\r\npredict stormwater quantity and quality from urban catchments.

\r\nHowever, due to a range of practical issues, most models produce

\r\ngross errors in simulating complex hydraulic and hydrologic systems.

\r\nDifficulty in finding a robust approach for model calibration is one of

\r\nthe main issues. Though automatic calibration techniques are

\r\navailable, they are rarely used in common commercial hydraulic and

\r\nhydrologic modelling software e.g. MIKE URBAN. This is partly

\r\ndue to the need for a large number of parameters and large datasets in

\r\nthe calibration process. To overcome this practical issue, a

\r\nframework for automatic calibration of a hydrologic model was

\r\ndeveloped in R platform and presented in this paper. The model was

\r\ndeveloped based on the time-area conceptualization. Four calibration

\r\nparameters, including initial loss, reduction factor, time of

\r\nconcentration and time-lag were considered as the primary set of

\r\nparameters. Using these parameters, automatic calibration was

\r\nperformed using Approximate Bayesian Computation (ABC). ABC is

\r\na simulation-based technique for performing Bayesian inference

\r\nwhen the likelihood is intractable or computationally expensive to

\r\ncompute. To test the performance and usefulness, the technique was

\r\nused to simulate three small catchments in Gold Coast. For

\r\ncomparison, simulation outcomes from the same three catchments

\r\nusing commercial modelling software, MIKE URBAN were used.

\r\nThe graphical comparison shows strong agreement of MIKE URBAN

\r\nresult within the upper and lower 95% credible intervals of posterior

\r\npredictions as obtained via ABC. Statistical validation for posterior

\r\npredictions of runoff result using coefficient of determination (CD),

\r\nroot mean square error (RMSE) and maximum error (ME) was found

\r\nreasonable for three study catchments. The main benefit of using

\r\nABC over MIKE URBAN is that ABC provides a posterior

\r\ndistribution for runoff flow prediction, and therefore associated

\r\nuncertainty in predictions can be obtained. In contrast, MIKE

\r\nURBAN just provides a point estimate. Based on the results of the

\r\nanalysis, it appears as though ABC the developed framework

\r\nperforms well for automatic calibration.", "references": null, "publisher": "World Academy of Science, Engineering and Technology", "index": "International Science Index 110, 2016" }