Guidance Report I.2: Model calibration and validation in model-based water management

Henrik Madsen, Ann van Griensven, Anker Lajer Højberg

Research output: Chapter in Book/Report/Conference proceedingChapter in bookResearchpeer-review


Any model-based river basin management project requires proper model calibration and validation. The overall objective of model calibration is to select model parameters so that the model as closely as possible simulates the behavior of the considered system. A calibration protocol is outlined that includes the following steps: (i) Choice of calibration parameters, (ii) Specification of calibration objectives, (iii) Estimation of model parameters, and (iv) Validation of the calibrated model.

Concepts such as non-identifiability/equifinality of overparameterised models are introduced and the use of sensitivity analysis to select parameters to be estimated is presented. Often leading to multiple objectives to be maximized, the estimation problem can be formulated in a general multi-objective framework, considering multi-variable measurements, multi-site measurements and multi-response modes. The process of parameter estimation is either done manually or automatically, the latter gaining increasing attention. Manual methods take full advantage of the modeler's experience and knowledge. However, experiences are not easily transferred, potentially leading to different calibration results by different modelers. Conversely, automatic procedures are objective and require considerably less labor. However, they require numerical performance measures that may not be that easy to formulate. A section is devoted to the optimization algorithms, distinguishing local and global search methods and discussing the trade-off between effectiveness of the algorithm and the required computational load.

Model validation is essential to a river basin modeling project in order to document the predictive capabilities and credibility of the calibrated model. A hierarchical validation test scheme with increasing power is presented that distinguishes between model predictions performed on the same river basin as used for parameter estimation or on a different river basin, and whether predictions are done under the same conditions as present in the calibration period or under changed conditions. Finally, uncertainty aspects of model calibration are highlighted, with special attention devoted to the issue of non-uniqueness of the parameter estimates.

The whole is illustrated with four calibration examples, fora lumped, conceptual rainfall-runoff model, for a groundwater model, for a distributed integrated model and for a water quality model. Finally, a number of available optimization algorithm packages are described.
Original languageEnglish
Title of host publicationModelling aspects of Water Framework Directive implementation Volume 1
EditorsPeter A. Vanrolleghem
Place of PublicationLondon, U.K.
PublisherIWA Publishing
Number of pages98
ISBN (Electronic)9781780401676
ISBN (Print)9781843392231
Publication statusPublished - 30 Apr 2010

Publication series

SeriesWater Framework Directive Series

Programme Area

  • Programme Area 2: Water Resources


Dive into the research topics of 'Guidance Report I.2: Model calibration and validation in model-based water management'. Together they form a unique fingerprint.

Cite this