Nitrogen Monitoring and Sugar Yield Estimation Analysis of Sugar Beet Based on Multisource and Multi-temporal Remote Sensing Data

This study aimed to explore the potential of multisource and multi-temporal UAV remote sensing data for sugar yield estimation and to investigate the relationship between different remote sensing features and nitrogen accumulation at various growth stages. UAV hyperspectral images, RGB images, and light detection and ranging (LiDAR) data were collected at different growth stages, and a comprehensive set of spectral, structural, and textural features reflecting the sugar beet canopy were extracted. Three machine learning algorithms, including multiple linear regression (MLR), random forest (RF), and support vector machine (SVM), were used to construct prediction models for nitrogen accumulation and sugar yield. The results showed the following. LiDAR features and textural features that characterize the canopy structure of sugar beet are essential for reflecting nitrogen accumulation, and LiDAR features play a key role in sugar yield prediction. For nitrogen accumulation prediction, the MLR model performed best during the rapid foliage growth period (R² = 0.70, RMSE = 0.44 ). For sugar yield prediction, the MLR model, when combined with multi-temporal data, achieved the highest accuracy (R² = 0.95, RMSE = 0.16), which was 21% higher than the best single-phase prediction result (sugar accumulation stage). The collaborative use of multisource remote sensing data significantly improved accuracy compared to single data sources, with nitrogen estimation accuracy increasing by 55% and sugar yield estimation accuracy increasing by 28%. These findings indicate that multisource remote sensing data can be used to diagnose nitrogen nutrition and predict sugar yield in sugar beet.