Authors: Abdul Qadir*, University of Delaware, Pinki Mondal, University of Delaware
Topics: Remote Sensing, Geographic Information Science and Systems, Land Use and Land Cover Change
Keywords: remote sensing, monsoon crop, radar, satellite
Session Type: Paper
Start / End Time: 9:35 AM / 10:50 AM
Room: Terrace, Sheraton, IM Pei Tower, Terrace Level
Presentation File: No File Uploaded
Monitoring the monsoon crops through optical satellite data in tropical regions is challenging due to many reasons including consistent cloud cover, small field size, and highly dynamic cropping pattern through space and time. Radar data provide an alternative for monsoon crop mapping and monitoring due to the sensor’s cloud penetrating capabilities. In this work, we utilized radar satellite data from Sentinel-1, and derived sowing dates for studying the impacts of monsoon rain variability on crops. We also combined radar data with Sentinel-2 optical data for improved mapping of monsoon crops. We proposed a novel method Radar Optical cross Masking (ROM) to generate high resolution monsoon crop mask. A machine learning based random forest (RF) classifier was used to generate the monsoon crop map. The proposed method was trained and tested in a contiguous region composed of five different Agro-Ecological Regions (AER) in India. It was observed that Sentinel-1 radar data could identify and map the sowing dates for monsoon crops at high accuracy. The classification accuracy obtained for monsoon crop mapping by combining radar and optical data was 93%. The proposed method for generating monsoon crop map using ROM-derived crop mask is particularly effective in regions with cropland mixed with plantation/mixed forest, typical of small-scale farms in tropical regions.