MID-TERM PERFORMANCE 
EVALUATION OF THE 
REGIONAL AGRICULTURE
DEVELOPMENT PROGRAM IN 
AFGHANISTAN
Estimating Percentage Change in Land Area 
Under Licit Agriculture in Program Areas
February 2019
MID-TERM PERFORMANCE EVALUATION OF 
THE REGIONAL AGRICULTURE 
DEVELOPMENT PROGRAM IN AFGHANISTAN
Estimating Percentage Change in Land Area Under Licit Agriculture in 
Program Areas
Task Order Contract No.: AID-OAA-TO-16-00008
Submitted to:
USAID/Feed the Future 
Contact: Evans Lartey
Email: elartey@usaid.gov
Prepared by:
Russ Jones, GIS/Remote Sensing Analysis Lead, Abt Associates 
Tulika Narayan, Project Quality Advisor, Abt Associates
Abigail Conrad, Project Director/Qualitative Analysis Lead, Abt Associates
Contractor:
Program Evaluation for Effectiveness and Learning (PEEL)
ME&A, Inc.
1020 19th Street NW, Suite 875
Washington, DC 20036
Tel: 240-762-6296 
E-mail: MEandAHQ@engl.com
DISCLAIMER
The authors’ views expressed in this publication do not necessarily reflect the views of the United States 
Agency for International Development or the United States Government.
CONTENTS
ABSTRACT.............................................................................................................................. i
1. Introduction....................................................................................................................... 1
2. Background ........................................................................................................................ 3
RADP Background ........................................................................................................................... 3
Literature Review............................................................................................................................. 4
2.2.1 Previous Research................................................................................................................. 5
3. Methods .............................................................................................................................. 9
General Approach and Domain .................................................................................................... 9
Satellite Imagery............................................................................................................................. 10
NDVI Index and Season............................................................................................................... 12
Data and Processing...................................................................................................................... 15
4. Results and Discussion ..................................................................................................19
Tables of Results and Discussion............................................................................................... 19
4.1.1 RADP-South Region .......................................................................................................... 19
4.1.2 RADP-North Region ......................................................................................................... 19
4.1.3 RADP-East Region.............................................................................................................. 22
4.1.4 RADP-West Region........................................................................................................... 23
Discussion ....................................................................................................................................... 24
References ............................................................................................................................34
Annex ................................................................................................................................ 35
Annex 1: Landsat Scenes Used in Analysis...................................................................................... 36
LIST OF TABLES
Table 1. Evaluation Questions........................................................................................................................................ 1
Table 2. RADP Characteristics....................................................................................................................................... 3
Table 3. RADP Results Framework............................................................................................................................... 4
Table 4. Pre-Implementation and Evaluation Periods for Each RADP Region.................................................. 10
Table 5. Crop Calendar Used to Guide Imagery Dates Needed to Differentiate between Poppy and 
Non-Poppy Crops (UNODC, 2017) .......................................................................................................................... 12
Table 6. NDVI Value Ranges Used for Classification.............................................................................................. 13
Table 7. Flowering and Harvest Season Criteria Used to Determine Final Land Cover Classification...... 16
Table 8. Change in Area under Poppy and Other Uses across the RADP-South Region ............................. 19
Table 9. Change in Area under Poppy and Other Uses across the RADP-North Region ............................ 20
Table 10. Change in Area under Poppy and Other Uses across the RADP-East Region .............................. 23
Table 11. Change in Area under Poppy and Other Uses across the RADP-West Region ........................... 23
Table 12. RADP-West Region, District-Level Comparison between UNODC 2017 Report and Pre￾Program Period................................................................................................................................................................ 27
Table 13. RADP-South Region, District-Level Comparison between UNODC 2017 Report and Pre￾Program Period................................................................................................................................................................ 28
Table 14. RADP-North Region, District-Level Comparison between UNODC 2017 Report and Pre￾Program Period................................................................................................................................................................ 29
Table 15. RADP-East Region, District-Level Comparison between UNODC 2017 Report and Pre￾Program Period................................................................................................................................................................ 30
Table 16. Province-Level Summaries of Poppy Cultivation, Showing Upper and Lower 95% Confidence 
Ranges................................................................................................................................................................................. 32
LIST OF FIGURES
Figure 1. Geographic Extent of the Four RADP Programmatic Regions and Districts Examined................. 9
Figure 2. Comparison of Resolution between High- and Moderate-Resolution Imagery.............................. 11
Figure 3. NDVI Profiles for Three Years Representing Mixed-Pixel Values from Different Combinations 
of Crop Types within a 250-Meter Pixel from the MODIS Satellite................................................................... 14
Figure 4. Spatial Extent of Landsat 8 Scenes Needed across the RADP Regions............................................ 16
Figure 5. Portion of the RADP-South Region Showing Landsat 8 Imagery Used in the Classification 
Process............................................................................................................................................................................... 17
Figure 6. Example of Final Classification Output for a Partial Scene in the RADP-South Region................ 18
Figure 7. Desert Area in the RADP-North Region Showing Expansion of Poppy Crop ............................... 20
Figure 8. Images of the RADP-North Region for the Evaluation Period (2018) Showing Expansion of 
Poppy Production into the Desert.............................................................................................................................. 21
Figure 9. Images of Desert Region of Poppy Expansion in the RADP-North Region to Highlight Potential 
Agricultural Fields............................................................................................................................................................ 22
Figure 10. Images of the RADP-West Region for the Evaluation Period (2018) Showing Expansion of 
Poppy Production into the Desert.............................................................................................................................. 24
ACRONYMS
Acronym Description
DAI Development Alternatives, Inc.
DAILs Departments of Agriculture, Irrigation, and Livestock
DO Development Objective
EQ Evaluation Question
ET Evaluation Team
GIS Geographic Information Systems
GPS Global Positioning System
HVC High-Value Crop
IR Intermediate Result
MCN Ministry of Counter Narcotics
MODIS Moderate Resolution Imaging Spectroradiometer
MRI Moderate Resolution Imagery
NDVI Normalized Difference Vegetation Index
NIR Near-Infrared Spectral Band
OBIA Object-Based Image Analysis
OLI Operational Land Imager
RADP Regional Agricultural Development Program
RADP-E RADP-East
RADP-N RADP-North
RADP-S RADP-South
RADP-W RADP-West
TIRS Thermal Infrared Sensor
TM Thematic Mapper
UNODC United Nations Office on Drugs and Crime
U.S. United States
USAID United States Agency for International Development
USGS United States Geological Survey
UTM Universal Transverse Mercator
VHR Very High-Resolution
i 
ABSTRACT 
For the mid-term performance evaluation of the United States Agency for International Development
Afghanistan’s Regional Agriculture Development Program (RADP) the Evaluation Team (ET) 
independently assessed the program’s influence on reducing land under poppy cultivation. The RADP 
aims to improve food and economic security of rural Afghans through four projects: RADP-North, 
RADP-West, RADP-South, and RADP-East. The ET used remote sensing to differentiate between illicit 
and licit agriculture and quantified the change in area under licit agriculture between pre-program 
implementation and the evaluation period. The ET used geographic information systems (GIS) data to 
identify programmatic districts and moderate resolution imagery (MRI) from the Landsat 8 Operational 
Land Imager and the Thermal Infrared Sensor satellite for comprehensive coverage. To differentiate land 
cover types, the ET used the Normalized Difference Vegetation Index (NDVI) and seasonal crop 
management differences. Results indicate that over the evaluated period, the area under poppy 
cultivation in RADP-South increased by 20 percent. As a percentage of total agriculture, the area
decreased from 7.7 percent in 2013 to 7.4 percent in 2017. In RADP-North, the ET calculated a 440 
percent increase in poppy production between the two periods—an increase of 2.1 percent in 2013 to 
13 percent in 2017. Water availability contributed to increased poppy cultivation, although RADP-North
showed evidence of poppy expanding to desert areas. In both RADP-East and RADP-West the area 
under poppy cultivation decreased by 78 percent and 57 percent, respectively. In both regions, the 
decline was attributed to environmental conditions more favorable to licit agriculture.
1 
1.0 INTRODUCTION
The Evaluation Team (ET) conducted a performance evaluation to assess the overall progress and 
accomplishments of the Regional Agricultural Development Program (RADP) to date in four regions of 
Afghanistan that aimed to improve the food and economic security of rural Afghans. This performance 
evaluation comes during implementation for the RADP-North and RADP-East regions and at the end of 
activities for the RADP-West and RADP-South regions. The ET used Earth observation data and GIS 
analysis to estimate the percentage change in land area under poppy and non-poppy crops in the target 
areas to answer evaluation question (EQ) 3d (see Table 1). The results of the GIS analysis are presented 
in this document. 
All other EQs are answered using a qualitative assessment (see Table 1; Conrad et al., 2018). The 
qualitative assessment identifies achievements, performance issues, and constraints related to 
implementation and effectiveness of the program. It also identifies results and lessons learned from 
implementation and provides recommendations to scale up or modify activities. The findings for the 
qualitative portion are presented in Conrad et al. (2018). 
Table 1. Evaluation Questions
Evaluation Questions
A. Implementation
1. Is the agribusiness support model likely to be sustainable?
a. Are we supporting businesses that largely have good potential for growth?
b. Are beneficiary enterprises flexible or adaptable enough to be responsive to market shifts?
2. Is the training model design and implementation effective in contributing to beneficiary adoption of
improved agricultural technologies and licit crops?
a. How did selection criteria, curricula, and pedagogy address farmer constraints?
b. To what level are the communities being reached (saturation)?
c. To what extent does the approach include considerations for varying socioeconomic status
among beneficiaries (e.g., security of land tenure, income)?
B. Results
3. Based on the implementation approach, status of implementation, and milestones achieved at the time of
mid-term evaluation, is there evidence that the program is likely to achieve its results by the end date?
Agribusiness model results:
a. Are activities creating more market linkages? Are those market linkages more efficient in terms of
maximizing returns to the smallholders while meeting the market needs?
b. Are beneficiary enterprises more responsive to market demand?
Training model results:
c. To what degree did beneficiary farmers adopt improved agricultural technologies?
d. To what degree did beneficiary farmers adopt licit crops? What was the percentage change in land
area under poppy and non-poppy crops in the target areas?
C. Cross-Cutting
4. To what extent have women and youth been integrated into the implementation of RADP, and to what
extent do women and youth respectively contribute to program goals?
a. Which approaches have been most effective in specifically reaching women and youth?
b. How have women and youth respectively benefited from RADP implementation, and how are
these benefits measured?
2 
c. Are there any missed opportunities with specifically incorporating women and youth into RADP
activities?
5. Are there any successes from RADP that can be adapted to specifically integrate women and youth?
D. Thematic
6. Is the regional approach to the RADP programming likely the most effective for achieving program goals?
For example, how might a regional approach compare to a commodity-specific approach?
7. To what extent will institutional and informational structures be sustainable after project
implementation?
The next section provides a brief review of previous work used to identify and quantify illicit production 
of poppy crop in Afghanistan using remote sensing techniques and imagery, followed by sections on the 
methods used, the results of the ET analysis, and finally a discussion and comparison to other work. 
3 
2.0 BACKGROUND
RADP Background
The United States Agency for International Development (USAID) funds RADP, the umbrella program in 
Afghanistan, with a budget of just over $241.7 million and operations in 4 regions across 21 provinces
and 116 districts. The program is run as four separate regional projects: RADP-North, RADP-West, 
RADP-South, and RADP-East (implementation areas shown in Table 2 below).
