Environmental data inference through fuzzy clustering

Retrieving the relevant information from noisy environmental data is a difficult task, which requires advanced filtering techniques. This paper presents a method for extracting representative prototypes from a set of data containing daily and seasonal fluctuations. It is based on a combination of wavelet filtering and fuzzy clustering. Discriminating features are extracted with a Fuzzy Maximum Likelihood Estimates (FMLE) clustering algorithm, which was selected for its variable metric enabling the adaptation of the cluster shape and volume to the data. The results show that the discriminating power of this algorithm is considerable, as demonstrated by the application to two differing domains: the discrimination of dissolved oxygen circadian cycles in the Orbetello lagoon and the daily and seasonal fluctuations in photosynthesis in the Arno river. In both cases the isolated patterns have a clear ecological meaning and reveal the relevant ecosystem variations on differing time-scales.