Comparing Multiple Machine Learning Models to Investigate Thermal Drivers in an Arid-Oasis Urban Park and Its Surroundings Using Mobile Monitoring

At present, the research on the microclimate of urban parks mainly focuses on the univariate or multivariate research contents of park design elements, and there are few analyses that can combine the park with the surrounding regional environment to jointly explore the cooling mechanism of park design elements. This study takes the People’s Park in Urumqi, a typical oasis city in an arid area, as the research object. Combined with different land use natures (park area/residential area), it analyzes the spatiotemporal variation law of temperature through mobile meteorological monitoring in different periods of summer and autumn and optimizes the buffer zone to further compare the performance of the multiple linear regression model and three machine learning models. The selection of the optimal model for collaborative analysis and comparison revealed the dominant variables and their threshold effects affecting the temperature of the park area and the residential area. The results show that: (1) In multi-scenario comparisons, a larger buffer has a better fitting effect. (2) The random forest model is the best model for temperature prediction in the study area. (3) The dominant factors of temperature in different seasons show significant differences, and only a few periods have cross-seasonal persistence. In the park area, the green coverage rate and road network density play a leading and influential role, while in the residential area, the influence of water cover ratio is more obvious. Furthermore, the influence direction of residential area indicators on temperature shows opposite trends in the morning and afternoon periods. (4) There are obvious limited-threshold effects on the influence of dominant factors on temperature in different regions. It is suggested that in the urban spatial layout, while considering the differences for different utilization Spaces, collaborative planning should be carried out. These findings offer new insights into temperature drivers and provide practical references for urban planners.