Simulating Urban Heat Island Footprint Using Machine Learning And Land Surface Temperature Data

Urban heat island is one of the hot issues in the study of urban environment and human-settlement environment.The urban heat island footprint(SUHIF)is a phenomenon that the urban warming effect superimposes on the background temperature field and indirectly leads to the increase of land surface temperature(LST)in the surrounding suburbs.At present,there are two kinds of simulation methods for SUHIF: one is the exponential fitting method which emphasizes on the simulation of the footprint value(D),The other is the geometric index method which emphasizes on the simulation of the whole space shape.In this paper,we use machine learning technology and land surface temperature data to simulate SUHIF,in order to provide a method and ideas for SUHIF simulation.The specific contents include:(1)Designing,constructing and training machine learning models using machine learning technology;(2)Analyzing and comparing the simulation performance of machine learning models and the exponential fitting model;(3)Testing the application of the best machine learning model to the case city.The results show that:(1)Under the current data and training conditions,the optimal model based on simple machine learning is ANN(15-BR),the optimal model based on feature engineering and machine learning is a series model composed of k NN and ANN(10-LM);(2)Machine learning technology has obvious effect on improving the simulation precision of footprint distance D,and the model based on feature engineering and machine learning is the best model among the three models;(3)The visual simulation of 12 case cities shows that the machine learning model has good data applicability,and the model can be used to improve the spatial accuracy of footprint morphology.