Quantifying relative uncertainties in the detection and attribution of human-induced climate change on winter streamflow

Abstract

In spite of recent popularity for investigating human-induced climate change in regional areas, understanding the contributors to the relative uncertainties in the process remains unclear. To remedy this, this study presents a statistical framework to quantify relative uncertainties in a detection and attribution study. Primary uncertainty contributors are categorized into three types: climate data, hydrologic, and detection uncertainties. While an ensemble of climate models is used to define climate data uncertainty, hydrologic uncertainty is defined using a Bayesian approach. Before relative uncertainties in the detection and attribution study are quantified, an optimal fingerprint-based detection and attribution analysis is employed to investigate changes in winter streamflow in the Connecticut River Basin, which is located in the Eastern United States. Results indicate that winter streamflow over a period of 64years (1950–2013) lies outside the range expected from natural variability of climate alone with a 90% confidence interval in the climate models. Investigation of relative uncertainties shows that the uncertainty linked to the climate data is greater than the uncertainty induced by hydrologic modeling. Detection uncertainty, defined as the uncertainty related to time evolution of the anthropogenic climate change in the historical data (signal) above the natural internal climate variability (noise), shows that uncertainties in natural internal climate variability (piControl) scenarios may be the source of the significant degree of uncertainty in the regional Detection and Attribution study.