Integrating Multi-Source Temperature Data and Explainable Deep Learning for Urban Microclimate Analysis

Abstract. Understanding the relationship between land surface temperature (LST) and near-surface air temperature is essential for fine-scale urban heat assessment. This study examines the spatial and temporal coupling between Landsat-8-derived LST and in-situ air temperature measured by a dense IoT sensor network across 19 sites on a university campus during June-August 2024. The campus includes varied building forms, surface materials, vegetation, and water bodies, providing a heterogeneous setting for evaluating surface–air thermal relationships. Rather than treating LST as a direct proxy for air temperature, the analysis compares spatial rankings, diurnal variations, and surface–air temperature differences to identify consistent and divergent thermal patterns. A convolutional neural network combined with Gradient-weighted Class Activation Mapping (Grad-CAM) was further used to test whether spatially reweighted LST information better corresponds to observed air temperature variability. Results show that LST presents stronger spatial differentiation than near-surface air temperature, whereas air temperature displays smoother spatial patterns and clear nighttime convergence. Surface–air temperature differences vary systematically across site environments, indicating heterogeneous coupling rather than random mismatch. The CAM-assisted regression analysis shows that emphasizing thermally relevant surface regions improves the correspondence between satellite-derived thermal information and ground observations. This study provides an interpretable framework for micro-scale analysis of surface–air temperature relationships and supports more reliable characterization of urban thermal environments through integrated satellite and sensor-based observations.