Urbanization altered land surfaces and urban thermal environments. However, previous studies focused on impervious surface expansion and single-scale assessments, overlooking the complex thermal responses affected by various impervious surface trajectories. The impacts of impervious surface changes on the intensification or mitigation of urban warming remain unquantified at different scales. Therefore, this study proposed assessment frameworks integrating MODIS data with machine learning to monitor impervious surface percentage (ISP) across Asia from 2000 to 2023, and classified ISP into four types (increased, unchanged, decreased, and other) to quantify land surface temperature (LST) changes. The multi-scale framework incorporates pixel and city perspectives, evaluating LST trends from four dimensions: diurnal contrast, seasonal patterns, climatic zone differences, and urban-rural variation. Results revealed gradient differences in LST trends induced across ISP trajectories types. Daytime warming with increased ISP reached up to 0.0648 degrees C/year (Terra), approximately 2 times higher than in areas with decreased ISP, which showed weaker or negative trends. Daytime thermal responses were spatially heterogeneous, with intensified warming in East and Southeast Asia, and Aqua showing stronger signals due to its afternoon overpass. Conversely, nighttime warming was more spatially consistent (0.0597 degrees C/year) across different ISP trajectories. Climatic analysis revealed stronger daytime LST gradient differences in temperate regions, while seasonal assessments showed significant responses and spatial heterogeneity in spring and summer. Moreover, some cities in South and Central Asia have opposite trends for different seasons. Urban-rural comparisons confirmed that urban heat island intensity generally increases with ISP expansion but may weaken in arid zones. The framework integrates ISP trajectory and multidimensional thermal assessment to support sustainable land management and understand urban thermal environmental processes.