Identifying Rice Transplantation Dates using Sentinel-1 Synthetic Aperture Radar data and Machine Learning

Abstract

This study evaluates the effectiveness of Sentinel-1 Synthetic Aperture Radar (SAR) data, acquired at a 10-m spatial resolution with 12-day intervals, for accurately mapping rice transplantation dates during the Kharif season of 2023 in Kakinada district, Andhra Pradesh. SAR’s ability to penetrate clouds and provide all-weather observations was particularly valuable for tracking rice cultivation under monsoon conditions. The study integrated the Random Forest algorithm to classify rice crop pixels, achieving precise identification of transplantation dates using Google Earth Engine (GEE) platform. A comparison between satellite-based estimates and the Agriculture Department's records demonstrated strong alignment. Sentinel-1 SAR observations from 28^th July, 9^th August, 21^st August, and 2^nd September 2023 closely matched Agriculture Department records from 2nd August, 9th August, 23rd August, and 6^th September. On 2^nd August, satellite data estimated 44,672.42 hectares, compared to the Department’s 43,414 hectares with a 2.9% deviation. By 9^th August, satellite estimates were 61,199.22 hectares, while the Department’s estimation was 62,563 hectares, showing a −2.18% deviation. By 23^rd August, estimates reached 81,064.94 hectares, with the Department recording 81,889 hectares with a −1.01% difference. Finally, on 6^th September, the satellite estimate was 84,049.53 hectares, closely aligning with the Department’s 83,685 hectares, reflecting a minimal 0.44% deviation. These minor variations, likely due to timing or reporting differences, underscore the reliability of SAR data for near-real-time monitoring. Accurately identifying transplantation dates and mapping is crucial, as it significantly aids in the precise estimation of rice crop biomass, which is a key parameter for forecasting rice yields.