Estimation of streamflow recession parameters: New insights from an analytic streamflow distribution model

Abstract

Streamflow recession analysis characterizes the storage-outflow relationship in catchments. This relationship, which typically follows a power law, summarizes all catchment-scale subsurface hydrological processes and has long been known to be a key descriptor of the hydrologic response. In this paper, we tested a range of common recession analysis methods (RAMs) andpropose the use of an analytic streamflow distribution model as an alternative method for recession parameter estimation and to objectively compare different RAMs. The used analytical model assumes power law recessions, in combination with a stochastic process for streamflow triggering rainfall events. This streamflow distribution model is used in the present framework to establish reference values for the recession parameters via maximum likelihood estimation.The model-based method has two main advantages: (a) joint estimation of both power law recession parameters (coefficient and exponent), which are known to be strongly correlated, and (b) parameter estimation based on all available streamflow data (no recession selection).The approach is applied to five rainfall-dominated catchments in Switzerland with 40 years of continuous streamflow observations. The results show that the estimated recession parameters are highly dependent on methodological choices and that some RAMs lead to biased estimates.The recession selection method is shown to be of prime importance for a reliable description of catchment-scale recession behaviour, in particular in presence of short streamflow records. The newly proposed model-based RAM yields robust results, which supports the further development of this method for comparative hydrology and opens new perspectives for understanding the recession behaviour of catchments.

Recession analysis revisited: impacts of climate on parameter estimation

“…The log-normal distributions for event magnitude and duration are chosen for the synthetic hydrographs because the distributions for Lookout Creek are skewed right and Fig. S1), which is also consistent with other skewed-right precipitation distributions in previous studies (Begueria et al, 2009;Selker and Haith, 1990). Recharge events were created with log-normally distributed inter-arrival times (µ = 2.5, σ = 1) and event magnitudes (µ = 1 d, σ = 1) where both values are normalized by timescale and the unit hydrograph respectively, resulting in dimensionless quantities.…”

Spatial and Temporal Patterns in Baseflow Recession in the Continental United States

“…Reference gages are identified as the least-disturbed watersheds within each of the 12 major ecoregions of the United States and are generally free of obstructions and major dams (Falcone, 2011). Though there is no general consensus on methods for defining and selecting recession events, robust analyses of sensitivity and uncertainty for event-based recession selection techniques for 16 watersheds in the Pacific Northwest, United States (Dralle et al, 2017), and 40 watersheds in Switzerland (Santos et al, 2019) have argued for a generally unrestrictive selection procedure, with the exception that both streamflow and the absolute value of its derivative must decrease monotonically (i.e., a concave hydrograph). We have broadly followed these recommendations.…”

“…However, it is becoming increasingly common to interpret the data point-cloud as a mathematical artifact (Jachens et al, 2019;Sánchez-Murillo et al, 2015) and to acknowledge that point-cloud based regression methods systematically underestimate the nonlinearity of observed recession events (Santos et al, 2019;Tashie et al, 2020). Instead, many researchers have begun to assess watersheds according to the typical values of recession parameters calculated using individual recession events (e.g., Dralle et al, 2017;Shaw & Riha, 2012).…”

Further Reply to First Reviewer

“…> We use dq/dt as a proxy for how the stream network contracts (recession) and expands in combination with overland flow. For recession analysis (the more classical use of dq/dt), the rate of flow change can be calculated in many ways (Santos et al 2019), the most frequent being the calculation of dq/dt from a streamflow record on a daily time step. We decided to use a two-day window to get a more robust estimate of when we are in a period of increasing streamflow versus a period of decreasing…”