Floods, as natural disasters, have a profound impact on society. Assessing them is highly complex due to the interplay of factors such as meteorology, topography, and land cover. Accurate forecasting is essential to reduce disaster risks, guide emergency response strategies, and minimize economic and social losses. Recent advancements in machine learning have significantly improved the accuracy of flood predictions, offered more cost-effective solutions and enhanced decision-making processes. This paper reviews the most common and recent advancements in machine learning applications for flood hazard assessment and forecasting and compares their performance with traditional approaches such as numerical modelling and remote sensing. While numerical models provide detailed predictions, they are computationally demanding and depend on precise data inputs. While remote sensing provides valuable large-scale data for flood monitoring, it often faces limitations in real-time responsiveness and accuracy, particularly under rapidly changing flood conditions. Machine learning addresses these limitations by leveraging historical data to identify patterns and refine predictions, improving both accuracy and efficiency. Challenges such as the variability of model performance across different regions and the requirement for high-quality data remain. This paper explores both long-term and short-term flood forecasting and the hazard assessment, shows that combining different methods in hybrid models can improve accuracy by reducing data uncertainties. Future research should prioritize refining machine learning algorithms for diverse environments, improving data processing techniques, and developing integrated methodologies. These advancements will lead to more reliable flood predictions, ultimately helping to mitigate the risks and impacts of flood disasters.
Zhang et al. (Tue,) studied this question.