Stochastic analysis of monthly streamflows

Monthly runoff sequences from catchments in southeastern Botswana were subjected to stochastic analysis. It was found that the stochastic component accounted for 63-80% of the variance in the log-transformed monthly runoff sequences. The stochastic component behaved as an autoregressive process of order 3, i.e. AR(3) for all the stations and was found to be greatly dominated by first-order persistence trends with the first-order autoregressive parameters ranging from 0.54 to 0.96. The contribution of higher-order persistence terms, by contrast, was found to be fairly uniform for all the stations. Total monthly runoff sequences, on the other hand, behaved as an integrated moving average process (IMA(0,1,2)) for all the stations. This model performed satisfactorily for forecasting total monthly flows for up to 12 months lead time.