Exploring the non-linear impacts of urban features on land surface temperature using explainable artificial intelligence

High land surface temperatures (LST) have emerged as crucial threats to urban ecosystems and sustainable urban development. To better understand and mitigate their impacts, it is essential to analyze the contributing urban features. Against this background, we developed a random forest model enhanced by Explainable Artificial Intelligence (XAI) to analyze the impact features of LST in Beijing, China. By applying the XAI method, our results suggest that the major impact features of LST in Beijing are elevation (44.19%), compactness of impervious surface (17.27%), Normalized Difference Vegetation Index (11.12%), proportion of impervious surface area (8.04%), and tree height (3.83%). Compactness of impervious surface exhibited an overall cooling effect, which became weaker at high values. LST increased with building height, and the trend became weaker as building height reached 5 m. The most important features impacting LST in the inner city are the proportion and height of buildings, whereas in the outer city these features are tree height and the compactness of impervious surfaces. The study applies XAI to explain the non-linear interactions between LST and urban features, offering innovative insights to policy-makers to develop sustainable urban planning strategies. Our findings suggest that increasing green spaces and water bodies as well as controlling building density and height can effectively mitigate heat in dense urban areas and enhance cooling effects.