Data-driven mapping of soil salinity in the Red River Delta: Sentinel-2 MSI and machine learning integration

Résumé fourni par la source

Understanding the spatial and temporal distribution of soil salinity is essential for assessing land degradation and developing effective mitigation strategies to support sustainable local development. However, previous research has predominantly focused on assessing soil salinity at isolated points in time, with limited attention given to its temporal variability. This gap is especially critical in the Red River Delta (RRD) in Vietnam, one of the regions most severely affected by salinization, posing challenges to agricultural productivity, food security, and environmental sustainability. Our study focuses on monitoring and assessing both the spatial and temporal distribution of soil salinity in the RRD, particularly in Tien Hai District, Thai Binh Province. To achieve this, we integrated remote sensing data from Sentinel-2A imagery with advanced machine learning algorithms, including CatBoost (CBR), AdaBoost (ADB), XGBoost (XGR), Decision Trees (DTR), and Gradient Boosting Regression (GBR), to model and predict salinity patterns over time. A total of 106 soil salinity samples were collected during the rainy season of May 2023 and the dry season of March 2025. Simultaneously, 45 environmental and spectral variables were extracted to serve as input for model training. Among the model tested, CBR achieved the highest performance with an R2 value of 0.823, followed by GBR (0.802), ADB (0.796), XGR (0.757), and DTR (0.746). Spatial mapping results indicated that in May 2023, approximately 71.9% of the area was in non-saline, 12.78% slightly saline, and 15.3% strongly saline. By March 2025 these figures shifted to 63.28%, 21.62%, and 15.01% respectively.

Ce résumé expose les affirmations des auteurs. BNTIC ne l’interprète pas comme une validation indépendante des résultats.