Estimation Winter Wheat Crop Yield through Multi-Modal Analysis using Satellite Imagery and Meteorological Data

In recent years, satellite and meteorological data have been extensively used separately as a single modality data source for crop yield prediction tasks, however, leveraging the multi-modality aspect by combining these two data sources can help us gain new insights and improve yield prediction accuracy. In this study, we follow a multi-modality approach, we examined and expanded upon the use of nine different satellite data types and five common climate variables to assess their impact on yield prediction. We compared the accuracy of various machine learning models and exponential function combination equations, ultimately identifying an efficient method for determining the optimal coefficients for yield prediction. The satellite data is obtained from the Sentinel-2 satellite, which includes NDVI, NDRE, EVI, NMDI, GNDVI, GDVI, GVMI, NDMI, and NDII across various winter wheat growth stages. Meteorological parameters encompass air temperature, rainfall, wind speed, relative humidity, and solar radiation. We employed five machine learning models: Support Vector Regression (SVR), Random Forest, Gradient Boosting Regression, AdaBoost Regression, and Histogram-based Gradient Boosting Regression. Additionally, we developed an Exponential Functions Least Squares Matrix (EFLSM) model for yield prediction using these datasets. The EFLSM model achieved the highest R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.771, while the Gradient Boosting Regression model recorded the lowest RMSE of 0.958. These two models demonstrated the best performance in predicting winter wheat yield.