Assessing the Impact of Uncertain Climate Inputs in Hydrology in a Warming Climate

Abstract

The Intergovernmental Panel on Climate Change (IPCC) has found that the fidelity of the current generation of climate model simulations is at a point where there is a good understanding of future changes to the global climate system. However, more effort is needed to quantify the consistency of climate models in evaluating local climate changes to understand the uncertainty that model agreement or disagreement introduces. This source of uncertainty in climate projections is the focus of this thesis. This research provides frameworks to quantify the uncertainty of climate modelling outputs by firstly developing new metrics for uncertainty and secondly to use the uncertainty of climate modelling effectively in climate change impact assessments. Uncertainty in climate change impact assessments is demonstrated through a statistical downscaling process in four steps. The first step applies an uncertainty metric named SREV (Square Root Error Variance) quantify uncertainty of General Circulation Models (GCMs) upper-level atmospheric variables as the conditional spread of GCM simulations. This is the first application of uncertainty measures to variables that are often used as downscaling predictors. SREV quantifies the contribution of different sources of uncertainty and it is shown that scenario and model uncertainties in general contribute reasonably evenly to total uncertainty, with smaller contributions from the initial condition ensembles. In the second step, a new metric called Variable Reliability Score (VRS) is developed to compare the reliability of different GCM variables and rank them based on their uncertainty. It is shown that the ranked reliability of the climate variables remains relatively similar globally. The third step proposes a new framework to incorporate GCM output uncertainty in impact assessment. This framework uses a statistical method called Second Order Approximation (SOA) to use the climate variables uncertainty in statistical downscaling of streamflow. It shows that ignoring uncertainty associated with climate projections can lead to significant errors, particularly in estimates of low flows. Finally the SOA framework is expanded to also consider the interactions between atmospheric variables and their uncertainties. In this fourth step, SREV is modified to a new metric, namely EVd (Error Variance dependence), to include variable dependence by estimating the covariance of the uncertainty associated with the different GCM outputs.