Assessment of empirical and regression methods for infilling missing streamflow data in Little Ruaha catchment Tanzania

Water resources and engineering projects are of major importance due to great demand of water and power supplies, irrigation needs, drought mitigation and flood control. In order to plan and design these projects, complete and reliable hydrological datasets are required. Missing data can severely compromise data quality and utility. The Rufiji Water Basin Office has developed and kept a database of daily streamflow records from 1950s to-date for at least 87 gauging stations. While the majority of the records are complete, data checks revealed significant gaps. Some stations are experiencing large gaps of up to 19 consecutive years. There is no dedicated study that looked at assessment of suitable methods for infilling such gaps. This study therefore made a contribution by appraising empirical and regression rainfall runoff based methods in the Little Ruaha River catchment, a sub-catchment of the Great Ruaha River sub-basin both within the Rufiji River Basin. The methods employed included simple linear regression (with untransformed and log-transformed data), multiple linear regression (with untransformed and log-transformed data), rainfall-runoff relationship using double mass curve technique, flow duration matching and drainage-area ratio. In addition, rainfall runoff modeling using HBV-Light was done for further comparison. With exception to the rainfall-runoff relationship and HBV-Light model, all other methods relied upon data transfer from donor stations (upstream & downstream station(s)) for infilling a downstream/upstream station. Data quality and consistency checks were performed, and performances of infilling methods were evaluated based on three performance criteria namely Nash-Sutcliffe efficiency coefficient (NSE), Coefficient of determination (R2) and standard error of estimate (SE) during calibration and validation periods. Four gauging stations (2 each upstream and downstream) were separately used to infill artificially created gaps to the target station. Overall, the calibration and validation daily results at 1KA21A indicated that the flow duration matching technique and multiple linear regression methods performed better than other methods with NSE (71%; 93%) and NSE (55%; 75%) respectively. These results have a potential for wide application in other basins of Tanzania for hydrological analysis and water resource management, where missing data is very common.