Search

Rainfall modelling is an essential component of different hydrological studies. However, rainfall modelling in tropical regions, especially urban areas, remains inadequate. To determine the applicability of different types of rainfall modelling approaches, simulations by two Markov models (Matlab-based weather generator (MulGETS) and multi-site rainfall simulator (MRS)) and a Neyman–Scott based Poisson cluster model (RainSim) were compared with a proposed modified k-nearest neighbour (KNN) model for 30 years rainfall in Singapore. The MRS model was determined to be suitable for single-site applications in tropical regions. However, for multi-site conditions, RainSim was adjudged the most suitable given its accuracy in preserving observed spatial information, despite its performance lagging in few statistical indicators. The KNN model was found to perform satisfactorily during the wet seasons, and was the only model that could repeat the extreme precipitation values closely. Although typical studies evaluate the performance of models based on a set of criteria, there exists a lacuna in understanding the quality of simulations by these models. Therefore, uncertainty analysis based on two different criteria was implemented to understand the performance of the stochastic processes within. Although the MulGETS model exhibited the lowest differences between Prediction Intervals, RainSim's Prediction Intervals were found to subsume observed data more often. The proposed study would be useful for users examining rainfall models for differing objectives.

A novel combined statistical downscaling and disaggregation framework (CSDD) based on Long Ashton Research Station-Weather Generator (LARS-WG) and K-nearest neighbour (KNN) was proposed to examine the climate change impact on regional extreme rainfall in Singapore. The approach could generate high-resolution rainfall sequences both spatially (e.g. station level) and temporally (e.g. 5-min) based on 12 emission scenarios under 4 general circulation models. Three different routes were proposed to generate high-resolution synthetic rainfall time series: (1) the rainfall records at all stations were combined and treated as one station (Route 1); (2) the rainfall record at each station was treated individually (Route 2); and (3) the total areal rainfall at daily and hourly timescales was used to keep inter-site correlations (Route 3). The results indicated that all routes under CSDD framework could effectively reproduce the regional rainfall at various timescales. This study further examined the impacts of climate change on regional extremes. The results show that the maximum increase rates of 100-year extremes at the end of this century at durations of 24-h, 1-h and 5-min would be 14.8, 16.3 and 16.8%, respectively. The proposed framework takes the full advantage of both LARS-WG and KNN and could effectively help provide continuous high-resolution synthetic rainfall data (with consideration of spatial dependence and uncertainty) under climate-change conditions for hydrological impact study.

A generalized likelihood uncertainty estimation (GLUE) method incorporating moving least squares (MLS) with entropy for stochastic sampling (denoted as GLUE-MLS-E) was proposed for uncertainty analysis of flood inundation modelling. The MLS with entropy (MLS-E) was established according to the pairs of parameters/likelihoods generated from a limited number of direct model executions. It was then applied to approximate the model evaluation to facilitate the target sample acceptance of GLUE during the Monte-Carlo-based stochastic simulation process. The results from a case study showed that the proposed GLUE-MLS-E method had a comparable performance as GLUE in terms of posterior parameter estimation and predicted confidence intervals; however, it could significantly reduce the computational cost. A comparison to other surrogate models, including MLS, quadratic response surface and artificial neural networks (ANN), revealed that the MLS-E outperformed others in light of both the predicted confidence interval and the most likely value of water depths. ANN was shown to be a viable alternative, which performed slightly poorer than MLS-E. The proposed surrogate method in stochastic sampling is of practical significance in computationally expensive problems like flood risk analysis, real-time forecasting, and simulation-based engineering design, and has a general applicability in many other numerical simulation fields that requires extensive efforts in uncertainty assessment. Copyright © 2014 John Wiley & Sons, Ltd.

A coupled K-nearest neighbour (KNN) and Bayesian neural network (BNN) model was developed for downscaling daily rainfall at a single site. The KNN was used for classification of dry/wet day and rainfall typing based on rainfall magnitude. The BNN was applied for prediction of rainfall amount. The proposed method was applied to rainfall downscaling at Singapore Island. The Climate Forecast System Reanalysis (CFSR) data were used for providing large-scale predictors at a high spatial resolution; 31-years daily rainfall record at two typical weather stations on the island was used as predictand. The performance of KNN–BNN was compared with two classical downscaling tools including automated statistical downscaling tool (ASD) and generalized linear model (GLM). The study results indicated that, the proposed model performed equally good or better than both ASD and GLM, in terms of prediction of basic statistical indicators (i.e. mean, SD, probability of wet days, 90th percentile rainfall amount, and maximum rainfall); it notably outperformed others in generating narrower uncertainty intervals for all indicators, especially for monthly mean and maximum rainfall. It was also demonstrated that separation of yearly data into monthly or seasonal could considerably enhance the performance of KNN–BNN.