Optimizing irrigation and nitrogen (N) management in saline soils is critical for sustainable cotton production in arid regions that have been subjected to climate change. In this study, a two-year factorial field experiment (3 salinity levels × 3 N rates × 3 irrigation quotas) is integrated with the RZWQM2 model to (1) identify water–N–salinity thresholds for cotton yield and (2) to project climate change impacts under SSP2.4-5 and SSP5.8-5 scenarios (2031–2090) in Xinjiang, China, a global cotton production hub. The results demonstrated that a moderate salinity (6 dS/m) combined with a reduced irrigation (3600 m³/hm²) and N input (210 kg/hm²) achieved a near-maximum yield (6918 kg/hm²), saving 20% more water and 33% more fertilizer compared to conventional practices. The model exhibited a robust performance (NRMSE: 5.94–12.88% for soil–crop variables) and revealed that warming shortened the cotton growing season by 1.2–9.5 days per decade. However, elevated CO₂ (832 ppm by 2090) levels under SSP5.8-5 increased yields by 22.6–42.1%, offsetting heat-induced declines through enhanced water use efficiency (WUE↑27.5%) and biomass accumulation. Critically, high-salinity soils (9 dS/m) required 25% additional irrigation (4500 m³/hm²) and a full N input (315 kg/hm²) to maintain yield stability. These findings provide actionable strategies for farmers to optimize irrigation schedules and nitrogen application, balancing water conservation with yield stability in saline-affected arid agroecosystems that have been subjected to climate change.

Net primary productivity (NPP) reflects the carbon sequestration capacity of terrestrial ecosystems and it is used as an important indicator for measuring ecosystem quality. However, due to the effects of “warming and humidification” and “oasisization”, the spatiotemporal evolution and driving mechanisms of the NPP of vegetation in the northern slope of the Tianshan Mountains (NSTM), a typical arid area in China, are still unclear. Thus, in this study, we used remote sensing data and meteorological data to construct a Carnegie–Ames–Stanford–Approach (CASA) model for estimating the NPP of vegetation in the study area. Trend analysis, partial correlation analysis, and optimal parameter-based geographic detector (OPGD) methods were combined to explore the spatiotemporal evolution and driving mechanisms to changes in the NPP. The results showed that from 2001 to 2020, the annual average NPP on the NSTM exhibited an overall significant upward trend, increasing from 107.33 gC⋅m⁻²⋅yr⁻¹ to 156.77 gC⋅m⁻²⋅yr⁻¹, with an increase of 2.47 gC⋅m⁻² per year and 46.06% year-on-year. Over the past 20 years, climate change and human activities generally positively affected the changes in NPP in the study area. Human activities in the study area are mainly manifested in the large-scale conversion of other land use types into farmland, with a total increase of 16,154 km² in farmland area, resulting in a net increase of 6.01 TgC in NPP. Precipitation has the strongest correlation with NPP in the study area, with a partial correlation coefficient of 0.30, temperature and solar radiation have partial correlation coefficients with NPPs of 0.17 and 0.09, respectively. Therefore, increases in precipitation, temperature, and solar radiation have a promoting effect on the growth of NPP on the NSTM. During the study period, the land use type and soil moisture were the main factors that affected the spatial differentiation of vegetation NPP, and the effects of human interference on natural environmental conditions had significant impacts on vegetation NPP in the area. Therefore, in this study, we accurately determined the spatiotemporal variations in the NPP on the NSTM and comprehensively explored the driving mechanisms to provide a theoretical basis for sustainable development in arid areas and achieving carbon neutrality goals.

Riverine–lacustrine ecosystems in river–lake continua face increasing threats, yet conventional vulnerability maps often overlook local degradation drivers. This study presents an advanced satellite-based mapping framework using Deep Attention Networks (DANets) for accurate, interpretable vulnerability assessment. In the Ebinur Lake Basin, a representative dryland river system, we first built a satellite-derived evidence map of ecosystem stress aligned with the IPCC’s vulnerability definition. We then optimized DANets via two nature-inspired algorithms: Genetic Algorithm (GA) and Grey Wolf Optimizer (GWO). The optimized models demonstrated strong predictive capacity, explaining a large share of vulnerability variance (R² = 0.78 for GA-DANets; R² = 0.76 for GWO-DANets). For high/low-vulnerability discrimination, GWO-DANets was most effective and stable, with a mean AUC = 0.960 ± 0.044. Factor importance analysis identified soil organic carbon (SOC; 0.29), precipitation seasonality (0.24), and aridity (0.22) as dominant drivers. Two distinct pathways emerged: chronic degradation in arid plains, driven by low SOC and poor water retention; and acute hydrological stress in wetlands, where carbon-rich soils are sensitive to drying. This insight shifts management from uniform to targeted approaches: soil restoration in plains and water-flow protection in wetlands. By integrating metaheuristically optimized deep learning with multi-sensor satellite data, the framework offers a scalable decision-support tool for safeguarding water-dependent ecosystems. The study confirms that vulnerability in the basin follows two predictable, process-based trajectories, which can be directly linked to measurable soil and hydrological conditions. These clear patterns allow managers to prioritize interventions where they will have the greatest effect under ongoing climate pressure.