224 to 3,333 between 2000 and 2014, and is 
projected to have a population of between 5,260 and 3,796 in 2050. Feydhoo (Addu) grew from 2,829 to 3,397 
between 2000 and 2014, and is projected to have a population of between 2,768 and 1,706 in 2050, the smallest 
island population projected. 
3 To give a sense of relative scale, there were 2.5 million Bangladeshi migrants in India in 2020 (MPI 2022).
Evaluating Climate-Related Migration Forecasting Models 23
Chen and Mueller examine household vulnerability to flooding and increased soil salination to explore 
whether households with greater human, social, or physical capital are more inclined to migrate. They 
also consider the impact of age, gender, and religion on migration outcomes. They find that wealthier 
households are less likely to migrate in response to gradual changes in soil salinity, perhaps because they 
are better equipped to diversify livelihoods locally, while poorer households are more likely to use 
international migration as an adaptation strategy as soil salinity increases. These and other related findings 
provide important evidence for how climate-migration interactions might vary for different social groups 
and more vulnerable households. Chen and Mueller do not use these more nuanced analyses, however, 
to forecast future trends based on household characteristics.
The authors are transparent about the limitations of their model. They lack information on the duration 
of each event and are unable to validate whether moves are temporary or permanent. The measure for 
soil salinity is relatively coarse and focuses on changes over 5 years. This misses the potential mobility 
implications of seasonal or annual variations. They also note that soil salinity can result from changes in 
landscape and deforestation along the coast, sea level rise and storm surges and ground management. It 
is not clear what factors are contributing most to increase soil salinization and thus where policy makers 
should prioritize funding for adaptive investments. Finally, they note one drawback of using a large 
administrative dataset in the absence of detailed survey questions that would be required to explore more 
carefully the reasons for migration and obstacles potential migrants face.
Report | December 2022 24
Table 2. Description of nine climate-related migration forecasting models
Model 
#
Brief title 
of paper
Coverage Country Model Precip Temp SLR* Storms Drought** Other Type of 
Migration
***
Time 
horizon of 
predictio
n
Other 
Factors
*****
Examples
1 Groundswell I 
& II
Global, 
Regional
Bangladesh, 
Mexico, 
Ethiopia, 
Morocco, 
Vietnam, Kyrgz 
Republic
Gravity ✓ ✓ water and crop 
production
Internal 2050, 2100 Dem, 
Dev
population projections, 
SSPs
2 The Great 
Climate 
Migration
Regional Central 
America, 
Mexico, United 
States
Gravity ✓ ✓ ✓ ecosystem 
impacts, flood, 
groundwater, 
extreme heat
Internal, 
International
2020-2050, 
2100
Pol, 
Econ, 
Dem, 
Soc, 
Env, 
Dev
diaspora, instability. 
Violence, corruption, 
GDP, SSPs
3 Universal 
Model
National Bangladesh Radiation ✓ Internal, 
International
2050, 2100
4 Systems 
Model
National Mali System 
Dynamics
✓ ✓ extreme events Internal, 
International
2060 Econ, 
Pol, 
Dem, 
Dev, 
Soc
labor supply, rural/urban, 
violence, governance, 
level of tech, food 
availability 
5 Dynamic 
Model
National, 
Sub-national
Thailand ABM ✓ crop production, 
floods
Internal, 
Return
25 
years****
Econ, 
Soc
social networks 
household assets, 
remittances
6 Inequality 
Model
Global CGEM***** ✓ ✓ International 2040, 2070, 
2100
Econ, 
Dem
education, age, wage 
rates, fertility projections, 
urbanization, GDP
7 Global Model Global, 
National
ABM ✓ Internal, 
International, 
Immobility
2008-2100, 
2081-2100
Dem, 
Pol
population projections, 
open borders 
8 Small Island 
Model
National Maldives ABM ✓ tsunami, erosion, 
damages to reef
Internal 2014-2050 Dem, 
Pol, Soc
governance, open/closed 
borders, networks, 
population projections
9 Statistical 
Extrapolation 
Model
National Bangladesh SEM ✓ ✓ flooding, soil 
salination, bright 
sun
International Unstated Dem, 
Econ, 
Cul
gender, age, assets, 
religion
*Sea level rise **We also coded for soil quality and desertification as hazards, but none of the models incorporated them. ***We also coded for cascading migration, but none of the models incorporated it. ****The 
authors do not look at specific years. Rather, they give data on how migration might expand over any 25 years under the climate scenarios tested. The data used was gathered in 2000 and 1990-2008, so a time 
horizon might be 2025 or 2033. ***** Economic (Econ), Political (Pol), Demographic (Dem), Social (Soc), Technological (Tech), Environmental Resilience (Env), Aggregate Development Indicators/Scenarios (Dev), 
Culture (Cul). ***** Computable General Equilibrium Model. 
Evaluating Climate-Related Migration Forecasting Models 25
Summary of the Nine Models
The forecasting models reviewed above provide insight into the wide range of modeling approaches being 
developed to forecast climate-related migration. Overall, the authors of these models and the experts 
interviewed emphasized the uncertainties inherent in the models and urged caution in interpreting their 
results. The models are best used as tools to explore potential migration scenarios under various climate, 
development, and policy futures; to test theories; and to explore how climate impacts may affect broader 
social systems within which migration is embedded. This summary section begins with a short review of 
some of the climate-related migration hotspots identified in the global models we analyzed, before 
considering key limitations and evidence gaps that hinder the real-world applicability of these projections. 
Future Climate-Related Migration Hotspots
Migration hotspots refer to locations more likely to experience higher degrees of climate-induced in- or 
out-migration. Hotspots of climate-driven out-migration typically identified include low-lying cities and 
coastlines vulnerable to sea level rise and areas of high water and agricultural stress. Climate in-migration 
hotspots tend to be locations with better climatic conditions for agriculture as well as cities able to provide 
greater livelihood opportunities (Rigaud et al. 2018). 
Three of the models provided a global-level overview of potential future climate-related migration. The 
Groundswell report provides numerical forecasts of internal migrants for six world regions (Model 1), 
with detailed maps illustrating potential future migration hotspots in East Africa, South Asia, Mexico and 
Central America in Part I and North Africa, Lower Mekong, and Central Asia in Part II. We encourage 
readers to refer to the Groundswell summaries for more specific projections of migration hot spots. 
However, a few general findings from the six regional analyses are summarized here. 
Within East Africa, the report predicts this region will see between 6.9-10.1 million climate migrants by 
2050 depending on the climate-development scenario. Some of the major origin areas include the coastal 
regions of Kenya and Tanzania, western Uganda, and the northern highlands of Ethiopia. Inland hotspots 
are driven by declining water availability and declining crop yields, while the coastal areas reflect rising sea 
levels and storm surges. Migrants are projected to settle in the regional countries of the Lake Victoria 
Basin, the eastern highlands of Ethiopia, and the area of Lilongwe (the capital of Malawi). These destinations 
tend to have more favorable climate conditions and more limited impacts from climate change compared 
to coastal areas. These areas also tend to be already densely populated and lie along national borders.
In South Asia, Groundswell forecasts between 11.4 and 35.7 million migrants depending on the climate 
scenario. Migrants will primarily leave the eastern and northern parts of Bangladesh, the northern Gangetic 
Plains, the corridor from Delhi, India to Lahore, Pakistan, and metropolitan coastal cities like Mumbai and 
Chennai, India and Dhaka, Bangladesh. The main areas for in-migration are the southern Indian highlands, 
particularly between Bangalore and Chennai, northwest India, and Nepal. In general, the report finds that 
irrigated and rice-growing areas will see increased out-migration while rainfed areas will see more in￾migration. The in-migration hotspots are driven by expected improved water availability in those areas, 
especially southern India. 
In the region of Central America and Mexico, the absolute number of climate migrants is expected 
to be much lower than the other two regions, with a range of 0.2-3.9 million depending on the scenario. 
Groundswell anticipates that most climate migrants will leave the low-lying areas of Central America and 
Mexico, particularly along the Gulf of Mexico and the Pacific Coast of Guatemala. Some cities like 
Report | December 2022 26
Monterrey and Guadalajara, Mexico will also be sites of out-migration. Generally, hotter, low-lying, and 
rain-fed areas of this region will be places of out-migration under climate change. Conversely, pastoral, 
rangeland, and highland areas will see populations increase due to climate migrants moving into them. The 
Central Plateau of Mexico and the highlands of Guatemala are projected to become hotspots of in￾migration under climate change.
In North Africa, Groundswell predicts the region of North Africa could have between 4.5 and 13 million 
climate migrants by 2050. Migrants will come from the coastal areas of the region including the Nile Delta, 
northeast of Tunisia, northwest of Algeria, and western Morocco. The major cities of these areas will be 
the primary sources of out-migration in North Africa. This will be due to rising sea levels and reduced 
water availability. The area around the Atlas foothills in Morocco will also see reduced water availability 
and thus increased out-migration. The areas that will see more in-migration will be those where water 
availability is unchanged or increased. These areas include primarily the Nile Valley and central Delta area, 
southern coasts of Tunisia, eastern coasts of Algeria, and the northern part of Morocco. These are home 
to the major cities of Cairo, Algiers, and Tunis. Because of their higher elevations, many of these cities 
will see increased in-migration, despite their locations near the coast. They also are projected to have 
greater water availability and crop productivity. 
In the Lower Mekong region, 3.3-6.3 million people are expected to be climate migrants by 2050. This 
is due to sea-level rise in the coastal areas and decreased water and crop productivity elsewhere. 
Specifically, the major areas for out-migration will be in northern Vietnam, the Vietnam Mekong Delta, Ho 
Chi Minh City, and central Thailand and Myanmar. In-migration will be concentrated in southern Thailand, 
the inner portions of the Mekong Delta in Vietnam, the Red River Delta in Vietnam, southern Myanmar, 
and parts of the Irrawaddy River. All of these in-migration locations are predicted to have greater water 
availability and crop productivity, but some also lie along the coast where sea-level rise and storm surges 
will affect them. This poses a problem for these areas with a greater population and climate change 
impacts. 
Within Central Asia, 1.7-2.4 million people are expected to be climate migrants by 2050. Those migrants 
are expected to come from the southern border of Kazakhstan, areas surrounding the Ferghana Valley of 
Uzbekistan and Tajikistan, and Bishek. Minor hotspots of out-migration are along the Amu Darya River in 
Turkmenistan and Uzbekistan. This is likely due to decreased water availability and crop productivity. In￾migration is expected in places with increased water availability like the Ferghana Valley, lower elevation 
areas of southern Tajikistan, and northern Kazakhstan.
