Precise Apple Yield Prediction Utilizing Differential Fusion of UAV and Satellite Multispectral Images

Accurate regional apple yield estimation is vital for orchard planning and management. However, background elements such as soil and shadows in satellite imagery often hinder the reliable extraction of canopy spectral information. This study proposes an integrated framework combining Unmanned Aerial Vehicle (UAV) and Sentinel-2 data to improve apple yield prediction. The Normalized Difference Canopy Shadow Index (NDCSI) was applied to UAV multispectral imagery to suppress background noise, while Non-negative Matrix Factorization (NMF) was used to fuse yield-sensitive variables from both UAV and satellite data. Four predictive models—Partial Least Squares Regression (PLSR), Support Vector Machine (SVM), Random Forest (RF), and Back Propagation Neural Network (BPNN)—were constructed and validated using field survey data from Qixia in 2023 and 2024. Results showed that NDCSI effectively enhanced reflectance in the red-edge and near-infrared bands. After variable fusion, the correlation with yield improved by 0.06–0.13. Among the models, the RF model based on fused variables achieved the highest accuracy, with a validation R² of 0.84, nRMSE of 0.14, and RPD of 2.01. Compared to the non-fusion model, R² and RPD increased by 0.12 and 0.64, respectively, demonstrating superior accuracy and robustness. This study provides a technical reference for multi-source remote sensing-based yield prediction in orchards.