Abstract
Outliers, which are essentially unexpected data items, affect the interpretation of model results and measurements alike. They are typically screened out through statistical procedures, though this approach does not provide a real justification for discarding values, other than they appear different from the rest of the data. On the other hand, remote sensing products are frequently derived by inverting a model against measurements, a procedure that naturally leads to the estimation of some ‘cost function’, a numerical value that quantitatively expresses the ability of the model to ‘fit’ the data. This paper argues that it may be more meaningful to identify and sift outliers on the basis of this cost function, than solely on the basis of being different from some measure of central tendency. One advantage of this approach is that it will filter out data points with an excessive mismatch between the model and the data, whether or not these appear to be outliers. This approach is demonstrated by analysing specific products derived from NASA’s Multi-angle Imaging SpectroRadiometer (MISR) data, though the method is applicable to any result generated through model inversion against observational data, and is therefore of general interest to a wide range of geographical applications.