The other two models that provided a global picture of future trends unfortunately do not provide similar 
outputs, making it difficult to compare findings across models. The second global model, Model 6 
(Burzyński et al.) provides numerical forecasts for local, regional, and international migration globally. But 
the findings presented in the paper do not identify migration hotspots at the same granular level as 
Groundswell. The paper focuses instead on detailing broad trends in future movements of working-age 
adults with higher or lower levels of education at the global level. The model identifies major corridors of 
international migration under an RCP7.0 scenario in 2040, 2070, and 2100. For high-skilled international 
migrants, the predominant corridors are from Asia to Europe (10 million people), from Asia to North 
America (6 million people), from Africa to Europe (6 million people), and from South America to North 
America (5.5 million people). For low-skilled migrants, predominant pathways are from Asia to North 
America (3.6 million people), from Asia to Europe (1.5 million people), and from South America to North 
America (1 million people). It should be noted that although the primary focus of the published article was 
not to identify migration hotspots at a more granular level, the model could be applied to do so. 
Evaluating Climate-Related Migration Forecasting Models 27
The third global model, Model 7 (Smirnov et al.), reports relative changes in percentage of internal and 
international migration and immobility by country, region and globally; it does not offer spatial information 
on where migration hotspots will be. Nevertheless, the model analyses RCP scenarios 8.5 and 4.5 and 
projects the greatest future displacement (over two percent of the world’s total displacement) in Egypt, 
Syria, Senegal, and Guatemala, and greatest immobility (over two percent of world’s immobile population) 
in Mali and Syria. It forecasts the top five international migration corridors (origin-destination) will be 
India-China, Venezuela-Colombia, Nigeria-Niger, Pakistan-India, and Morocco-Algeria.
Key Limitations and Evidence Gaps
Each model type has its own distinct strengths and limitations, which are reviewed in greater detail in the 
above sections. The following summarizes the general limitations and data gaps common across forecasting 
models reviewed in this project. 
Regarding climate hazards, the nine forecasting models we reviewed focus primarily on slow-onset 
climate changes, rather than sudden-onset events. From a modeling perspective, this is understandable 
considering the consistency of data across hazards such as temperature or precipitation and the difficulty 
of predicting major natural disasters. However, sudden-onset events are likely to compound the 
socioeconomic and environmental impacts of ongoing slow-onset events and will be key drivers (often 
temporary) of displacement. Relatedly, many models forecast migration in relation to one or two climate 
hazards only (e.g., sea level rise, drought), when in reality, several interrelated slow- and sudden-onset 
climate changes will affect future population mobility or immobility. In the broader climate-migration 
literature, discrete history event modeling has been used to model historical responses to both sudden￾onset events and multi-hazard environments, and these findings can ground future projections, even if the 
exact timing or severity of future natural disasters remains unknown.
Non-climate related drivers of migration remain underrepresented in the models we reviewed, 
including political, economic, and cultural factors, and attitudes toward destinations or host countries. 
Only half of the selected models include covariates for economic and political drivers, and the measures 
used to capture these conditions often differ significantly for each model. Economic indicators used include 
GDP, labor supply, employment, or wages. Political variables include indicators for governance, violence, 
corruption, and instability. Other dimensions related to infrastructure or technology were far less 
common. Indicators capturing cultural or social norms (e.g., gendered norms, place attachment) are 
completely omitted and remain difficult to measure, though these play an important role in shaping a given 
population’s attitudes towards migration. 
It is notable that almost all models that take a scenarios-based approach use RCP and/or SSP scenarios 
derived from the IPCC Fifth Assessment Report. These have become the go-to pathways for forecasting 
models to consider a range of climate and development scenarios and thus a range of potential climate￾related migrations. The strengths, limitations, and uncertainties entailed in generating these climate and 
development pathways are beyond the scope of this report, but each brings its own set of assumptions 
and uncertainties to the scenarios used for migration modeling. For example, there are still many 
unknowns about future temperature responses to carbon dioxide emissions. Similarly, the pace of sea 
level rise remains challenging to predict given uncertainties about ocean heat uptake and the rate of glacier 
and ice sheet melt. 
Regarding intervening obstacles or facilitators, of the nine models examined, only the dynamic (agent￾based) model accounts for social networks. Social networks are a fundamental migration facilitator, 
impacting who is most likely to move and where they will migrate, yet few models are able to incorporate 
Report | December 2022 28
this dimension of migration processes. Also known as the “friends and family effect,” social networks 
significantly reduce the informational, capital, and social costs of moving (Hatton and Williamson 1994). 
Some models included indicators for border controls (an intervening obstacle), but this was mostly 
explored in terms of open or closed border scenarios, neither of which are likely in the coming decades. 
Forecasting models still tend to assume a relatively linear relationship between climate change and 
migration, and although we now know climate change and migration interact in a non-linear way, it is 
difficult to adequately capture future tipping points or thresholds in climate-migration relationships. 
Relatedly, climate-related migration forecasting models are not yet able to reliably identify or integrate 
potential cascading effects.
4 Even if research shows that cascading effects are real and occurring, experts 
shared the extreme difficulty of forecasting those effects with confidence. 
Regarding model outputs, future research would benefit from exploring how internal and international 
migration act as complements or substitutes. Internal migration can often turn into international migration. 
Conversely, the migration of one individual can enable a household to avoid migration, bolstered by the 
value of remittances from the individual (Stark and Bloom 1985). Only one model included return 
migration, though research suggests the pattern is common following sudden-onset disasters and 
displacements, and thus has important implications for post-disaster planning. 
Only one of the nine models attempts to forecast suppressed mobility, immobility or “trapped 
populations”—those who aspire to migrate but lack the ability to move. In addition to the push and pull 
factors that encourage migration, migration trends are also shaped by retain and repel factors, as well 
as political, economic, or other constraints that deprive individuals and households of the capability to 
move (Schewel 2020). Retain and repel factors are not yet incorporated into the theoretical frameworks 
that underlie these models, which remain focused almost exclusively on push and pull factors. 
Finally, no model can predict major unforeseen events, such as pandemics, wars, or technological 
revolutions such as the internet that may fundamentally reshape migration flows. This rather obvious point 
is arguably the most consequential factor undermining the accuracy of all the models’ long-term 
projections. 
Areas of Future Research
The statistician George Box famously said, “All models are wrong, but some are useful.” Statistical models 
always fall short of the complexities of reality but that does not mean they are useless. Models can reduce 
uncertainty, even if they do not eliminate it completely. The models described above are some of the best 
early attempts to forecast climate-related migration. Modelers are well aware of existing limitations, and 
they are making consistent efforts to push the field forward. The field is progressing quickly, and many 
promising advancements have yet to be published. 
The Africa Climate Mobility Initiative, a collaboration of the African Union Commission, the United 
Nations, and World Bank, is one ongoing initiative that explicitly builds on the Groundswell model by 
adding maximum rural/urban population densities, slow onset ecosystem impacts, and rapid onset events 
like flood risk projections. It includes additional SSP scenarios running in five-year increments and at a 
resolution of 4km grid cells (de Sherbinin et al. 2022). Researchers are also seeking to incorporate different 
forms of adaptation into the assumptions of SSPs, which currently fail to make nuanced assumptions about 
4
 See De Lellis et al. (2021) for one attempt to model cascading migration.
Evaluating Climate-Related Migration Forecasting Models 29
the various ways humans are adjusting to climate change (Reimann 2022). The African Climate Mobility 
Initiative is also notable for its deliberate efforts to pair modeling with qualitative research and stakeholder 
engagement. 
Beyond specific projects, expert interviews highlighted four general areas for further research. The first is 
establishing a multi-level modeling approach—for example, combining gravity models with agent-based 
models—so the strengths of one can compensate for the limitations of the other. Specifically, gravity 
models could be used to identify potential hotspots of migration, and agent-based models could be used 
in those hotspots to model rates of migration for different social groups. System dynamics models could 
also be integrated to better understand the effects of various policies in those locations. A second area of 
interest is machine learning. Machine learning offers the prospect of identifying thresholds or tipping points 
in climate-migration interactions–that is, identifying at what point a particular climate stress tips an existing 
migration system in a new direction. But because machine learning requires very large data inputs, it is 
difficult to apply to many regions of the world where reliable data is scarce. 
Third, experts highlighted that many models are equipped to explore the impact of adaptation or 
development interventions, and more research attention could be directed towards this aim—for 
example, modeling the migration impacts of sudden flows of financial capital or the adoption of air￾conditioning. Fourth, experts highlighted the need for greater trans-disciplinary collaboration between 
climate scientists and social scientists. Measurements and assumptions around exposure and vulnerability 
to climate change remain relatively coarse in current forecasting models. Regarding sea level rise, for 
example, it is often assumed that all households have the same level of exposure and vulnerability to 
inundation, whereas the duration, intensity and depth of flooding can vary significantly across a single 
village, and the adaptive capacity of households in that same community can be highly unequal, leading to 
different migration and immobility outcomes. 
The most important advances in forecasting may come from modelers working on population or migration 
projections, not necessarily from those looking at climate-related migration. In this regard, Simini et al. 
(2021) recently developed a “Deep Gravity” model that generates flow probabilities, exploits many 
features (e.g., land use, road network, transport, food, and health facilities) extracted from voluntary 
geographic data, and uses deep neural networks to discover non-linear relationships between those 
features and mobility flows in England, Italy, and New York. Just as the earlier radiation model developed 
by Simini et al. (2012) influenced the Davis et al. (2018) model for Bangladesh, new advances in population 
or mobility modeling may set the stage for more advanced climate-related migration modeling. 
Policy Implications
Modelers urge caution in interpreting the results of climate-related migration forecasts. At this stage of 
understanding of climate science, models and their accompanying projections are best understood as 
explorations of systems dynamics, ways to test theories or consider potential policy effects, and tools for 
considering a range of possible futures contingent on particular assumptions—what Jones (2020) calls a 
“a set of ‘what if’ scenarios.” Lustgarten, the author of the New York Times Magazine report, writes, “As 
with much modeling work, the point here is not to provide concrete numerical predictions so much as it 
is to provide glimpses into possible futures. Human movement is notoriously hard to model, and as many 
climate researchers have noted, it is important not to add a false precision to the political battles that 
inevitably surround any discussion of migration.”
Report | December 2022 30
Bryan Jones (2020) suggests “it is the variation across the different scenarios that should be considered as 
a starting point for discussions of potential policy intervention, areas that require additional research, or 
simply to start focusing on questions that will be critical to anticipating and planning for climate-induced 
migration appropriately.” Other experts echoed this opinion. In theory, if models come to a similar 
conclusion about potential hotspots for out-migration or in-migration under various scenarios, or across 
multiple models, policy-makers could have more confidence in targeting these regions with additional 
investment, programming, and support. Yet at this stage there are not enough models being developed 
for the same geographic region to make multi-model comparisons and validation possible for many world 
regions where USAID works. One important exception is Bangladesh, where at least seven models have 
attempted to forecast climate-related migration, and a future study could systematically compare findings 
across these models and begin validating projections. 