Table 2. RADP Characteristics
Area RADP-North RADP-West RADP-South RADP-East
Award number 306-C-14-00002 306-C-14-00007 306-C-13-00018 306-C-16-0001
Award date May 15, 2014 August 10, 2014 October 2, 2013 July 2016
Activity end 
date
Anticipated May 2019 September 27, 2016a October 2017* July 2021
Funding amount $78,429,714 $27,658,205 $107,621,129 $28,000,000
Activity funding AG 75%, AD 25%
Implementing 
organization
Development 
Alternatives, Inc. (DAI)
Chemonics 
International, Inc.
Chemonics 
International, Inc.
DAI
Subcontractors/ 
grantees
ACDI/VOCA, Alcis, 
Development and 
Training Services, Inc., 
Dutch Committee for 
Afghanistan, Joint 
Development 
Associates, Pax Mondial 
Risk Management 
Company, 
, Afghan 
Public Protection Force
Agency for 
Rehabilitation and 
Energy Conservation 
in Afghanistan, 
Coordination of 
Humanitarian 
Assistance, Dutch 
Committee for 
Afghanistan, Equal 
Access, Services 
International LLC, 
Afghan Public 
Protection Force
Afghan 
Development 
Association, Dutch 
Committee for 
Afghanistan, Equal 
Access
Organization for 
Sustainable 
Development 
and Research, 
Relief 
International, 
Crimson Capital 
and Mondial Risk 
Management 
Company
Province 
coverage
Badakhshan, Baghlan, 
Balkh, Jawzjan, Kunduz, 
and Samangan
Herat, Farah, and 
Badghis
Kandahar, 
Helmand, Uruzgan, 
and Zabul
Kabul, 
Nangarhar, 
Laghman, 
Ghazni, Logar, 
Wardak, 
Parwan, and 
Kapisa
Targeted value 
chains
Wheat, livestock, high-value crops Livestock, high￾value crops a. Terminated for convenience.
Working in alignment with Government of Afghanistan goals and the Ministry of Agriculture, Irrigation, 
and Livestock as well as the Departments of Agriculture, Irrigation, and Livestock (DAILs) at the 
subnational level, RADP aims to improve the food and economic security of rural Afghans. The 
program’s development hypothesis is that if all value-chain actors have access to appropriate training, 
information, inputs, and technology within a more competitive, efficient, and profitable agricultural value 
chain as well as a conducive policy and regulatory environment, they will increase their incomes and 
food security.
Each RADP activity employs a value-chain facilitation methodology that works in all segments of 
agriculture supply chains from the farmers (i.e., producers) to the consumers, addressing bottlenecks in 
4 
each targeted value chain. Specifically, activities focus on improving productivity of wheat, investment in 
and production of high-value horticultural crops (e.g., grapes, raisins, pomegranates, stone fruits, nuts, 
vegetables), and production of livestock (see Table 3 below). Increasing women’s participation in and 
benefit from the agricultural sector is a critical component of each activity. To develop targeted value 
chains, RADP activities are strengthening relationships and trust among private sector entities. Each 
RADP also focuses on critical cross-cutting themes such as on-farm water management, nutrition, 
alternative development, finance, and support for DAILs. Another cross-cutting theme is to strengthen 
the enabling environment to support strengthening the private sector by promoting an improved legal 
and regulatory framework, and by using policy advocacy and dialogue to strengthen the private sector 
via a mix of direct implementation and a $10-million grant fund for each RADP. These enabling 
environment activities were led by the RADP-South region until its termination, and after an interlude, 
was taken up by the RADP-North region. Table 2 shows how intended (cross-cutting) outcomes of the
program will lead to an Intermediate Result (IR), Sub-Intermediate Results (Sub-IRs), and finally to the 
Development Objective (DO) to expand sustainable, agriculture-led economic growth.
Table 3. RADP Results Framework
DO 1: Sustainable, Agriculture-Led Economic Growth Expanded
IR 1.2: Vibrant and prosperous agriculture sector developed
Sub-IR 1.2.1: Productivity of key agriculture crops increased Sub-IR 1.2.2: Commercial viability of 
agribusinesses increased
Outcome 1: Increased agriculture 
sector productivity and profitability in 
target region
Outcome 4: Increased and 
sustained adoption of licit 
crops
Outcome 2: Increased profitability of 
small, medium, and large agribusinesses
Cross-Cutting Outcome 3: Increased women’s participation in agriculture value-chain activities
Cross-Cutting Outcome 5: Improved enabling environment for farmers and agribusiness 
While each regional project uses similar approaches—blending direct technical assistance and training 
with strengthening value-chain linkages—variation in climate, capacity, security, and shifting government 
priorities have led to minor differences in activity implementation. The RADP-South, RADP-West, and 
RADP-North regions provide technical assistance to wheat producers and value-chain actors, whereas 
the RADP-East region does not focus any of its activities on the wheat sector. All four RADPs support 
producers and value-chain actors of high-value crops (HVCs), but specific crops vary widely by regional 
climate. Likewise, all RADPs support livestock value chains, but the challenges for different species vary 
widely by region. RADP-South put relatively more resources toward supporting veterinary units than 
the other regions. Although not explicit in the design phase of all RADPs, providing opportunities for 
licit farming and employment became a government and donor priority over the course of 
implementation, particularly for the RADP-South region, which was active in areas traditionally known 
to be the most active in poppy production. The RADPs aimed to provide opportunities to encourage 
farmers to shift from illicit crops to licit crops by improving farmers’ abilities to produce HVCs and 
livestock products and by facilitating market linkages to increase their product sales.
2.1 LITERATURE REVIEW
In this analysis, the ET used a combination of remote sensing analysis with GIS to differentiate between 
illicit and licit agriculture and quantify the change in illicit agriculture between pre-program 
implementation and the evaluation period for participating districts in the four regions. Previous efforts 
to use remotely sensed imagery to quantify the production of poppy and non-poppy agriculture are 
discussed below, with a primary focus on methods that use moderate resolution imagery for 
5 
comprehensive coverage across Afghanistan rather than a sample-based approach using very high￾resolution imagery in limited areas and extrapolation to the wider region.
2.1.1 Previous Research
The work by Sader (1990) represents one of the first detailed assessments of the ability to detect and 
quantify poppy crops using satellite imagery. Sader noted that detailed mapping had never been 
attempted in any country, although two pilot projects using satellite imagery for monitoring poppy had 
been conducted—one in Thailand and one in India. However, Sader’s work laid the groundwork in 
detailing the data and methods needed to support such an analysis. This included the need for “good 
quality” multispectral satellite data [such as SPOT and the Landsat Thematic Mapper (TM)], a regional
crop calendar (detailing the flowering and post-harvest periods and taking into account regional and 
environmental differences), ground-truthed samples roughly coinciding temporally with satellite imagery 
acquisition dates, and fields of sufficiently large size and with minimal intercropping to avoid having the 
pixel straddle different fields or land cover types (resulting in a “mixed pixel” spectral signal). 
In terms of the crop calendar, Sader (1990) noted that poppy in the Southern region of Afghanistan is
planted in the fall, flowers in March or April—with the flower phase lasting two to three weeks—and 
harvested between mid-April to early May. He also noted that the crop calendar varies strongly with 
altitude and region.
Sader (1990) suggested a combination of different classification techniques: a supervised classification 
using field samples from multiple types of crops, co-located with the imagery as training samples to 
identify pixels with similar spectral signals; and an unsupervised classification separating agricultural lands 
from other land cover types. Sader also suggested that signatures needed to be developed independently 
for each satellite image (i.e., scene) due to variation in vegetation conditions and imagery properties. 
Building on the work of Sader (1990), Simms (2016) evaluated different alternative methods used by the 
United Nations Office on Drugs and Crime (UNODC) and the United States (U.S.) Government to 
monitor poppy production in Afghanistan. His extensive research included descriptions of sampling 
techniques using very high-resolution (VHR) imagery extrapolated to the wider region as well as a 
comprehensive classification across Afghanistan using low-resolution imagery.
Simms (2016) examined the spectral profiles of reflectance values from homogenous (“pure”) samples of 
poppy, alfalfa, and wheat, noting that there is good separation between the crops within certain 
wavelength ranges (“bands”), which could be used to differentiate between the crops from remotely 
sensed imagery. However, he described various factors that can cause deviation from this ideal, including 
pixel resolution, crop spacing, temporal factors such as image acquisition date and frequency of imaging, 
radiometric resolution, and atmospheric conditions and corrections of the imagery.
Simms (2016) described a number of previous attempts to quantify poppy from satellite imagery, 
including different resolutions. Ultimately, he summarized three key factors that are needed for a robust 
analysis: the timing and revisit1 time of imagery, the resolution of the imagery, and the use of supporting 
ancillary data.
Simms (2016) evaluated the potential of using lower-resolution imagery for country-wide assessments of 
poppy crops. His evaluation used Moderate Resolution Imaging Spectroradiometer (MODIS) 250-meter 
resolution imagery for classification of poppy using Normalized Difference Vegetation Index (NDVI) 
values. NDVI values are an indication of the amount of chlorophyll content in the vegetation and 
therefore can be used to indicate the relative degree of plant vigor.
1 The frequency that imagery for the same location is captured.
6 
To better understand the geographical relationship between NDVI values and crop cycles, Simms (2016) 
collected hundreds of images across the country, ranging from VHR to lower-resolution imagery. From 
this effort, he obtained spectral profiles from irrigated and non-irrigated areas through time to identify 
distinct temporal cropping patterns and associated NDVI values. In Helmand Province, cropping patterns 
indicated two crop cycles for cereal crops (e.g., wheat, barley, oats), while poppy is only green in the 
first cycle, as it is not suited for late summer heat. Additionally, he found NDVI values before February 
were very low. By March, poppy and wheat both have stem elongation and therefore higher NDVI 
values. By early April, poppy and cereal crops both start flowering. Peak NDVI values were found to 
correspond to around the end of poppy flowering. However, Simms noted that the flowering itself was 
not observed in the MODIS 250-meter spectral values due to the coarse pixel size (i.e., the flowering 
signal represented a small portion of the reflectance value). After opium harvest, poppy plants quickly 
dry and die back and by early June, poppy locations had very low NDVI values, while there were higher 
values for the other crops.
However, Simms (2016) also noted that there is wide variation in NDVI values of agriculture due to 
differences in latitude and elevation that control the timing of crop cycles, including a two-month 
difference in first-peak NDVI between adjacent valleys in Badakhshan Province, due to differences in 
elevation. Lastly, Simms also noted the sizeable NDVI variations annually (especially in the Northern 
region) in dryland crop production due to periods of drought, disease, or cooler spring conditions, 
which affect factors such as the date of green-up and senescence, as well as the establishment of plants. 
This latter thereby lowers the NDVI value within a pixel due to mixes of bare ground and canopy cover. 
Conversely, he found higher NDVI values in some of the higher valleys with good water supply.
Because of the importance of the date of the imagery and the difficulty in finding comprehensive cloud￾free imagery within a specific time window, Simms (2016) also examined the potential of using 16-day 
composited imagery from MODIS—with the pixel value representing the highest NDVI value from the 
image stack. However, he found that the composite imagery masked some of the variation that is 
important (e.g., changes in plant phenology), especially given that the flowering period of poppy is only 
two weeks. Simms therefore decided against this approach.
Lastly, because of the coarse resolution (250 meters) of MODIS imagery and the fact that field sizes are 
commonly less than one hectare in size, Simms (2016) evaluated the impact of different mixes of crops 
within a single pixel on resulting NDVI values. To do so, Simms generated annual spectral profiles of 
NDVI values derived from MODIS data for three years and highlighted the variation in NDVI values 
across those years. His findings showed that different mixes of poppy, non-poppy agriculture, and barren 
lands effect both the timing and magnitude of the pixel NDVI values.
