Calibration and Validation Parameter of Hydrologic Model HEC-HMS using Particle Swarm Optimization Algorithms – Single Objective

• Published in: Majallah-i āb va khāk Vol. 29; no. 3; pp. 615 - 626
• Main Authors: R. Garmeh, A. Faridhosseini, S. M. Hasheminia, A. Hojjati
• Format: Journal Article
• Language: Persian
• Published: Ferdowsi University of Mashhad 01.02.2016
• Subjects: HEC-HMS; Particle Swarm Optimization Algorithm; Validation
• ISSN: 2008-4757; 2423-396X
• DOI: 10.22067/jsw.v0i0.37239

Introduction: Planning and management of water resource and river basins needs use of conceptual hydrologic models which play a significant role in predicting basins response to different climatic and meteorological processes. Evaluating watershed response through mathematical hydrologic models requires finding a set of parameter values of the model which provides thebest fit between observed and estimated hydrographs in a procedure called calibration. Asmanual calibration is tedious, time consuming and requires personal experience, automaticcalibration methods make application of more significant CRR models which are based onusing a systematic search procedure to find good parameter sets in terms of at least oneobjective function. Materials and Methods: Conceptual hydrologic models play a significant role inpredicting a basin’s response to different climatic and meteorological processes within natural systems. However, these models require a number of estimated parameters. Model calibration is the procedure of adjusting the parametervalues until the model predictions match the observed data. Manual calibration of high-fidelity hydrologic (simulation) models is tedious, time consuming and sometimesimpractical, especially when the number of parameters islarge. Moreover, the high degrees of nonlinearity involved in different hydrologic processes and non-uniqueness ofinverse-type calibration problems make it difficult to find asingle set of parameter values. In this research, the conceptual HEC-HMS model is integrated with the Particle Swarm Optimization (PSO) algorithm.The HEC-HMS model was developed as areplacement for HEC-1, which has long been considered as astandard model for hydrologic simulation. Most of thehydrologic models employed in HEC-HMS are event-basedmodels simulating a single storm requiring the specificationof all conditions at the beginning of the simulation. The soil moistureaccounting model in the HEC-HMS is the onlycontinuous model that simulates both wet and dry weatherbehavior.Programming of HEC –HMS has been done by MATLAB and techniques such as elite mutation and creating confusion have been used in order to strengthen the algorithm and improve the results. The event-based HEC-HMS model simulatesthe precipitation-runoff process for each set of parameter values generated by PSO. Turbulentand elitism with mutation are also employed to deal with PSO premature convergence. The integrated PSO-HMS model is tested on the Kardeh dam basin located in the Khorasan Razavi province. Results and Discussion: Input parameters of hydrologic models are seldomknown with certainty. Therefore, they are not capable ofdescribing the exact hydrologic processes. Input data andstructural uncertainties related to scale and approximationsin system processes are different sources of uncertainty thatmake it difficult to model exact hydrologic phenomena.In automatic calibration, the parameter values dependon the objective function of the search or optimization algorithm.In characterizing a runoff hydrograph, threecharacteristics of time-to-peak, peak of discharge and totalrunoff volume are of the most importance. It is thereforeimportant that we simulate and observe hydrographs matchas much as possible in terms of those characteristics. Calibration was carried out in single objective cases. Model calibration in single-objective approach with regard to the objective function in the event of NASH and RMSE were conducted separately.The results indicated that the capability of the model was calibrated to an acceptable level of events. Continuing calibration results were evaluated by four different criteria.Finally, to validate the model parameters with those obtained from the calibration, tests perfomed indicated poor results. Although, based on the calibration and verification of individual events one event remains, suggesting set is a possible parameter. Conclusion: All events were evaluated by validations and the results show that the performance model is not desirable. The results emphasized the impossibility of obtaining unique parameters for a basin. This method of solution, because of non-single solutions of calibration, could be helpful as an inverse problem that could limit the number of candidates. The above analysis revealed the existence of differentparameter sets that can altogether simulate verificationevents quite well, which shows the non-uniqueness featureof the calibration problem under study. However, the methodologyhas benefited from that feature by finding newparameter intervals that should be fine-tuned further inorder to decrease input and model prediction uncertainties.The proposed methodology performed well in the automatedcalibration of an event-based hydrologic model;however, the authors are aware of a drawback of the presentedanalysis – this undertakingwas not a completely fair validationprocedure. It is because validation events represent possiblefuture scenarios and thus are not available at the time ofmodel calibration. Hence, an event being selected as a validationevent should not be used to receive any morefeedback for adjusting parameter values and ranges.However,this remark was not fully taken into consideration, mostlybecause of being seriously short of enough observed eventsin this calibration study. Therefore, the proposed methodology,although sound and useful, should be validated inother case studies with more observed flood events.