Application of Artificial Neural Networks in Short-Term Rainfall Forecasting
Short-term rainfall is important in agriculture, industry, the energy sector, and any other water-dependent activities where profitability depends on climatic conditions. The scarcity of reliable prediction models encouraged the authors of the present study to develop a modeling platform using a neurogenetic model to estimate rainfall occurrence within a short-term duration. The data on both the quantity and the probability of occurrence of rainfall based on the previous 1–5 days were used to predict the quantity and occurrence of rainfall 1–4 days hence. The potential of neurogenetic models to predict short-term rainfall on the basis of such a small-scale data set was analyzed with the aim of developing a software platform for laypeople and to help related professionals maintain the profitability of their organization by reducing the likelihood of wastage resulting from large-scale prediction errors, which are common with the available linear models. The results indicate that neurogenetic models can reliably predict rainfall 1, 3, and 4 days in advance, but not 2 and 5 days, if the models are trained with a suitable algorithm. The subpar performance of the 2- and 5-day rainfall prediction models was attributed to the choice of training algorithms and length of time, although the reliable prediction of rainfall even 1 day in advance warrants pursuing further development of the present investigation.
KeywordsRainfall Event Neural Network Model Kappa Index Stochastic Neural Network Quick Propagation
- Lekouch I, Lekouch K, Muselli M, Mongruel A, Kabbachi B, Beysens D (2012) Rooftop dew, fog and rain collection in southwest Morocco and predictive dew modeling using neural networks. J Hydrol (in press), Accepted manuscript, Available online 13 Apr 2012Google Scholar