Forecasting monthly maximum 5-day precipitation using artificial neural networks with initial lags


Junaida Sulaiman, Herdianti Darwis, Hideo Hirose


International Symposium on Computational Intelligence and Design (SCID2013), October 28-29, 2013, Hangzhou, China.

Successive days of precipitation are known to cause flood in monsoon-type countries. Forecasting of daily precipitation helps to predict the occurrences of rainfall and number of wet days while with a maximum 5-day precipitation, we can predict the magnitude of precipitation within a specified period that may signified the precipitation extremes. This paper describes a method to forecast the trend of maximum 5-day precipitation (MX5d) in the next month using artificial neural networks (ANN). The purpose is to predict the trend of maximum precipitation using a descriptive index outlined by World Meteorological Organization (WMO). The index is used by WMO for evaluating changes in precipitation extremes. The analysis of extreme precipitation trend is important for future prediction of high precipitations events in the area of interest. ANN is widely applied in the hydrology field due to non-linearity ability in prediction to non-stationary and seasonal data. Here, ANN is compared with seasonal autoregressive integrated moving average (ARIMA) in forecasting next month maximum 5-day precipitation. We have compared ANN with seasonal ARIMA to measure their performances. Prior to model development, the significant input lags are determined using linear correlation analysis (LCA) and stepwise regression method, respectively. The ANN method is feasible in forecasting precipitation extremes when it is trained with the particle swarm optimization.

Key Words
artificial neural networks; particle swarm optimization; extreme precipitation; seasonal autoregressive integrated moving average



Times Cited in Web of Science:

Cited in Books: