Application of multi-spectral remote sensing for crop discrimination in Afghanistan

The spectral properties of poppy and other annual crops vary considerably throughout their growth and development. Until the publication of this research the spectral signature of poppy and its contrast with neighbouring crops in Afghanistan was undefined. The aim of this work was to investigate the application of remote sensing to discriminate poppy from other cover types using spectral signatures obtained from the analysis of multi-spectral imagery. The consistency of discrimination through time for different geographical regions was of particular interest. A review of previous poppy studies identified weaknesses with existing methods used to monitor poppy and provide reference data to validate resulting maps. Weaknesses were in the main due to the limited availability of quantifiable knowledge on the spectral-temporal properties of cover types and the lack of accuracy measures necessary to validate poppy identification. To overcome the lack of quantitative knowledge this research characterises the spatial and temporal variability of poppy spectral response patterns. A methodology was developed to acquire multi-temporal IKONOS images, aerial photographs and ground data covering two growth cycles across a range of sites in Afghanistan. Optimum techniques were developed to facilitate the collection of training pixels for each cover type to satisfy the assumptions of the supervised Maximum Likelihood classification (MLC). Spectral signatures of cover types were examined using the Jeffries Matusita distance measure to identify signature separability and predict classification accuracy. The accuracy of each MLC was assessed using error matrices, Kappa statistics and regression. Results confirm that sufficient spectral contrast exists between poppy and other crops during poppy flowering which enables accurate discrimination. A relationship was found between overall spectral separability and classification accuracy, showing separability can be used to predict classification accuracy at flowering. At other times insufficient differences exist between the spectral reflectance of other crops and poppy. Multi-temporal image classifications achieved greater accuracy than their corresponding single date classifications in the majority of cases studied.