Context: The Songnen Plain is a crucial grain-producing region in China, and its western area, influenced by natural conditions and long-term human activities, faces severe issues of soil salinisation. In recent years, soil organic carbon (SOC) in saline-alkali soils has gained increasing attention as a material foundation for grain production and an essential carbon sink for mitigating climate change. However, the driving factors behind regional-scale changes in SOC in saline-alkali soils remain unclear. Aims: This research was conducted to assess changes in soil organic carbon stocks and identify main driving factors in saline soils at 40-years scale. Methods: Taking Tongyu County in the western part of the Songnen Plain as a case study, this research explored the spatiotemporal variation of soil organic carbon density (SOCD) and soil organic carbon storage (SOCS) from 1982 to 2022 using GIS. Random forest regression and structural equation modelling were employed to identify environmental factors influencing SOCD distribution in different soil layers. Key results: (1) From 1982 to 2022, the average SOCD in the surface, subsurface, and bottom soil layers decreased by 0.65, 0.34, and 0.46 kg m−2, respectively, resulting in a total carbon storage reduction of 12.93 Tg C. (2) In 1982, the vertical distribution of SOCD was higher at the top and bottom layers and lower in the middle; however, by 2022, it exhibited a gradual decrease layer by layer. (3) Topographic factors only influenced surface SOCD, while the influence of environmental humidity and land use on SOCD decreased with increasing depth. Conclusions: These findings provided valuable scientific insights for implementing regional soil carbon sequestration and soil nutrient conservation measures.

Data uncertainty and limited model generalization remain critical bottlenecks in large-scale remote sensing of soil salinization. Although the integration of multi-source data has improved predictive potential, conventional deterministic feature selection methods often overlook stochastic noise inherent in environmental variables, leading to models that overfit spurious correlations rather than learning stable physical signals. To address this limitation, this study proposes a Bootstrap–Boruta feature stability assessment framework that shifts feature selection from deterministic “feature importance” ranking to probabilistic “feature stability” evaluation, explicitly accounting for uncertainty induced by data perturbations. The proposed framework is evaluated by integrating stability-driven feature sets with multiple machine learning models, including a Back-Propagation Neural Network (BPNN) optimized using the Red-billed Blue Magpie Optimization (RBMO) algorithm as a representative optimization strategy. Using the Minqin Lake region as a case study, the results demonstrate that the stability-based framework effectively filters unstable noise features, reduces systematic estimation bias, and improves predictive robustness across different modeling approaches. Among the tested models, the RBMO-optimized BPNN achieved the highest accuracy. Under a rigorous bootstrap validation framework, the quality-controlled ensemble model yielded a robust mean R² of 0.657 ± 0.05 and an RMSE of 1.957 ± 0.289 dS/m. The framework further identifies eleven physically robust predictors, confirming the dominant diagnostic role of shortwave infrared (SWIR) indices in arid saline environments. Spatial mapping based on these stable features reveals that 30.7% of the study area is affected by varying degrees of soil salinization. Overall, this study provides a mechanism-driven, promising, within-region framework that enhances the reliability of remote-sensing-based soil salinity inversion under heterogeneous environmental conditions.

In arid and semi-arid regions, severe salinity imposes strong abiotic constraints on farmland restoration. This study evaluated the one-season effects of organic amendments on soil quality, maize productivity, and short-term ecosystem multifunctionality (EMF) responses under severe salinity stress. We conducted a field experiment with biofertilizer (targeting plant–soil biological regulation) and composted manure (targeting direct soil amelioration) applied at different rates. The high-rate composted manure treatment (T6) showed the largest short-term improvements: compared with the chemical-fertilizer-only control under the same irrigation and plastic-mulch management (CK; soil quality index (SQI) = 0.716, yield = 6.657 t ha⁻¹, EMF_a = 0.456), SQI increased by 165%, maize yield increased by 41%, and EMF_a increased by 104% relative to CK within one growing season. Partial least squares path modeling (PLS-PM) suggested that under biofertilizer treatments, a plant-related association pathway was observed but relatively weak (β = 0.215), whereas under composted manure treatments, EMF variation was mainly associated with soil quality improvement (β = 0.992). Overall, these short-term results suggest that, under the tested application rates and severe salinity stress, high-rate composted manure can more effectively improve baseline soil conditions than biofertilizers during the initial season. These findings offer a preliminary conceptual perspective for a phased management strategy, serving strictly as a preliminary hypothesis where priority is given to soil amelioration in the initial phase and gradual integration of biologically oriented interventions as baseline conditions improve. However, future multi-year and multi-site studies are strictly required to validate the long-term viability of this proposed framework and to test whether these association patterns persist across longer time scales and broader regional contexts.