Abstract:
This work investigates differences in the survey methodologies of the monitoring
programmes of the United Nations Office on Drugs and Crime (UNODC) and the
US Government that lead to discrepancies in quantitative information about poppy
cultivation. The aim of the research is to improve annual estimates of opium production.
Scientific trials conducted for the UK Government (2006–2009) revealed differences
between the two surveys that could account for the inconsistency in results.
These related to the image interpretation of poppy from very high resolution satellite
imagery, the mapping of the total area of agriculture and stratification using full
coverage medium resolution imagery. MODIS time-series profiles of Normalised
Difference Vegetation Index (NDVI), used to monitor Afghanistan’s agricultural
system, revealed significant variation in the agriculture area between years caused
by land management practices and expansion into new areas.
Image interpretation of crops was investigated as a source of bias within the sample
using increasing levels of generalisation in sample interpretations. Automatic
segmentation and object-based classification were tested as methods to improve
consistency. Generalisation was found to bias final estimates of poppy up to 14%.
Segments were consistent with manual field delineations but object-based classification
caused a systematic labelling error. The findings show differences in survey
estimates based on interpretation keys and the resolution of imagery, which is compounded
in areas of marginal agriculture or years with poor crop establishment.
Stratified and unstratified poppy cultivation estimates were made using buffered
and unbuffered agricultural masks at resolutions of 20, 30 and 60 m, resampled from
SPOT-5 10 m data. The number of strata (1, 4, 8, 13, 23, 40) and sample fraction (0.2
to 2%) used in the estimate were also investigated. Decreasing the resolution of the
imagery and buffering increased unstratified estimates. Stratified estimates were
more robust to changes in sample size and distribution. The mapping of the agricultural
area explained differences in cultivation figures of the opium monitoring
programmes in Afghanistan.
Supporting methods for yield estimation for opium poppy were investigated at
field sites in the UK in 2004, 2005 and 2010. Good empirical relationships were
found between NDVI and the yield indicators of mature capsule volume and dry
capsule yield. The results suggested a generalised relationship across all sampled
fields and years (R2 >0.70) during the 3–4 week period including poppy flowering.
The application of this approach in Afghanistan was investigated using VHR satellite
imagery and yield data from the UNODC’s annual survey. Initial results indicated
the potential of improved yield estimates using a smaller and targeted collection
of ground observations as an alternative to random sampling.
The recommendations for poppy cultivation surveys are: the use of image-based
stratification for improved precision and reducing differences in the agricultural
mask, and use of automatic segmentation for improved consistency in field delineation
of poppy crops. The findings have wider implications for improved confidence
in statistical estimates from remote sensing methodologies.