Potential evapotranspiration (PET) is an important input variable of many ecohydrological models, but commonly used empirical models usually input numerous meteorological factors. In consideration of machine learning for complex nonlinear learning, we evaluated the applicability of three machine learning algorithms in PET estimation in the Yellow River basin (YRB), in addition to determining significant factors affecting the accuracy of machine learning. Furthermore, the importance of meteorological factors at varying altitudes and drought index grades for PET simulation were evaluated. The results show that the accuracy of PET simulation in the YRB depends on the input of various meteorological factors; however, machine learning models including average temperature (T_mean) and sunshine hours (n) as input achieved satisfactory accuracy in the absence of complete meteorological data. Random forest generally performed best among all investigated models, followed by extreme learning machine, whereas empirical models overestimated or underestimated PET. The importance index shows that T_mean is the most influential factor with respect to PET, followed by n, and the influence of T_mean on PET gradually decreased with increased altitude and drier climate, whereas the influence of n shows the opposite trend.

Based on MODIS NDVI and a meteorological dataset, this study analyzed the spatial and temporal variation characteristics of vegetation cover in different geomorphic zones of Loess Plateau (LP) from 2000 to 2020 with trend analysis, partial correlation, residual analysis and the CA–Markov method and discussed the driving factors. The research results show that: (1) There are spatial differences in vegetation coverage in different geomorphic regions. The Loess Hills and Forests zone (LF) exhibits the highest coverage, with a multi-year average of 86.64%, and the Arid Grassland (AG) has the poorest vegetation with only 8.53%. Overall, there has been significant improvement in vegetation coverage over the past two decades, although certain geomorphic zones, particularly the Highland Steppe zone (HS) and Alluvial Plains zone (AP), show signs of degradation. (2) Relative humidity has the greatest impact on vegetation among the three climate factors, i.e., relative humidity, precipitation and temperature. Relative humidity predominantly promotes vegetation in all geomorphic zones. Temperature generally inhibits vegetation growth, except in the Wind Sandy zone (WA) and AG. The impact of precipitation on vegetation depends on the region. A lag effect is observed, with temperature and humidity showing a one-month lag and precipitation showing a two-month lag on vegetation response. (3) Human activities play a crucial role in promoting vegetation, particularly in the WA zone, in which the percentage of area where human activities contribute to vegetation has changed from 13.80% to 86.85%, an increase of 73.05%, while the HS experiences an inhibitory effect due to overgrazing and water resource overutilization. Similarly, the AP zone’s vegetation growth is hindered by urban development and land use changes. (4) Land use change significantly impacts vegetation dynamics on the LP. Over the past two decades, the area of forest lawn increased by 122,800 km², which is 1.5 times more than the area of reduction. However, conversion to building land has hindered vegetation growth in certain regions. A comprehensive strategy is required to conserve land resources and promote healthy vegetation growth on the LP.

Prediction of whole-plant short-term water use efficiency (WUE_s,P) is essential to indicate plant performance and facilitate comparison across different temporal and spatial scales. In this study, an isotope model was scaled up from the leaf to the whole-plant level, in order to simulate the variation in WUE_s,P in response to different CO₂ concentrations (C_a; 400, 600, and 800 μmol·mol⁻¹) and soil water content (SWC; 35–100% of field capacity). For WUE_s,P modelling, leaf gas exchange information, plant respiration, and “unproductive” water loss were taken into account. Specifically, in shaping the expression of the WUE_s,P, we emphasized the role of both stomatal (g_sw) and mesophyll conductance (g_m). Simulations were compared with the measured results to check the model’s applicability. The verification showed that estimates of g_sw from the coupled photosynthesis (P_n,L)-g_sw model accounting for the effect of soil water stress slightly outperformed the model neglecting the soil water status effect. The established coupled P_n,L-g_m model also proved more effective in estimating g_m than the previously proposed model. Introducing the two diffusion control functions into the whole-plant model, the developed model for WUE_s,P effectively captured its response pattern to different C_a and SWC conditions. Overall, this study confirmed that the accurate estimation of WUE_s,P requires an improved predictive accuracy of g_sw and g_m. These results have important implications for predicting how plants respond to climate change.

The mechanism of soil detachment on steep slopes is obviously different from that on gentle slopes. However, the slope effect of soil detachment remains unclear. The objective of this study was to quantify the slope effect of soil detachment capacity at the varying hydrodynamic characteristics. In this study, the soil detachment capacity (Dc) on clay loam and hydrodynamic characteristics were measured by conducting the runoff scouring experiments at 10 slope gradients (1.7–57.7%) and 5 unit flow discharges (0.022–0.089 m²·min⁻¹). The results showed that the relationships between Dc and hydrodynamic parameters were affected by slope gradient. Based on the optimal functional relationship, the hydrodynamic characteristics (flow velocity, flow shear stress, stream power, unit stream power, and unit energy) calculated by maximum and minimum Dc in this study changed by 19.91–95138.10%, and the Dc calculated by the maximum and minimum hydrodynamic characteristics could differ by up to nine orders of magnitude. Overall, the power function of hydrodynamic parameters was superior to the linear function in different slope gradients. The stream power was the best predictor for Dc compared with other hydrodynamic parameters. For all combinations of slope gradients, the adjusted coefficient of determination (Adj. R²) of the power relationship between Dc and stream power was 9.41–27.40% higher than it was between Dc and other hydrodynamic parameters. The coefficient and index of power function for different hydrodynamic parameters showed a trend change with increasing slope gradient, indicating that there was a slope effect on Dc. Further analysis found that Dc could be well predicted using a power combination equation of slope gradient, flow velocity, and flow depth (Adj. R² = 0.96). This study helps to better understand the mechanism of soil detachment and emphasizes that the slope effect should be considered when establishing a soil detachment equation.