Rainfall Forecasting Using Artificial Neural Networks

This research explores a data-driven approach to short-term rainfall forecasting using feed-forward, back-propagation artificial neural networks. Instead of relying on physical atmospheric models, the study leverages ground-level meteorological data collected in Colombo, Sri Lanka, to train and validate predictive models. Several neural network architectures were investigated, each tailored to capture seasonal rainfall dynamics and optimize forecast accuracy for one-day-ahead predictions. The research demonstrates that dividing the annual cycle into season-specific networks, as opposed to a single unified model, yields improved predictive performance. The findings highlight the value of neural networks as robust tools for data-driven hydrometeorological forecasting and provide methodological insights that may be extended to similar applications in other regions or with alternative data sources.

M.Sc. Thesis, Department of Mathematics, University of Peradeniya, Sri Lanka
M.Sc. research conducted at the Department of Mathematics, University of Peradeniya, Sri Lanka (2012–2013).
Artificial Neural Networks Rainfall Forecasting Hydrometeorology Machine Learning Data-Driven Modeling