A novel ANN-feature-enhanced nested machine learning method for soil salinity retrieval from Landsat data

Soil salinization poses a growing threat to agricultural productivity and food security in irrigated regions. To support sustainable land management in these vulnerable areas, this study developed an Artificial Neural Network (ANN) Feature-Enhanced Nested Machine Learning framework for soil salinity mapping using Landsat satellite imagery in the Lower Yellow River Irrigation District (LYRID), one of China’s most important grain production zones. The proposed framework effectively leverages the advantages of ANN to enhance feature representation, thereby significantly improving model accuracy under limited-sample conditions. The support vector regression model constructed using ANN-enhanced features (AEF_SVR) achieved high predictive performance (R² = 0.72 ± 0.05, RMSE = 2.83 ± 0.35 dS/m, and MAE = 1.77 ± 0.13 dS/m), showing a clear advantage over both the replicated feature-space model and existing global soil salinity mapping products. The salinity mapping results indicate a distinct southwest–northeast gradient, with approximately 41.72% of salinized croplands concentrated near the Yellow River Delta, likely influenced by shallow groundwater and marine intrusion. Meanwhile, localized salinity risks persist in several inland irrigation zones, underscoring the spatial heterogeneity of soil salinization in the region. These findings demonstrate that the proposed framework provides a reliable and interpretable tool for precision salinity monitoring and targeted soil management, offering scientific support for salinization control and sustainable irrigation practices in the LYRID and other vulnerable agricultural regions. • A novel nested learning framework with feature enhancement was developed. • The framework improved salinity retrieval under limited sample conditions. • The framework captured regional salinity gradients and localized inland hotspots. • The framework outperformed two existing approaches in the study area. • The maps support differentiated irrigation and salinity management.