Machine-Learning-Algorithm-Based Prediction of Land Use/Land Cover and Land Surface Temperature Changes to Characterize the Surface Urban Heat Island Phenomena over Harbin, China

Abstract

The rapid expansion of urban land has altered land use/land cover (LULC) types, affecting land surface temperatures (LSTs) and intensifying the urban heat island (UHI) effect, a prominent consequence of urbanization. This study, which focuses on Harbin, a representative city in a cold region, employs the patch-generating land use simulation (PLUS) model to predict LULC changes and a Bidirectional Long Short-Term Memory (Bi-LSTM) model to predict LST. The PLUS model exhibits a high prediction accuracy, evidenced by its FoM coefficient of 0.15. And the Bi-LSTM model also achieved high accuracy, with an R2 value of 0.995 and 0.950 and a root mean square error (RMSE) of 0.199 and 0.390 for predictions in winter and summer, respectively, surpassing existing methods. This study analyzed the trends in LULC, LST, and the urban thermal field variance index (UTFVI) to assess the relationships among LST, LULC, and UTFVI. The results show that urban land increased by 27.81%, and woodland and grassland decreased by 61.07% from 2005 to 2030. Areas with high temperatures increased by 40.86% in winter and 60.90% in summer. The proportion of the medium UTFVI zone (0.005–0.010) in urban land increased by 50.71%, and the proportion of areas with medium UTFVI values and above (>0.005) decreased at a rate of 84.70%. This finding suggests that the area affected by the UHI has decreased, while the UHI intensity in some regions has increased. This study provides a technical reference for future urban development and thermal environment management in cold regions.