Exposure Modeling
Given the limitations of accepting the results of current models, one strategy is to return to the origins 
of climate-related migration forecasting: exposure modeling. Exposure modeling, or hazard mapping, was 
one of the first approaches to identifying populations at risk of climate-induced displacement. When 
modelers assumed all populations in at-risk regions would be forced to migrate, this gave rise to 
catastrophic, unrealistic migration forecasts. We now know that many populations in environmentally 
stressed regions do not migrate. Many will choose to stay and adapt in place; others may become trapped 
in place; still others may move such short distances or for such short time-periods that we do not assess 
this mobility as migration. Recognizing this, exposure mapping remains an important tool to identify 
populations whose lives and livelihoods are threatened by climate change and to target development 
assistance accordingly. Exposure mapping could become more sophisticated as natural scientists, social 
scientists, and other stakeholders work together to develop more refined maps of potential exposure, 
vulnerability, and adaptive capacity. 
Although the precise consequences of climate stress on human mobility or immobility are difficult to 
predict, more general research at the climate-migration nexus support three areas of investment: 1) in￾situ adaptation in climate-affected areas, recognizing that many households will prefer to stay in their home 
communities; 2) facilitating migration as an important adaptation strategy to cope with climate change, 
whether seasonally or permanently; and 3) expanding access to housing, employment, and services in 
urban areas and major cities neighboring climate-affected regions, as these are the most likely destinations 
for distress migrants. Research suggests the destinations of climate-related migrants are often within close 
proximity to their origin (Fussell et al. 2014). In places where international migration systems are 
increasingly robust – for example, between Central America and the United States, Burkina Faso and Côte 
d’Ivoire, or Bangladesh and India – it can be assumed that climate-related migration will also follow those 
established migration pathways. 
Foresight Research
Foresight research constitutes another tool that may be used to anticipate potential climate-related 
migration trends for particular regions.5 Foresight exercises, also referred to as “qualitative migration 
scenarios,” bring together migration experts, stakeholders, and scholars to consider potential scenarios 
5
See ‘Future Migration Trends’ in the Migration Data Portal: https://www.migrationdataportal.org/themes/future￾migration-trends
Evaluating Climate-Related Migration Forecasting Models 31
for future migration (Vezzoli et al. 2017). These exercises have been conducted by the Global Migration 
Futures project and contributed to the Future of International Migration to OECD Countries report and 
the British Government’s influential Foresight Report (Black et al. 2011). Foresight research could be 
particularly useful in exploring future climate-related migration trends in particular regions to inform 
USAID policy. With climate migration modelers working alongside academics from a range of fields, 
including the natural sciences, policy-makers, and other stakeholders, migration forecasting model findings 
can enhance more qualitative knowledge of regional dynamics. 
The benefit of foresight research is that scholars who specialize in a region may be able to identify gaps 
and inaccuracies in the inputs or assumptions of forecasting models, offer cultural and historical insights 
that might improve them, and offer key caveats about the applicability of the models’ findings for particular 
populations or sub-regions. For example, one expert shared an anecdote about a forecasting model for 
Nigeria that predicted greater movement to the North of the country where land and climate conditions 
could sustain larger populations. The model did not account for the control of these same areas by militant 
groups, making it unlikely them unlikely destinations for voluntary inflows of migrants. Experts also 
mentioned that it is common for forecasting models to assume a standard level of exposure and 
vulnerability across a particular region. Natural scientists may be better equipped to look at differences in 
exposure and vulnerability at the household or community-level, while social scientists and migration 
scholars may be able to add nuance to who is most likely to use migration as a coping mechanism versus 
adapting in place. Where models exclude social networks, migration scholars can share information on 
established migration pathways that new movements are more likely to follow.
Improved Data Analytics
To further improve climate-related migration forecasting, resources may be best utilized upstream in the 
modeling process, where all experts agree there is a dearth of resources dedicated to the most important 
modeling input: data. Migration poses unique challenges to data collection, as mobility makes respondents 
difficult to locate, particularly for large-scale surveys. Existing longitudinal surveys with high quality tracking 
tend to be limited in size and scope, both temporally and geographically (Chen and Mueller 2019). 
Collecting high quality data at adequate scale is often not feasible in terms of time and financial resources.
Innovative ways to collect migration and mobility data are emerging in the form of cell-phone data and 
experiments with remote sensing and earth observation data. However, investments in high-quality and 
reliable censuses remain critical for larger modeling efforts; much data collected in censuses globally is 
inaccurate and unreliable yet forms the basis for most population projections. Investments in the statistical 
bureaus of host countries is one important way to build capacity for regular and reliable census data 
collection. In parallel, international organizations like the United Nations Department of Economic and 
Social Affairs, the International Organization for Migration and the United Nations High Commissioner 
for Refugees remain important sources of global population, migration, and displacement data, including 
flow monitoring.
In addition to census data, more high-quality individual and household survey data is needed to uncover 
migration histories, intentions, and perceptions of climate change. A major challenge of surveying to 
anticipate climate-related migration is that many people do not conceptualize climate change as a driver 
of their migration; rather, the downstream effects of climate change (e.g., loss of income) as well as other 
political, socioeconomic, and other factors are more often cited as the primary motivations for moving. 
As such, surveys must be conscious to address climate change impacts alongside other drivers of migration. 
Report | December 2022 32
Given the need to enhance understanding about the climate-migration-development nexus, USAID could 
also consider systematically including questions about migration decision-making and behavior into regular 
monitoring and evaluation surveys in climate-stressed regions. This would expand the evidence base 
available to assess which population groups are more likely to migrate or to stay, and how different 
development interventions affect migration aspirations and ability in climate-stressed contexts. 
Enhancing predictive tools also requires more foundational research, not only future-oriented research. 
A more robust knowledge base is needed to clarify how societies have historically and are currently 
responding to climate change through movement. This will improve the conceptual understanding and the 
theoretical frameworks that shape forecasting models. In this sense, we echo the recommendations of the 
2021 United States government Report on the Impact of Climate Change on Migration to consider investments 
in research, analysis, and programming. Such investments will help build understanding and address 
important questions on the likely evolution and consequences of climate related migration, including but 
not limited to: How do different social, economic, geographic, political, and other characteristics mitigate 
or exacerbate the effects of climate change on migration? What geographies’ conditions are most 
associated with the risk of immobility and trapped populations? What is the role of migration in supporting 
adaptation and resilience from the household to national scales?
Key Takeaways
• The field of climate-related migration forecasting is still in its infancy. The volume and 
direction of climate-related migration predicted by these models to horizons like 2050 should be 
taken as notional at best. Experts urged caution in using numerical estimates to inform policy and 
programming. Instead, they suggested, models may be better suited to explore possible future 
pathways, and policies oriented towards encouraging best-case scenarios. 
• The scenario-based approach is preferable to single narratives in climate-related 
migration forecasting, precisely because of uncertainty surrounding many inputs and 
the outcomes. Modeling human behavior, including migration, is an exercise fraught with 
uncertainty, and accounting for climate change only compounds this uncertainty. Therefore, the full 
spectrum of future scenarios should be considered when using migration forecasts.
• Climate change will affect migration, but it is one of many drivers. Climate variation and 
change will increasingly affect migration decision-making and migration patterns in coming decades, 
but climate impacts are heavily mediated by political, economic, social, technological, and cultural 
factors. Although models are advancing quickly, current approaches lack the sophistication to 
adequately capture how climate-related factors intersect with other non-climate related drivers of 
migration and immobility for different social groups. The most promising advancements may come 
from improving general migration models that better integrate climate-related impacts.
• Migration and immobility outcomes specific to gender or marginalized populations are 
not yet a major focus of climate-related migration forecasting models. Statistical 
extrapolation approaches and ABMs have begun to explore these differences more robustly. 
Targeted data collection efforts to capture migration aspirations, plans for adaptation and migration, 
and migration behavior in climate-stressed contexts will contribute to more nuanced forecasting 
specific to gender and marginalized populations. 
Evaluating Climate-Related Migration Forecasting Models 33
• Short-term projections tend to be more accurate than long-term estimates. When using 
gravity or radiation models to project future population distributions, projections in the one, five, or 
10 year range are more likely to be reliable than those forecasting to 2050 or 2100. Accuracy 
degrades with the time horizon. We chose models forecasting to 2050 rather than 2100 for this 
reason, but if concrete projections are to influence policy and programming, it may be more helpful 
to consider projections for the coming decade. 
• Exposure mapping can highlight populations most vulnerable to climate change, and 
development interventions in these regions should consider both in-situ adaptation and 
facilitating migration as adaptation. Distress migrants tend to move shorter distances, and 
supporting investments in infrastructure, housing, jobs, and social services in towns and cities of 
vulnerable regions will remain important. 
• Significant data collection improvements are needed to enhance forecasting. This 
includes investments in statistical bureaus to improve both the frequency and accuracy of census 
data collection, investments in longitudinal (ideally panel) surveys to capture migration and 
immobility dynamics in local contexts and for different social groups, and investments in innovative 
ways to track migration and displacement following sudden-onset events (e.g., cell phone data). 
USAID could also consider mainstreaming questions about migration aspirations, plans, and ability 
into monitoring and evaluation surveys to expand the evidence base available to assess the impact of 
development interventions on migration decision-making in climate-stressed regions.
• Foresight exercises may help policymakers anticipate future climate-related migration 
better than forecasting models alone. Convening discussions with modelers, migration 
scholars, natural scientists, development practitioners, and other stakeholders may lead to a more 
comprehensive assessment of potential future mobility trends to inform policy and programming for 
a given region. Migration and regional scholars can identify gaps or inaccuracies in the inputs or 
assumptions of forecasting models and offer caveats about the applicability of the models’ findings 
for particular populations or sub-regions. Natural scientists can offer greater insight into differences 
in exposure and risk at the household or community-level. 
Report | December 2022 34
References 
Abrahm Lustgarten. 2020. “The Great Climate Migration.” New York Times Magazine. July 32, 2020.
Ajzin, Icek. 1991. “The Theory of Planned Behavior.” Organizational Behavior and Human Decision Processes
50(2):179-211.
Benveniste, Hélène, Michael Oppenheimer, and Marc Fleurbaey. 2022. "Climate Change Increases 
Resource-Constrained International Immobility." Nature Climate Change 12 (7):634-641. doi: 
10.1038/s41558-022-01401-w.
Black, Richard, W. Neil Adger, Nigel W. Arnell, Stefan Dercon, Andrew Geddes, and David Thomas. 2011. 
Foresight: Migration and Global Environmental Change. Final Project Report. The Government Office for 
Science, London.
Boas, Ingrid, Carol Farbotko, Helen Adams, Harald Sterly, Simon Bush, Kees van der Geest, Hanne Wiegel, 
Hasan Ashraf, Andrew Baldwin, Giovanni Bettini, Suzy Blondin, Mirjam de Bruijn, David Durand￾Delacre, Christiane Fröhlich, Giovanna Gioli, Lucia Guaita, Elodie Hut, Francis X. Jarawura, Machiel 
Lamers, Samuel Lietaer, Sarah L. Nash, Etienne Piguet, Delf Rothe, Patrick Sakdapolrak, Lothar Smith, 
Basundhara Tripathy Furlong, Ethemcan Turhan, Jeroen Warner, Caroline Zickgraf, Richard Black, and 
Mike Hulme. 2019. "Climate Migration Myths." Nature Climate Change 9 (12):901-903. doi: 
10.1038/s41558-019-0633-3.
