Origin and variability of statistical dependencies between peak, volume, and duration of rainfall-driven flood events

Floods are among the most common and impactful natural events. The hazard of a flood event depends on its peak (Q), volume (V) and duration (D), which are interconnected to each other. Here, we used a worldwide dataset of daily discharge, two statistics (Kendall’s tau and Spearman’s rho) and a conceptual hydrological rainfall-runoff model as model-dependent realism, to investigate the factors controlling and the origin of the dependence between each couple of flood characteristics, with the focus to rainfall-driven events. From the statistical analysis of worldwide dataset, we found that the catchment area is ineffective in controlling the dependence between Q and V, while the dependencies between Q and D, and V and D show an increasing behavior with the catchment area. From the modeling activity, on the U.S. subdataset, we obtained that the conceptual hydrological model is able to represent the observed dependencies between each couple of variables for rainfall-driven flood events, and for such events, the pairwise dependence of each couple is not causal, is of spurious kind, coming from the “Principle of Common Cause”.

Evaluating the Suitability of Century-Long Gridded Meteorological Datasets for Hydrological Modeling

Recent advances in climate reanalyses have led to the development of meteorological products providing information from the beginning of the last century or even before. As these data sources might be of interest to practitioners in the event of missing data from meteorological stations, it is important to assess their usefulness for different applications. The main objective of this study is to investigate the ability of two long-term reanalysis datasets (CERA-20C and 20CR) and one long-term interpolated dataset (Livneh) for supporting hydrological modeling. The precipitation and temperature data of the three datasets were first compared, downscaled, and then used as inputs to the conceptual hydrological model HBV in 168 basins in the United States. The findings suggest that the quality of all three datasets decreases the further we go back in time. Models calibrated at the beginning of the time series, where the data quality is worse, are only able to capture the general properties of the time series and thus do not show a decrease in performance as the period between calibration and validation becomes larger. The opposite is true for models calibrated at the end of the time series, which show a clear decrease in performance toward the beginning of the century. While the hydrological model driven with the interpolated datasets achieved the best performance, the results obtained with the reanalysis datasets were still informative (i.e., better than the long-term monthly mean), and they matched the performance of the interpolated dataset in a few catchments in the northwestern United States.