Applicability of Partially Observed Runoff Data in Parameter Calibration

Conceptual water balance models are typically calibrated using continuously observed runoff data of a basin. Most African river basins, however, have limited hydrologic data especially runoff data. When selecting the years of continuously observed runoff data only, the length of calibration data may be significantly reduced. To avoid the loss of valuable hydrologic data embedded in partially observed streamflow data, this study proposes to use these partially observed runoff data in parameter calibration instead of using continuously observed data only. The proposed methodology is applied to the upper Blue Nile River Basin of Ethiopia where data are limited and hydrologic importance is high to the riparian Nile Basin system. This study evaluates the value of information of using partially observed hydrographs by comparing calibration and verification efficiencies obtained using an existing water balance model, and continuous and random calibration data sets. Continuous calibration data sets are sampled using 1 to 25 years of moving windows from the 30 year observed hydrograph while the corresponding lengths of random calibration data sets are randomly sampled from the same hydrograph. The results reveal that it requires at least 10 years (120 months) of continuous calibration data and 36 months of random calibration data for robust parameter calibration with low parameter uncertainty. It is likely that random calibration data need a shorter period to include various runoff dynamics of the basin than continuous calibration data. It is also found that parameter calibration becomes robust when including more than 30% of high flows. The proposed methodology and findings of the study are essential to estimate runoff of sparsely monitored sub-basins and to develop a hydrologic monitoring strategy for the upper Blue Nile River Basin.