Quantifying uncertainties in temperature projections: A factorial-analysis-based multi-ensemble downscaling (FAMED) method

Abstract In this study, a factorial-analysis-based multi-ensemble downscaling (FAMED) method is developed through incorporating multi-level factorial analysis and multiple statistical methods within a general framework. FAMED can effectively downscale climate variables as an ensemble from a global scale to a local scale, as well as disclose the individual and interactive effects of global climate model (GCM), emission scenario (ES), statistical downscaling method (SDM) on climate projection responses. Then, FAMED is applied to the City of Nur Sultan (the capital of Kazakhstan) for projecting the future changes of daily maximum, mean and minimum temperatures (Tmax, Tmean and Tmin). The mean, extreme and trend indices of temperature projections under 60 simulation chains with five GCMs, three ESs and four SDMs are examined for the period of 2021–2100. Major findings are: (i) ensemble simulations under the sixty simulation chains show that Nur Sultan would experience a warming trend (0.00046–0.00566 °C/month of Tmax, 0.00049–0.00540 °C/month of Tmean, and 0.00056–0.00542 °C/month of Tmin) in 2021–2100; (ii) comparing with the base period (1979–2004), monthly maximum, mean and minimum temperatures would increase by 3.25–7.41 °C, 1.96–5.87 °C, 2.10–6.00 °C in the future period (2075–2100); (iii) GCM is the main factor affecting the mean values of temperature projections (its contribution >65%), ES is the primary factor for the trend of temperature projections (its contribution >80%), and both GCM and SDM have important effects on the extreme values of temperature projections (the total contribution >70%); (iv) among all GCMs and SDMs, IPSL-CM5A-LR and stepwise cluster analysis (SCA) have the best performances in the model validation, demonstrating that the two tools are applicable to other cities and regions in Central Asia.