Curran et al. (2013) examined the potential of using object-based image analysis (OBIA) techniques to 
identify different crop types from VHR multispectral imagery. Their study utilized Quickbird high￾resolution (2.8 meters) multispectral imagery along lands adjacent to the Helmand River within Helmand 
Province in mid-April 2008 (noting that this was the best period to differentiate poppy from other 
crops). Realizing the difficulty in identifying crops by the spectral signature alone, they included additional 
characteristics of poppy crops to improve their outcome. For example, they noted that the poppy 
canopy is non-continuous with gaps between plants exposing soil, while other crops such as wheat have 
a more continuous canopy that results from broadcast seeding. While these textural differences are too 
fine to pick up through moderate resolution imagery such as MODIS (250 meters) or Landsat 
(30 meters), they are identifiable using high-resolution imagery. The Curran et al. team noted that for 
high-resolution imagery, OBIA was shown to produce more accurate classification than traditional pixel￾based classification approaches.
Curran et al. (2013) used an OBIA approach to feature extraction. The analysis first segments the image 
into objects, and then the software calculates various attributes (spectral, textural, etc.) for each object. 
The feature extraction tool allows image classification using either a supervised or rule-based approach. 
7 
In the supervised approach, the user selects individual objects to guide the software for classification 
parameters, while the rule-based extraction allows the user to manually select classification parameters.
These parameters include band minimum/maximum/average, area, compactness, roundness, elongation, 
convexity, rectangular fit, texture mean/range/variance/entropy, hue, intensity, and NDVI.
For the supervised classification, Curran et al. (2013) used 110 samples for poppy, 120 samples for 
wheat, and 55 samples for other agriculture. However, they noted that even in this small study area they 
did not have enough ground reference samples to be statistically valid for an accuracy assessment. They 
used crop type maps from aerial imagery and discontinuous ground reference information for their 
ground-truthed information, which they used for both training and accuracy assessments.
Curran et al. (2013) noted a difference of about 13 percent of the amount of land classified as poppy 
between the two approaches. Their accuracy assessment showed there was low accuracy using either 
the traditional supervised (50 percent overall accuracy) or rule-based (47 percent overall accuracy) 
classification techniques without OBIA. The overall accuracy improved by incorporating OBIA into 
either approach: 76 percent for the OBIA supervised approach and 73 percent for the OBIA rule-based 
approach. However, the kappa coefficients (a statistical test to show agreement between reference and 
classified data) for OBIA are low to moderate—therefore the confidence in the accuracy for either 
method is low. It is also important to note that Curran et al. (2013) conducted their analysis for a single 
time period rather than taking seasonal approaches used by others.
Finally, the Ministry of Counter Narcotics (MCN) and the UNODC conduct annual surveys of the 
location and extent of poppy production across Afghanistan. The UNODC 2017 report shows the 
annual amount of poppy production from 1994 to 2017. They have used remote sensing in their 
estimates since 2002, though the methods have changed over the years. Since 2015, they have used a 
combination of approaches based on the location of likely poppy production by area. In provinces where 
the most poppy is found, they use a sampling approach; in provinces with a low level of poppy 
production, they use a targeted approach using full coverage of satellite imagery for portions of the 
province. In provinces with no indication of poppy cultivation, they use a village survey approach. For 
2017, out of the 34 provinces, 15 were sampled, 13 were targeted, and 6 were considered poppy-free. 
They note that their criteria for being poppy-free is any area with less than 100 hectares of opium poppy 
cultivation. In all locations where the sampling approach was used in both 2016 and 2017, the same 
sample locations were utilized to maintain consistency. While some provinces had increased poppy 
production, the methodology change (from a targeted approach in 2016 to a sampling approach in 2017) 
could have influenced the results.
The UNODC 2017 sampling regime was changed after 2015 from using 10-kilometer by 10-kilometer
grids to 5-kilometer by 5-kilometer grids. The final sampling frame used was composed of 6,498 cells 
from which images were randomly selected. The domain of the sample area consisted of both irrigated 
and non-irrigated lands and represented approximately 38 percent of the total potential agricultural 
lands within the 15 provinces that used the sample approach. In addition, they excluded any lands less 
than 0.25 square kilometersfrom the sampling domain to reduce the chance of selecting sampling cells 
with low amounts of arable lands. The sample size was determined by the number of available images, 
the number of provinces, and the agricultural area of the province. In addition, they set the sample size 
to no fewer than 20 samples per province. In total, the 2017 report relied on VHR imagery for 
673 5-kilometer by 5-kilometer grids over the 15 provinces.
The UNODC 2017 report relied on satellite imagery following a crop calendar aligned to the flowering 
phase and the post-harvest phase, noting that poppy is plowed immediately after harvest whereas wheat 
fields are not. As such, they utilized imagery in the RADP-South, -East, and -West regions from March 
or April; and used imagery in May, June, or July for the RADP-North region. They then used imagery 
approximately two months later for the second date imagery. They used visual interpretation to 
delineate opium poppy fields from 0.5-meter resolution (pan-sharpened) PLEIADES multispectral 
8 
satellite images covering a 5-kilometer by 5-kilometer area. They combined that with ground-truth 
information from maps, global positioning system (GPS) points, ground photographs tagged with latitude 
and longitude coordinates, and visual interpretation of the second-dated images to differentiate poppy 
from other crops. The visual interpretation was based on a combination of natural color and false-color 
infrared. They also had 1,854 GPS points for poppy fields with pictures. UNODC researchers also noted 
that they had some difficulty getting images for the dates needed given cloud cover.
Finally, the UNODC 2017 report used a statistical relationship based on the ratio of poppy to potential 
agricultural land—taking into account different sampling regimes for the different provinces. In their 
quantification, they report at the 95 percent confidence level. An examination of the results shows a 
wide range (up to 6.7 times) between the low and high estimates of poppy area (at the province level). 
However, overall the average of the 15 sampled provinces had an average of 1.7 times the difference 
and only 1.2 times the national level (which included the target provinces). As with other researchers, 
UNODC was unable to conduct an accuracy assessment due to security conditions that prevented the 
collection of the number of reference samples needed for a valid assessment.
In the targeted provinces (areas with little poppy production), UNODC relied on a surveillance system 
and information from village “headmen” within the provinces to identify poppy production locations. 
Field crews also visited potential poppy sites and collected GPS coordinates for those confirmed 
locations, which were used for reference samples. Imagery was collected for areas of concentrated 
poppy production. They used the poppy classified area in the imagery for the total estimate of poppy 
cultivation. UNODC considered these estimates minimums as some areas would undoubtedly be 
missed.
9 
3.0 METHODS 
3.1 GENERAL APPROACH AND DOMAIN
As noted above, other researchers such as UNODC have used a statistically based approach to estimate 
the area under poppy cultivation drawn from a set of sample locations within the study area domain. In 
their analysis, they utilize a select set of high-resolution images aligned with field-referenced data for 
their classification and then extrapolate the results to the broader region. In contrast, in this analysis the 
ET used a non-statistically based approach to classify moderate-resolution imagery covering the entire 
study domain. The ET’s basis for this approach is described in more detail below.
While the imagery available covers the full extent of the RADP regions, the ET limited their estimation 
of poppy cultivation to a subset of districts within the four regions that participated in the program. This 
represents 21 provinces out of a total of 34 provinces in Afghanistan. Figure 1 shows the districts the ET
examined as well as the subset of districts within those provinces.
Figure 1. Geographic Extent of the Four RADP Programmatic Regions and Districts 
Examined
Boundary data source: ESRI, 2018a.
10 
To assess the potential impact of the program on poppy production, the ET classified the landscape into 
poppy, non-poppy agriculture, and other land use types for each of the RADP programmatic areas. The 
pre- and interim evaluation dates for each region are listed in Table 4. 
Table 4. Pre-Implementation and Evaluation Periods for Each RADP Region
Region Pre-Implementation Period Evaluation Period
North 2013 2017 
South 2013 2017 
West 2014 2017 
East 2015 2018 
3.2 SATELLITE IMAGERY
As noted above, the ET used moderate resolution imagery in this analysis. As such, they utilized imagery 
from the Landsat 8 Operational Land Imager (OLI) and Thermal Infrared Sensor (TIRS) satellite.
2 The 
Landsat imagery is available cost-free from the United States Geological Survey (USGS) 
(https://landsatlook.usgs.gov) and covers all of Afghanistan from February 2013 to the present. Each 
scene (i.e., individual image) covers a swath of approximately 17 kilometers north-south by 
183 kilometers east-west and contains 11 bands: eight spectral bands (bands 1 to 7 and 9) at 30-meter 
resolution (30 by 30-meter pixel), one panchromatic (band 8) at 15-meter resolution, and two thermal 
infrared bands at 100-meter resolution. Additionally, each scene was radiometrically calibrated and 
corrected for terrain displacement (i.e., orthorectified) and therefore can be used directly for surface 
reflectance analysis.
In contrast to Landsat data, high-resolution imagery is available from Digital Globe extending back to 
1999. Digital Globe provides access to multi-spectral imagery across Afghanistan at resolutions from 
sub-meter to approximately 2-meter resolution from their archives. Depending on the specific satellite, 
one panchromatic and four to eight bands of multispectral imagery are collected covering the visible and 
near-infrared portion of the electromagnetic spectrum, which is consistent with the bands that are 
needed for land cover differentiation. Each scene covers a width ranging from 13 to 18 kilometers and 
the length is based on the project’s needs. The level of processing needed for analysis (atmospheric and 
terrain correction) is variable.
While both sets of data offer imagery that is suitable for the land cover analysis, there are pros and cons 
to each set of imagery. The main advantage of the Landsat data is the combination of complete spatial 
coverage across Afghanistan, the availability of imagery extending back to 2013, a 16-day revisit time for 
each location, and the spectral bands needed to classify different land cover types. However, the 30-
meter resolution of the data means that an individual pixel may cross multiple land cover types 
(e.g., crossing adjacent fields). In these cases, the pixel value would represent the average of land cover 
types (e.g., a pixel crossing the boundary between a field of poppy and an adjacent fallow field) and is 
therefore harder to classify correctly. Thus, this “mixed pixel” value is harder to classify than a pixel that 
covers a single land cover type.
2 While data are available for Landsat 7 and Landsat 8 satellites, the Landsat 7 data were not suitable for this analysis due to a 
sensor failure that resulted in swaths of missing data throughout each image.
11 
The main advantage of using Digital Globe data is the availability of multispectral imagery at extremely 
high resolutions. The small cell size increases the likelihood that a single pixel will cover a single land 
cover type rather than spanning multiple cover types (Figure 2).
Figure 2. Comparison of Resolution between High- and Moderate-Resolution Imagery
Left: false-color infrared imagery from high-resolution imagery from Digital Globe (2-meter cell size). Right: false-color 
infrared moderate resolution imagery from Landsat 8 (30-meter cell size). Sources: GeoEye-1, 2017, Digital Globe, Inc., 
2018; Landsat 8 OLI, 2017, USGS, 2018. 
Additionally, the resolution is high enough that the imagery can identify textural patterns that can be 
used to differentiate between crop types (e.g., wheat fields are often more homogenous than poppy 
fields). However, the imagery available from Digital Globe is limited to the spatial and temporal 
requirements of the projects for which the data were collected. Therefore, the imagery the ET was able 
to identify was very spotty. While some of this limitation can be offset by the sample and extrapolation 
techniques noted earlier, a combination of factors made this approach not feasible, including: 
• The small number of ground reference samples were limited to the RADP-North region;
• The imagery coincident with the sample data in space and time was difficult to obtain or not
available;
• The number of samples available was not enough to be statistically significant;
• Easily obtainable imagery was limited to natural color bands and thus less suitable for
discrimination of crop types;
12 
• The amount of individual scenes needed for coverage across all four regions was beyond the
resources available; and
• The atmospheric calibration and topographic corrections needed were beyond the resources
available.
