Urban heat island distribution observation by integrating remote sensing technology and deep learning

Using particle swarm optimisation algorithm to optimise support vector machines enhances urban heat island observation methods, while remote sensing technology aids in selecting temperature estimation parameters. Then the two are combined to construct a model for estimating urban near-surface temperature. A contribution study is conducted on the selected parameters. The selected parameters have contributions in the near-surface temperature estimation. The determination coefficient of the constructed urban near-surface temperature estimation model was 0.892. The root mean square error was 0.42 degrees C, the F1 value is 0.82, and the running time is 0.41 seconds, which was superior to other comparison models. Additionally, this model was applied to observe the urban heat island in Xi'an. The overall spatial distribution was low in the south and high in the north, with the central area being higher than the surrounding area, the highest temperature is 23.51 degrees C, and the lowest temperature is 19.05 degrees C. Moreover, the intensity level in the high-temperature area accounted for 16.9%. Based on the above results, the near-surface temperature estimation model constructed in the study has shown high accuracy and efficiency in urban heat island observation. It can be applied in practice, providing theoretical reference for urban planning and ecological environment protection.