Predicting the flood susceptibility under land use and climate change scenarios using deep learning algorithms

Area of Science:

• Hydrology
• Environmental Science
• Climate Change Studies

Background:

• Semi-arid, snow-fed basins face escalating flood risks due to combined land-use and climate change.
• The synergistic impacts of these changes on flood generation potential remain inadequately quantified.

Purpose of the Study:

• To predict future flood generation potential (FGP) in the Gharesou Watershed (Iran) under integrated land-use and climate change scenarios.
• To quantitatively attribute the contributions of land-use and climate change to future flood risk.

Main Methods:

• Future land-use scenarios (2034-2054) simulated using Markov chain.
• Climate variables downscaled using the change-factor method under three Shared Socioeconomic Pathways (SSP) scenarios.
• Flood generation potential mapped using Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), and Deep Neural Network (DNN) algorithms, validated with observed discharge data.

Main Results:

• Projected decline in natural vegetation by 20.9% and expansion of agricultural/residential lands by 2054.
• Anticipated temperature rise of 3.5-4.5°C, with more frequent extreme precipitation events despite a potential annual maximum decline.
• High- to very-high flood risk zones projected to expand from 62% to 87% of the watershed under the optimized CNN model.

Conclusions:

• Land-use change is identified as the dominant driver (60-70%) of increased flood risk, with climate change acting as an intensifier (30-40%).
• Conservation of natural vegetation and strict land-use conversion controls in high-risk areas are more critical than climate adaptation alone.
• Proactive policies, including rangeland restoration, integrating climate scenarios into spatial planning, and enforcing land-use regulations, are essential for enhancing watershed resilience.