Stochastic Characteristics of Rainfall and River Flow of Kainji Reservoir System

The analysis of time series is essential to detect intrinsic stochastic characteristics of hydrologic variables which are cardinal in understanding the reliability and uncertainty associated with water supply yields derived from surface water reservoirs which are central for planning purposes. To this end analysis of stochastic characteristics like moments and dependence structure of the data series were done to be able to evaluate randomness and trend pattern of the time series. Analytical tools employed were moment equations; Autocorrelation for serial dependency, trend, and Hurst Phenomenon. Analysis for serial dependency was done employing Durbin Watson standard test on the rainfall of the basin and inflow. The lag 1 autocorrelation value obtained was (0.19). The DW value computed was (1.62) for seasonal rainfall which indicates that there is no serial autocorrelation in the seasonal rainfall time series. Also the lag 1 autocorrelation value (0.667) obtained and DW value (0.67) for the seasonal inflow depicted positive autocorrelation. Analysis of long term dependency on the inflows employing Modified Rescaled Range Statistic (R/S)using Lo's(1991) q opt indicates long term dependency of the inflow time series with computed QN, q value of 3161.6.Mann-Kendall and step trend test were employed for the trend analysis. The trend test analysis for the rainfall, draw-down and inflow time series showed that the rainfall time series exhibit positive trend with Z value of 1.98 at 0.05 level of significance and Sens's slope Q of 3.235.Also the z value was greater than 1.96 i.e. the chosen significant level of 5%, with Z 0.025= ±1.96, indicating the null hypothesis that trend does not exist is void. The analysis of the step or jump trend of the reservoir draw-down shows that the critical value t_0.05 (18) = 1.734, t = 4.46 so that the hypothesis H1 is accepted, as t (μ_1 ≠ μ_2), meaning that the step or jump trend is significant and positive at 5% probability. From the annual MannKendall trend analysis of the flow, the Z statistic was determined to be -0.23, the Sen's slope Q was determined to be-0.012 indicating the alternative hypothesis that trend does exist is void.