On the Use of Different Efficiency Criteria for the Validation of a Heavy Metal Balancing Tool

Abstract

The validation of metal balancing tools is usually based on the comparison of simulated versus observed data. In our study, we applied a set of different relative and absolute criteria to evaluate the performance of the model Assessment Tool for Metals in Soils. In this process, the uncertainty of the model output and the sensitivity of model parameters were also assessed. The study includes data from 123 agricultural used top soils which are characterized by the application of different fertilizers (mineral and farmyard fertilizers, sewage sludge) resulting in diverse metal inputs into the soil. Although the most common validation criteria (coefficient of determination, error ratio between prediction and observation) indicated a good model performance in predicting the metal contents over a simulation period, the absolute measure (mean absolute difference between prediction and observation) showed that the informational value of the validation results was limited for several sites. Therefore sites with short simulation periods and/or low metal inputs are not suitable for validation, because the model uncertainty covers the metal concentration changes. Excluding such sites from the validation statistics led to evaluable and quite better validation results. Although the calculated output uncertainty was low, a further reduction can be realized by improving the database for the identified sensitive parameters (initial soil metal content and fertilizers metal concentration).