Completion Report for Statistical analysis of Biscayne Bay Water Quality Data

Nnaji, Soronadi

An observational study was performed on a set of water quality data obtained in time and space over the Biscayne Bay. In all, about 131 sampling episodes were carried out over a period of about 12 years. Data summaries of eleven variables were calculated for all the sampling stations and the categories of sampled depths. Based on a multivariate analysis of the variances it was concluded that there is no significant variation in the sampled variables with depth. on the average, the ammonium ion, turbidity and dissolved oxygen were the most variable while the salinity and the conductivity were the least variable. Heuristic approaches that rely on standard hypothesis testing were applied to investigate the significance of a 1979 change in water quality variables measuring techniques. The correlation ·matrix of the before change and that of the after change data were used in this regard. It was not established that the change was significant at the 5% level. Two clustering methods were applied to the data collected between 1979 and 1984. The Single Linkage method provided a dendrogram from which the number of clusters required was determined. The K-Means method, with the number of cluster k=3, was also applied. An additional application of this method for k=2 was made for comparison. The conclusion was that there were basically two broad clusters of stations - those at the mouth of the canals and near shore and those that are off shore. A secondary analysis of the off shore stations provided a basis for deleting some of these stations without losing valuable information on the microecosystem around the deleted station. The records of the Moody and Mowry canal discharges into the Bay were combined with the sampled bay water quality data in order to investigate the impacts of such discharges. Thresholds were assumed for the discharge and water quality variables. These assumed values were used to define four categories of discharge variable value combination. Contingency tables conditioned on the canal and water quality variable were then formulated using the resulting categorical data. Because of the arbitrary choice of thresholds, only relative inferences could be made concerning the pattern of frequencies of the table of sampling stations by categories. Definite trends in the percent frequencies of the different categories with distance from the canal discharge point were observed for salinity and conductivity. Temperature showed a fairly uniform distribution of percent frequencies while Turbidity did not show any patterns. The Kendall rank correlation method was used to investigate possible trend in the data sets and to determine an optimal sampling frequency. Independence between sucssesive observations was assumed to be established· if the calculated rank score exceedance probability was greater than 0.05. Based on this criterion, the monthly observations were found to be independent with an average probability of 0.48. The probability increased to 0.65 for sampling frequency of once in two months and did not change significantly with increasing frequency. Thus, the analyses did not indicate any trend or seasonality. This, however, may be because measurements were made just once, or in a few cases only several times in any one month.