Estimating near-surface air temperature in urban functional zones in China using spatial-temporal attention

Abstract

Near-surface air temperature (T_a) plays a critical role in land-atmosphere interactions, influencing human health, agricultural productivity, and ecosystem stability. Urban functional zones (UFZs), with distinct social and economic roles, exhibit unique temperature characteristics shaped by variations in land use, building density, and spatial configuration, thereby significantly affecting T_a distribution in urban areas. This study introduces a Functional-Spatial-Temporal Graph Convolutional Network (FST-GCN) model, enhanced with a spatial-temporal attention mechanism, for precise estimation of T_a across UFZs. By integrating high-resolution remote sensing imagery and multi-source meteorological data, the model was applied to estimate T_a in thirteen major Chinese cities from 2015 to 2021. The model's performance was evaluated using five-fold cross-validation against meteorological station data, achieving a mean absolute error (MAE) of approximately 1.44 °C and a coefficient of determination (R²) exceeding 0.9, demonstrating significant improvements over traditional methods. The analysis reveals a strong dependency of temperature characteristics on UFZ types and seasonal variations. High-density urban zones, such as industrial and commercial areas, exhibit higher temperatures, particularly during warmer months, due to factors like intensified human activity and limited vegetation cover. Conversely, green zones and residential areas show a notable cooling effect, emphasizing the critical role of urban greenery in mitigating heat retention. Additionally, distinct seasonal temperature variations are observed between northern and southern cities, driven by regional climatic differences and urban spatial configurations.