Convolutional Long Short-Term Memory network for generating 100 m daily near-surface air temperature

Global warming and urbanization serve as critical research themes in fine-scale climate studies, particularly in developed cities. This study aims to provide a high spatiotemporal resolution dataset of near-surface air temperatures for densely developed urban areas. The dataset comprises daily maximum, minimum, and mean temperatures for the summer months (June to August) from 2019 to 2023, at a spatial resolution of 100 m, across the Jiangbei climate zone in China. We applied the Convolutional Long Short-Term Memory (ConvLSTM) deep learning model with multi-source data, including ERA5 temperature data, topography, landcover and vegetation fraction cover. Model evaluation indicates high accuracy, with mean absolute errors (MAE) ranging from 0.564 to 0.784 °C, root mean square errors (RMSE) from 0.733 to 1.027 °C, and coefficients of determination (R²) from 0.892 to 0.943. Our dataset is distinguished by the 100 m spatial resolution and the inclusion of recent summer data from 2023 at a daily scale. This work is valuable for urban or inner-urban climate studies on heatwave mitigation policies and adaptation strategies.