3.3 NDVI INDEX AND SEASON
Due to limitations in the availability of high-resolution imagery and existing reference samples, as well as 
the inability to collect new reference samples due to security concerns, the ET relied on the Landsat￾based approach noted above. In this approach, the ET relied upon a combination of the NDVI and 
seasonal differences in crop management to differentiate between land cover types—an approach used 
by other researchers (Sader, 1990; Curran et al., 2013; Simms, 2016; USAID, 2016; UNODC, 2017). 
This multi-temporal approach takes advantage of the difference between pre- and post-harvesting 
schedules to differentiate between poppy and non-poppy fields. Poppy fields are often plowed after 
harvest, while other crops such as wheat are not (Sader, 1990; UNODC, 2017). In addition, the ET used 
NDVI values derived from the literature to help differentiate between poppy and non-poppy crops. 
NDVI is a measure of the amount of chlorophyll content in vegetation. The NDVI value takes advantage 
of the difference in surface reflectance between the near-infrared and red spectral bands. Equation 1 
shows the NDVI calculation:
Equation 1. NDVI calculation
NDVI = (NIR – Red) / (NIR + Red)
Where:
NDVI = Normalized Difference Vegetation Index;
NIR = Near-infrared spectral band – Landsat 8, band 5 (0.851–0.879 µm); and
Red = Red spectral band – Landsat 8, band 4 (0.636–0.673 µm).
Fields with vigorous growth (growing season) will have a high NDVI, while less-vigorous or fallow fields 
(post-harvest) will have a low or zero NDVI (UNODC, 2017). The UNODC 2017 report also noted 
that it is key to utilize imagery coincident with the flowering stage to the maximum degree possible to
allow for better differentiation from other crops, although the flowering stage only lasts for a few weeks 
and getting cloud-free imagery that coincides with that period can be challenging.
To differentiate between poppy and non-poppy crops, the ET used NDVI values relative to the season—
flowering stage or post-harvest stage. The ET used crop calendars specific to each region that were 
based on the UNODC 2017 report, which are shown in Table 5. 
Table 5. Crop Calendar Used to Guide Imagery Dates Needed to Differentiate between 
Poppy and Non-Poppy Crops (UNODC, 2017)
Region Project Period
Season
Growing/Flowering Harvest
North Pre-Implementation May 2013 July 2013
Evaluation Period May 2017 July 2017
South Pre-Implementation March/April 2013 June 2013
Evaluation Period March/April 2017 June 2017
West Pre-Implementation March/April 2014 June 2014
Evaluation Period March/April 2017 June 2017
13 
Region Project Period
Season
Growing/Flowering Harvest
East Pre-Implementation March/April 2015 May/June 2015
Evaluation Period March/April 2018 May/June 2018
Many of the published NDVI values for poppy vs. other crops are derived from high-resolution imagery 
and thus represent a more homogenous signal than what would be expected from the moderate￾resolution Landsat imagery, which in some cases represents a mixed pixel of different fields or land use 
types (e.g., poppy, non-poppy crops, and barren). To address this issue, the ET used a range of NDVI 
values derived from Simms (2016) for both the flowering and post-harvest phase of poppy. Simms 
(2016) generated spectral profiles of NDVI values from 250-meter MODIS satellite data for different 
mixes of poppy and non-poppy land cover types for three years and plotted the mixed-pixel values 
against time (Figure 3).
Figure 3 shows peak NDVI values in 2006 (which was a mix of crop types but had the greatest amount 
of poppy of the three years reported): about 0.55 in late March and a slightly higher value (about 0.6) in 
2005, which had a higher mix of wheat. This is similar to values in the UNODC 2017 study that showed 
an approximately 0.6 value for poppy and a 0.7 value for wheat in mid-March. Again, the UNODC values 
were derived from high-resolution imagery. However, during the flowering phase of poppy (early/mid￾April), Figure 3 shows NDVI values of about 0.4–0.5 in Helmand Province. The ET compared these 
values to the NDVI values and those in the Landsat data for a scene covering an agricultural area in the 
Southern region and found the maximum NDVI value of about 0.49. Taking into account pixels that 
would represent different mixes of poppy and non-poppy crops, the ET chose an NDVI range of 0.30–
0.45 for the flowering season—the lower value capturing pixels with a fairly vigorous vegetative growth 
and the higher value not exceeding what one might expect for a pure poppy field during the flowering 
phase (e.g., a value > 0.45 might indicate a majority of wheat). For the harvest period, the ET used a 
range of NDVI values from 0 to 0.135 for captured fields where the majority of land cover was either 
fallow (plowed) or had scarce vegetation. Table 6 shows the NDVI values used in the ET’s classification 
scheme for different land cover types, noting that it is the difference in NDVI bins between seasons (and 
therefore harvest practices) that differentiates poppy from either other agriculture or other land cover 
types. Therefore, to be classified as poppy, the NDVI value for a pixel must fall between 0.3 and 0.45 in 
the flowering season and must be between 0 and 0.135 in the harvest season as poppy is plowed
following harvest while other crops such as wheat are not.
Table 6. NDVI Value Ranges Used for Classification
Land Cover Type
Flowering Season
NDVI Range (Bin Value)
Harvest Season
NDVI Range (Bin Value)
Other agricultural > 0.45 (1) > 0.135 (2)
Poppy 0.3–0.45 (1)
Other land (non 0–0.135 (1) -agricultural 
vegetation, barren, urban, etc.)
0.0 – < 0.3 (0)
Water/snow/ice < 0.0 (-1) < 0.0 (-1)
14 
Figure 3. NDVI Profiles for Three Years Representing Mixed-Pixel Values from Different 
Combinations of Crop Types within a 250-Meter Pixel from the MODIS Satellite
Source: Simms, 2016, Figure 3.5, p. 30. (Note: the top-right Ikonos image was originally incorrectly labeled as 
“IKONOS 25 April 2007, b,” but was corrected to read as “IKONOS 25 April 2007, c.”
15 
Although the date of flowering or harvest differed by region, the ET used these same NDVI values for 
classification across all scenes and regions. While ordinarily the ET would adjust the NDVI values used 
based on the specific environmental conditions within a scene (or sub-scene), without access to field￾verified sample data or information, the ET did not feel comfortable adjusting the criteria. Additionally, 
the ET’s approach maintained the consistency needed across the programmatic periods to increase the 
likelihood that changes between periods resulted from program implementation rather than differences 
in relative NDVI values used. The ET also notes that the degree that pixels fall within a single field or 
span multiple fields also depends on field size, which can vary substantially between regions and even 
within regions. Lastly, the ET notes that in the vast majority of imagery it was possible to align the 
imagery date with the crop calendar date needed (the middle of the flowering season or the beginning of 
the harvest season), but it was not always possible due to cloud cover or poor image quality. A full list 
of the Landsat 8 scenes used is provided in Annex 1. 
3.4 DATA AND PROCESSING
As noted above, the ET downloaded Landsat 8 OLI multispectral (11-band) imagery covering the full 
extent of the RADP programmatic regions. The province/district GIS boundary data were obtained from 
the ESRI3 ArcGIS hub portal (ESRI, 2018a) and linked to programmatic province and district attribute 
data provided by USAID implementing partners. The ET downloaded imagery data from two USGS 
portals: the Landsatlook website (https://landsatlook.usgs.gov/) and the USGS Earth Explorer website
(https://earthexplorer.usgs.gov/). The imagery is available for the full archive of Landsat 8 OLI scenes and 
has been radiometrically calibrated, orthorectified, and projected into the Universal Transverse 
Mercator (UTM) coordinate system (processing level L1TP) and is thus suitable for analytic purposes. 
The ET also queried the imagery for the amount of cloud cover and overall image quality, and 
downloaded scenes aligned to the year and season shown in the crop calendar by region (Figure 4). In 
many cases, multiple dates for a single scene for the respective season were available. In these cases, the 
ET utilized the best imagery based on the following criteria:
• The image fell close to the middle of the flowering season or in the first two weeks of the
harvest season;
• The image was cloud-free over the areas of interest;
• The image covered the full extent of the scene boundary (in some cases scenes were clipped);
and
• The image date was as consistent as possible with other scenes in the season and region, and
across program years to maximize consistency.
Figure 4 shows the spatial footprints of the individual Landsat 8 scenes needed.
For each image footprint shown on the map, the ET obtained imagery for four images—corresponding 
to the two seasons (flowering and harvest) and two time periods (pre-program and program interim 
evaluation). The ET examined each scene downloaded using a false-color infrared band combination 
(near infrared/red/green) to highlight the vegetative growth and evaluated relative to the criteria above. 
Taking into account some scenes that extended slightly beyond their boundary, the ET ultimately relied 
upon 141 scenes (see Annex 1).
The ET then converted the imagery into NDVI raster layers using Equation 1 shown above and then 
reclassed the NDVI output layers using the classes indicated in Table 6. Figure 5 shows an example of 
false-color infrared and NDVI imagery for a portion of the RADP-South region.
3 ESRI is an international supplier of GIS software, web GIS, and geodatabase management applications. 
https://www.esri.com/en-us/home. 
16 
Figure 4. Spatial Extent of Landsat 8 Scenes Needed across the RADP Regions
The ET then overlaid the resulting flowering and harvest layers within a GIS and masked the output to 
limit the analysis area to the programmatic districts within each RADP region (i.e., excluding the area 
outside of the programmatic districts). Finally, the ET compared the attribute table of final overlay to
classify pixels into a final land cover type using the combination of flowering and harvest NDVI class 
values, as shown in Table 7. 
Table 7. Flowering and Harvest Season Criteria Used to Determine Final Land Cover 
Classification
Flowering Season NDVI Bin Harvest Season NDVI Bin Final Land Cover
0, 1 (barren or sparse vegetation) -1 (water/snow) “Water/snow”a
-1 (water/snow) 0, 1, 2 (barren, sparse, vigorous vegetation) “Water/snow”a
0 (barren or sparse vegetation) 1 (barren or sparse vegetation) “Other”
0 (barren or sparse vegetation) 2 (vigorous vegetation) “Other crop”
1 (moderate vegetative growth) 1 (barren or sparse vegetation) “Likely poppy”
1 (moderate vegetative growth) 2 (vigorous vegetation) “Other crop”
2 (vigorous vegetation) 1 (barren or sparse vegetation) “Other crop”
2 (vigorous vegetation) 2 (vigorous vegetation) “Other crop”
a Denoted as “water/snow” as it contained water in either season but could indicate a marsh in cases where it was water in 
one season and had no or some vegetation in the other season.
17 
Figure 5. Portion of the RADP-South Region Showing Landsat 8 Imagery Used in the 
Classification Process
Left: false-color infrared imagery (bands 5, 4, and 3). Right: NDVI layer. 
Figure 6 shows an example of output from the classification process for a partial Landsat scene in the 
RADP-South region. The left side of the figure shows the licit and illicit classification for the pre￾program period and the right side shows the output for the evaluation period. The ET conducted all 
analysis using ArcGIS Pro (v. 2.0 and 2.2.2) software for all image and GIS processing (ESRI, 2018b). 
Following the final classification of land cover type, the ET combined the final layer with the district layer 
and calculated the total area by land cover type by summing the cell count in each final classification 
within each district. This workflow was conducted for both the pre-program and program interim￾evaluation periods and tabulated to compare the change in land cover types. 
18 
Figure 6. Example of Final Classification Output for a Partial Scene in the RADP-South 
Region
Left: pre-program (2013). Right: evaluation period (2017). Red = “likely poppy,” green = “other crop,” gray = “other” (non￾vegetated or sparsely vegetated), and light blue = “water/snow.”
