Knowledge Inference from a Small Water Quality Dataset with Multivariate Statistics and Data-Mining

摘要

Multivariate analysis (MV) and data mining (DM) techniques were applied to a small water quality dataset obtained from the surface waters at three water quality monitoring stations in the Petaquilla River Basin, Panama, during the hydrological period of 2008 through 2011 for the assessment and understanding of the ongoing environmental stress within the river basin. From Factor Analysis (PCA/FA), results indicated that the factors which changed the quality of the water for the two seasons differed. During the dry (low flows) season, water quality showed to be strongly influenced by turbidity (NTU) and total suspended solids (TSS) concentrations. In contrast, during the wet (high flows) season the main changes on water quality sources were characterized by an inverse relation of NTU and TSS with the electrical conductivity (EC) and chlorides (CL), followed by significant sources of agricultural pollution. To complement the MV analysis, DM techniques like cluster analysis (CA) and classification (CLA) was applied to the data. Cluster analysis was used to separate the stations based on their levels of pollution and the classification of stations was implemented by C5.0 algorithm to classify stations of unknown origin into one of the several known groups of water quality constituents. The study demonstrated that the major water pollution threats to the Petaquilla River Basin are industrial and urban development in character and uses of agricultural and grazing land which are defined as non-point sources. The use of DM techniques was to complement the MV analysis. Taking into account the limited data, the usage of these methodologies is regarded useful in aiding water managers for implementing water monitoring campaigns and in setting priorities for improving and protecting water quality sources that are impaired due to land disturbances from anthropogenic activities.