Abstract:

[Objective] Urban flooding severely threatens social security and development, and identifying key influencing factors of urban flooding is fundamental to studying flood disasters. Many studies have used the Geographically Weighted Regression (GWR) model to analyze the causes of urban flooding, but its limitation lies in ignoring the scale variations in spatial heterogeneity in various influencing factors. The Multi-scale Geographically Weighted Regression (MGWR) model overcomes this limitation, but few studies have applied MGWR to analyze the relationship between the degree of urban flooding and key influencing factors. This study uses the MGWR model to analyze the key influencing factors of urban flooding in Wuhan and explores the spatial differences in the correlation between these factors and flood severity.[Methods] The central urban area of Wuhan was selected as the study area, and flood point data from 2016 were collected. Elevation and slope were chosen to reflect the impact of terrain on urban flooding, while eight land use types (farmland, forest, grassland, wetland, water body, impervious surface, shrubland, and bare land) were selected to reflect the impact of land use on urban flooding. River density was chosen to represent the impact of the river network on flooding. Global Ordinary Least Square (OLS), GWR, and MGWR were used to analyze the relationships between influencing factors and flood severity.[Results] After screening using OLS, the selected influencing factors for GWR and MGWR analysis were farmland area, grassland area, and impervious surface area. The model performance comparison revealed that the MGWR model outperformed both GWR and OLS. The MGWR showed that the correlation between influencing factors varied across spatial scales. The impact of impervious surface area had the smallest spatial scale, with a bandwidth of 43; the impact of farmland area had a smaller spatial scale, with a bandwidth of 71; and the impact of grassland area and the constant term was close to the global scale, with a bandwidth of 163. Impervious surface area positively influenced the degree of flooding, and it was the most significant factor affecting flooding, with a mean regression coefficient of 0.934. The largest regression coefficients were found in the Wuchang and the central areas of Hongshan District, indicating the highest flood risk there. Farmland area and grassland area negatively influenced the degree of flooding, with the mean regression coefficient for grassland area at -0.280 and for farmland area at -0.241.[Conclusion] The MGWR model considers the varying impact scales of different variables, and the degree of flooding is highly sensitive to impervious surface area, with strong spatial heterogeneity. Impervious surface area positively affects the degree of flooding, while farmland area and grassland area negatively affect it. Among all influencing factors, impervious surface area is the most significant factor affecting the degree of flooding, followed by grassland and farmland areas. The study demonstrates that the MGWR model provides significant improvements over the GWR model and is well-suited for studying the influencing factors of urban flooding.

Key words: urban flooding, Geographically Weighted Regression (GWR) analysis, influencing factor, land use types