Review article

Validation of biophysical models: issues and methodologies. A review

¹ Agriculture Research Council, via di Corticella 133, 40128 Bologna, Italy
² Macaulay Institute, Craigiebuckler AB15 8QH, Aberdeen, UK

^* Corresponding author: g.bellocchi@isci.it

Accepted: 20 January 2009

Abstract

The potential of mathematical models is widely acknowledged for examining components and interactions of natural systems, estimating the changes and uncertainties on outcomes, and fostering communication between scientists with different backgrounds and between scientists, managers and the community. For favourable reception of models, a systematic accrual of a good knowledge base is crucial for both science and decision-making. As the roles of models grow in importance, there is an increase in the need for appropriate methods with which to test their quality and performance. For biophysical models, the heterogeneity of data and the range of factors influencing usefulness of their outputs often make it difficult for full analysis and assessment. As a result, modelling studies in the domain of natural sciences often lack elements of good modelling practice related to model validation, that is correspondence of models to its intended purpose. Here we review validation issues and methods currently available for assessing the quality of biophysical models. The review covers issues of validation purpose, the robustness of model results, data quality, model prediction and model complexity. The importance of assessing input data quality and interpretation of phenomena is also addressed. Details are then provided on the range of measures commonly used for validation. Requirements for a methodology for assessment during the entire model-cycle are synthesised. Examples are used from a variety of modelling studies which mainly include agronomic modelling, e.g. crop growth and development, climatic modelling, e.g. climate scenarios, and hydrological modelling, e.g. soil hydrology, but the principles are essentially applicable to any area. It is shown that conducting detailed validation requires multi-faceted knowledge, and poses substantial scientific and technical challenges. Special emphasis is placed on using combined multiple statistics to expand our horizons in validation whilst also tailoring the validation requirements to the specific objectives of the application.