Modelling of Alfalfa Yield Forecasting Based on Earth Remote Sensing (ERS) Data and Remote Sensing Methods

Abstract

This study aims to develop a method for modelling early forecasting of alfalfa yield on a farm scale located in East Kazakhstan. The authors evaluated the correlation coefficient between forage crop yield and different data sets, including weather data, climate indices, spectral indices from drones and satellite observations. An ensemble machine learning model was developed by combining three commonly used basic training modules: random forest (RF), support vector method (SVM), and multiple linear regression (MLR). It is found that the best yield prediction algorithm in this study is the Random Forest (RF) algorithm, which predicts yields with R2 = 0.94 and RMSE = 0.25 t/ha. The results of this study showed that combining remote sensing drought indices with climatic and weather variables from UAV and satellite imagery using machine learning is a promising approach for alfalfa yield prediction.