Role of base flows on the mathematical structure of river flows

In this study, daily flows recorded at Billoris (2633) located on the Tigris River-Botan Creek for the period 1955-2004 are separated into surface and baseflowcomponents by the method of digital filtering. Then, the monthly surface flow andbaseflow time series are made up by aggregating the estimated daily flowcomponents at each month. In the all stages of this research it is assumed that thesurface and baseflow components estimated through digital filtering procedure wererepresentative to the actual situation.Impacts of the surface and baseflow components on the periodic (seasonal)structure of total flows are investigated by relative cumulative periodogram (linespectrum)analyses applied on the seasonal sample statistics, such as the twelvemonthly means, standard deviations, skewnesses and lag-one seasonalautocorrelations. This research have shown us that the seasonal structure of thebaseflow process is considerably effective not only on the amplitudes and angularphases of total flows but also on its distributional (such as the skewnesses) and serialdependence structure.At the second stage of the study non-seasonal stochastic time series models(namely, the Box-Jenkins type stationary mathematical models) are fitted for thefully standardized original and normalized (transformed) time series of total flowsand surface and baseflow components. These modeling attempts revealed thatbaseflow components cause to increase the serial dependence of total flows as twiceas the serial dependence of surface flow components especially at small lags. Hence,the ARMA(1,1) structure of the surface flow components is dominated by the AR(3)structure of the baseflow components, resulting an intermediate AR(3) modelstructure for the total flows.At the last stage of the study the transfer function models (TFM) with rainfallinputs, namely the ARX (na, nb, nk) models, are fitted for the monthly total flows andfor the flow components in order to compare the responses of the surface flow andbaseflow components to the areal rainfall inputs. It is found that the areal monthlyrainfall series is almost a noise, and therefore prewhitening of the input and outputsis not necessary. The TFM analyses have shown us that all the three input-outputrelationships can be represented by an ARX(2,2,0) model. However, the mostpowerful TFM is found to be the ARX(2,2,0) model of baseflow components,probably because of the strong serial dependence structure of this components.