Assessing flood risk in a rapidly urbanizing basin: a logistic regression approach for Istanbul (Türkiye)
Year: 2025
Type: preprint
Abstract: The complex topography and rapid urbanization of Istanbul along the Bosphorus Strait have increased flood risk by modifying drainage and increasing runoff. Considering the challenging natural geomorphology and rapid urban growth, the study applies a logistic regression to identify flood-prone areas within an area of ~ 670 km² across the city. Independent variables include elevation, slope, aspect, curvature, stream channel, and Topographic Wetness Index (TWI), all derived from a high-resolution 5m Digital Elevation Model (DEM), along with land use data (Istanbul Metropolitan Municipality). Accordingly, historical flood occurrences between 1995 and 2020 served as the binary dependent variable. Pre-processing included extracting physical variables from a 5-meter DEM, converting data to raster, standardizing inputs, and converting flood records from points to polygons. Data were analyzed using Spyder (Anaconda 2.6), and the logistic regression model was calibrated with 80% training and 20% testing data using a balanced dataset to address imbalance. The resulting model demonstrated an overall accuracy of 76%, with sensitivity and specificity of 77.2% and 74.9%, respectively. The Area Under the ROC Curve (AUC) obtained at 0.76, indicating moderate to good predictive performance. Cross-validation (k=5) confirmed robustness, with the AUC scores ranging from 0.71 to 0.94. The analysis showed that low elevation, high TWI, proximity to streams, and convex curvature strongly correlate with higher flood risk. Conversely, steep slopes, high elevation, and natural land cover (especially dense vegetation) reduce risk. Elevation had the strongest negative effect (coefficient = -1.58), while curvature and TWI contributed most positively. Using the retrieved coefficients, urban flood susceptibility was delineated for Istanbul and categorized into five risk levels ranging from very low to very high. High and very high susceptibility zones primarily align with low-lying, densely urbanized regions, which alarms the necessity of employing adapting measures to buffer flood hazard and exposure. These findings demonstrate logistic regression’s potential in assessing floods in complex urban environments by evaluating multiple spatial factors.
Language: English
Subfield: Global and Planetary Change
Field: Environmental Science
Domain: Physical Sciences
Open Access status: gold
OpenAlex ID: W4417048482
Cites:
Cited by:
Related to: