Research On The Urban Surface Thermal Environment By Considering The 3D Spatial Information

With the rapid expansion of cities and the intensification of human activities,urban heat island effect has become one of the most significant features of urban climate.The urban heat island not only impacts urban microclimate,human settlement and public heat,but also influences other urban climate effects such as changes in vegetation phenology.As the rapid development of geographical information system and remote sensing,it provides an alternative way to achive the data covering area.A series of researches has been carried out to understand the urban surface thermal environment,including spatial-temporal characteristics,driving forces and mechanisms of urban surface thermal environment.The researches,related to the spatial-temporal change of urban surface thermal environment,mainly focus on analyzing the distribution and intensity variation of urban surface thermal environment in two or three-dimensional space.The previous studies were mainly based on "urban-rural" binary cognition system,which could reveal the facts of urban heat island phenomenon.However,it can not provide practical gudiance for alleviating urban heat island effect,due to lack of consideration of urban heat island in the local areas.In order to overcome these deficiencies and defects,some studies have been dedicated to engaging in investigating the relationship between Local Climate Zones and land surface temperature.However,researches on urban surface thermal environemt,which is based on local climate zones,are still at the initial stage.How to acquire the local climate zones with high precision?Are there seasonal variations in the relationships between local climate zones and land surface temperature?Are there significant difference between different local climate zones in terms of mean land surface temperature?These problems have been an urgent question to be solved in the urban surface thermal environment based on local climate zones.Driving forces and mechanisms of urban surface thermal environment are of great importance to understand spatial-temporal characteristics of urban surface thermal environment.A significant number of researches have explored the relationships between land surface temperature and the proportion of land cover features,such as impervious surface area,green vegetation,human activity,and water.It is noted that despite many studies have proved the positive relationship between impervious surface area and land surface temperature,individual influences of buildings and roads on land surface temperature were not well-examined.A few scholars have dedicated to examining the relationship between landscape pattern of building and urban surface heat island,there are still many shortcomings.First,there is littlle consideration is given to the impact of urban three-dimensional morphology on the land surface temperature,for instance,buidling height and its variation.Second,there is lack of researches to reveal the extent to which the impacts of building on urban surface thermal environment vary across different spatial scale.Finally,the study is short in seasonal variations of the impact of building on urban surface thermal environment.As a consequence,the key objectives of this work are outlined as follows:(i)to produce the urban land cover map and 3D spatial information of surface elements with high precision,and further classify the land surface in the basis of Local Climate Zones classification system by combining multi-source multi-resolution satellite images;(ii)to analyze the relationship between Local Climate Zones and land surface temperature as well as seasonal variations;(iii)to reveal how landscape pattern of building and urban three-dimensional morphology impact land surface temperature in different seasons as well as scaling effect.The main contents and conclusions of this dissertation are summarized:1)The Random Forest classification model,concerning spectral features,spatial features and three-dimensional structure of surface objects,was built in order to gain better land cover classificaiton results in the urban area.LiDARderived height,intensity and multiple-return features were extracted from airborne LiDAR data,while spectral and spatial information were obtained from multispectral data.Given that different feature sets can contribute to improving classification accuracy in a variety of ways.To gain a detailed understanding of what level of classification result could be obtained by using different scenarios of input features,seven different scenarios were employed for the image classification experiments.This study investigated how much classification accuracy can be acquired by using different feature combinations and classifiers,quantify the relative importance of all input variables and explore the contribution of each feature to the classification accuracy,and assess the influnce of the integration of LiDAR and multispectral data on the classification uncertainty.Experimental results demonstrated that the exclusive use of LiDAR-derived height features produced the land cover map with the lowest map accuracy,and the best result was obtained by the combination of SPOT-5 and LiDAR data using all input features.Random Forest can not only yield better classification results,but also is more stable,when compared to Rotation Forest and Stacked Autoencoder.Analysis of feature relevance indicated that LiDAR-derived height features were more conducive to the classification of urban area when compared to LiDAR-derived intensity and multiplereturn features.While the nDSM was the most useful feature in improving the classification performance of urban land cover,the feature importance scores for the following LiDAR-derived height features was also very high,including variance,SD and so on.As for feature importance per class,the variable importance varied to a very large extent.Results of classification uncertainty suggested that feature combination can tend to decrease classification uncertainty for different land cover classes,but there is