Estimation of regional-scale maize plant nitrogen content based on multi-source remote sensing data

This study aims to systematically analyze the challenges of water scarcity and low nitrogen use efficiency in maize production in the arid Hexi Corridor. It provides a scientific basis for efficient water and fertilizer management. This study innovatively integrates multi-source data from satellite and Unmanned Aerial Vehicle (UAV) remote sensing. The datasets include Sentinel-2A imagery, UAV-based multispectral images, and ground-based observations. Based on these data, a comprehensive data fusion framework was established. Data were collected across four key growth stages of maize in 2024, with 66 sampling points established in the main experimental area and 48 sampling points in the auxiliary validation area for model training and validation. Pearson correlation analysis was employed to identify the optimal combination of vegetation indices (VIs). The inversion accuracy of various models at different growth stages was systematically analyzed. Notably, a novel region-scale maize Plant Nitrogen Content (PNC) inversion method based on band correction was proposed. This method not only achieves the harmonization of multi-source remote sensing data but also optimizes the PNC inversion at the regional scale, accounting for inter-sensor spectral response differences and leveraging multi-growth-stage data to enhance the model's robustness and generalization capability. Furthermore, the applicability and reliability of this model for crop growth monitoring in arid regions were thoroughly evaluated. The results showed that: (1) The PNC prediction model based on Convolutional Neural Networks (CNN) demonstrated significant performance advantages. It achieved a coefficient of determination (R²) of 0.80. Compared with traditional machine learning models, such as Support Vector Machines (SVM) and Random Forest (RF), the prediction accuracy improved by more than 10%. (2) Band correction significantly enhanced the modeling performance of Sentinel-2A data in PNC retrieval. The R² of the prediction model increasing from 0.35-0.45 (uncorrected) to 0.70-0.80. This confirmed the positive impact of band correction on model accuracy. (3) The prediction accuracy in the auxiliary validation area was highly consistent with that in the main validation area, further confirming the stability and reliability of the proposed method under varying regional conditions. This study provides an effective approach for rapid and precise monitoring of maize nitrogen status in arid regions. It also offers scientific support for regional-scale crop nitrogen management and precision fertilization decisions. The findings have significant theoretical and practical implications.