Abstract: The temperature is a crucial climatic resource.To obtain spatially continuous temperature data under the complex terrain of Chongqing,this study utilized multi-source data,including topography,nighttime light,FY-2G satellite observations,and meteorological station temperature records.Three machine learning methods Linear Regression (LR),Extreme Learning Machine (ELM),and Random Forest (RF) were employed to establish models for spatially interpolating and validating the daily average temperature on the 15th of each month in 2022 across Chongqing.The results demonstrate:The variables screened in this study can serve as effective feature variables for machine learning-based spatial interpolation of daily average temperature in Chongqing.RF significantly outperformed the other two methods,with an average expected real error of less than 0.5 ℃,indicating reliable spatial interpolation results.In terms of evaluation metrics,Mean Absolute Error (MAE),Root Mean Square Error (RMSE),and Coefficient of Determination (R²),RF achieved the best performance (0.24 ℃,0.34 ℃,and 0.98).However,interpolation accuracy was slightly lower in summer compared to spring,autumn,and winter.The interpolation accuracy of all three methods improved with increasing elevation.All three methods remained effective for temperature interpolation even in regions with missing feature variables.The RF interpolation results for January,April,July,and October were notably superior to the temperature products from the China Meteorological Administration's High-Resolution Land Data Assimilation System (HRCLDAS),with finer spatial resolution.