Automatic model calibration of combined hydrologic, hydraulic and stormwater quality models using approximate Bayesian computation. (8th July 2022)

• Record Type: Journal Article
• Authors: Chowdhury, Anupam; Egodawatta, Prasanna
• Is Part Of: Water science and technology. Volume 86:Number 2(2022)
• Page Start: 321
• Page End: 332
• Publication Date: 2022-07-08
• Subjects: approximate Bayesian computation -- automatic calibration -- stormwater pollutant processes -- stormwater quality -- water quality modeling; Water -- Pollution; Sewage -- Purification; Water quality management; Periodicals; 628.168
• DOI: 10.2166/wst.2022.207
• Languages: English
• ISSNs: 0273-1223

Abstract

A range of automatic model calibration techniques are used in water engineering practice. However, use of these techniques can be problematic due to the requirement of evaluating the likelihood function. This paper presents an innovative approach for overcoming this issue using a calibration framework developed based on Approximate Bayesian Computation (ABC) technique. Use of ABC in automatic model calibration was undertaken for a combined urban hydrologic, hydraulic and stormwater quality model. The simulated runoff hydrograph and total suspended solid (TSS) pollutograph were compared with observed data for multiple events from three different catchments, and found to be within 95% confidence intervals of the simulated results. The R programmed model was validated by comparing simulated flow with similar commercially available modeling software, MIKE URBAN output determined using mean value of parameters obtained from the calibration exercise, and performed well by satisfying statistical criteria's such as coefficient of determination (CD), root mean square error (RMSE) and maximum error (ME). The developed framework is useful for automatic calibration and uncertainty estimation using ABC approach in complex hydrologic, hydraulic and stormwater quality models with multi-input-output systems. HIGHLIGHTS: Superior performance in calibrating complex models with multiple outputs. Avoids the use of likelihood function by comparing with observed data. Applicable for urban hydrologic and stormwater quantity/quality models. Graphical Abstract