Brown, Oli, and Robert McLeman. 2013. “Climate Change and Migration: An Overview.” The Encyclopedia 
of Global Human Migration. Immanuel Ness, editor: Blackwell Publishing Ltd.
Brown, Oli. 2008. "Migration and Climate Change." International Organization for Migration’s Migration 
Research Series 31.
Burzyński, Michał, Christoph Deuster, Frédéric Docquier, and Jaime De Melo. 2022. "Climate Change, 
Inequality, and Human Migration." Journal of the European Economic Association (0(0)):1-53 (Global). 
Carling, Jørgen. 2002. "Migration in the Age of Involuntary Immobility: Theoretical Reflections and Cape 
Verdean Experiences." Journal of Ethnic and Migration Studies 28 (1):5-42.
Carling, Jørgen, and Kerilyn Schewel. 2018. "Revisiting Aspiration and Ability in International Migration." 
Journal of Ethnic and Migration Studies 44 (6):945-963. 
Chen, Joyce, and Valerie Mueller. 2018. "Coastal Climate Change, Soil Salinity and Human Migration in 
Bangladesh." Nature Climate Change 8 (11):981-985. 
Chen, Joyce, and Valerie Mueller. 2019. "Climate-Induced Cross-Border Migration and Change in 
Demographic Structure." Population and Environment 41 (2):98-125. 
Clemens, Michael. 2020. "The Emigration Life Cycle: How Development Shapes Emigration from Poor 
Countries." Center for Global Development Working Paper Series (Paper 540).
Clement, Viviane, Kanta Kumari Rigaud, Alex de Sherbinin, Bryan Jones, Susana Adamo, Jacob Schewe, 
Nian Sadiq, and Elham Shabahat. 2021. Groundswell Part 2. World Bank, Washington, D.C. 
Evaluating Climate-Related Migration Forecasting Models 35
Cundill, Georgina, Chandni Singh, William Neil Adger, Ricardo Safra de Campos, Katharine Vincent, Mark 
Tebboth, and Amina Maharjan. 2021. "Toward a Climate Mobilities Research Agenda: Intersectionality, 
Immobility, and Policy Responses." Global Environmental Change 69:102315. 
Czaika, Mathias and Constantin Reinprecht. 2022. "Why Do People Stay Put in Environmentally Stressful 
Regions? Cognitive Bias and Heuristics in Migration Decision-Making." Regional Environmental Change
22 (3):84. doi: 10.1007/s10113-022-01934-y.
Davis, Kyle Frankel, Abinash Bhattachan, Paolo D’Odorico, and Samir Suweis. 2018. "A Universal Model 
for Predicting Human Migration under Climate Change: Examining Future Sea Level Rise in 
Bangladesh." Environmental Research Letters 13 (6):064030.
de Haas, Hein. 2010. “Migration Transitions: A Theoretical and Empirical Inquiry into the Developmental 
Drivers of International Migration.” IMI Working Paper Series, WP24. International Migration Institute.
De Lellis, Pietro, Manuel Ruiz Marin, and Maurizio Porfiri. 2021. “Modeling Human Migration Under 
Environmental Change: A Case Study of the Effect of Sea Level Rise in Bangladesh.” Earth’s Future 9 
(4): 1-14.
de Sherbinin, Alex. 2020. “Climate Impacts as Drivers of Migration.” Migration Policy Institute.
de Sherbinin, Alex. 2022. “Internal Climate Migration Scenarios Using Population Gravity Modeling 
Approach.” Scenarios Forum 2022.
Disney, George, Arkadiusz Wiśniowski, Jonathan J Forster, Peter W. F. Smith, and Jakub Bijak. 2015 
“Evaluation of Existing Migration Forecasting Methods and Models,” 65.
Entwisle, B., A. Verdery, and N. Williams. 2020. “Climate change and migration: New insights from a 
dynamic model of out-migration and return migration.” American Journal of Sociology 125(6), 1469-
1512. 
Farbotko, Carol and Celia McMichael. 2019. "Voluntary Immobility and Existential Security in a Changing 
Climate in the Pacific." Asia Pacific Viewpoint 60 (2):148-162. doi: https://doi.org/10.1111/apv.12231.
Farbotko, Carol, Olivia Dun, Fanny Thornton, Karen E McNamara, and Celia McMichael. 2020. "Relocation 
Planning Must Address Voluntary Immobility." Nature Climate Change 10 (8):702-704.
Feng, Shuaizhang, Alan B. Krueger, and Michael Oppenheimer. 2010. "Linkages among Climate Change, 
Crop Yields and Mexico–Us Cross-Border Migration." Proceedings of the National Academy of 
Sciences 107 (32):14257-14262. 
Forrester, Jay W. 1971. “Counterintuitive behavior of social systems.” Theory and Decision 2: 109-140.
Fussell, E., Hunter, L., & Gray, C. (2014). Measuring the environmental dimensions of human migration: 
the demographer’s toolkit. Global Environmental Change, 28, 182–191. 
Gray, Clark L., and Valerie Mueller. 2012. "Natural Disasters and Population Mobility in Bangladesh." 
Proceedings of the National Academy of Sciences 109 (16):6000-6005. 
Gray, Clark, and Erika Wise. 2016. "Country-Specific Effects of Climate Variability on Human Migration." 
Climatic change 135 (3-4):555-568.
Report | December 2022 36
Henry, Sabine, Bruno Schoumaker, and Cris Beauchemin. 2004. "The Impact of Rainfall on the First out￾Migration: A Multi-Level Event-History Analysis in Burkina Faso." Population and Environment 25 
(5):423-460. 
International Organisation for Migration. 2019. Glossary on Migration. Geneva: International Organization 
for Migration. 
Jones, Bryan. 2020. Modeling Climate Change-Induced Migration in Central America & Mexico: 
Methodological Report. Marxe School of Public and International Affairs, Baruch College and the 
CUNY Institute for Demographic Research (CIDR).
Lutz, Wolfgang and Joshua R. Goldstein. 2004. “Introduction: How to Deal with Uncertainty in Population 
Forecasting?” International Statistical Review / Revue Internationale de Statistique 72 (1): 1–4.
McLeman, Robert A. 2014. Climate and Human Migration: Past Experiences, Future Challenges: 
Cambridge University Press.
McLeman, Robert. 2018. "Thresholds in Climate Migration." Population and Environment 39 (4):319-338. 
doi: 10.1007/s11111-017-0290-2.
Naugle, Asmeret Bier, George A. Backus, Vincent C. Tidwell, Elizabeth Kistin-Keller, and Daniel L. Villa. 
2022. "A Regional Model of Climate Change and Human Migration." Research Anthology on 
Environmental and Societal Impacts of Climate Change 449-471. IGI Global. (Mali and African region)
Nawrotzki, Raphael J., Fernando Riosmena, Lori M. Hunter, and Daniel M. Runfola. 2015. "Amplification 
or Suppression: Social Networks and the Climate Change—Migration Association in Rural Mexico." 
Global Environmental Change 35:463-474. 
New Immigrant Survey. (2016). NIS - The New Immigrant Survey. Available from: https://nis.princeton.edu/
Organisation for Economic Co-operation and Development. (2008). International Migration Outlook 2008—
OECD. Available from: https://www.oecd.org/els/mig/internationalmigrationoutlook2008.htm
Reimann, Lena. 2022. “Exploring spatial feedbacks between sea level rise, adaptation policies, and 
migration.” Scenarios Forum 2022.
Rigaud, Kanta Kumari, Alex de Sherbinin, Bryan Jones, Jonas Bergmann, Viviane Clement, Kayly Ober, 
Jacob Schewe, Susana Adamo, Brent McCusker, and Silke Heuser. 2018. Groundswell: Preparing for 
Internal Climate Migration. World Bank, Washington, D.C. 
Schewel, Kerilyn. 2020. “Understanding Immobility: Moving Beyond the Mobility Bias in Migration Studies.” 
International Migration Review 54(2): 328-355.
Schewel, Kerilyn and Legass Bahir Asmamaw. 2021. "Migration and Development in Ethiopia: Exploring 
the Mechanisms Behind an Emerging Mobility Transition." Migration Studies 9 (4):1673-1707. 
Schewel, Kerilyn. 2021. Staying Put: Why It’s Time to Pay More Attention to Mixed Immobility. Mixed 
Migration Review 2021: Mixed Migration Center.
Shryock, Henry S. and Jacob S. Siegel. 1980. The Methods and Materials of Demography. Department of 
Commerce, Bureau of the Census.
Evaluating Climate-Related Migration Forecasting Models 37
Simini, Filippo, Gianni Barlacchi, Massimilano Luca, and Luca Pappalardo. 2021. "A Deep Gravity Model for 
Mobility Flows Generation." Nature Communications 12 (1):6576. 
Simini, Filippo, Marta C. González, Amos Maritan, and Albert-László Barabási. 2012. "A Universal Model 
for Mobility and Migration Patterns." Nature 484 (7392):96-100. 
Singh, Ruchi. 2022. ‘Origin of World’s Largest Migrant Population, India Seeks to Leverage Immigration,’ 
Migration Policy Institute. 
Smirnov, Oleg, Gallya Lahav, John Orbell, Minghua Zhang, and Tingyin Xiao. 2022. "Climate Change, 
Drought, and Potential Environmental Migration Flows under Different Policy Scenarios." International 
Migration Review:01979183221079850.
Sofuoğlu, Emrah and Ahmet Ay. 2020. “The Relationship between Climate Change and Political Instability: 
The Case of MENA Countries (1985:01–2016:12).” Environmental Science and Pollution Research 27(12): 
14033–43.
Speelman, L.H., R.J. Nicholls, and R.S. de Campos. 2021. “The role of migration and demographic change 
in small island futures.” Asian and Pacific Migration Journal 30, no. 3 (September 2021): 282–311. 
Stark, Oded, and David E. Bloom. 1985. "The New Economics of Labor Migration." American Economic 
Review 75:173-178.
Stefanouli, Maria and Serafeim Polyzos. 2017. “Gravity vs radiation model: two approaches on commuting 
in Greece.” Transportation Research Procedia 24: 65-72.
Thober, Jule, Nina Schwarz and Kathleen Hermans. 2018. “Agent-based modeling of environment￾migration linkages: a review.” Ecology and Society 23 (2): 22-68
Van Praag, Lore. 2021. "Can I Move or Can I Stay? Applying a Life Course Perspective on Immobility When 
Facing Gradual Environmental Changes in Morocco." Climate Risk Management 31:100274. 
Vezzoli, Simona, Ayla Bonfiglio, and Hein De Haas. 2017. Global Migration Futures: Exploring the Future 
of International Migration with a Scenario Methodology.” International Migration Institute Working Paper 
Series, WP No. 135. International Migration Institute.
Wahba, J. 2021. “Who benefits from return migration to developing countries?” IZA World of Labor. 
