Métodos multivariados aplicados no monitoramento da qualidade da água de cisternas de placas na região do Pajeú - PE

Abstract

The monitoring of water quality is of utmost importance for humanity, whether in rural or urban areas, since all living things depend on this liquid for survival, and it is getting increasingly scarce. The studies on water quality variables still benefits very little statistical methods to detect problems the same. This work aimed to develop a new approach to cluster analysis, from the combination of hierarchical methods and non-hierarchical clustering, showing that these statistical techniques can contribute effectively in studies related to quality variables gives water. This study will use a database related to quality variables gives water collected in cisterns plaques in four settlements located in the city of Serra Talhada Pajeu in the region in the interior of Pernambuco. The methodology used here were the multivariate statistical methods specifically the hierarchical cluster analysis and non-hierarchical and the silhouette of statistical indices as validation criteria of the obtained clusters. The idea is basically to compare the various clustering methods as the "quality" of the groups obtained, relating the quality of water variables of the plaques of cisterns, the various methods used. The groups obtained for cisterns now studied showed that in general the cisterns C14, C49 and C90, C98 are similar among themselves, and dissimilar to other groups according to hierarchical methods of grouping. Regarding the size of the groups, which is generally observed Group 1 scored the highest number of cisterns both in the hierarchical method with the Euclidean distance as the non-hierarchical method k - medoid, while the grouping of hierarchical methods with the metric of canberra and non hierarchical clustering k - mean and fuzzy grouping this behavior was not observed, and consequently the cisterns were better distributed among all groups. The Correlation cophenetic has indicated as better grouping of the average bond method. In fuzzy clustering only two groups were left with seven end eight cisterns respectively, as in the non-hierarchical clustering k - mean hear the formation of three groups which roughly classified the same number of cisterns, and yet it has been found that this grouping method rated a group of dissimilar cisterns of other cisterns classified in other groups where these tanks are easily identified, the is the cisterns C64, C65, C66, C71, C79 and C100 have rates fecal coliforms beyond abnormality in other quality variables gives water. The k method - medoid also formed three groups, wherein the group 1 allocated many cisterns while groups 2 and 3 were with few observations where cisterns C75, C64 and C53 are the centroids or cisterns which represent each group respectively. Therefore, it is concluded that the hierarchical methods and non-hierarchical clustering aided by the silhouette of statistical index can, quite satisfactorily monitor the water quality boards tanks in a semiarid region of Nordeste Brasileiro.