Accurate evapotranspiration (ET) estimation is vital for the ecology and water resource management of arid and semiarid regions. This study focused on the Kolqin Sandy Land Inland Closed Basin. The simulation of energy flux data for the area was conducted by applying Landsat 8 data from 2018 to 2022 during the growing season, the Two-Source Energy Balance (TSEB), and the Mapping Evapotranspiration at High Resolution with Internalized Calibration (METRIC) models. The validation of both models was performed via eddy covariance systems and large-aperture scintillometers, and the applicability of both models was compared. At the ecosystem scale, the TSEB model exhibited excellent accuracy of latent heat flux simulation in dune and meadow ecosystems (R²>0.8, RMSE<45.74 W/m2, and MAPE<26%), while the METRIC model excelled in meadow ecosystems (R2 = 0.89, RMSE = 27.79 W/m2, and MAPE = 12%). In contrast, at the regional scale, the TSEB model outperformed METRIC for sensible heat flux simulation (R2: 0.65-0.68, RMSE: 22.83-43.28 W/m2, MAPE: 14%-17%). The TSEB model ranked evapotranspiration in the order of lakes>meadow wetlands>farmland>dunes, peaking in August and following a seasonal pattern of summer>spring>autumn. Additionally, evapotranspiration varied with LAI during the pre-middle growth period and responded significantly to △T in the late vegetation period.

Land evapotranspiration (ET) is essential for the hydrological cycle and surface energy balance. Investigating the spatiotemporal evolution and response to land cover changes is of significance in socioeconomic development and regional water resource management. However, consistent estimation of ET is challenging due to “space-time” conflicts in optical remote sensing data and susceptibility to cloud contamination. This study adopted the GF-SG model for reconstructing high-resolution NDVI time-series data from the Yellow River Basin of Inner Mongolia. These data were then input into the Priestly-Taylor Jet Propulsion Laboratory (PT-JPL) model for improving the resolution of ET retrieval. ET was estimated at an 8-day 30m resolution from 2000 to 2022, and its spatiotemporal patterns were analyzed. The model sensitivity parameters were optimized and validated based on in-situ observations from the eddy covariance stations. The optimized PT-JPL model demonstrated excellent simulation results in the basin, with calibration period accuracy ranging from the R2 of 0.88 to 0.91, RMSE between 0.57 and 0.60 mm/d, and MAE from 0.38 to 0.46 mm/d. The validation period accuracy ranged from the R2 of 0.85 to 0.87, RMSE between 0.54 and 0.72 mm/d, MAE from 0.36 to 0.47 mm/d. Over the past 23 years, the mean basin-wide ET has been 304.18 mm, with an increasing trend of 2.587 mm/year. Spatially, ET exhibited west-low and east-high distributions. Across the basin, most (82%) of the trends in ET change were not significant, with noticeable increases mainly in the eastern regions, such as the Daheihe and Hunhe River Basins. Future ET trends mainly showed increasing or non-continuing patterns. The sequence of ET increase in various land use transfer areas was transfer to forest land > transfer to cropland > transfer to construction land > transfer to grassland > transfer to unused land.

Net primary productivity (NPP) is a critical indicator of carbon sequestration and biomass accumulation in terrestrial ecosystems, directly reflecting ecosystem carbon sink capacity. Existing NPP studies have primarily emphasized climate-driven interannual variability. Spatially explicit analyses that jointly quantify multi-factor driving mechanisms, thresholds, and land-use transition risks remain limited. Here, we develop an integrated multi-method analytical workflow (DRIBS) that integrates Distributional Response, Informative Boundary constraints, and Spatial Interpretability Optimization, and apply it to the Jiziwan region in the Yellow River Basin, one of China’s major ecological restoration hotspot regions. From 2000 to 2020, the annual increasing rate of NPP was 5.80 gC·m⁻²·yr⁻¹, and 78% of the area showed a significant increasing trend. Among them, grasslands and croplands in the eastern and western parts exhibited strong fluctuations and low long-term stability. Evapotranspiration (ET) and fractional vegetation cover (FVC) were the dominant drivers of NPP spatial heterogeneity, and precipitation around ~220 mm marked a critical water-stress threshold. Population density and nighttime lights showed a non-linear “ecological adaptation window”, implying both disturbance and management potential. Land-use transitions exhibited divergent risk signatures: grassland/cropland-to-forest transitions produced stable enhancement (priority restoration zones), whereas cropland/unused-to-urban transitions were associated with degradation risk (urgent management). Overall, DRIBS provides an interpretable “change-mechanism-threshold-risk” assessment to support carbon-sink regulation and restoration prioritization in arid and semi-arid regions.