Data-Driven Modeling of Flows of Antalya Basin and Reconstruction of Missing Data

Abstract

The length and completeness of any hydrologic observation series increase the reliability of the results obtained by methods using the series. But the observed series generally contain gaps caused by many factors. Therefore, the imputation of missing data with appropriate methods is an important step in hydrologic analysis. This paper presents the implementation of the frequency-based imputation method in the estimation of 385 missing and 5543 observed total flow values of 23 stations in Antalya basin, Turkey. The data-driven method calculates the estimations for missing values based on the temporal and quantitative associations in existing observations by considering the frequencies of the observed value ranges. The observations of the stations in the basin show significant variations in climatic behavior. The correlations between the estimations of the frequency-based imputation method and the observations for each station vary between 0.62 and 0.98, while 87% of the correlations are higher than 0.80 and 65% of them are higher than 0.9. The obtained high correlations show that the frequency-based imputation method can be used reliably in estimating streamflow in Southern Turkey within the Mediterranean climate zone. Other statistical measures (Nash–Sutcliffe efficiency coefficient, normalized root-mean-square error and mean absolute scaled error) and the comparison with the kriging method also validate the success or the obtained results.