no "one-feature-combinationfits-all" solution.The values of classification uncertainty showed marked differences between the land cover classes.Furthermore,using all input variables usually resulted in relatively lower classification uncertainty values for most of the classes as compared to other input features scenarios.2)An integtive GIS-based approach was proposed to classify land surface through the utilization of a climate scheme called Local Climate Zones in the central part of Nanjing,China.Furthermore,land surface temperatures in different seasons between pairs of LCZ types was investigated thoroughly.LCZ classification was divided into building forms and land cover types.the building forms and land cover types were obtained,respectively,according to the calculated data,including land use/cover classification,building height and so on.This study discussed the relationship between Local Climate Zones and land surface temperatures in different seasons,and explored whether temperature differences in different LCZ is statistically-significant or not.In addition,relative number of multiple comparison tests indicating siginficant differences in land surface temperatures among LCZ was also examined.Experimental results indicated that open building types covers larger areas than compact builiding types do.LCZ 5(i.e.,open midrise building)accounts for the largest proportion of the main urban area among the building types.As for land cover types,the proportions of the study area were LCZ AB(i.e.,trees),LCZ G(i.e.,water),LCZ E(i.e.,bare rock or paved),LCZ F(bare soil or sand)and LCZ D(i.e.,low plants)in descending order.Concerning the spring and summer,all local climate zones except for LCZ AB and LCZ G showed higher average land surface temperature,which is above the average land surface temperature of the study area.However,as far as autumn and winter are concerned,the mean land surface temperatures of LCZ,which is lower than the average land surface temperature of the study area,include not only land cover types but also building types.Furthermore,what we can conclude from the results of the oneway ANOVA is there are significant differences among the averages of LCZ land surface temperatures,regardless of seasons.The multiple comparison results suggested that land surface temperature in summer differentiated better between LCZ,in contrast to other three seasons.It is worth noting that pairs of LCZ,for which mean land surface temperature are significantly different,were susceptible to season.3)This section is aimed at revealing how landscape pattern of building and urban three-dimensional morphology influence land surface temperature as well as seasonal variation,and explore the resolution sensitivity of the relationships between land surface temperature and these two driving factors.First,appropriate landscape metrics were adopted to measure the spaitial pattern of building,involving in 1 composition metrics and 6 configuration metrics,and 4 quantified parameters(i.e.,building height,building height variation,3-D urbanization index and sky view factor)were selected to describe urban three-dimensional morphology.Second,the quantitative relationships between land surface temperature and building fraction was firstly built at multiple resolutions,with the aim of examining the role of building fraction in explaining variation of land surface temperature in terms of different seasons.Futhremore,spatial configuration of building and/or urban three-dimensional morphology in combination with building fraction were utilized to predict land surface temperatures in different seasons.Third,the influencing factors,which significantly impact land surface temperature in different seasons,were explored at multiple resolution.Finally,relative importance of building surface,spatial configuration of building and urban three-dimensional on land surface temperature in different seasons were analyzed in the context of multiple resolutions.Experimental results suggested that,as for the main urban area of nanjing,the areas with high values of building height and building heigh variation are mainly concentrated on both sides of the main roads and around the traffic squares.When the size of analytical unit is 1080 m,the average values of building height and building height variation are 14.9 m and 10.97 m,respectively.Results from the OLS linear regression showed that the variations of land surface temperature,as explained by building fraction alone,were affected by seasonal factors.The explained variations of land surface temperature was summer,spring,autumn winter in descending order.It should be noted that with the increasement of the size of analytical unit,the relationships between building fraction and land surface temperature tend to be stronger.While the size of analytical unit is 1080 m,57.9%of variation of land surface temperature in summer can be explained by building fraction,and building fraction can only account for 2.4%variation of land surface temperature in winter.Besides building fraction,spatial congfiguration of building and urban three-dimensional morphology can also affect land surface temperature.Combining building fraction,spatial configuration of building and urban three-dimensional can explain the most variations of land surface temperature free of seasons.Furthermore,the combination of these three influencing factors can explain more variations of land surface temperature in spirng and summer than in autumn and winter.With the increasement of the size of analytical unit,the ability of the combination of these three influncing factors to explain variability of land surface temperature gradually increased for spring,summer and autumn,while the ability to interpret variation of land surface temperature in winter decreaed first and then increased.Building fraction played more important role in predicting land surface temperature in spring,summer and winter,regradless of the size of analytical unit.As for land surface temperature in winter,the influencing factors with the highest realtive importance vary with the size of analytical unit.