Object-based agricultural land use map of Khorezm

Abstract

In recent decades, multi-spectral and hyper-spectral remotely sensed imageries with high and modern spatial resolutions at sufficient time-series interval have been developed. This allows for detecting crop types and its distribution over large areas and at short time intervals. Among the advantages of remote sensing technologies are its cost effective evaluation over extensive areas and the ability to provide reliable information on land surface conditions. This is useful also for areas with sporadic information on the spatial extent of croplands effected by for instance water scarcity. The elaboration of sustainable natural resource management that demands a judicious management of land and fresh water, requires accurate information on status of these croplands. For classifying on field basis, agricultural fields were digitized based on very high spatial resolution SPOT 5 imageries. For the actual land use classification, 5 time-series images were used for the growing period in 2013. In order to consider accuracy assessment of classified training data, the random forest confusion matrix was implemented and training data allowed to classify an accuracy of 93 percent.