Abstract: Objectives: The construction of the urban thermal environment network for resource-depleted cities can contribute to urban transformation and sustainable development. Constructing a rational resistance surface for the flow and transmission of thermal environments is a crucial area worthy of in-depth exploration. Methods: By utilizing multi-source data such as Landsat remote sensing images, digital elevation model, point of interest, road networks, and water systems, and integrating regression analysis theory, this study applies machine learning methods to construct urban thermal environment resistance surfaces under three scenarios: natural, human activity, and combined natural + human activity. Based on these resistance surfaces, the spatial network of the thermal environment is identified. Results: The high-temperature area in the study region accounts for 18.22% of the total area, with the core high-temperature zone covering more than 70% of the high-temperature area. In nonlinear models for different scenarios, the contribution of multiple resistance factors to the thermal environment differs significantly, with the model under the "human activity + natural" scenario showing the highest accuracy. The length of thermal corridors is longer under the human activity and human activity + natural scenarios, measuring 95.86 km and 94.34 km, respectively. The areas of pinch points and obstacle points are highest under the human activity and human activity + natural scenarios, respectively, and a synergistic variation trend is observed between the pinch point area and the mean resistance. Conclusions: By discussing the spatial network of thermal environments under different scenarios in resource-exhausted cities, constructive guidance is provided for accurately mitigating urban thermal environments.