Reanalysis Data in Hydrological Applications: A Case Study from Georgia

Reanalysis data have proven to be a valuable support for hydrologic modeling and calculation of standardized climate indices, useful tools for characterizing local climate regimes and improving water resource management in areas with limited availability of observational data. This study examines the use of ERA5 dataset emphasizing bias correction techniques to enhance their applicability and understanding their limits in a case study in Georgia. The work assesses the effectiveness of five bias correction techniques - Linear Scaling (LS), Empirical Quantile Mapping (QM-EMP), Quantile Mapping Spline Bias Correction (QM-SBC), Mean Bias Subtraction (MBS), and Simple Linear Regression (SLR) - each examined through two different bias correction approaches: classical and sliding window, applied to daily and monthly reanalysis time series. Observational climate data are scarce in Georgia, therefore the opportunity of using reanalysis data for hydrological studies is of great interest for engineering applications. In this study, performed in collaboration with Idrostudi S.r.l., one of the foremost European engineering professional services consulting firms, the extraction of ERA5 data for the entire nation of Georgia was performed automatically by developed algorithms that also allowed to do bias correction. The algorithms, developed using the open-source programming language R, employ observed data collected by five meteorological stations across diverse climatic zones of Georgia to test and compare different bias correction methodologies. The aim is to validate the performance of bias correction methods to improve the accuracy of rainfall data generated by ERA5 reanalysis model at daily and monthly scales. The techniques were evaluated carrying out two experiments, i.e. using (i) the complete datasets and (ii) the series that were split into a calibration and validation subset; metrics such as Root Mean Square Error (RMSE) and Nash-Sutcliffe Efficiency (NSE) were used to assess the performance. The dataset undergoes a calibration phase using 70% of the data to tune the bias correction methods, followed by a validation phase with the remaining 30% to test their effectiveness. Results demonstrate that bias correction improves the quality of reanalysis data, dealing to enhanced reliability for hydrological modelling and climate index computation. The LS method has emerged as the most effective among classical techniques for bias correction in daily-scale reanalysis data when all data are available. The introduction of the Sliding Window approach has further enhanced the performance of all techniques, adapting the correction to local variations and improving accuracy for daily precipitation events. It is important to note, however, that at a monthly scale, the classic approach to bias correction already proves to be sufficiently reliable. Therefore, further enhancements through the sliding window approach are not deemed necessary for monthly corrections. In the experiment (ii), techniques such as QM-EMP, QM-SBC, and SLR proved to be more suitable for applications in climatic contexts with high variability and fragmentation. This underlines the importance of selecting the appropriate bias correction technique based on the quality and availability of data, as well as the specific objectives of the analysis. Further studies are needed for a further optimization of bias correction approaches.