Assessing the impact of land use–land cover change on flood susceptibility using frequency ratio and binomial logistic regression models in the Kallada River Basin, India

ABSTRACT

Graphical abstract illustrating a workflow for assessing the impact of land use–land cover (LULC) change on flood susceptibility in the Kallada River Basin, Kerala, India. The figure shows the study area, a methodological framework, twelve flood conditioning factors (slope, elevation, LULC, landforms, soil drainage, soil texture, surface runoff, topographic position index, lithology, distance from streams, drainage density, and road network), multi-temporal LULC maps for 1997, 2007, 2017, and a projected 2037 scenario, and flood susceptibility maps generated using Frequency Ratio (FR) and Binomial Logistic Regression (BLR) models. The workflow begins with a 2018 flood inventory and environmental datasets, followed by LULC prediction using a Cellular Automata–Artificial Neural Network (CA–ANN) model, susceptibility modelling, and ROC–AUC validation. Results indicate increasing built-up areas and declining agricultural and vegetated land over time, leading to greater flood susceptibility, particularly in low-elevation areas, gentle slopes, floodplains, and locations near river channels. Both FR and BLR models produce comparable spatial patterns with moderate predictive performance (AUC approximately 0.68), supporting their application for flood risk management, land use planning, and climate-resilient development in tropical river basins.

This study evaluates the influence of land use–land cover (LULC) changes on flood susceptibility in the Kallada River Basin, India, using an integrated geographic information system (GIS)-based framework. Twelve flood-conditioning factors, including slope, elevation, landforms, soil drainage, soil texture, surface runoff, topographic position index, lithology, distance from stream, drainage density, road network, and LULC, were incorporated into frequency ratio (FR) and binomial logistic regression (BLR) models to develop flood susceptibility maps. Multitemporal LULC maps for 1997, 2007, and 2017 were derived from Landsat imagery, and a future LULC scenario for 2037 was projected using a cellular automata–artificial neural network model. The analysis shows substantial expansion of built-up areas and declines in agricultural and vegetated lands, increasing runoff and flood susceptibility, especially in midland and coastal regions. Low elevation, gentle slopes, floodplain and alluvial landforms, proximity to river channels, and urban land use are the dominant controls on basin flood susceptibility. Receiver operating characteristic-based validation shows moderate area under the curve values of 0.685 (FR) and 0.679 (BLR), suggesting that while the models capture general spatial patterns of flood susceptibility, their predictive capability remains limited, partly due to data constraints and reliance on a single-event inventory. This study offers a transparent, data-efficient approach to flood susceptibility assessment, supporting flood risk management, land-use planning, and climate-resilient development in tropical river basins.