A modified geographical weighted regression model for better flood risk assessment and management of immovable cultural heritage sites at large spatial scales

Abstract

With the increase in extreme climatic events globally in recent years, the increased frequency of flood hazards has had a great impact on immovable cultural heritage sites (ICHs) due to their prolonged exposure to the disaster environment. This poses a risk management challenge, especially on large scales. Most existing flood risk assessment models for ICHs are determined using common natural hazard methods directly and focus less on the characteristics of ICHs. In this paper, we propose a modified geographical weighted regression (MGWR) model to assess flood risk at ICHs, and this model considers the spatial and age properties of the ICHs. These two properties were used for the construction of the weight matrix in the MGWR model. Eleven selected indices and loss survey data with 417 sample points, including 5 types of ICHs, were utilized for model training and testing in Shanxi Province, China. The results showed that the MGWR model had good accuracy with an R2 of 0.928. A comparison between the MGWR and normal GWR models indicated that the accuracies of the older ICHs improved more in the MGWR than in the GWR. We also found that the proposed model performed better than the normal GWR model using age as an index. Moreover, in comparison with three machine learning methods (decision tree, logistic regression, and random forest), the MGWR model still performed better and was less limited by the number of training samples. This paper provides evidence that the characteristics of ICHs are crucial in the construction of flood risk assessment models, and the proposed model can benefit the risk management of various types of ICHs at large spatial scales.