Available from: https://doi.org/10.15185/izawol.123
Walsh, S. J., G. P. Malanson, B. Entwisle, R. R. Rindfuss, P. J. Mucha, B. W. Heumann, P. M. McDaniel, B. 
G. Frizzelle, A. M. Verdery, N. Williams, Y. Xiaozheng, and D. Ding. 2013. "Design of an Agent-Based 
Model to Examine Population-Environment Interactions in Nang Rong District, Thailand." Applied 
Geography 39. doi: 10.1016/j.apgeog.2012.12.010.
Wiśniowski, Arkadiusz, Peter W. F. Smith, Jakub Bijak, James Raymer, and Jonathan J. Forster. 2015. 
“Bayesian Population Forecasting: Extending the Lee-Carter Method.” Demography 52 (3): 1035–59.
Zelinsky, Wilbur. 1971. "The Hypothesis of the Mobility Transition." Geographical Review 61 (2):219-249.
Zickgraf, Caroline. 2021. "Theorizing (Im)Mobility in the Face of Environmental Change." Regional 
Environmental Change 21 (4):126. doi: 10.1007/s10113-021-01839-2.
Report | December 2022 38
Appendix A.1: Background on Search 
Strategy
Phase 1
The science of forecasting climate-related migration is still nascent. Given that almost all models have 
emerged only since the 2010s, we did not apply a time limit in choosing models to assess. Our search 
strategy was three pronged. First, we used the search engines Scopus, Web of Science, Proquest, and the 
Duke Library system to find papers that use the terms “climate,” “migration,” and “model.” This yielded 
269 papers, primarily from academic literature. After an initial round of sorting, we identified 29 potentially 
relevant articles that focused on forecasting migration (rather than modeling historical trends), with the 
majority providing estimates at the country or regional level.
Second, to review the gray literature, we used the PAIS Index, which includes reports from the World 
Bank, the Brookings Institution, and the RAND Corporation. We also searched the International 
Monetary Fund E-Library, the OECD iLibrary, and the World Bank e-Library. In order to ensure the most 
comprehensive search, we utilized Google search results, particularly those at .gov or .org sites, and those 
mentioned in media reports. Combined, these searches yielded 64 additional papers and reports. Once 
duplicates were removed and each model was assessed, the gray literature search resulted in five 
additional models for consideration.
Third, we did a broader search of major migration forecasting models (not specifically climate-related) to 
see whether some included climate variables, including projections from the United Nations Population 
Division and Department of Economic and Social Affairs, the OECD, the International Organisation for 
Migration, and the Vienna Institute of Demography. This resulted in one additional model (Sander et al. 
2013, 2017). Unfortunately, many estimates and projections are based on observed changes in populations 
or migration trends and include neither climate-related nor significant political or economic variables. We 
are more confident that models that do include climate-related variables would include this focus in the 
title or description/abstract, which would have been found via our first two search strategies.
Through this three-pronged search, we initially identified 115 potentially relevant models. Upon closer 
inspection we found that only 33 of these articles actually forecasted migration relating to climate change. 
In our sorting, we removed articles theoretical in nature or that simply reviewed other models, leaving 
26 models that made future-oriented projections. The vast majority of papers modeled climate-migration 
interactions based on existing or historical data, rather than forecasting migration, which is why there is 
such a steep drop between the 115 potentially relevant models and the final 26 we included from this 
portion of the search. Figure 2 details our search strategy according to the Preferred Reporting Items for 
Systematic Reviews and Meta-Analyses (PRISMA) guidelines.
Evaluating Climate-Related Migration Forecasting Models 39
Figure 2. Literature Search Process
Phase 2
After meeting with USAID, we were asked to look more closely at whether we could find models for 
Europe and Eurasia, the Middle East, and small island states. We conducted another literature search using 
Duke library database and Google Scholar, using the following broad search terms:
• In database: climate OR migration OR model OR drought OR water OR storm OR flood OR 
displacement OR mobility OR population movement AND country name 
• In Google: migration model [country name]
Report | December 2022 40
This resulted in two additional models for small island states. We did not find any articles that met our 
criteria for the Middle East or Europe and Eurasia. This brought the total number of climate-related 
migration forecasting models to 30.
In order to yield a broad sense of the state of climate migration modeling, we first coded all 30 models 
on a variety of factors including the type of model, model prediction, geography, intervening variables, and 
data sources. In order to fit the USAID scope, we removed models focused solely on developed countries 
such as the United States and Australia. We focused on models that provide projections to 2050 (or 
between 2040 and 2060). We removed those that provided predictions for the year 2100 and beyond, 
because since we judged that date to be too distant to inform USAID programming. Finally, we added one 
statistical extrapolation model to ensure we have an example of this promising approach. This was an 
omission highlighted by one of our expert reviewers. Statistical extrapolation models (also called discrete 
history event modeling) were not initially included in our review for several reasons. Many discrete history 
event models tend to focus on modeling historical trends; those that use the relationships they uncover 
between climate variables and migration outcomes to project future trends do not always emphasize these 
projections in the title or abstract. In one paper, for example, it is a footnote. Others that focus more 
directly on forecasting often do not offer projections within a specific time horizon. Instead, they may 
examine how migration might change in relation to the degree of change of a climate-related variables 
(e.g., flooding, soil salinity, or a hurricane of a certain degree of severity). We decided to add one model 
to our review that exemplifies this approach. The final analysis included 20 models. 
Phase 3
To illustrate more clearly the various approaches, strengths, and weaknesses of forecasting models, we 
conducted a deep analysis of nine models purposely selected for variation across world regions, geographic 
scope or scale (global, regional, or country-level forecasts), source (academic or gray literature), migration 
type (internal, international, and trapped populations), model types (gravity, ABM, radiation, etc.), and 
impact. The list was shared and further refined in conversation with USAID. 
In selecting models for in-depth study, we originally adopted a purposive sampling approach–a non￾probability approach that selects cases in a strategic way, based on their relevance to the research 
question. There are different types of purposive sampling techniques (typical case sampling, extreme or 
deviant case sampling, criterion sampling, theoretical sampling, etc.). In our first proposal, we took a 
variation sampling approach, meaning we wanted to identify models that would familiarize USAID with the 
strength and weaknesses of a wide variety of approaches to climate-related migration forecasting. We 
looked for variation across a range of areas:
• Geographic coverage (what regions of the world are represented)
• Geographic scope or scale (global, regional, or country-level forecasts)
• Source (academic or gray literature)
• Migration type (internal, international, and trapped populations)
• Model types (gravity, radiation, ABM, other)
Based on these focus areas, and in conversation with USAID, we selected nine models for in-depth review. 
Evaluating Climate-Related Migration Forecasting Models 41
Appendix A.2: Background on Expert 
Interviews
In addition to this evaluation of the proposed models, our team held three individual expert interviews 
and an additional panel workshop. The final workshop invited the authors of the climate-related migration 
forecasting models to workshops to better understand what they see as the strengths, limitations, 
applicability, scalability, and policy-relevance of their respective models. Interviews and panel workshops 
addressed common questions and themes, including: 
• The strengths, weaknesses, and scope of applicability of different modeling approaches
• How to interpret the results of climate-related migration forecasting models and the implications for 
policy 
• Recent advancements in the field and areas for future research 
The results of these discussions are included in the findings of the report. Participants include: 
Bryan Jones, an Assistant Professor of Sustainability at the Marxe School of Public and International 
Affairs, Baruch College, City University of New York. His work explores the relationship between spatial 
population dynamics, urbanization, and climate change vulnerability. He contributed to modeling for the 
Groundswell report and the Great Climate Migration (ProPublica) project. (Individual interview)
Robert McLeman, a Professor in the Department of Geography and Environmental Studies at Wilfrid 
Laurier University. His research focuses on the human dimensions of environmental change, with 
particular attention to the relationship between environment and human migration, community adaptation 
to climatic variability and change, and fostering citizen participation in environmental science. He is a 
Coordinating Lead Author for the Intergovernmental Panel on Climate Change’s working group on 
impacts, vulnerability and adaptation. (Individual interview)
Alex de Sherbinin, Deputy Director and Senior Research Scientist at the Center for International Earth 
Science Information Network (CIESIN), a spatial data and analysis center within the Columbia Climate 
School of Columbia University specializing in the human aspects of global environmental change. He is also 
Deputy Manager at the NASA Socioeconomic Data and Applications Center. His research focuses on the 
human aspects of global environmental change and geospatial data applications, integration, and 
dissemination. His research and teaching address climate-related mobility, climate vulnerability mapping, 
urban climate vulnerability, population dynamics and the environment, and environmental indicators. He 
contributed to modeling for Groundswell and is currently involved with modeling for the Africa Climate 
Migration Initiative. (Individual interview and Panel discussion)
Helene Beneviste, a postdoctoral Environmental Fellow at the Harvard University Center for the 
Environment, based at the Harvard Kennedy School of Government. Her research explores interactions 
between climate change impacts, human migration, and inequality, using Integrated Assessment Models 
and a scenarios-based approach. She is a lead author on a recent model forecasting resource constrained 
immobility for climate change scenarios. (Panel discussion)
Report | December 2022 42
Michael Burzyński, a researcher at Luxembourg Institute of Socio-Economic Research. His research is 
focused on the economics of international migration, the economic impact of climate change, selection of 
workers and matching on labor markets, and quantitative economic theory. He is a lead author on the 
“Climate Change, Inequality, and Human Migration” model assessed in this report. (Panel discussion)
Fabien Cottier, a Postdoctoral Research Scientist at Lamont-Doherty Earth Observatory of Columbia 
University. His research focuses on how global warming reshapes migration patterns and the implications 
for the risk of conflict associated with the movement of people in developing countries. (Panel Discussion)
The final report benefited from the expert review of Arizona State University Professor Valerie 
Mueller. Valerie Mueller is an associate professor in the School of Politics and Global Studies. Her 
research focuses on quantifying rural household vulnerability to climate variability, focusing on migration, 
nutrition, and health markers in Africa and Asia. Additionally, she uses randomized controlled trials to 
identify mechanisms to improve the delivery of rural services (legal justice for women, agricultural 
extension, and the equitable allocation of irrigation water) in East African countries.
Finally, the authors benefited from presentations and insights offered during an expert consultation on 
forecasting climate-related migration for small island states in September 2022. The workshop was 
organized by the Global Centre for Climate Mobility (GCCM), the Association of Caribbean States 
(ACS), Columbia University's Climate School, and the World Bank’s Global Knowledge Partnership for 
Migration and Development (KNOMAD) Working Group on Environmental Migration. 
Evaluating Climate-Related Migration Forecasting Models 43
Appendix A.3: Details of the Nine Models
Model 1. Groundswell Model (Rigaud et al. 2018; Clement et al. 2021)
Model type Gravity
Type of Migration Internal; permanent
Geographic Coverage Regional (Sub-Saharan Africa, South Asia, Latin America, East Asia and the Pacific, North 
Africa, Eastern Europe, and Central Asia)
National (Ethiopia, Bangladesh, Mexico, Morocco, Vietnam, Kyrgyz Republic) 
Data Inputs
Scenarios thinking Three scenarios based on a combination of two climate scenarios and two development 
scenarios.
