Evaluating dynamical downscaling and bias correction methods for hydrological impact assessments

Abstract

<p>Climate change is predicted to affect the availability of water resources, including changes in the frequency and severity of hydrological extremes, such as droughts or extreme precipitation. Hydrological impact studies are critical, for example, for ensuring sustainable water resource management, food production, and economic prosperity into the future. Such impact studies are commonly based on hydrological models forced with outputs of global climate models (GCMs) that simulate future climate conditions under a range of greenhouse gas emission scenarios. Generally, global climate models are run at relatively coarse resolution – coarser than what would be required to force hydrological models. In addition, small-scale processes that are below the climate model resolution are approximated using parameterisations, leading to potential biases in some variables or processes. A range of bias-correction and downscaling methods have been developed to remove systemic biases in GCM outputs and to increase the resolution of the model output to match the spatial resolution required by the impact models.</p><p>The Bureau of Meteorology (BoM) has recently released a National Hydrological Projections service as part of the new Australian Water Outlook (https://awo.bom.gov.au). This new projections service provides estimates of future climate change impacts on Australian water resources based on an ensemble of two greenhouse gas concentration pathways, four global climate models and a total of four statistical and dynamical bias-correction and downscaling methods (one dynamical downscaling and three bias-correction methods). This presentation provides an overview of the four bias correction and downscaling methods employed as part of the service and the evaluation of these methods for hydrological impact assessments in Australia.</p><p>The following methods have been applied to raw GCM outputs: 1) a trend-preserving quantile matching approach developed for the Intersectoral Impacts Model Intercomparison Project (ISIMIP2b) (Hempel et al., 2013), 2) a multi-variate recursive nested bias-correction method (MRNBC) (Johnson & Sharma, 2012; Mehrotra & Sharma, 2016; Nahar et al., 2017), and 3) a quantile matching method optimised for preserving extreme events (Dowdy, 2019). Additionally, dynamically downscaled projections based on the CCAM regional climate model (Watterson & Rafter, 2017) were bias corrected using the ISIMIP2b method. The Australian Water Availability Project data (AWAP; Jones et al., 2009), a gridded dataset that contains climate observations (including precipitation, temperature) at 0.5 km grid resolution, was used as target dataset for the bias-correction methods. Subsequently, we forced the gridded land surface water balance model AWRA-L (Frost et al., 2018) with the bias-corrected and downscaled outputs to produce hydrological simulations for the historical period (1950-2005). We evaluated the outputs against a historical reference run using AWAP data as climate inputs. Here, we present the evaluation of bias-corrected and downscaled climate inputs (particularly precipitation and temperature) as well as impact-model simulated soil moisture, evapotranspiration and runoff over a 30-year period (1976-2005). The evaluation includes assessments of mean biases, cross-correlations, and temporal autocorrelations, as well as biases in variability and extremes at multiple time scales (monthly to multi-annual). We discuss implications of our findings for impact assessments for water resource management and outline potential uses of these methods.</p>