Modeling of winter wheat yield prediction based on solar-induced chlorophyll fluorescence by machine learning methods

Timely and accurate prediction of large-scale crop yields is critical for national food security. Solar-induced chlorophyll fluorescence (SIF), an indicator of photosynthesis, has emerged as a promising predictor of crop yields. However, it remains unclear to what extent satellite-based SIF data can predict crop yields at the regional scale compared to the newly proposed Near-Infrared Reflectance of Vegetation (NIRv). Using multiple statistical machine learning (ML) methods, this study investigated the predictive abilities of SIF and NIRv by combining climate data to predict winter wheat yields in five provinces in the North China Plain (NCP). Results showed that: (a) SIF outperformed NIRv in predicting winter wheat yields. However, in the Extreme Gradient Boosting (XGB) model, SIF's predictive performance was better than that of the combination of SIF and NIRv, indicating that combining SIF and NIRv could not completely enhance SIF's predictive performance. (b) Random Forest (RF) and XGB models were significantly better than the other models in yield prediction; specifically, the RF model had high stability. The results highlighted the benefits of combining multiple sources of data and revealed the advantages of RF and XGB models in crop yield prediction in the major grain production region.