Model performance and variable-importance data for DRIS-based UAV-LST analysis

This dataset contains the model performance and variable-importance results underlying the random forest and generalized additive model analyses reported in the associated manuscript, "Representing the spatiotemporal modulation of pedestrian-scale urban thermal environments by dynamic radiative processes." The dataset includes three components: (1) test-set performance metrics (R², RMSE, MAE) and permutation-based variable-importance scores for the nested and structurally matched random forest models (Models A–I, C3, C4) built on 64,025 object-based near-ground thermal units extracted from 2024 UAV thermal imagery; (2) cross-validated performance metrics and variable-importance scores for the ground-shadow and entity-based obstruction models applied to an independent 2022 near-ground surface-temperature dataset (N = 221); and (3) summary statistics from the generalized additive model fitted to 66 synchronized 2024 UAV–ground matched samples, examining the association between entity-based obstruction indicators (JOI, TM_JOI) and the temperature difference between top-view UAV-derived and ground-measured near-ground surface temperatures. These results correspond to Table 3 and Figs. 3–5 of the manuscript