Monte Carlo Deep Learning Model for Quantitative Inversion of Total Nitrogen Concentration in the Source Area of the Yellow River Using Google Earth Engine

As a vital ecological barrier in China, the Yellow River source area's water changes significantly impact the regional environment. Traditional remote sensing inversion methods face challenges like limited accuracy and complex data processing. This study uses Sentinel-2 remote sensing data and ground-based hyperspectral data, combined with an improved deep learning model (MC-DL), to establish an efficient framework for key water parameter inversion. Focusing on Ruoergai County, the MC-DL model, enhanced by Monte Carlo dropout, quantitatively inverts total nitrogen (TN) concentration. The MC-DL model outperforms Support Vector Regression (SVR) and Convolutional Neural Network (CNN) in accuracy and stability (R² = 0.95, MAE = 0.08, MBE = -0.004, RMSE = 0.13).This study provides a new technological approach for water monitoring in the Yellow River source area and supports ecological management and protection.