WHEAT AREA MAPPING AND PHENOLOGY DETECTION USING SYNTHETIC APERTURE RADAR AND MULTI MULTI-SPECTRAL REMOTE SENSING OBSERVATIONS
Keywords: Wheat area mapping, Sentinel-1, Sentinel-2, Random Forest, Classification, Google Earth Engine
Abstract. In season crop area mapping is of significant importance for multiple reasons such as monitoring if crop health and residue burning areas, etc. Wheat is one of the important cereal crop cultivated all across the India, with Punjab-Haryana being the prime contributors to the total production. In this study we propose a method for early season Wheat area mapping using the combined use of temporal Sentinel-1 and 2 observations. Further, we propose a method to estimate the crop phenology parameter viz. sowing date using the early time series of Normalized Difference Vegetation Index (NDVI). Few districts from Haryana and Punjab have been selected. The Wheat sowing starts in month of Oct.–Nov. Considering the sowing window, images available during Oct.–Dec. 2017 have been chosen for early season Wheat area mapping. The field data for Wheat, other crops, forest, water and settlements classes is gathered using human participatory sensing and Google Earth Engine (GEE) platform and used for data analysis. We have assessed the performance of random forest classifier using 1. NDVI derived from Sentinel-2, 2. VV and VH backscatter obtained from Sentinel-1 and 3. Both NDVI and VV-VH backscatter. Results show the maximum classification accuracy of 88.31 % when using combination of NDVI, VV and VH. However, accuracy drops to 87.19 % and 79.16 % while using NDVI and VV-VH respectively. Further, to estimate the sowing date we have considered the NDVI time-series during Oct.–Dec. for Wheat pixels. A method based on NDVI compositing is used with gradual increase of 0.1–0.15 at every 12 days for subsequent two images. We have found a good agreement between the estimated sowing dates and actual sowing dates.