Development of Multivariate Flood Loss Functions Using Geographically Weighted Regression

Development of Multivariate Flood Loss Functions Using Geographically Weighted Regression

Abstract Flood damage estimation should be the focus of disaster prevention and response especially considering the global effect of climate change. Generally, the loss function is established by Ordinary Least Squares (OLS) in a Global Regression Model (GRM). This study collects flood depths, duration time, flood area, income and land price from a GIS-based flood inundation map. The data for the damagesof local facilities were acquired from the flood damages report collected by the local government after the August 2012 flood event in Gunsan City. The OLS is used to estimate damages of residential, commercial and agricultural facilities. The estimates are assessed using coefficient of determination and spatial autocorrelation. The assessment results show that the Local Regression Model (LRM) is more adaptive than GRM in estimating the damages. This is because the spatial patterns of the residuals exhibit spatial autocorrelation. The Geographically Weighted Regression (GWR) developed in this study as one of the LRMs do not just capture the spatial variations of the flood damage factors but it also modify the OLS.

Abstract Flood damage estimation should be the focus of disaster prevention and response especially considering the global effect of climate change. Generally, the loss function is established by Ordinary Least Squares (OLS) in a Global Regression Model (GRM). This study collects flood depths, duration time, flood area, income and land price from a GIS-based flood inundation map. The data for the damagesof local facilities were acquired from the flood damages report collected by the local government after the August 2012 flood event in Gunsan City. The OLS is used to estimate damages of residential, commercial and agricultural facilities. The estimates are assessed using coefficient of determination and spatial autocorrelation. The assessment results show that the Local Regression Model (LRM) is more adaptive than GRM in estimating the damages. This is because the spatial patterns of the residuals exhibit spatial autocorrelation. The Geographically Weighted Regression (GWR) developed in this study as one of the LRMs do not just capture the spatial variations of the flood damage factors but it also modify the OLS.