19 
4.0 RESULTS AND DISCUSSION 
4.1 TABLES OF RESULTS AND DISCUSSION
The results of the ET’s analysis are shown in Tables 8–11. 
4.1.1 RADP-South Region 
For the districts participating in the program in the RADP-South region (Table 8), the ET estimated that 
just over 23,600 hectares were under poppy in the pre-program period (2013), which rose to
28,401 hectares in the program evaluation period (2017), an increase of about 20 percent. For other
agricultural lands, the ET estimated a similar increase: from about 284,000 hectares to about
353,500 hectares, a roughly 24 percent increase in area. However, the area of poppy as a percentage of
total agriculture decreased slightly from approximately 7.7 percent in 2013 to 7.4 percent in 2017. It is 
interesting to note that the amount of water/snow was estimated to be approximately 120 percent
higher in 2017 than 2013. Therefore, it could be that the increase in the area under poppy cultivation 
was due to increased water availability. This argument is supported by the fact that the area under other 
crops increased as well. The reduction in area under poppy as a percentage of total area under 
agriculture is encouraging though.
Table 8. Change in Area under Poppy and Other Uses across the RADP-South Region
Classification
Hectares Change
2013 2017
Change in 
Hectares
% Change in 
Hectares
Poppy as % of Total 
Agriculture
2013 2017
Poppy 23,663 28,401 4,738 +20.0% 7.7% 7.4%
Other agriculture 284,285 353,499 69,214 +24.3% N/A
Water/snow/ice 9,155 20,201 11,046 +120.7% N/A
4.1.2 RADP-North Region
In the RADP-North region (Table 9), the ET estimated a dramatic increase in poppy production 
between the two periods. In 2013, the ET estimated there were about 15,000 hectares of poppy, which 
increased to almost 81,000 hectares by 2017, an increase of approximately 440 percent. In contrast, in 
the RADP-North region, the area under other agriculture decreased about 22 percent, from about
700,000 hectares in 2013 to about 548,000 hectares in 2017. The ET also saw an increase of about 
67 percent in land cover classified as water/snow. Therefore, the increase could be due to increased 
water availability. However, the ET saw areas of poppy production in the desert areas in the evaluation 
period but not in the pre-program period (Figures 7–9), implying that the increase could be due to a 
shift from licit to illicit agriculture—especially as the decrease in absolute amount of other agriculture 
(151,673 hectares) was roughly twice as much as the increase in poppy area (65,827 hectares). Figure 7
shows the desert area in the pre-program and evaluation periods, highlighting the expansion of poppy 
crop. Figure 8 shows the same area of poppy cultivation (left) and natural color high-resolution imagery 
(right). The ET’s initial thought was that the evaluation period classification was in error. However, a 
large-scale (zoomed in) portion of the same area (Figure 9) clearly shows agricultural fields, which gave
the ET confidence in their classification as poppy. 
20 
Table 9. Change in Area under Poppy and Other Uses across the RADP-North Region
Classification
Hectares Change
2013 2017
Change in 
Hectares
% Change 
in Hectares
Poppy as % of 
Total Agriculture
2013 2017
Poppy 14,967 80,794 65,827 +439.8% 2.1% 12.9% 
Other agriculture 699,315 547,642 (151,673) -21.7% N/A
Water/snow/ice 9,909 16,581 6,672 +67.3% N/A
Figure 7. Desert Area in the RADP-North Region Showing Expansion of Poppy Crop
Left: pre-program period. Right: evaluation period. Red = poppy, green = other agriculture, and beige = other land cover 
(e.g., desert).
21 
Figure 8. Images of the RADP-North Region for the Evaluation Period (2018) Showing 
Expansion of Poppy Production into the Desert
Left: evaluation period (2018) in the RADP-North region showing expansion of poppy production into the desert. Red = 
poppy, green = other crop, beige = other land cover (e.g., desert). Right: high-resolution natural color imagery of same area 
shown on left. Imagery source: ESRI, 2018c. 
Mansfield (2018) found a similar expansion of poppy into desert regions in the RADP-South region in 
response to eradication efforts. Mansfield also noted that as the cost of production is high in these 
desert areas due to the need to pump water for irrigation, any agricultural production would necessarily 
be a high-value crop such as poppy. As a percentage of total agriculture, poppy production increased 
from 2.1 percent in 2013 to almost 13 percent by 2017. Alternatively, it could be that the low amount 
of production in 2013 was due to environmental conditions—such as a cold spring.
22 
Figure 9. Images of Desert Region of Poppy Expansion in the RADP-North Region to 
Highlight Potential Agricultural Fields
Left: the area at a small scale. Right: the area at large scale. 
Imagery source: ESRI, 2018c. 
4.1.3 RADP-East Region
In the RADP-East region (Table 10), there was a large decrease in poppy production between the post￾program (2018) and pre-program (2015) periods. The ET estimated that there were about 
12,000 hectares in 2015, which decreased to approximately 2,600 hectares in 2018; the reduction of 
about 9,400 hectares represented a decrease in crop area of about 78 percent. The ET saw a large 
decrease in other agricultural production between the evaluation and pre-program periods as well, with 
over 1 million hectares estimated in 2015 and about 786,000 hectares in 2018, a reduction of about
23 percent. As a percentage of total agriculture, poppy represented a very small amount, but decreased 
from 1.1 percent to 0.3 percent between the time periods. Similar to the RADP-North and RADP-South 
regions, the ET saw an increase in land cover classified as water/snow—with a doubling (202 percent) of 
area between the two periods. There are a number of potential reasons for the large reduction in illicit 
crop production between the program periods. However, given that other agriculture decreased by 
about 240,000 hectares and the fact that there was over a 200 percent increase in water (mainly in the 
form of snow) in 2018 compared to 2015, the idea that environmental conditions could have been less 
favorable for agriculture in the latter time period is supported. 
23 
Table 10. Change in Area under Poppy and Other Uses across the RADP-East Region
Classification
Hectares Change
2015 2018
Change in 
Hectares
% Change in 
Hectares
Poppy as % of Total 
Agriculture
2015 2018
Poppy 11,873 2,593 (9,280) -78.2% 1.1% 0.3%
Other agriculture 1,026,206 785,874 (240,331) -23.4% N/A
Water/snow/ice 164,726 496,768 332,043 +201.6% N/A
4.1.4 RADP-West Region
Lastly, the RADP-West region (Table 11) saw a large reduction in area under poppy cultivation between 
the 2014 pre-program period and the 2017 program evaluation period. In 2014, the ET estimated 
approximately 31,500 hectares of poppy, while in 2017 the ET’s estimate was approximately 
13,500 hectares, a decrease of about 57 percent. Other agricultural areas increased by approximately 
150 percent, from over 257,000 hectares in 2014 to more than 640,000 hectares in 2017. The 
percentage of poppy to total agriculture also dropped from about 11 percent to 2 percent between the 
time periods. This, along with the fact that total licit agriculture increased by about 383,000 hectares, 
could plausibly be explained as a shift from illicit to licit crops, although other environmental factors 
could be at play. This is supported by the increase in area classified as snow/water (about
3,600 hectares) between the time periods that could have made conditions more favorable for licit 
agriculture. Similar to the RADP-North region, the ET saw areas in the non-traditional agricultural areas 
(e.g., desert) that met the criteria for poppy cultivation (Figure 10). However, these same areas were 
not classified as poppy in the evaluation period, which is one reason for the reduction in overall poppy 
area.
Table 11. Change in Area under Poppy and Other Uses across the RADP-West Region
Classification
Hectares Change
2014 2017
Change in 
Hectares
% Change in 
Hectares
Poppy as % of Total 
Agriculture
2014 2017
Poppy 31,518 13,490 (18,028) -57.2% 10.9% 2.1%
Other agriculture 257,369 640,210 382,841 148.8% N/A
Water/snow/ice 1,660 5,284 3,624 +218.4% N/A
24 
Figure 10. Images of the RADP-West Region for the Evaluation Period (2018) Showing 
Expansion of Poppy Production into the Desert
Left: evaluation period (2018) in the RADP-West region showing poppy production into the desert. Red = poppy, green = 
other vegetation, beige = other land cover (e.g., unvegetated). Right: high-resolution natural color imagery of same area 
shown on left.
Imagery source: ESRI, 2018c. 
4.2 DISCUSSION
The differences the ET found between time periods are undoubtedly the result of a variety of limitations
to the data and approach, as well as other factors that may have affected changes in crop production. In 
terms of limitations, the ET faced a number of issues that may have led to misclassification of poppy vs. 
other types of agriculture. The limitations to the ET’s analysis included things that might cause 
differences between the crop calendar used and on-the-ground conditions, such as crop management 
practices (e.g., when poppies are planted or harvested, whether the field is plowed after harvest, and 
whether fields are intercropped or not); climatic anomalies (e.g., colder spring, hotter summer, drought, 
or higher snowpack that would result in changes timing of flowering or harvest); and variation in crop 
calendars between latitudes and elevation. As Simms (2016) noted, there was a two-month difference in 
the timing of flowering between adjacent valleys in the same province due to differences in valley 
elevations. In addition, there were other limitations such as the quality of imagery (including cloud cover, 
haze, etc.), dates of the imagery available, the inability to customize thresholds to every variation of 
mixed-pixel values resulting from different underlying land cover types (due to field sizes, intercropping, 
and vegetative health), and the sparsity of ground reference samples aligned with different phenologic 
stages for the specific programmatic years. 
25 
As noted in Section 3, Methods, without better information on any of the above issues, the ET chose to 
use a classification approach that relied on spectral ranges gleaned by other research and applied the 
method consistently across time periods to minimize the impact of subjective changes by the analyst.
Ultimately, the thresholds the ET used to classify areas of poppy cultivation assume that the crops are 
flowering on the first-dated imagery and have been harvested on the second-dated imagery. As the 
flowering period lasts approximately two weeks, if the image used was not aligned with the flowering 
period, the ET would most likely not classify the crop as poppy, which would result in an 
underestimation of poppy area. Additionally, as described in Section 3, Methods, the NDVI values used 
were based partially on estimates from the MODIS (250-meter resolution) mixed-pixel analysis 
conducted by Simms (2016). While the ET did adjust the specific values to account for the difference in 
pixel size (Landsat 30-meter resolution) and the effect of different thresholds on the amount of poppy 
classified, the ET sensitivity analysis was based on a limited agricultural region in the RADP-South region. 
The ET emphasizes that even minor changes in the NDVI thresholds used (both in flowering and post￾harvest periods) can have a substantial impact on what is classified as poppy or other crop type. In the 
end, the ET believes that the ranges used in this analysis are conservative.
To account for the regional and topographic influences that affect the timing of flowering and harvesting, 
it could be possible to adjust the crop calendar both regionally and locally by incorporating gridded data 
of monthly temperature and precipitation anomalies (difference from the average) or the elevation of 
agricultural fields into the analysis. This would help guide the choice of image dates to use. That said, 
understanding the relationship between the timing of flowering and harvesting based on these climatic 
and topographic differences could be challenging without an analog location with similar environmental 
conditions.
In addition to the technical limitations, other factors were likely to have caused actual changes in the 
area of poppy cultivation. For instance, the onset of disease or outbreak of pests (or avoidance of such) 
could also affect the amount of poppy cultivation either directly or indirectly—directly by reducing 
cultivated area outright or indirectly by increasing the cultivated area as a result of the adoption of 
herbicides or pesticides. The latter effect is supported by other research. According to the Afghanistan 
Research and Evaluation Unit Issues Paper (Mansfield, 2018), the use of herbicides in Southwest 
Afghanistan in 2013 was limited but by 2017 herbicides were almost universal.
