Quantifying uncertainty in estimation of hydrologic metrics for ecohydrological studies

Abstract

Hydrologic metrics have been used extensively in ecology and hydrology to summarize the characteristics of riverine flow regimes at various temporal scales but there has been limited evaluation of the sources and magnitude of uncertainty involved in their computation. Variation in bias, precision and overall accuracy of these metrics influences the ability to correctly describe flow regimes, detect meaningful differences in hydrologic characteristics through time and space, and define flow-ecological response relationships. Here, we examine the effects of two primary factors-discharge record length and time period of record-on uncertainty in the estimation of 120 separate hydrologic metrics commonly used by researchers to describe ecologically relevant components of the hydrologic regime. Metric bias rapidly decreased and precision and overall accuracy markedly increased with increasing record length, but tended to stabilize >15 years and did not change substantially >30 years. We found a strong positive relationship between the degree of overlap of discharge record and similarity in hydrologic metrics when based on 15- and 30-year discharge periods calculated within a 36-year temporal window (1965-2000), although hydrologic metrics calculated for a given stream gauge tended to vary only within a restricted range through time. Our study provides critical guidance for selecting an appropriate record length and temporal period of record given a degree of metric bias and precision deemed acceptable by a researcher. We conclude that: (1) estimation of hydrologic metrics based on at least 15 years of discharge record is suitable for use in hydrologic analyses that aim to detect important spatial variation in hydrologic characteristics; (2) metric estimation should be based on overlapping discharge records contained within a discrete temporal window (ideally >50% overlap among records); and (3) metric uncertainty varies greatly and should be accounted for in future analyses.