The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Articles | Volume XLVIII-M-1-2023
21 Apr 2023
 | 21 Apr 2023


J. D. Mohite, S. A. Sawant, A. Pandit, R. Agrawal, and S. Pappula

Keywords: Yield Forecasting, Soybean Crop, Remote Sensing, Weather Data, Random Forest Regression

Abstract. The main objective of this study is the in-season forecasting of soybean crop yield using the integration of satellite remote sensing and weather observations. The study was carried out in the Paran´a state of Brazil. The soybean crop in the study region is sown during Oct.–Nov. month and harvested between Feb.–Mar. of the next year. Municipality-level soybean yield data for 15 municipalities was obtained from the AGROLINK portal of Brazil, from the 2005–06 season to the 2020–21 season. The crop yield data constituted yearly municipality-wise yield in kg/ha. Remote sensing-based indicators such as the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EVI) and Land Surface Temperature (LST), and Rainfall data from CHIRPS was considered in the study. Regression modelling was carried out between municipality-level yield as the dependent variable and features generated from remote sensing and weather observations as independent variables. Performance evaluation of tuned random forest regression (RFR) and tuned support vector regression (SVR) were performed against multiple linear regression (MLR). A comparison of results in terms of algorithms shows that RFR performed better than SVR and MLR. Further, a rootmean- square-error (RMSE) of 414 kg/ha and an R2 value of 0.748 were achieved by the best RFR model. Validation of developed RFR model was performed on the data from the new soybean season, i.e., 2020–21. We have achieved an R2 value of 0.693 with a RMSE of 585 kg/ha. Although the model performance on the data of 2020-21 season is slightly reduced, R2 and RMSE are in good agreement with test results. This study showed that, integration of remote sensing and weather observations would be useful for in-season yield forecasting of soybean at municipality level.