Development scenarios: Shared socioeconomic pathways (SSP2 and SSP4). Under 
SSP4 only low income countries experience high population growth, coupled with 
substantial inequality leading to adaptation challenges. Middle income countries 
experience low population growth much like high income countries. SSP2 is a moderate 
development scenario between SSP1 (“sustainability”) and SSP3 (“fragmentation”) with a 
slow reduction in inequalities among world regions and more moderate trends in 
population growth, urbanization, income, and education.
Climate scenarios: Forecasts are based on two greenhouse gas concentration 
trajectories using representative concentration pathways (RCP8.5 and RCP2.6). For the 
higher emissions scenarios, temperatures rise by 1.4°–2.6°C by 2050, and by 2.6°–4.8°C 
by 2100 (IPCC 2014). For the lower emissions pathway, temperatures peak at 0.25°–
1.5°C above recent baseline levels by 2050 and then stabilize through the end of the 
century. In the higher emissions scenario, temperatures rise by 0.5°–2°C by 2050 and by 
3°–5.5°C by 2100.
Environmental and 
Climate Impacts
Slow-onset changes;
Crop yields, water availability, sea level rise
Forecasts for water availability and crop yields using data from the Inter-Sectoral Impact 
Model Intercomparison Project, which uses computer model simulations of biophysical 
climate impacts. The water sector model outputs represent river discharge, measured in 
cubic meters per second in daily/monthly time increments. The crop sector model 
outputs represent crop yield in tons per hectare on an annual time step at a 0.5° x 0.5° 
grid cell resolution. Crops include maize, wheat, rice, and soybeans. For regions with 
multiple cropping cycles, yield reflects only the major crop production period. The data 
were converted to decadal average water availability and crop production (in tons) per 
grid cell. An index was then calculated that compares those values with the 40-year 
average for water availability and crop production for 1970–2010. Sea level rise 
projections from the IPCC Fifth Assessment Report, augmented by an increment for 
storm surges.
Report | December 2022 44
Population data The population baseline used is the 2010 baseline in the Center for International Earth 
Science Information Network (CIESIN) Gridded Population of the World Version 4 
(GPWv4) (CIESIN 2016) (Map 3.1). The gravity model was calibrated based on 
population change estimates for 1990 to 2000 from GPW version 3 (CIESIN and others 
2005) and for 2000 to 2010 from GPWv4. GPW versions 3 and 4 model the distribution 
of the population on a continuous global surface based on the highest spatial resolution 
census data available from the 2000 and 2010 rounds of censuses, respectively. 
Population count grids were used that were adjusted to national-level estimates from 
the United Nations World Population Prospects reports. GPWv3 and v4 are gridded 
data products with output resolutions of 2.5 arc-minutes (a square approximately four 
kilometers on a side at the equator) and 30 arc-seconds (a square approximately one 
kilometer on a side at the equator), respectively. For model calibration and baseline 
population for future projections, the data were aggregated to 7.5 arc-minutes (a square 
approximately 14 kilometers on a side at the equator [i.e. grid cells with an area of 196 
square kilometers]).
Non-climate drivers Shared Socioeconomic Pathways
Intervening 
Obstacles/Facilitators
Not directly included.
Personal and HH 
characteristics
Not directly included.
Model Description Deviations between population distributions in model runs that include crop and water 
impacts and development-only (also referred to as the SSP or “no climate impact”) model 
runs are assumed to be driven primarily by differences in climate change–induced internal 
migration. The model assumes that fertility and mortality rates are relatively consistent 
across populations in a locale.
Outputs Changes in population distribution (and indirectly in internal migration). 
Scalability/Adaptability The model may be customized and expanded at different scales. Gravity modeling is one 
of the few approaches available to take climate migration modeling to scale.
Limitations The model cannot forecast all future adaptation efforts or conflict, cultural, political, 
institutional, or technological changes. Discontinuities are likely to arise as a result of 
political events and upheavals that can heavily influence migration behavior. The scenario 
framework is not designed to predict shocks to any socioeconomic or political system, 
such as war or market collapse. The models also cannot anticipate new technologies that 
may dramatically affect adaptation efforts to the degree that climate impacts become 
negligible. SSPs, as well as output from the global climate model and the Inter-Sectoral 
Impact Model Intercomparison Project, reflect plausible futures that span a wide range 
of global trajectories, with the caveat that extremely unpredictable or unprecedented 
events are explicitly excluded.
Model 2. Great Climate Migration Model (Jones 2020)
Model type Gravity
Type of Migration Internal, international; permanent
Geographic Coverage Regional (Latin America)
Data Inputs
Evaluating Climate-Related Migration Forecasting Models 45
Scenarios Development scenarios: Shared socioeconomic pathways (SSP1, SSP3, and SSP5) 
Climate scenarios: Climate output consistent with three representative 
concentration pathways (2.6, 4.5, and 8.5) are incorporated in this work as drivers of 
the vulnerability and sectoral-change indicators
Five plausible socioeconomic and climate futures are considered: An 
optimistic/reference scenario (SSP1 and RCP2.6), a pessimistic scenario (SSP3 and 
RCP8.5), a more climate-friendly scenario (SSP3 and RCP4.5), a more development￾friendly scenario (SSP5 and RCP8.5), and a moderate scenario (SSP5 and RCP4.5). 
Environmental & 
Climate Impacts
Water Availability, Agriculture/Crop Yields, Biomes/Ecosystem Productivity, and Flood 
Hazards (Inter-Sectoral Impact Model Intercomparison Project)
Freshwater Availability, Sea level rise, Elevation (Socio-Economic Data and Applications 
Center, NASA/Columbia University Earth Institute)
Extreme Heat Days (Community Earth Systems Model, National Center for 
Atmospheric Research) 
Slope 
Water Bodies (Environmental Systems Research Institute)
Population data Gridded Population of the World v4 (GPW) (CIESIN), modified by the Global Human 
Settlement Population Grid (GHS-Pop) (European Commission Joint Research Center)
Historical Bilateral Migration Flows (Migration Policy Institute, dataset by Abel and 
Cohen 2019)
Non-climate drivers Shared Socioeconomic Pathways
Political Stability, Control of Corruption (Worldwide Governance Indicators, World 
Bank)
Gross Domestic Product (OECD)
Intervening 
Obstacles/Facilitators
Diaspora (United Nations Population Division)
Man-made structures (“Built-up”) (European Commission Joint Research Center)
World Database on Protected Areas (International Union for Conservation of Nature)
Personal and HH 
characteristics
Population (Age and Sex Structure) (Gridded Population of the World Version 4.10 
Basic Demographic Characteristics)
Model Description To forecast international migration, existing bilateral flow data is used to train and 
project the model as a function of sectoral impacts (crops, water, NPP), political 
instability and corruption, global income levels (GDP per capita), and the existing 
diaspora, to estimate potential changes in origin-destination flows under the five 
alternative futures (RCP/SSP combinations) noted above. 
To forecast internal migration, the model authors use a modified version of the 
Groundswell model. The model estimates changes in the spatial population distribution 
(including the impact of climate change) in five-year time steps by (1) calculating a 
population potential surface (a distribution of values reflecting the relative 
attractiveness of each grid cell), and (2) allocating population change to grid cells 
proportionally based on potential. To generate estimates of internal migration under 
climate change, they then run a set of scenarios that exclude the impacts of climate 
change. They hold the values for all variables that are influenced by climate change 
constant at current day values (crop, water, NPP, heat extremes, flood hazard, and sea￾level). The difference in population distribution between the five primary scenarios and 
these “no-climate” scenarios is attributed to migration induced by changing conditions, 
as the only variables that have changed are those impacted by a shifting climate. 
Report | December 2022 46
Outputs Changes in bilateral flow data (international migration) and population distribution (and 
indirectly internal migration). 
Scalability/Adaptability The model may be customized and expanded at different scales. Calibration of the 
model, particularly for international migration, requires reliable, historical data on 
bilateral migration flows. Reliable data is rare. 
Limitations Some assumptions/inputs are unrealistic (e.g., age structure, sex ratio, build-up land, 
groundwater, political stability, among others remain constant in future projections) 
Does not incorporate non-linear changes in how populations respond to climate stress. 
Like all other models, it cannot forecast all future adaptation efforts or conflict, cultural, 
political, institutional, or technological changes. 
Model 3. Universal Model (Davis et al. 2018)
Model type Radiation
Type of Migration Internal, permanent
Geographic Coverage Country (Bangladesh)
Data Inputs
Scenarios Not for 2050, where the model using RCP8.5 estimates of sea level rise. The model 
does use two climate scenarios for 2100 estimates. 
Environmental & 
Climate Impacts
Sea level rise (RCP8.5 projection used for 2050 estimates; global mean sea level rise 
for four RCP pathways used for 2100 projection)
Elevation data (Shuttle Radar Topography Mission SRTM Global 1 arc second Version 
3.0)
Population data District-(zila-)level population for the year 2010 was estimated by aggregating the 
Gridded Population of the World (GPWv4) dataset (30 arcsecond grid) and adjusting 
to United Nations medium-variant population estimates.
Non-climate drivers Not included.
Intervening 
Obstacles/Facilitators
Not included. 
Personal and HH 
characteristics
Not included. 
Model Description In the baseline model (based on Simini et al. 2012) every individual, X, leaving from 
location i is associated with a positive number, representing the absorption threshold 
for that individual. On average, individuals leaving from highly populated regions have a 
higher absorption threshold than those emitted from a scarcely populated location. 
The surrounding cities have a certain probability to absorb individual X. The individual 
stops in the closest location that has an absorbance greater than the individual’s 
absorption threshold. By repeating these steps for a given number of out-going 
individuals (predicted emigrants), the model calculates flux among the different 
locations. To simulate migrations among districts of Bangladesh that are each affected 
differently by sea level rise, Simini et al.’s baseline model was adapted to account for 
the fact that migration fluxes toward inundated areas are less likely.
To estimate additional food, housing, and jobs required at migrant destinations, authors 
multiplied the number of arriving migrants by average calorie consumption in 2010 
(FAO Food Balances Sheets), average household size, and ratio of employed to total 
population (World Development Indicators). 
Evaluating Climate-Related Migration Forecasting Models 47
Outputs Estimates of sources, destinations, and fluxes of migrants displaced under projected sea 
level rise.
Includes estimates of lifetime migrants and five-year migrants.
Scalability/Adaptability The limited data inputs and parameter free model means this approach is easier to 
apply to other areas, particularly data-scarce regions.
The model has been adapted by others to include a parameter for baseline migration 
rates; this addition significantly changes the results of the model (see De Lellis et al. 
2021) 
Limitations Does not incorporate adaptation or temporary flows. It assumes no return migration. 
