On The Scale Effect Of Relationship Identification Between Urban Heat Island And 3D Landscape Pattern In Beijing: A Study Based On Random Forest

With the acceleration of urbanization,human activities have greatly changed the urban landscape pattern and urban thermal environment.Urban heat island(UHI)is one of the urban ecological problems caused by the continuous deterioration of urban thermal environment.The affection of UHI to the urban atmospheric environment,living environment and material circulation usually negatively affects human physical and mental health in urban areas.Significant multi-scale correlation exists between the land surface temperature(LST)and landscape pattern,and scale effect is one of the important issues in landscape pattern research.In the past decades,most studies on the relationship between UHI and landscape pattern were conducted at a single scale,and few studies on multiple scales were involved.Compared with the traditional multi-scale methods,the regression model based on random forest has the advantages of higher accuracy and better learning ability,and the model can remove the linear correlation between independent variables in the regression process.Taking the central urban area of Beijing as an example,this paper calculated the 2D and 3D landscape metrics of the second and fourth ring roads of Beijing by using the moving window method,and analyzed the multi-scale relationship between landscape pattern and LST in Beijing by using Pearson correlation coefficient,multiple linear regression and random forest regression.The results showed that LST in the main urban area of Beijing was relatively high and tends to decrease from the center to the surrounding areas.The effection of 3D landscape metrics on the interpretation of LST was stronger than of 2D landscape metrics.The Pearson correlation coefficient between landscape composition and configuration metrics and LST was higher than between the roughness metrics.The effection of landscape diversity(SHDI)and evenness(SHEI)on the change of LST were more obvious than other metrics.The multiple linear regression accuracy and random forest regression accuracy of 3D landscape metrics and LST were higher than that of 2D metrics with similar trends.However,due to the change of moving window size,the sample number changed greatly.In the case of small windows,the accuracy of random forest regression was higher than that of multiple linear regression.When the window size became larger,the accuracy of random forest regression fluctuated.In the fourth Ring Road of Beijing,the multi-scale relationship between LST and landscape pattern was obvious,and the change of extent on landscape pattern was stronger than the change of grain size.The interpretation of landscape metrics on LST and the correlation between landscape metrics and LST increased with the increase of moving window.Impervious surface can significantly increase LST.Because of the solar radiation during the daytime,the covering area of buildings to the surface makes the effect of high-rise building area on the decrease of LST more obvious than that of lowrise building area.The negative correlation between sky view factor(SVF)and LST suggested that increasing the distance between buildings can effectively reduce daytime LST by increasing the energy exchange rate between urban and rural areas.Vegetation and water can effectively reduce LST,but large and clustered irregular patches have a better effect on surface cooling than small and discrete patches.The results of power function fitting of the coefficient of rectangle variation of 3D landscape metrics showed that the optimal window size to study the relationship between LST and 3D landscape pattern is about 700 m.In general,the random forest regression algorithm is an analytical method that can be used to study the multi-scale relationship between LST and landscape pattern.Our study is useful for future urban planning and provides references to mitigate the daytime(UHI)effect.