Wheat Yield Prediction based on Sentinel-2, Regression and Machine Learning Models in Hamedan, Iran

An accurate forecast of wheat yield prior to harvest is great importance ensure the sustainability food production in Iran. The primary objective this study determine best remote sensing features and regression model for prediction Hamedan, In addition, effects various time windows on different models are verified. For purpose, several vegetation indices (VIs) reflectance values obtained from Sentinel-2, as input models, used windows. As a result, Gaussian process (GPR) random forest (RF) represented top two methods, results were achieved GPR with SAVI, NDVI, EVI2, WDRVI, SR, GNDVI GCVI corresponding image captured at end May. yielded root mean square error (RMSE) 0.228 t/ha coefficient determination R^2 = 0.73. Moreover, methods regarding number training data compared. neural network linear most stepwise was affected least by samples. Experimental provide technical reference estimating large scale yield.