Chinese cities are experiencing rapid urban expansion and being transformed into more dispersed urban form which necessitate the spatially explicit analysis and modeling of urban patterns.In our research,spatio-temporal models are developed for measuring urban sprawl in Wuhan,a typical metropolitan area in Central China using multi-temporal remote sensing images.We first extract urban built-up land from Landsat TM (1995 and 2005),ETM+ (2000) and ALOS (2010) satellite images over four periods.Object-oriented segmentation,SVM classification algorithm and post classification correction based on auxiliary data such as land use survey data and DEM are employed to improve accuracy.The interpreted built-up land,as the proxy for urban sprawl indicator and dependent variable for further analysis,has been decomposed to district level to produce a spatio-temporal dataset for modeling.Then exploratory data analysis techniques are used to identify the relationships between urban sprawl and its causative factors such as population,GDP,fixed assessment investment and revenue.After that,we perform spatial autoregressive models: mixed spatial autoregressive model (SAR),spatial error model (SEM) and spatial dublin model (SDM) to establish these relationships quantitatively.We argue that these models can be more effective in modeling urban sprawl when prior information about spatial and temporal characteristics of urban development can be applied.As a result,we devise four types of weight matrices with values of 0,0.5 and 1 to identify the spatial and temporal relationships among these districts in these four time periods and applied them to SAR,SEM and SDM respectively.In the meantime,we also undertake the casetti distance expansion model,which employs a vector of distances to accommodate spatial heterogeneity.The distance is measured from the mean center of district weighted by their built-up area to the urban center.The results reveal that population density (PopD),non-agricultural population ratio (NAPR) and per capita GDP (PGDP) are the primary causes for urban sprawl and they are significantly positively correlated to PB in most cases.Spatial regression models with the combination of different weight matrices have produced different results and we are not attempting to recommend one than the other,since they have demonstrated their respective strengths and drawbacks.For example,SDM has higher R2 but the log-likelihood is lower than SEM,and present inconsistent coefficient for the independent variable NAPR.However,the spatio-temporal model shows superiority than the empirical OLS regression model,not only in the goodness of fit,but also in exploring the spatio-temporal lagged effect.Comparatively,casetti distance spatial expansion produce better result with R2 reaching 0.96.PopD,NAPR and PerGDP are positively correlated to PB,whereas population has negative effect on urban sprawl.It is concluded that 1) built-up land in Wuhan have expanded and dispersed tremendously in the past two decades and most of them are around the urban core area; 2) to curb urban sprawl,regulations or policies are not confined to the prevention of further encroachment from non-urban land to urban land,but also extend to achieve a compact urban form; 3) spatial regression models,especially the distance-based one,show superiority in exploring the relationships between urban sprawl and its factors.