On the uncertainty in estimation of statistical moments when analysing climate models

Sammanfattning

Climate projection ensembles are used to estimate future climate-change signals and their asso-ciated uncertainty, but finite ensemble size can affect the stability of estimated ensemble statistics. 1This thesis evaluates how ensemble size influences the estimation of regional climate-change signalsover Sweden in the CMIP5-driven EURO-CORDEX ensemble. The analysis focuses on projectedchanges in near-surface air temperature (tas) and precipitation (pr) relative to 1971–2000, usingthree Swedish regions and the RCP2.6, RCP4.5, and RCP8.5 scenarios.First, paired GCM–RCM temperature-change signals are compared descriptively to assess howthe downscaled regional signal relates to the driving global model signal. Second, a subsampling-based ensemble evaluation is applied to estimate how many ensemble members are needed to repro-duce the full-ensemble mean change and standard deviation. The evaluation uses combined mean–standard-deviation criteria under 5 % and 10 % relative-error thresholds. The same framework isapplied to the main fixed-period ensemble, a 2 ◦C warming-level ensemble, and modified ensembleconstructions.The results show that estimated sufficient ensemble size depends strongly on variable, threshold,scenario, region, and ensemble construction. Precipitation generally required larger ensemble sharesthan temperature, and the 5 % threshold was substantially more restrictive than the 10 % threshold.The modified ensembles were less stable than the main and 2◦C ensembles, implying that ensembleadequacy cannot be assessed from nominal ensemble size alone. Overall, the thesis demonstrates thatstatistical evaluation of ensemble size is necessary for interpreting regional climate-change estimatesand provides a reference point for assessing future CMIP6-based downscaling ensembles.