A growing wave of urban thermal studies has used machine learning models and post-hoc explanation tools, particularly Shapley additive explanations (SHAP), to rank landscape factors and detect nonlinear thermal responses. Yet such predict-then-explain frameworks are designed to explain fitted models rather than real-world interventions, making them vulnerable to spurious correlations, unstable importance rankings, and ambiguous planning implications. This study applies a causal-diagram-informed generalized random forest (GRF) framework to estimate the effects of landscape interventions on land surface temperature (LST) at the city-block scale in Beijing. Built on adaptive local weighting, GRF constructs randomized-controlled-trial-like comparisons among observationally comparable city blocks, enabling the estimation of unit-specific dose-response intervention effects. The results show that replacing impervious surfaces with water bodies, trees, and croplands significantly reduced LST by 0.050 ± 0.023, 0.042 ± 0.011, and 0.034 ± 0.012 °C per 1% increase, respectively, while increasing building coverage raised LST by 0.022 ± 0.014 °C per 1%. Building height, building-height variability, and surface albedo also generated strong cooling effects, whereas all landscape configuration metrics showed no significant effects after accounting for land-cover composition. We also detected strong nonlinear and spatially heterogeneous effects: vegetation, water, and albedo interventions generally showed diminishing marginal cooling returns, and the cooling effects of trees, building-height variability, and albedo were strongest in highly built-up or central urban areas. This GRF workflow can support unit-specific and spatially heterogeneous landscape strategies for targeted urban heat mitigation.

Drought and nitrogen deposition are major drivers of global change that can influence forest ecosystems and plant–microbe interactions, yet their combined effects on endophytic fungal communities and the roles of dark septate endophytes (DSE) remain unclear. In this study, we examined the diversity of culturable endophytic fungi in leaves and roots of Quercus dentata under different drought and nitrogen deposition treatments and evaluated the functional effects of representative DSE strains on host growth and physiology. A total of 1488 fungal isolates were obtained, revealing distinct organ-specific community patterns. Root-associated communities showed greater compositional stability across treatments, whereas leaf communities were more responsive to environmental variation. Severe drought reduced the dominance of several genera and promoted community restructuring, while nitrogen deposition had contrasting effects on α-diversity in leaves and roots. Beta diversity analyses indicated significant interaction effects between drought and nitrogen addition. Inoculation with four DSE strains produced strain-dependent effects on plant biomass, photosynthesis, water-use efficiency, physiological traits, and nutrient contents. These results indicate that drought and nitrogen deposition jointly influence endophytic fungal communities and that functional differences among DSE strains may affect host responses to combined stress.