It uses scenarios of global mean sea level rise and national average population growth, 
which in reality will occur with greater spatial heterogeneity. The model does not 
include the potential for adaptation, non-linear change, or political, economic, and 
social change. The model is calibrated based on current migration patterns and 
decisions, and the relative importance of the factors influencing migration will likely 
change as climate change progresses. 
Model 4. Systems Model (Naugle et al. 2022) 
Model type Systems Dynamics Model
Type of Migration Internal, International, Permanent 
Geographic Coverage National (Mali, including potential destinations outside Mali)
Data Inputs (Note: sources of data inputs often not clear from article)
Scenarios No. 
Environmental & 
Climate Impacts
Temperature and precipitation
(The climate change scenario assumes a linear temperature increase from 2010 through 
2100.)
Population data The model creators calibrate the model with historical bilateral migration data from 
the World Bank (2017) and future migration data from World Population Prospects 
(United Nations 2017). 
Non-climate drivers Food availability 
Generic resource availability
Resource utilization 
(Un)Employment 
Labor Supply (skilled and common)
Indexed level of technology 
Existing Violence
Governance Effectiveness
Infrastructure and services 
Intervening 
Obstacles/Facilitators
Not included. 
Personal and HH 
characteristics
Gender, age, type of labor (skilled, common), births, mortality, and aging flows
Report | December 2022 48
Model Description The model couples a model of migration choice with a multi-sectoral model of the 
environment in which the potential migrant functions, including feedback of migration 
decisions on both the sending and receiving communities. A system dynamics model 
simulates the influence of climate on human migration, focusing on the causal factors 
of economy, labor, population, violence, governance, water, food availability, and 
disease. The model has an annual time step extending over a 70-year time horizon, 
beginning in 1990 (for historical reproduction) and projecting out to 2060.
Decisions about where to live are calculated using a cognitive formulation based on 
qualitative choice theory (McFaddin 1982). The “Migration Utility” is calculated based 
on a number of factors: labor type, gender, age, wage income, food availability, 
governance effectiveness, infrastructure and services, disease mortality, violence 
incongruity, income incongruity, food incongruity, unemployment rate, population, and 
national disaster index. 
The model simulates the economic situation in each region by calculating the potential 
gross regional product, and then adjusts this using other relevant factors to calculate 
the realized gross regional product for each region, which along with population 
dynamics, determines labor and wage dynamics.
Analyses were referenced to a base case taken from population and economic 
projections of World Development Indicators (World Bank 2017) and World 
Population Prospects (UN 2017).
Five policy options for reducing pressures to migrate internationally were explored: 
increasing contraception availability, increasing governance effectiveness, improving 
infrastructure and services provided by the government, increasing foreign aid to urban 
Mali, and increasing foreign aid to rural Mali.
Outputs The model determines the fraction of the Malian population which chooses to live in 
each simulated region. Potential locations are rural Mali, urban Mali, neighboring 
countries (Burkina Faso, Cote d’Ivoire, Gabon, Gambia, Ghana, Guinea, Mauritania, 
Niger, and Senegal), the United States, and the rest of the world. 
Scalability/Adaptability The model could be applied to other world regions but requires significant data inputs. 
Limitations Researchers applied a data-intensive model to what is typically considered a data-scarce 
region of the world, raising questions about the reliability of the underlying data used. 
The authors were not clear in the article about the sources of the data. 
Model 5. Dynamic Systems Model (Entwisle, Williams and Verdery 2020)
Model type Agent-based Model 
Type of Migration Internal, permanent, return
Geographic Coverage Sub-National (Rural region in Northeast Thailand)
Data Inputs
Scenarios Four climate scenarios were created based on monthly rainfall data for Nang Rong 
from 1900 to 2008 (accessed from the University of Delaware Center for Climate and 
Land Surface Change).
The first scenario focuses on droughts; its middle years are marked by a seven-year 
period of extremely dry weather. The second focuses on floods and contains a seven￾year period of extremely wet weather in the middle years. The third examines 
variability with a scenario whose middle years fluctuate between severe droughts and 
floods. Each is compared to a reference scenario, containing normal-normal weather 
during the middle years, which serves as a benchmark.
Evaluating Climate-Related Migration Forecasting Models 49
Environmental & 
Climate Impacts
Georeferenced villages, households, and the locations and attributes of plots farmed 
and crops grown by each household (survey data) 
Time Series of satellite images classified for land cover/land use
Digital elevation model constructed from topographic maps
Soil depth and drainage maps
Observation in the field (qualitative data) 
Population data Longitudinal panel survey data that followed all individuals in 51 villages, including out￾and in-migrants and return migrants (Nang Rong Project, see Walsh et al. 2013). Final 
model on 41 villages. 
Non-climate drivers
Intervening 
Obstacles/Facilitators
Social network data (kin and exchange) in villages of origin (survey data)
Remittances received (survey data) 
Personal and HH 
characteristics
Age, gender, marital status, place of residence
Model Description The ABM simulation includes multiple types of agents: individuals, land parcels, and 
households. The primary pathways through which extreme climate events can influence 
migration patterns in this ABM include the effect on crop yields of timing and amount 
of rainfall (rice, cassava, and sugar are modeled separately), the way crop yields affect 
household assets, and how household assets and the characteristics of current and 
prior household members affect out-migration and return migration (households with 
more assets are better able to finance migration and to afford the loss of labor 
associated with it). 
There are feedbacks from migration to household assets (remittances increase assets), 
migration to crop yields (through labor availability), and household assets to crop yields 
(through inputs such as fertilizer, which must be purchased). Rainfall is assumed to be 
exogenous, a reasonable assumption given the focus on the experiences of individual 
villages over time. Because plots farmed by households vary in their vulnerability to 
floods and droughts in terms of elevation, distance from rivers, and soil suitability, the 
impact of climate shocks can vary within villages (see Walsh et al. 2013 for a complete 
description of this part of the model).
The rules for out-migration and return migration are based on a probabilistic approach 
and were derived from a statistical analysis of previous survey data and as well as 
relevant substantive and theoretical literature.
The coefficients from regression models are used in the agent-based simulation to 
determine individual specific probabilities of out- migration and return migration in 
each simulated year. Individuals are randomly selected to migrate, with the chance of 
doing so proportional to probabilities defined by regression-based out-migration and 
return migration equations. These coefficients capture the aforementioned feedback 
loops as well as the effects of the personal and household characteristics.
Outputs Changes in rate of out-migration (defined as the proportion leaving of those eligible to 
leave in each year of the model run compared to the reference scenario) induced by 
different climate scenarios in 41 villages. 
Rates of return migration (defined as the proportion returning among those eligible to 
return in each model year, in comparison with the reference scenario) for all 41 villages.
Note: models increases/decreases in migration response (not direction/destination)
Scalability/Adaptability Not easily scalable. The model focuses on a small area in Northeast Thailand and a 
similar approach could be used only in other areas with similar panel survey data. 
Report | December 2022 50
Limitations The model does not derive information on migrant destinations, only climate impacts 
on rates of out- and return-migration. The model is calibrated based on previous 
migration patterns and decisions, and the relative importance of the factors influencing 
migration may change under future climate changes. Like other models, it does not 
incorporate the potential for adaptation or other political, economic, technological, or 
cultural changes. 
Model 6. Inequality Model (Burzyński et al. 2022) 
Model type Computable General Equilibrium Model 
Type of Migration Internal (local and regional)
International 
Geographic Coverage Global 
Data Inputs
Scenarios Yes. They consider RCP7.0 as the benchmark scenario, and RCP4.5 and RCP8.5 
delineate the spectrum of more optimistic and more pessimistic possibilities. 
Environmental & 
Climate Impacts
Changes in average temperature (slow-onset) 
Extreme events and disasters (sudden-onset) (floods, storms, droughts, extreme heat 
waves) (Worldclim.org)
Sea level rise projections (Jackson et al. 2016)
Climate damage (SEDAC, NASA and EM-DAT)
Elevation (SEDAC, NASA)
Population data The world is divided into 198 countries, divided into 2.319 administrative regions [64 
small states], divided into +7.100.000 pixels of 5km x 5km per side. 
Population, age and education structure by pixel (from WorldPop.org and IHME) 
International migration stock data by education level (OECD DIOC-E database) and 
South-South migration stocks imputed by using the United Nations Population Division 
data set
Country-specific regional migration stocks by education levels are constructed using 
census data (IPUMS International), the WorldPop.org data (Sorichetta et al. 2016), and 
the Labor Force Survey data by Eurostat. 
Non-climate drivers Economic sector of locality (agricultural or non-agricultural)
Wage rates endogenized with CES production function and total factor productivity 
(TFP)
Gross Domestic Product estimates by pixels used to calculate TFP residuals (Kummu, 
Taka, and Guillaume 2018)
Urbanization by pixel as related to whether the pixel is based in agriculture or not 
(WorldPop.org)
Fertility projections (UNDP)
Education wage gaps (Global Jobs Indicators Database (JoIn) by the World Bank)
GDP levels and shares of agriculture in consumption and PPP rates (World Bank)
Shares of High Skilled and educated population (Barro and Lee 2013, IHME)
Intervening 
Obstacles/Facilitators
Nothing is explicitly modeled as an intervening obstacle, but it does estimate a cost for 
migrating to approximate legal, monetary, and psychological tolls of moving. 
Evaluating Climate-Related Migration Forecasting Models 51
Personal and HH 
characteristics
Age (under 30 and between 30 and 60)
Education (college education and less education) (Inst. for Health Metrics and Eval.)
Model Description First, the researchers model slow-onset trends and productivity losses; sea level rise, 
forced displacement and productivity losses; and fast-onset shocks, productivity, and 
utility losses. Second, they consider behavior and market responses to climate change. 
These behavior and market responses and productivity losses are then incorporated 
into a utility model as part of variables to gauge how climate change affects economic 
and non-economic incentives to move. 
They model migration decisions as an outcome of a RUM (random utility maximization) 
model. The utility model looks at migration as a set of three choices: a person can 
move pixels while staying in the same region, move regions while staying in the same 
country, or move to a new country. This choice is dependent on economic differences 
across the three levels, the costs of migrating, external effects of population congestion, 
and individual preference. For each choice a utility is calculated using the four 
components just mentioned. Each choice’s utility is then used to construct aggregated 
probability for making that specific migration choice. These probabilities are used to 
estimate how many people are expected to move pixels, regions, or countries. Once 
a new country or region is selected, migrants are assigned to a new pixel based on 
expected utilities across the new region or country. 
When considering forcibly displaced individuals due to sea level rise, the formula is 
slightly adjusted because choosing not to migrate is no longer an option. 
Outputs The number of migrants leaving different areas of the world and their destinations 
relative to a no climate change scenario. The model further breaks these numbers 
down by the type of migration (local, regional, international) and the type of migrant 
(low or high skilled). It gives projections for migration flows for parts of the world.
The model also gives an expected loss in productivity and GDP due to increased 
temperatures. Connected to this, the model also gives projections for how these 
changes will affect income distribution and inequality. Part of this includes changes in 
college-educated and urbanization.
