Proceedings Article | 7 October 2011
Proc. SPIE. 8174, Remote Sensing for Agriculture, Ecosystems, and Hydrology XIII
KEYWORDS: Statistical analysis, Data modeling, Remote sensing, Error analysis, Vegetation, Biological research, Analytical research, Statistical modeling, Performance modeling, Current controlled current source
The capability of models to predict vegetation biophysical variables is usually evaluated by means of one or several
goodness-of-fit measures, ranging from absolute error indices (e.g. the root mean square error, RMSE) over correlation
based measures (e.g. coefficient of determination, R<sup>2</sup>) to a group of dimensionless evaluation indices (e.g. relative
RMSE). Hence, the greatest difficulty for the readers is the lack of comparability between the different models'
accuracies. Therefore, the objective of our study was to provide an overview about the quantitative assessment of
biophysical variable retrieval performance. Furthermore, we aimed to suggest an optimal set of statistical measures. This
optimum set of statistics should be insensitive to the magnitude of values, range and outliers. For this purpose, a
literature review was carried out, summarizing the statistical measures that have been used to evaluate model
performances. Followed by this literature review and supported by some exemplary datasets, a range of statistical
measures was calculated and their interrelationships analyzed. From the results of the literature review and the test
analyses, we recommend an optimum statistic set, including RMSE, R², the normalized RMSE and some other
indicators. Using at least the recommended statistics, comparability of model prediction accuracies is guaranteed. If
applied, this will enable a better intercomparison of scientific results urgently needed in times of increasing data
availability for current and upcoming EO missions.
