Modeling of convective rains for predicting flash floods

Abstract

Intense convective rain cells are often responsible for extreme hydrometeorological events including the majority of flash flood episodes, which is one of the most common and destructive weather-related phenomena especially in urban areas of Malaysia. Both ground and radar data from the Klang Valley were the inputs of this study on the spatial and temporal characteristics of convective rains. A classification based on the β value was used to differentiate the slightly, moderately and strongly convective rains. The areal reduction factor (ARF) obtained from this study is comparable with ARF values obtained earlier by other researchers. An intensity duration frequency (IDF) curve plotted based only on convective storms generally result in higher storm intensity compared to the existing IDF curve and is potentially more appropriate for determining design storms for urban areas with high occurrence of convective events. Synthetic rainfall data series was generated to overcome lack of short duration data series. Two predominant stochastic rainfall model namely a point-process model based on the Neyman-Scott Rectangular Pulses (NSRP) stochastic process and the Markov Chain Mixed Exponential (MCME) was employed. Results of the model evaluation using a 10-year hourly rainfall record at station 3217001 in the Wilayah Persekutuan indicated that NSRP models describe adequately various statistical and physical properties at different timescales (1, 6, and 24-hour). Qualitative and numerical evaluation between the NSRP and MCME models indicated both models have comparable abilities in preserving the properties at the hourly scales, even though the models’ descriptive ability fared better than their predictive ability. However, they were able to preserve the seasonal trend of the observed properties. For forecasting hourly rainfall series, the Multivariate Autoregressive Integrated Moving Average (MARIMA) model was employed. A comparison with an autoregressive moving average model (ARMA) showed comparable results which highlights the potential of the MARIMA model as a forecasting method.