Mapping the confusion between classes - 'fuzzy' maps

Habitat classes often show considerable overlap in the environmental conditions within which they occur, and, where mapping is based on acoustic properties, it is not always possible to distinguish habitats based on characteristics of acoustic reflectance. This is shown by the distribution of incorrect classifications in the error matrix. If the error matrix is used in this way, it is termed a ‘confusion matrix’. These matrices are useful tools for measuring the overlap of classes caused through confusion between signatures.
 
Overlap is particularly marked, not surprisingly, between similar habitats. This situation reflects the fact that the natural environment is best represented as continua rather than discrete and separate units. Although we cannot map with such multidimensional continua, we at least have to acknowledge the ‘fuzzy’ boundaries between habitat classes.
 
This has implications for accuracy measures because instead of predictions being either right or wrong, predictions can be nearly right. Although there are ways to accommodate this fuzziness, computing this is convoluted and representing it can be confusing. This fuzziness can be demonstrated for a map by showing which habitat classes are confused and by how much through confusion matrices. Although instructive, the quantification of fuzziness is probably not easily incorporated into any assessment of accuracy or confidence.
 
GMHM5-06_fuzzy_maps
 
Comparing the use of a ‘hard classifer’ (left) with a fuzzy classifier (right) where the fuzzy approach shows where alternative habitats may be present
 
The map on the left has used a hard classifier so that only the most likely class is shown. The map on the right has second choices (where these have a high probability) as a hatched overlay. It might be a more informative map, but is it easier for a user to read? Whilst the measure of success of a map can be increased by using fuzzy procedures and also making allowances for near misses, eventually so much allowance for ‘near’ misses can be made that the resulting maps become unreliable.

All material variously copyrighted by MESH project partners 2004-2010

| List Access Keys | Mapping European Seabed Habitats (MESH) - Home | Overview | News | Work areas in the MESH Project | Partners | Contact Us | Search | Site Map | Partner Extranet |