Implementation of robust statistics in the calibration, verification and validation step of model evaluation to better reflect processes concerning total phosphorus load occurring in the catchment
•The outliers data have got great impact on the modeling results.•Results indicate the importance of the data interpretation and considering an impact of outliers on modeling results.•The article presents the results of implementation of robust statistics in the calibration, verification and validat...
Gespeichert in:
Veröffentlicht in: | Ecological modelling 2016-07, Vol.332, p.83-93 |
---|---|
Hauptverfasser: | , , , |
Format: | Artikel |
Sprache: | eng |
Schlagworte: | |
Online-Zugang: | Volltext |
Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
Zusammenfassung: | •The outliers data have got great impact on the modeling results.•Results indicate the importance of the data interpretation and considering an impact of outliers on modeling results.•The article presents the results of implementation of robust statistics in the calibration, verification and validation of model.
Obtaining a high quality of environmental modeling in the calibration, verification and validation processes causes problems for modelers all over the world. Uncontrolled variability of environmental conditions (diversity of natural phenomena, meteorological and climatic conditions occurring within the drainage basin), inaccuracy of measurement equipment, averaging the data obtained and the errors of human work vitiates data population to a greater or lesser extent. These factors impede the evaluation of modeling, described as the “goodness of fit” of modeling results to that of observational data.
Therefore, from the point of view of the correct interpretation of the results of calibration processes, and verification and validation of the model, it is essential to find a statistical tool which would allow limiting the influence of strongly outlying data on the final result of the correspondence of the model to the actual conditions. Such a tool may be constituted by robust statistics as presented in the article, or the L-estimators to be more precise, which appeared to be effective in the case of an environmental data set, which are characterized by certain measurement unreliability, and in addition, a limited number of result populations. In the research Marcomodel DNS was used, with a SWAT module, for two catchments located in Northern and Central Poland. The evaluation of modeling quality was conducted by the use of classical statistics and robust statistics. The results of the classical statistics were based on three statistical measurements: coefficient of determination (R2), were compared with percent bias (PBIAS) and Nash Sutcliffe efficiency (NSE), and with the results of these measurements after the process of winsorization. In the robust estimates calculation the strength of outliers was decreased. It may be a beneficial phenomenon worth considering in the evaluation of quality modeling of surface waters. |
---|---|
ISSN: | 0304-3800 1872-7026 |
DOI: | 10.1016/j.ecolmodel.2016.04.004 |