A ranking-based sub-selection of the climate ensemble for climate impact studies

Impact modelling requires fine-scale climate information to simulate possible impacts of climate change on different sectors such as agriculture, water management or food production. Such impact models are run at a much finer spatial and temporal resolution than global or regional climate models, and therefore a pre-selection of climate model chains is required due to the computational limitations of these models. To date, there is no structured guidance for practitioners and impact modelers on how to select climate model chains. This is also the case for the Swiss Climate Scenarios (CH2018), which main products are usually communicated to the users as median, upper and lower estimates calculated for each product and time slice individually.

In this work, we present a new sub-selection climate ensemble method tailored to the users’ needs and the desired emission scenario (Representative Concentration Pathways, RCP). The method builds on the core statements of the CH2018, i.e., droughts, heat waves, heavy rainfalls, and snow-scarce winters, and complements them with three further application cases, i.e., temperature, precipitation, combined temperature and precipitation. For each application case and each RCP, three representative climate model chains are selected from the full ensemble to cover the range of the climate change signal. These include one chain corresponding to the upper, middle and lower limits of the ensemble range. The selection of climate model chains is based on the climate change signals calculated for a set of pre-selected climate indicators (e.g., mean temperature or number of hot days). Next, each climate model chain is ranked for each climate indicator according to its climate change signal calculated between the end of the century and the CH2018 reference period (i.e., 2070-2099 vs. 1981-2010). This ranking is used to divide the models into three terciles, representing the upper, lower and middle bounds of the ensemble. For each tercile, one climate model chain is next selected that best meets the selection criteria. As a result, a sub-selected ensemble with three climate model chains is proposed to the users.

The method has been developed for Switzerland and five major Swiss regions using the CH2018 GRIDDED dataset, which contains of 68 daily, transient and bias-corrected simulations of climate model chains covering the simulation period of 1981-2099. The method allows the CH2018 users to choose from three RCPs (RCP2.6, RCP4.5 and RCP8.5) and seven application cases to obtain a set of three representative climate model chains. The selected climate model chains were next successfully implemented in a hydrological impact model to assess their applicability for assessing climate impacts on hydrological variables. The method is very flexible and can easily be applied to a new or an extended climate model ensemble or to newly defined application cases.