A Review of Machine Learning Algorithms and Weather Forecasting Integration for Enhancing Flood Prediction in the Nilwala River Basin

Abstract

In Sri Lanka, the Nilwala River Basin is very vulnerable to severe flooding that often places local lives, property, and livelihoods at risk. The current review evaluates the integration of complex machine learning models with weather forecasting methodologies, in particular rainfall data, aiming for substantial improvements in timely and accurate predictions of floods within this very fragile region. Timing and intensity of rainfall are crucial information for flood prediction, in which traditional forecasting often fails due to incorrect predictions. This review conducts a detailed study analysing ten years of daily rainfall records and utilises multiple machine learning models such as Artificial Neural Networks (ANN), Support Vector Machines (SVM), Convolutional Neural Networks (CNN) and Long Short-Term Memory networks (LSTM) to learn which predictive algorithms ultimately outperform others. The superiority of the LSTM network for predicting flood events in comparison to other models reveals the ability of LSTM networks, which are capable of detecting patterns over time and sequence data. By examining crucial factors including timing and intensity, the integration of rainfall data with machine learning algorithms improves the accuracy of flood prediction. In particular in flood-prone places like the Nilwala River Basin, this strategy helps lower risks for vulnerable people by bolstering early warning systems and disaster preparedness. This paper highlights the enabling role of machine learning in improving flood prediction and hazard assessment by working alongside conventional constitutive weather forecasting.