Statistical analysis
Principal components analysis (PCA) was used to group the years (2008-2020) based on physico-chemical variables and water level fluctuations. Linear regression (y = ax + c) and waveform sine regression (y = a*sin (2*p*x/b + c) analyses were performed to determine the best-fitting model to explain the patterns of lake level fluctuations (WLFs) over the years. Where both models were not significant, a Locally Weighted Scatter Plot Smoother (LOWESS, Cleveland, 1979) was used to describe the pattern of lake level fluctuations. LOWESS is based on a weighted least squares algorithm that gives local weights the most influence while minimizing the effects of outliers. A smoothness parameter (f) of 0.2 was found to adequately smooth the data without distorting the temporal patterns.
Pearson’s correlation coefficient was used to determine the concordance between WLF indicators (DLTM and Amplitude, WLamp) and fisheries variables and with water quality parameters (conductivity, turbidity, chlorophyll-a, depth, DO, temperature, TP, PO43–, NO3, TN, SiO44–, NH4+ and WQI). Both Pearson’s and linear regression analyses were conducted on log (x + 1) transformed data to meet the required assumption of normality of the dataset (Zar, 2010). The frequency distribution displayed as a histogram of pixel depth was used to group the sampling period in years from 1956 to 2021 based on the increased or decreased rate of the depth while, the lake water level – lake surface area relationship was determined using a linear regression model.
Linear regression and non-linear Gaussian distribution (Zar, 2010) were used to determine the influence of WLFs on the lake’s fishery yields and condition factor as a measure of growth. The Gaussian distribution follows a unimodal pattern and tested the hypothesis that the lake fisheries production and fish condition will correspond to optimum WLFs levels below and above which a decline is realized. All the graphical plots were implemented in the Sigma Plot software package.
RESULTS