Landsat Based Multitemporal Approach for Wheat Mapping in Semi-arid Region of Punjab, Pakistan

Abstract

Conventional ways of crop area and production estimation such as area list frame are not designed for pre-harvest results, while requiring huge human, logistics and financial resources.  An alternative method that can provide pre-harvest crop statistics is the need of the time. Data available from satellite monitoring of the earth has promising application in pre-harvest crop statistics. However, accuracy of information derived for heterogenous cropping systems and small land holding agriculture using multitemporal satellite data needs to be evaluated. We used multitemporal Landsat imagery for our research study that were available free of cost for pre-harvest wheat area estimation in Faisalabad. Different bands and indices of multitemporal Landsat imagery was used as input variables into random forest algorithm for wheat classification. The estimated results were then validated through accuracy assessment using field survey. Landsat-based wheat area estimated is within 0.6% of our sample-based reference estimate. The findings was 14% lesser as compared to official recorded disclosed by Punjab province. Overall accuracy of wall-to-wall wheat map was 86% (SE = 1.6).