A Parsimonious Empirical Approach to Streamflow Recession Analysis and Forecasting

Abstract

For more than a century, the study of streamflow recession has been dominated by seemingly physically based parametric methods that make assumptions on the nonlinear nature of the hydrograph recession. In practice, several studies have shown that various degrees of nonlinearity occur in the same time series and that parametric methods can underfit nonlinear recession patterns. As a result, these methods are often applied empirically to each recession segment. We propose a parsimonious data‐driven model, EDM‐Simplex, with two objectives: forecasting recession and characterizing its nonlinear behavior. We evaluate the new model through a global sensitivity analysis applied to three distinctive hydrograph series from a heterogeneous karstic catchment. The results show excellent 1‐day‐ahead forecasting performance (median Nash and Sutcliffe efficiency > 0.99) for all time series with four recession extraction methods. The sensitivity analysis also showed that empirical nonlinearity, that is, sensitivity to initial conditions, is best estimated through the absolute forecast performance and its decline over time. This indicator leads to different interpretations of nonlinearity compared to previous methods but is just as sensitive to the choice of recession extraction method. In particular, when forecasts were made for recession segments containing early stages of recession or flow anomalies, the upstream recession was significantly more linear than the downstream recession hydrographs affected by the karst. Consequently, our results support future research to interpret observed nonlinearities as a function of the catchment hydrological states for better integration of empirical, physical‐based, and operational approaches to recession analysis.

Understanding the role of media in the formation of public sentiment towards the police

“…The embedding dimensions should be picked such that the attractor used in the reconstruction is best unfolded, which is accomplished with a simplex projection 84 . Herein, CCM analysis was performed using a Python code by Delforge et al 93 Detection of latent variables. Transfer entropy cannot dismiss the presence of confounding variables 56 as it would require conditioning on all sources in the set of causal information contributors 94 , a virtually unfeasible task.…”

Identifying Causal Interactions Between Groundwater and Streamflow Using Convergent Cross‐Mapping

“…Jayawardena and Lai (1994) applied it to predict rainfall and SF and Delforge et al. (2020) to analyze and forecast SF recessions. Erkyihun et al.…”

“…For instance, Rodriguez-Iturbe et al (1989) has investigated the chaotic dynamics and limits of predictability of rainfall. Jayawardena and Lai (1994) applied it to predict rainfall and SF and Delforge et al (2020) to analyze and forecast SF recessions. Erkyihun et al (2016), Rajagopalan et al (2019), and Sangoyomi et al (1996) utilized it for hydroclimatic forecasting and simulation.…”