Random Forest for rice yield mapping and prediction using Sentinel-2 data with Google Earth Engine

Accurate information on crop yield prediction is essential for farmers, governments, scientists, and agricultural agencies to make well-informed decisions. Majority of yield prediction methods have been based on data assimilation, which incorporates consecutive obser-vation of canopy development from remote sensing data into model simulations of crop growth processes. But this study used high res-olution Sentinel-2 data with combination of different types of secondary data in Random Forest (RF) regression model on different phases of the crop growing season for higher accurate rice yield prediction. For that First, computed crop/non-crop and rice/non-rice crops through RF classifiers were applied on seasonal median composites of Sentinel-2 data for each pixel in the region. Thousands of crop/non-crop labels were collected using an in-house google earth engine (GEE) labeler, and several crop type labels were obtained from various sources during the crop growing seasons. Results demonstrate that sentinel-2 imagery is useful to detect crop/non-crop classes from cropland with more than 85% accuracy, thus it can be used for crop prediction. Furthermore, the Sentinel-2 imagery with secondary data such as environmental, soil and topographic data perform higher accuracy for yield prediction. Its show 0.40 to 1.01 t/ha yield production range at a landscape level. Overall, this study illustrates the Sentinel-2 imagery, GEE platform, advanced classification and rice yield mapping algorithms are enhance the understanding of precision agricultural systems.(c) 2022 COSPAR. Published by Elsevier B.V. All rights reserved.