Study on Urbanisation Impact on Singapore’s Catchment Runoff

The main objective of this project was to evaluate the applicability and the robustness of (1) catchment models coupled with an optimization model for model calibration; and (2) Genetic Programming (GP) as a forecasting tool to predict the impact of urbanisation on catchment runoff. The completion of the above evaluations enables the tools to be readily used for (1) impact evaluation study of future planned urbanisation on catchment runoff; and (2) flow or flood level forecasting. 

Figure 1: Comparison between observed and SWMM-ACGA simulated hydrographs.

A coupled SWMM-ACGA (Storm Water Management Model - Accelerated Convergence Genetic Algorithm) model, for example, was first tested for its applicability and robustness on the Upper Bukit Timah catchment area since this catchment has a long historical data record. Some data records were used to determine the optimal values of the model calibration parameters while other data records were reserved for verification of the resulting optimal calibration parameter set. The proposed coupled model yields accurate prediction as shown in Figure 1. The coupled model was then applied to the Marina Bay/Kallang Basin catchments which were constrained with very few recorded data. The study shows that calibration works can be successfully guided by predetermined criteria and explicit measures replacing the conventional practice of time consuming trial-and-error manual calibration process. It should be noted that in the proposed approach, the level of calibration success can be expressed objectively whereas subjective judgment is the norm in the trial-and-error approach.

Figure 2: Comparison between observed and GP simulated hydrographs. 

On the real-time forecasting front, a relatively new evolutionary computation technique known as genetic programming (GP) was used. The GP application shows symbolic regression can approximate the functional relationship to a given set of input and output paradigms for which a functional relationship exists but is unknown. In contrast to physically based models, which require a vast amount of catchment characteristics data, GP requires only a rainfall and runoff data set. GP induced rainfall runoff relationships may serve as alternatives to conventional (physically based and conceptual) rainfall runoff models. Careful observation of GP induced relationships shows that, the cause and effect relationship between future runoff and both past rainfall and past runoff is consistent with the hydrologic process. The result shows that the runoff prediction accuracy of symbolic regression based models, measured in terms of root mean square error and correlation coefficient, is reasonably high (Figure 2). Thus, GP induced rainfall runoff relationships can be considered as a viable alternative to traditional rainfall runoff models specially for real-time flood forecasting. Another advantage of a well trained GP model is that it requires only a fraction of the computational time of the catchment model such as SWMM (Storm Water Management Model). The time reduction is even more pronounced when the catchment size is very large. The significant reduction in computational time together with its high prediction accuracy makes GP highly suitable for forecasting of flow or flood; this would give the much needed time for flood warning and flood evacuation measures.

While the well-calibrated catchment model (SWMM) is ready to be applied in urban planning and management, a well trained GP model can serve to forecast flow or flood level in advance. 

The project is a collaborative research project between Drainage Division of the Ministry of Environment and National University of Singapore.