In addition to the environmental issues that may have affected the ET’s classification results, a number of 
socioeconomic factors are likely to have affected the changes found. These include:
• Economic factors that influence both the cost of production and/or market value of poppy
vs. other crops;
• Civil unrest; and
• Eradication efforts (including foreign military intervention).
This last bullet has resulted in expansion into desert lands away from traditional agricultural areas. It is 
likely that this is what the ET saw in the RADP-North region and potentially the RADP-West region, 
although it is difficult to be certain without ground reference or ancillary data. However, Mansfield 
(2018) reported a large expansion into desert lands since 2013, due to a combination of the eradication 
efforts in conjunction with technological innovation in the RADP-South region, including:
• Deep wells for irrigation (made possible due to price of poppy);
• Reduction in well costs due to solar expansion (80 percent increase in reservoirs and pump
stations between 2016 and 2017); and
• Use of herbicides.
Similarly, the UNODC 2017 report noted that the use of solar panels, fertilizers, and pesticides may 
have increased the profitability of poppy cultivation even in less-favorable environmental conditions. 
26 
It is difficult to compare the ET’s findings with those of the UNODC 2017 report for a number of 
reasons. First, the ET notes that the years in common between the two studies are limited somewhat by 
region. While the RADP-North, -South, and -West regions have results from both pre-program and 
evaluation period years, the evaluation period for the RADP-East region was for the year 2018—which 
has yet to be evaluated by UNODC. As noted by Simms (2016) and others, there is a high variability in 
poppy cultivation between years, so without having estimates for the exact same years, a direct 
comparison is hard to support.
Second, at the province level, the ET’s estimates include only those districts that participated in the 
program and thus represents a small subset of a province (see Figure 1), while the estimates from the 
UNODC report are for the entire province area.
A third difference is that the UNODC 2017 analysis reported only areas of poppy cultivation of at least 
100 hectares, whereas the results of this analysis include all areas.
Lastly, the UNODC 2017 report provides a central estimate of poppy area as well as the upper and 
lower bounds of the 95 percent confidence interval at the province level. As the spread between the 
upper and lower bounds can be up to 6.7 times different and represents the entirety of the province, it 
makes province comparison between the studies difficult.
However, with these caveats, the ET compared the changes for the UNODC 2017 report (2016–2017 
change) and their changes between the pre-program and evaluation periods. The UNODC 2017 report 
found a substantial increase in poppy cultivation for all regions, while the ET found a mix of increases 
and decreases in their analysis. In the RADP-South region, UNODC found a 67 percent increase in 
poppy, while the ET found a 20 percent increase (2013–2017). In the RADP-West region, the UNODC 
reported a 6 percent increase compared to a 57 percent reduction in this study (2014–2017). In the 
RADP-East region, the UNODC reported a 36 percent increase in poppy while the ET found a 
78 percent reduction (2015–2018). Finally, in the RADP-North region, the UNODC report found an 
increase of 441 percent, while the ET found a 439 percent increase. Again, the years and total domains 
differ, but the results were in agreement in the direction of the sign (i.e., positive or negative) in 
two regions but opposite for the other two. The large reduction in 2018 in the RADP-East region in 
particular is interesting as it could indicate less-than-ideal climate conditions for poppy (such as a colder￾than-average spring) or a decrease in demand.
The ET also notes that the sampling domain used in the UNODC 2017 report did not include lands 
containing less than 0.25 square kilometers of arable land. This may therefore have excluded areas of 
new poppy expansion (e.g., desert areas), which could result in an underestimate of the amount of 
poppy crop estimated.
However, as noted above, the UNODC 2017 report also provides results at the individual district level 
and includes estimates for the pre-program and implementation years for all but the RADP-East region 
(for which only the pre-program results from UNODC are available). Tables 12–15 show the results of 
the district level comparison as well as the summation of the districts within the region. 
Comparing the results across districts for the same year shows substantial differences between the 
estimates—both in terms of absolute numbers and percent change. The likely reason is due to a 
combination of factors, including the classification methodology and the extent and comprehensiveness 
of the analysis. However, without having a statistically valid accuracy assessment across all regions, it is 
difficult to say which is more accurate, especially given the range from the upper and lower bounds 
(95 percent confidence intervals) noted at the province level in the UNODC 2017 report (Table 16). It 
is also probable that the accuracy estimates are quite variable at a sub-province level. As noted above, 
the UNODC 2017 analysis uses a different methodology based on where poppy production has 
traditionally occurred. In these traditional agricultural areas, the UNODC relies on a statistically based 
sampling approach, using visual classification of very high-resolution imagery along with ancillary 
27 
information (ground reference data) as the basis for the identification of poppy. While the use of high￾resolution imagery is undoubtedly superior to moderate-resolution image classification in these areas
where attributes such as field texture and hue can be exploited, there will undoubtedly be errors due to 
misinterpretation of imagery (especially if the image dates are not aligned with the flowering period). 
These errors would then be reflected in the ratios used in the UNODC estimates. 
Additionally, while the UNODC 2017 assessment utilizes a high number of well-distributed samples, 
there are likely areas where the relationships they used would not apply. The ET’s analysis showed a 
high degree of variation between different agricultural areas—with some areas showing the salt-and￾pepper pattern (e.g., Figure 6), while other areas were relatively homogenous and devoid of poppy. 
While these latter areas could be the result of misinterpretation based on differences between the crop 
calendar dates and on-the-ground timing of flowering/harvest, other areas are likely classified correctly, 
with little poppy production. In these cases, estimates based on the ratios from other areas would be 
incorrectly applied. 
As noted above, in areas with a low level of poppy production, the UNODC 2017 analysis used a 
targeted approach using full coverage of satellite imagery for portions of the province that were 
identified as likely to have poppy cultivation (based on a surveillance system and information by village 
“headmen”). In areas with no indication of poppy cultivation, they used a village survey approach without 
any image analysis. In both of these cases, the ET feels that the UNODC is likely missing many outlying 
areas of poppy cultivation. In using a comprehensive set of imagery across all the study areas, the ET 
found that some districts that the UNODC found devoid of poppy cultivation had substantial poppy
cultivation in the ET’s estimates. These seem to occur in outlying areas of deserts where poppy 
cultivation has expanded (Figures 7–10). In these cases, it could be that the UNODC methodology 
missed areas that were not either sampled or targeted as they did not conduct a wall-to-wall estimation. 
It could also be that the ET’s approach misclassified the image as poppy. However, a number of factors 
support the ET’s assessment: 1) these are areas of vigorous vegetative growth that match the seasonal 
management associated with poppy (Figure 10); 2) the higher cost of production (irrigation) means that 
any crops grown are likely to be of high value, which further supports that they are poppy; and 3) high￾resolution imagery of the region shows agricultural activity (Figures 7–10). 
Table 12. RADP-West Region, District-Level Comparison between UNODC 2017 Report 
and Pre-Program Period
Region Province District
Poppy 
(hectares) 
UNODC 
(2014)
Poppy 
(hectares) 
UNODC 
(2017)
Likely Poppy 
(hectares) 
(2014–Pre￾Program)
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
West Badghis Ab Kamari 0 281 1,906 112 NA -94%
West Badghis Muqur 47 2,097 992 374 4,362% -62%
West Badghis Qadis 57 1,802 12,217 4,924 3,061% -60%
West Badghis Qala-I-Naw 49 38 750 3 -22% -100%
West Farah Anar Dara 104 1 155 189 -99% 22%
West Farah Farah 4,760 47 1,700 1,883 -99% 11%
West Farah Pusht Rod 2,214 1,499 1,790 2,225 -32% 24%
West Hirat Guzara 0 0 1,409 171 NA -88%
West Hirat Injil 0 0 402 74 NA -82%
West Hirat Karukh 0 0 1,295 253 NA -80%
28 
Region Province District
Poppy 
(hectares) 
UNODC 
(2014)
Poppy 
(hectares) 
UNODC 
(2017)
Likely Poppy 
(hectares) 
(2014–Pre￾Program)
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
West Hirat Koshk 0 575 2,116 531 NA -75%
West Hirat Pashtun 
Zarghun 0 0 1,923 364 NA -81%
West Hirat Shindand 729 517 4,863 2,389 -29% -51%
Total 7,960 6,857 31,518 13,490 -14% -57%
Note: Totals may not sum due to rounding.
