Seasonal soil-salinity mapping improves when satellite data and ML are matched to conditions

A study using 580 soil samples in central Shaanxi compared geostatistical and machine-learning approaches. The authors report that an artificial neural network performed best in winter (R² 0.76), while random forest combined with Sentinel-2 data performed best in summer (R² 0.89).

Why it matters: It supports season-aware remote-sensing pipelines for agricultural land and water management. Results are regional and require validation across other soils, climates and collection practices.

Paper: Scientific Reports