The Role of Uni- and Multivariate Bias Adjustment Methods for Future Hydrological Projections and Subsequent Decision-Making

Sammanfattning: Climate models are essential for generating future climate projections. However, due to simplifications, the models can produce systematic differences between output and reality, which is referred to as model bias. Bias adjustment methods aim to reduce this error, which is important for making future projections more reliable. Here, the suitability of four different bias adjustment methods was tested: distribution-based (Distribution Scaling (DS), Quantile Delta Mapping (QDM)) and non-distribution- based methods (Copula, Multivariate Bias Correction (MBCn)), of which each one univariate and one multivariate approach. The methods were assessed on climate future projections together with a non bias adjusted data set, focusing on their impacts on hydrological modelling simulations. For this, 16 hydrological signatures were analysed and categorized into: 1) water balance and flow dynamics, 2) seasonal behaviour of the flow, 3) low flow characteristics and 4) high flow characteristics. The assessment was carried out based on 50 catchments in Sweden, 10 climate models and one hydrological model. Most noticeable differences were observed between distribution-based and non-distribution-based methods, rather than between univariate and multivariate methods. Bias adjustment methods introduce half as much variation as climate models, catchments contribute substantially more to the projected signatures. Specific hydrological signatures differed regionally, such as changes in the average spring streamflow magnitude and greater bias adjustment variations in low- and high-flow frequencies, compared to varia- tions among catchments, suggesting a shift in the frequency of extreme streamflow events in the future. The choice of bias adjustment method impacted ’High flow characteristics’ the strongest. The Copula method deviated in the trend analysis by utilizing an existing trend. This research prompts further exploration of variation between current and projected future climate, or the inclusion of other variables that might impact projections, to determine the necessity of the methods.