Table 13. RADP-South Region, District-Level Comparison between UNODC 2017 Report 
and Pre-Program Period
Region Province District
Poppy 
(hectares) 
UNODC 
(2013)
Poppy 
(hectares) 
UNODC 
(2017)
Likely Poppy 
(hectares) 
(2013–Pre￾Program) 
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
South Hilmand Garmser 4,527 13,211 5,813.1 205.29 192% -96%
South Hilmand Lashkar Gah 1,828 4,669 1,486.8 616.14 155% -59%
South Hilmand Nad Ali 19,136 27,398 4,709.16 14,028.84 43% 198%
South Hilmand Nahri Sarraj 18,701 18,464 5,983.47 7,384.14 -1% 23%
South Hilmand Nawa-I￾Barak Zayi 97 4,064 1,382.13 1,958.04 4,090% 42%
South Kandahar Arghandab 18 1,183 22.86 21.6 6,472% -6%
South Kandahar Daman 0 157 204.84 27 NA -87%
South Kandahar Kandahar 46 113 180.99 65.43 146% -64%
South Kandahar Khakrez 1,006 416 17.19 23.58 -59% 37%
South Kandahar Maywand 16,382 9,284 1,131.57 876.15 -43% -23%
South Kandahar Panjwayi 984 2,141 600.3 28.62 118% -95%
South Kandahar Spin Boldak 207 1,880 392.31 616.5 808% 57%
South Kandahar Zhari 7,017 7,605 828 151.74 8% -82%
South Uruzgan Chora 611 1,263 27.18 15.75 107% -42%
South Uruzgan Dihrawud 3,321 5,648 294.57 1,210.77 70% 311%
South Uruzgan Tirin Kot 1,936 8,368 341.01 1,126.71 332% 230%
South Zabul Qalat 28 10 90.99 13.5 -64% -85%
South Zabul Shahjoy 96 0 15.75 1.71 -100% -89%
South Zabul Shinkay 0 0 107.64 24.21 NA -78%
South Zabul Tarnak Wa 
Jaldak 0 959 32.85 5.13 NA -84%
29 
Region Province District
Poppy 
(hectares) 
UNODC 
(2013)
Poppy 
(hectares) 
UNODC 
(2017)
Likely Poppy 
(hectares) 
(2013–Pre￾Program) 
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
Total 75,941 106,833 23,663 28,401 41% 20%
Table 14. RADP-North Region, District-Level Comparison between UNODC 2017 Report 
and Pre-Program Period
Region Province District
Poppy 
(hectares) 
UNODC 
(2013)
Poppy 
(hectares) 
UNODC 
(2017)
Likely 
Poppy 
(hectares) 
(2013–Pre￾Program)
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
North Badakhshan Baharak 322 0 110.43 103.68 -100% -6%
North Badakhshan Fayzabad 48 10 1,002.78 515.88 -79% -49%
North Badakhshan Kishim 141 1,128 2,170.8 3,312.45 700% 53%
North Badakhshan Yaftal Sufla 18 52 647.91 751.05 189% 16%
North Baghlan Baghlani Jadid 0 0 680.4 10,415.79 NA 1431%
North Baghlan Dahana-I￾Ghuri 0 0 1,676.07 7,784.73 NA 364%
North Baghlan Puli Khumri 0 0 62.82 482.04 NA 667%
North Balkh Balkh 0 2,334 416.52 1,434.15 NA 244%
North Balkh Chimtal 400 5,768 150.57 4,418.91 1342% 2835%
North Balkh Dawlatabad 0 1 658.17 4,170.96 NA 534%
North Balkh Dihdadi 0 6 16.11 1,540.44 NA 9462%
North Balkh Khulm 0 0 2,016.27 3,664.53 NA 82%
North Balkh Mazari Sharif 0 0 0.09 0.63 NA 600%
North Balkh Nahri Shahi 0 0 144.9 3,438.09 NA 2273%
North Balkh Sholgara 0 0 159.93 12,877.47 NA 7952%
North Jawzjan Aqcha 0 20 0.54 31.14 NA 5667%
North Jawzjan Fayzabad 0 396 11.34 320.31 NA 2725%
North Jawzjan Shibirghan 0 867 5.04 2,669.76 NA 52871%
North Kunduz Aliabad 0 0 20.88 1,427.94 NA 6739%
North Kunduz Chahar Dara 0 0 1,759.41 2,150.55 NA 22%
North Kunduz Khanabad 0 0 2,056.86 13,258.44 NA 545%
North Kunduz Kunduz 0 0 582.93 2,076.66 NA 256%
North Samangan Aybak 0 0 184.59 1,225.62 NA 564%
North Samangan Feroz 
Nakhchir 0 0 4.5 708.12 NA 15636%
North Samangan Hazrati 
Sultan 0 0 55.17 982.17 NA 1680%
North Samangan Khuram Wa 
Sarbagh 0 0 371.7 1,024.11 NA 176%
30 
Region Province District
Poppy 
(hectares) 
UNODC 
(2013)
Poppy 
(hectares) 
UNODC 
(2017)
Likely 
Poppy 
(hectares) 
(2013–Pre￾Program)
Likely Poppy 
(hectares) 
(2017–Post￾Program) 
UNODC 
Change
RADP 
Change
Total 929 10,582 14,967 80,786 1039% 440%
Table 15. RADP-East Region, District-Level Comparison between UNODC 2017 Report 
and Pre-Program Period
Region Province District
Poppy 
(hectares) 
UNODC 
(2015)
Poppy 
(hectares) 
UNODC 
(2017)*
Likely 
Poppy 
(hectares) 
(2015–Pre￾Program)
Likely 
Poppy 
(hectares) 
(2018–Post￾Program) 
UNODC 
Change
RADP 
Change
East Ghazni Andar 0 0 0.81 0.09 NA -89%
East Ghazni Bahrami Shahid 
(Jaghatu) 0 0 10.35 0 NA -100%
East Ghazni Dih Yak 0 0 10.71 0.09 NA -99%
East Ghazni Ghazni 0 0 26.46 0.09 NA -100%
East Ghazni Jaghuri 0 0 30.96 0.36 NA -99%
East Ghazni Nawur 0 0 7.56 0 NA -100%
East Ghazni Qarabagh 0 0 41.22 0.18 NA -100%
East Ghazni Waghaz 0 0 3.96 0 NA -100%
East Kabul Bagrami 0 0 19.26 0.36 NA -98%
East Kabul Chahar Asyab 0 0 0.45 0 NA -100%
East Kabul Dih Sabz 0 0 14.76 0.27 NA -98%
East Kabul Farza 0 0 0 0 NA NA
East Kabul Guldara 0 0 0 0 NA NA
East Kabul Kabul 0 0 3.6 0 NA -100%
East Kabul Kalakan 0 0 3.33 0 NA -100%
East Kabul Khaki Jabbar 0 0 0.18 0 NA -100%
East Kabul Mir Bacha Kot 0 0 0.27 0 NA -100%
East Kabul Musayi 0 0 8.1 1.17 NA -86%
East Kabul Paghman 0 0 0.18 0.09 NA -50%
East Kabul Qarabagh 0 0 18.09 0.18 NA -99%
East Kabul Surobi 321 435 186.93 16.92 36% -91%
East Kapisa Hisa-i-Awali 
Kohistan 0 0 62.37 0.36 NA -99%
East Kapisa Koh Band 10 29 8.1 0 190% -100%
East Kapisa Mahmudi Raqi 0 0 38.52 7.02 NA -82%
31 
Region Province District
Poppy 
(hectares) 
UNODC 
(2015)
Poppy 
(hectares) 
UNODC 
(2017)*
Likely 
Poppy 
(hectares) 
(2015–Pre￾Program)
Likely 
Poppy 
(hectares) 
(2018–Post￾Program) 
UNODC 
Change
RADP 
Change
East Kapisa Nijrab 21 57 2.43 0 171% -100%
East Laghman Alingar 277 575 575.37 88.83 108% -85%
East Laghman Mihtarlam 123 281 1,218.42 146.25 128% -88%
East Laghman Qarghayi 4 137 598.05 135.54 3325% -77%
East Logar Baraki Barak 0 0 14.49 41.76 NA 188%
East Logar Charkh 0 0 2.16 3.15 NA 46%
East Logar Mohammad 
Agha 0 0 5.31 1.35 NA -75%
East Logar Puli Alam 0 0 4.77 1.26 NA -74%
East Nangarhar Acheen 1,090 1,364 634.23 46.35 25% -93%
East Nangarhar Bati Kot 4 757 855.63 448.11 18825% -48%
East Nangarhar Bihsud 0 0 302.85 52.74 NA -83%
East Nangarhar Chaparhar 1,504 2,337 805.05 78.66 55% -90%
East Nangarhar Goshta 6 0 542.61 13.68 -100% -97%
East Nangarhar Jalalabad 0 0 5.67 0.72 NA -87%
East Nangarhar Kama 0 0 463.32 426.69 NA -8%
East Nangarhar Khogayani 2,996 4,728 1,736.01 246.69 58% -86%
East Nangarhar Kot 872 49 590.67 113.85 -94% -81%
East Nangarhar Muhmand Dara 19 505 342.9 31.86 2558% -91%
East Nangarhar Pachier Agam 1,066 1,231 191.16 47.25 15% -75%
East Nangarhar Rodat 389 1,802 1,079.82 222.57 363% -79%
East Nangarhar Shinwar 70 1,245 833.04 53.1 1679% -94%
East Nangarhar Surkh Rod 188 816 208.44 346.59 334% 66%
East Parwan Bagram 0 0 341.46 6.21 NA -98%
East Parwan Chaharikar 0 0 9.72 2.97 NA -69%
East Parwan Salang 0 0 0.09 0 NA -100%
East Parwan Sayd Khel 0 0 6.93 5.58 NA -19%
East Parwan Shekh Ali 0 0 0 0 NA NA
East Parwan Shinwari 0 0 1.44 0.54 NA -63%
East Parwan Sia Gird 
(Ghorbund) 0 0 0.09 0 NA -100%
East Maydan 
Wardak
Chaki Wardak 0 0 0.81 0.09 NA -89%
32 
Region Province District
Poppy 
(hectares) 
UNODC 
(2015)
Poppy 
(hectares) 
UNODC 
(2017)*
Likely 
Poppy 
(hectares) 
(2015–Pre￾Program)
Likely 
Poppy 
(hectares) 
(2018–Post￾Program) 
UNODC 
Change
RADP 
Change
East Maydan 
Wardak
Jaghatu 0 0 0.09 0 NA -100%
East Maydan 
Wardak
Jalrez 0 0 0 0 NA NA
East Maydan 
Wardak
Maydan Shahr 0 0 0 0 NA NA
East Maydan 
Wardak
Markazi Bihsud 0 0 0.09 0 NA -100%
East Maydan 
Wardak
Saydabad 0 0 4.14 3.51 NA -15%
Total 8,960 16,348 11,873 2,593 82% -78%
Table 16. Province-Level Summaries of Poppy Cultivation, Showing Upper and Lower 95% 
Confidence Ranges
Province Point Estimate 
(hectares) 
Lower Bound 
(hectares) 
Upper Bound 
(hectares) 
Badakhshan 8,311 3,215 13,406
Badghis 24,723 11,518 37,928
Day-Kundi 1,508 605 2,412
Farah 12,846 7,210 18,482
Fayrab 22,797 15,171 30,422
Ghor 4,228 1,949 6,507
Helmand 144,018 129,371 158,664
Jawzjan 3,236 1,778 4,695
Kandahar 28,010 18,967 37,052
Kunar 1,634.91 741 2,528
Laghman 2,257 591 3,924
Nangarhar 18,976 13,154 24,799
Nimroz 11,466 7,314 15,619
Uruzgan 21,541 13,297 29,785
Zabul 2,131 527 3,735
Target provinces 20,620 - - 
National 328,304 301,472 355,135
Rounded 328,000 301,000 355,000
Source: UNODC, 2017, Table 26, p. 54.
33 
In summary, given the differences in methodologies and uncertainties in classification techniques, 
especially with the number of environmental, economic, and social variables that may affect the 
classification results, it is not surprising that the ET estimates may differ (sometimes substantially) from 
the results that others have found. However, as noted above, many of the environmental uncertainties 
could be reduced by incorporating climate and topography GIS data into the analysis. These results 
would undoubtedly allow the ET to better align the imagery with crop phenology and thereby improve 
estimates of poppy and non-poppy crops. Additionally, by combining these enhancements and running 
the classification on both programmatic and non-programmatic districts, the ET may be better able to 
understand the impact of the program.
Having said this, in the ET’s final assessment, the team believes that using high-resolution imagery, along 
with statistical sampling, is superior to classifying poppy crop than using an NDVI threshold approach—
at least in the agricultural area traditionally associated with poppy cultivation. However, the sampling 
and targeted approaches used by UNODC may miss areas that could be captured from the Landsat 
approach used here. It may therefore be useful to augment the UNODC’s sample/target approach to 
include the Landsat/NDVI-based analysis used by the ET to identify areas of poppy cultivation in outlying 
regions.
34 
REFERENCES
Conrad, Abigail, Charlotte Maxwell-Jones, Marius Meijerink, Betsy Ness-Edelstein, Phomdaen Souvanna, 
and Tulika Narayan. 2018. “Mid-term Performance Evaluation of the Regional Agricultural 
Development Program in Afghanistan.” Washington, DC: Program Evaluation for Effectiveness 
and Learning, ME&A. 
Curran, R.W., M. Campbell, S. Blundell, K. Kash, N. Kruskamp, and M. Voss. 2013. “Object-Based 
Feature Extraction of Poppy Fields in Afghanistan.” American Society of Photogrammetry and 
Remote Sensing (ASPRS) Annual Conference, Baltimore, MD. March 24–28. 
ESRI. 2018a. World Boundaries and Places. Digital vector data. Credits: ESRI, HERE, Garmin, Open 
Streetmap contributors, and the GIS user community. Environmental Systems Research 
Institute, Redlands, CA.
———. 2018b. ArcGIS Pro (versions: 2.0 and 2.2.2). Digital software. Environmental Systems Research 
Institute, Redlands, CA. https://www.esri.com/en-us/home. 
———. 2018c. World Imagery. Digital raster imagery data. Source: ESRI, Digital Globe, GeoEye, 
Earthstar Geographics, CNES/Airbus DS, USDA, USGS, AeroGRID, IGN, and the GIS User 
Community. Environmental Systems Research Institute, Redlands, CA.
Mansfield, D. 2018. “Still Water Runs Deep: Illicit Poppy and the Transformation of the Deserts of 
Southwest Afghanistan.” Kabul: Afghanistan Research and Evaluation Unit.
Sader, A. 1990. “Remote Sensing of Narcotics: With Special Reference to Techniques for Detection and 
Monitoring of Poppy Production in Afghanistan.” Prepared for the Narcotics Awareness Control 
Project.
Simms, D.M. 2016. “Remote Sensing of Opium Poppy Cultivation in Afghanistan.” Doctoral Thesis, 
Cranfield University.