Scalability/Adaptability This global model could be adapted as needed (and as relevant data is available) and 
applied to particular regions. 
Limitations The model keeps climate-related policies constant, foreseeing no investment in greener 
technologies. The model accounts only for those direct and indirect costs that can be 
calibrated using existing empirical studies and time-series data. 
Model 7. Global Model (Smirnov et al. 2021) 
Model type Agent-based Model 
Type of Migration Internal, international, involuntary immobility 
Geographic Coverage Global (all countries of the world) 
Data Inputs
Scenarios Yes. 
Three climate scenarios: (1) constant climate fixed at 2008–2017 levels (any changes in 
this scenario are due to population growth), (2) low emissions RCP 4.5 scenario, and (3) 
high emissions RCP 8.5 scenario
Report | December 2022 52
Environmental & 
Climate Impacts
16 Global Climate Models (GCM) used to model extreme drought projections on the 
basis of the standardized precipitation evapotranspiration (drought) index (24 months, 
SPEI <−2)
Spatial mask (movement constraints): Water
Population data Baseline population distribution from LandScan (1 km spatial resolution)
Population growth from United Nations medium fertility and high fertility scenarios
National identification
Non-climate drivers Not included. 
Intervening 
Obstacles/Facilitators
No border restrictions. 
Personal and HH 
characteristics
Not included.
Model Description Agent-based satisficing model, based on a random choice of an acceptable destination 
within a feasible distance area. The population movement algorithm has five distinct 
steps: 
1) For each grid cell of the world map, examine if the cell is occupied and affected 
by drought given the drought projections based on the climate model and 
emissions scenario used in the simulation
2) Calculate the proportion of the cell population migrating given the simulation 
parameters and national identification of the cell
3) Divide the migrating population into smaller groups. The groups attempt to 
migrate in all possible directions constrained by the maximum possible 
migration distance.
4) A destination cell is available for migration if it is not water, not affected by 
extreme drought, and the destination country is either the same (internal 
migration) or not significantly less developed. 
5) If the initial destination is not available, search for a next available destination 
in the direction chosen randomly. If no destinations are available, then the 
population trying to migrate becomes “immobile.” 
Outputs “Potential migration pressures” expressed as percentage increases or decreases in total 
migration. 
Scalability/Adaptability The model is scalable and other inputs could be included into the ABM. 
Limitations “The behavioral assumptions that we introduce in our model are coarse: humans are 
essentially automata following very basic rules [...] The agents in our model do not have 
age, gender, resources, social capital, diaspora networks, or other characteristics, all of 
which undoubtedly influence migration behavior” (3). 
The model does not incorporate institutional change, adaptation, or future shocks. 
Model 8. Small Island Model (Speelman et al. 2021)
Model type Agent-based Model 
Type of Migration Internal
Geographic Coverage National (Maldives)
Data Inputs
Evaluating Climate-Related Migration Forecasting Models 53
Scenarios Population projections (low growth and high growth scenarios) from the Population 
Division of the United Nations Department of Economic and Social Affairs (UN DESA 
2015).
Governance: No intervention vs. strong intervention. These dynamics are based on the 
existing “population consolidation” and “Safer Islands” policies of the government of 
the Maldives. For details, see Speelman (2016)
Globalization: closed borders vs. open borders. The distribution of international 
migrants is based on Speelman et al. (2017).
High emissions, high climate impact scenario versus low emissions, low climate impact 
scenario. 
The authors use six future scenarios, taken from the United Kingdom Foresight project 
on migration and global environmental change (Government Office for Science 2011). 
Environmental & 
Climate Impacts
Source for climate impacts (high and low impact scenarios) taken from the scenarios 
in the IPCC Special Report on Emission Scenarios (SRES) used by the United Kingdom 
Foresight Report (Black et al. 2011) 
Population data Census data from 1985, 1990, 1995, 2000, 2006, and 2014 
Aggregated data at island level including gender, population and age structure are 
available at the national level for all census datasets.
Annual nationally registered births and deaths
Two datasets are used for the 2014 census: (1) aggregated, island level, population data 
(Maldives Bureau of Statistics 2015) and (2) individual data (Maldives Bureau of Statistics 
2015). 
Note: Expatriates who (temporarily) reside in the Maldives (∼64,000 residents) for 
employment purposes are excluded from the analysis.
Non-climate drivers Indices for island characteristics (no further specificities found) 
Governance (see above)
Intervening 
Obstacles/Facilitators
Open vs. Closed Borders 
Note: the effects of social networks (as a proxy for social norms) are simulated in the 
model, but they are not based on “real” data inputs on social networks. Each agent in 
the model is connected to 50 other agents at the model start up. Information is shared 
about their migration moves and agents store this information for two years.
Personal and HH 
characteristics
For the 2006 and 2014 census datasets were made available that include a full 
anonymized list of the Maldivian population and corresponding characteristics such as 
age, level of education, marital status and migration history (Maldives Bureau of 
Statistics).
Report | December 2022 54
Model Description Agents can develop intentions to migrate to three potential destinations: (1) to the 
capital Malé, (2) within an Atoll District or (3) to another island in the Maldives. The 
intention to migrate is shaped by three factors: (1) the attitude of an agent towards 
migration to different destinations, (2) personal norms and past migration behavior of 
their peers and (3) the diversity of agent attitudes to migration, each based on a set of 
statistical attributes and simulated dynamics between agents and their environment. 
For each simulated time step of one year, each agent determines their intention to 
migrate. Migration behavior of individual agents is then implemented on the basis of 
probability functions. These result in a new population distribution. Birth and death 
rates are included in each time step based on historic data or population projections, 
as well as aging of the population. Data on government relocation of individual 
populations to other islands are also included in the model. These steps are repeated 
for the duration of the simulation. Historic simulations run from 1985 to 2014. 
Demographic futures are simulated from 2014 to 2050.
Outputs Simulated population size in 2050 for all 10 islands under six future scenarios. 
Scalability/Adaptability The model could be applied to other small island states with reliable census data.
Limitations Does not include impacts like tsunamis, which could also lead to significant 
displacement. It is not clear from the paper how climate impacts are measured (e.g., 
whether sea level rise is given a direct focus or rather inputs relate only to high/low 
emissions). The model estimates social networks in a rather coarse and uniform way, 
which is unlikely to reflect real-world dynamics. Overall, the conceptual model is more 
sophisticated than the data inputs available as proxies for the most important factors. 
Model 9. Statistical Extrapolation Approach (Chen and Mueller 2019)
Model type Statistical Extrapolation Model 
Type of Migration International (cross-border from Bangladesh to India, Pakistan, Nepal, Sri Lanka, and 
Bhutan)
Geographic Coverage National (Bangladesh)
Data Inputs
Scenarios No. 
Evaluating Climate-Related Migration Forecasting Models 55
Environmental & 
Climate Impacts
Remote sensing measures of sub-district flooding and rainfall and in situ measures of 
monsoon onset, temperature, radiation, and soil salinity. With the exception of 
salinization, all environmental variables are reflected as anomalies over 1-, 2-, and 3-
year periods to characterize migration patterns under different durations of exposure.
Data on inundation are constructed from NASA’s Moderate Resolution Imaging 
Spectro-radiometer (MODIS) satellite. Images are aggregated into 8-day composites 
that provide the best possible observation during the period, and each pixel in an image 
captures an area of 500 m2
. Inundation is represented by the Modified Normalized 
Difference Water Index (MNDWI). 
Data on rainfall are drawn from NASA’s Tropical Rainfall Measuring Mission (TRMM), 
which generates precipitation values of 0.25 x 0.25° resolution. They focus on monthly 
precipitation values extracted from TRMM, aggregated up to annual measures. 
Annual averages of minimum and maximum temperature and bright sun exposure
from the Bangladesh Meteorological Department. 
To capture monsoon onset, they use daily rainfall data (500+ weather stations) from 
the Bangladesh Water Development Board to generate an explanatory variable for 
monsoon onset. 
Measures of soil salinity are based on field surveys conducted in 18 of the 64 districts 
of Bangladesh by the Soil Resource Development Institute, an agency of Bangladesh’s 
Ministry of Agriculture.
Population data Nationally representative migration data from 2005 to 2011 are collected through the 
Bangladesh Bureau of Statistics vital registration records (SVRS). Sampling is stratified 
at the locality level, to achieve representation across rural, urban, and metropolitan 
areas. Approximately 200,000 households (1 million individuals) are surveyed each 
year. 
Non-climate drivers
Intervening 
Obstacles/Facilitators
Personal and HH 
characteristics
Demographic and wealth variables: age, age-squared, literacy, and religion of the 
household head; the number of household members in eight age-sex categories 
(number of male/female household members 0–5, 6–16, 17–54, and greater than 54 
years old); indicators for whether the household is in the coastal zone or the drought￾prone areas of the Northwest; indicators for whether the household has improved 
water and latrine facilities (primary/secondary water source comes from tap/well, has 
own water source, has modern or sanitary latrine); and sources of energy (has 
kerosene/electricity as a source of light/fuel, has gas as a source of fuel) (from SVRS 
data)
Migration information is reported only for individuals who have been away for at least 
6 months. They focus on two migration-dependent variables, the number of migrants 
in the household going to South Asia and the number of migrants in the household 
going to India. 
Report | December 2022 56
Model Description The model estimates the effect of environmental factors on the flow of migrants using 
a negative binomial specification. The forecasting dimension of this model uses these 
observed relationships to predict the change in the number of migrants from 
Bangladesh to India (alone) and South Asian countries (together) for a 1 standard 
deviation increase in flooding and a 1 standard deviation increase in soil salinity. 
They also compare households grouped by religion, household gender composition, 
age groups and assets. They examine the relationship between these characteristics 
and environmental-migration dynamics, but do not use those estimates to project 
numerical estimates based on household characteristics in this paper. 
Outputs Cross-border migration (over 6 months duration) by district. 
Model predicts a total of 17,874 more migrants moving to India in response to a 1 
standard deviation increase in soil salinity and a total of 5,754 fewer migrants moving 
to India in response to a one- standard deviation increase in flooding. 
Scalability/Adaptability The model can be adapted to other regions provided the same in-situ measures are 
available as data inputs. 
Limitations The authors are transparent about the limitations in the paper: 
“First, we lack confirmation of the duration of each event and, therefore, are unable to 
validate whether the moves are temporary or permanent. Second, the absence of 
spatial and temporal variation represented by our measure of soil salinity affects our 
ability to measure other important aspects of climate migration. By limiting the focus 
to changes in soil salinity over 5 years, for example, we are unable to express how 
mobility may be affected by seasonal or annual variations in soil salinity. Furthermore, 
the coarseness of our measure of salinity exposure affects our capacity to inform how 
policymakers should prioritize funding for adaptive investments. Soil salinity is likely a 
direct result of changes in landscape and deforestation along the coast, sea level rise 
and storm surges, and groundwater management. It remains an open question which 
contributing factor to soil salinity is driving the mobility patterns observed in the paper. 
Future research would benefit from including more refined measures of soil salinity” 
(118-119).