UNODC. 2017. “Afghanistan Opium Survey 2017: Cultivation and Production.” UNODC and Islamic 
Republic of Afghanistan Ministry of Counter Narcotics. United Nations Office on Drugs and 
Crime.
USAID. 2016. “Afghanistan Alternative Development Options Assessment – Final Report, March 2016. 
Regional agricultural development program – South (RADP-S).” United States Agency for 
International Development. 
35 
ANNEX
36 
ANNEX 1: LANDSAT SCENES USED IN ANALYSIS
Region Collection Data Type Path/Row Acquisition Date Year Month Day Period Season
East LC08 L1TP 151036 20180412 2018 04 12 Post-Program Flowering
East LC08 L1TP 152036 20180403 2018 04 03 Post-Program Flowering
East LC08 L1TP 153035 20180325 2018 03 25 Post-Program Flowering
East LC08 L1TP 153036 20180325 2018 03 25 Post-Program Flowering
East LC08 L1TP 153037 20180325 2018 03 25 Post-Program Flowering
East LC08 L1TP 154036 20180401 2018 04 01 Post-Program Flowering
East LC08 L1TP 154037 20180401 2018 04 01 Post-Program Flowering
East LC08 L1TP 151036 20180530 2018 05 30 Post-Program Harvesting
East LC08 L1TP 152036 20180606 2018 06 06 Post-Program Harvesting
East LC08 L1TP 153035 20180613 2018 06 13 Post-Program Harvesting
East LC08 L1TP 153036 20180613 2018 06 13 Post-Program Harvesting
East LC08 L1TP 153037 20180613 2018 06 13 Post-Program Harvesting
East LC08 L1TP 154036 20180604 2018 06 04 Post-Program Harvesting
East LC08 L1TP 154037 20180620 2018 06 20 Post-Program Harvesting
East LC08 L1TP 151036 20150404 2015 04 04 Pre-Program Flowering
East LC08 L1TP 152036 20150427 2015 04 27 Pre-Program Flowering
East LC08 L1TP 153035 20150504 2015 05 04 Pre-Program Flowering
East LC08 L1TP 153036 20150504 2015 05 04 Pre-Program Flowering
East LC08 L1TP 153037 20150504 2015 05 04 Pre-Program Flowering
East LC08 L1TP 154036 20150409 2015 04 09 Pre-Program Flowering
East LC08 L1TP 154037 20150409 2015 04 09 Pre-Program Flowering
East LC08 L1TP 151036 20150607 2015 06 07 Pre-Program Harvesting
East LC08 L1TP 152036 20150614 2015 06 14 Pre-Program Harvesting
East LC08 L1TP 153035 20150605 2015 06 05 Pre-Program Harvesting
East LC08 L1TP 153036 20150605 2015 06 05 Pre-Program Harvesting
East LC08 L1TP 153037 20150605 2015 06 05 Pre-Program Harvesting
East LC08 L1TP 154036 20150612 2015 06 12 Pre-Program Harvesting
East LC08 L1TP 154037 20150612 2015 06 12 Pre-Program Harvesting
West LC08 L1TP 156036 20170412 2017 04 12 Post-Program Flowering
West LC08 L1TP 156037 20170412 2017 04 12 Post-Program Flowering
West LC08 L1TP 157035 20170419 2017 04 19 Post-Program Flowering
West LC08 L1TP 157036 20170419 2017 04 19 Post-Program Flowering
West LC08 L1TP 157037 20170403 2017 04 03 Post-Program Flowering
37 
Region Collection Data Type Path/Row Acquisition Date Year Month Day Period Season
West LC08 L1TP 157038 20170419 2017 04 19 Post-Program Flowering
West LC08 L1TP 158037 20170309 2017 03 09 Post-Program Flowering
West LC08 L1TP 156036 20170615 2017 06 15 Post-Program Harvesting
West LC08 L1TP 157035 20170606 2017 06 06 Post-Program Harvesting
West LC08 L1TP 157036 20170606 2017 06 06 Post-Program Harvesting
West LC08 L1TP 157037 20170606 2017 06 06 Post-Program Harvesting
West LC08 L1TP 157038 20170606 2017 06 06 Post-Program Harvesting
West LC08 L1TP 158037 20170613 2017 06 13 Post-Program Harvesting
West LC08 L1TP 156036 20140420 2014 04 20 Pre-Program Flowering
West LC08 L1TP 156037 20140319 2014 03 19 Pre-Program Flowering
West LC08 L1TP 157035 20140427 2014 04 27 Pre-Program Flowering
West LC08 L1TP 157036 20140427 2014 04 27 Pre-Program Flowering
West LC08 L1TP 157037 20140427 2014 04 27 Pre-Program Flowering
West LC08 L1TP 157038 20140427 2014 04 27 Pre-Program Flowering
West LC08 L1TP 158037 20140418 2014 04 18 Pre-Program Flowering
West LC08 L1TP 156036 20140623 2014 06 23 Pre-Program Harvesting
West LC08 L1TP 157035 20140630 2014 06 30 Pre-Program Harvesting
West LC08 L1TP 157036 20140630 2014 06 30 Pre-Program Harvesting
West LC08 L1TP 157037 20140630 2014 06 30 Pre-Program Harvesting
West LC08 L1TP 157038 20140630 2014 06 30 Pre-Program Harvesting
West LC08 L1TP 158037 20140621 2014 06 21 Pre-Program Harvesting
North LC08 L1TP 152034 20170502 2017 05 02 Post-Program Flowering
North LC08 L1TP 152035 20170518 2017 05 18 Post-Program Flowering
North LC08 L1TP 153034 20170509 2017 05 09 Post-Program Flowering
North LC08 L1TP 153035 20170509 2017 05 09 Post-Program Flowering
North LC08 L1TP 154034 20170516 2017 05 16 Post-Program Flowering
North LC08 L1TP 154035 20170516 2017 05 16 Post-Program Flowering
North LC08 L1TP 155034 20170523 2017 05 23 Post-Program Flowering
North LC08 L1TP 155035 20170507 2017 05 07 Post-Program Flowering
North LC08 L1TP 152034 20170721 2017 07 21 Post-Program Harvesting
North LC08 L1TP 152035 20170721 2017 07 21 Post-Program Harvesting
North LC08 L1TP 153034 20170712 2017 07 12 Post-Program Harvesting
North LC08 L1TP 153035 20170712 2017 07 12 Post-Program Harvesting
North LC08 L1TP 154034 20170719 2017 07 19 Post-Program Harvesting
38 
Region Collection Data Type Path/Row Acquisition Date Year Month Day Period Season
North LC08 L1TP 154035 20170719 2017 07 19 Post-Program Harvesting
North LC08 L1TP 155034 20170710 2017 07 10 Post-Program Harvesting
North LC08 L1TP 155035 20170710 2017 07 10 Post-Program Harvesting
North LC08 L1TP 152034 20130523 2013 05 23 Pre-Program Flowering
North LC08 L1TP 152035 20130523 2013 05 23 Pre-Program Flowering
North LC08 L1TP 153034 20130514 2013 05 14 Pre-Program Flowering
North LC08 L1TP 153035 20130530 2013 05 30 Pre-Program Flowering
North LC08 L1TP 154034 20130505 2013 05 05 Pre-Program Flowering
North LC08 L1TP 154035 20130521 2013 05 21 Pre-Program Flowering
North LC08 L1TP 155034 20130613 2013 06 13 Pre-Program Flowering
North LC08 L1TP 155035 20130528 2013 05 28 Pre-Program Flowering
North LC08 L1TP 152034 20130726 2013 07 26 Pre-Program Harvesting
North LC08 L1TP 153034 20130717 2013 07 17 Pre-Program Harvesting
North LC08 L1TP 153035 20130717 2013 07 17 Pre-Program Harvesting
North LC08 L1TP 154034 20130708 2013 07 08 Pre-Program Harvesting
North LC08 L1TP 154035 20130708 2013 07 08 Pre-Program Harvesting
North LC08 L1TP 155034 20130715 2013 07 15 Pre-Program Harvesting
North LC08 L1TP 155035 20130715 2013 07 15 Pre-Program Harvesting
South LC08 L1TP 153037 20170407 2017 04 07 Post-Program Flowering
South LC08 L1TP 153038 20170407 2017 04 07 Post-Program Flowering
South LC08 L1TP 154037 20170414 2017 04 14 Post-Program Flowering
South LC08 L1TP 154038 20170414 2017 04 14 Post-Program Flowering
South LC08 L1TP 154039 20170414 2017 04 14 Post-Program Flowering
South LC08 L1TP 155037 20170421 2017 04 21 Post-Program Flowering
South LC08 L1TP 155038 20170304 2017 03 04 Post-Program Flowering
South LC08 L1TP 155039 20170507 2017 05 07 Post-Program Flowering
South LC08 L1TP 156038 20170412 2017 04 12 Post-Program Flowering
South LC08 L1TP 156039 20170412 2017 04 12 Post-Program Flowering
South LC08 L1TP 153037 20170610 2017 06 10 Post-Program Harvesting
South LC08 L1TP 153038 20170610 2017 06 10 Post-Program Harvesting
South LC08 L1TP 154037 20170601 2017 06 01 Post-Program Harvesting
South LC08 L1TP 154038 20170601 2017 06 01 Post-Program Harvesting
South LC08 L1TP 154039 20170601 2017 06 01 Post-Program Harvesting
South LC08 L1TP 155037 20170608 2017 06 08 Post-Program Harvesting
39 
Region Collection Data Type Path/Row Acquisition Date Year Month Day Period Season
South LC08 L1TP 155038 20170608 2017 06 08 Post-Program Harvesting
South LC08 L1TP 155039 20170608 2017 06 08 Post-Program Harvesting
South LC08 L1TP 156038 20170615 2017 06 15 Post-Program Harvesting
South LC08 L1TP 156039 20170615 2017 06 15 Post-Program Harvesting
South LC08 L1TP 155038 20130426 2013 04 26 Pre-Program Flowering
South LC08 L1TP 153037 20130428 2013 04 28 Pre-Program Flowering
South LC08 L1TP 153038 20130428 2013 04 28 Pre-Program Flowering
South LC08 L1TP 154037 20130505 2013 05 05 Pre-Program Flowering
South LC08 L1TP 154038 20130419 2013 04 19 Pre-Program Flowering
South LC08 L1TP 154039 20130419 2013 04 19 Pre-Program Flowering
South LC08 L1TP 155037 20130426 2013 04 26 Pre-Program Flowering
South LC08 L1TP 155039 20130426 2013 04 26 Pre-Program Flowering
South LC08 L1TP 156038 20130417 2013 04 17 Pre-Program Flowering
South LC08 L1TP 156039 20130410 2013 04 10 Pre-Program Flowering
South LC08 L1TP 153037 20130615 2013 06 15 Pre-Program Harvesting
South LC08 L1TP 153038 20130615 2013 06 15 Pre-Program Harvesting
South LC08 L1TP 154037 20130606 2013 06 06 Pre-Program Harvesting
South LC08 L1TP 154038 20130606 2013 06 06 Pre-Program Harvesting
South LC08 L1TP 154039 20130606 2013 06 06 Pre-Program Harvesting
South LC08 L1TP 155037 20130613 2013 06 13 Pre-Program Harvesting
South LC08 L1TP 155038 20130613 2013 06 13 Pre-Program Harvesting
South LC08 L1TP 155039 20130613 2013 06 13 Pre-Program Harvesting
South LC08 L1TP 156038 20130604 2013 06 04 Pre-Program Harvesting
South LC08 L1TP 156039 20130604 2013 06 04 Pre-Program Harvesting