Deriving probabilistic based climate scenarios using pattern scaling and statistically downscaled data

Abstract

This paper adopts a technique common in the dynamical climate modelling literature, that of pattern scaling, and applies it to previously available statistically downscaled station level data for Ireland for two climatically relevant variables, that of temperature and precipitation. This technique allows for the rapid development of climate scenarios for additional emissions scenarios not previously available from the GCM modelling centres. Having derived the end of century (2080s) change in both these variables for four marker emissions scenarios (A1FI, A2, B2, B1), regional response rates, or the regional rate of warming per °C global warming at each station, were calculated. The estimated ranges in regional responses at each station were then compared to regional response rates for the Irish ‘grid box’ derived from a larger sample of 14 GCMs, in order to determine if the calculated response rates were illustrative of a wider suite of GCMs. A Monte Carlo (MC) resampling approach was then employed to sample regional response rates for selected stations and for different estimates of future warming. On the basis of the MC approach, probability distribution functions (pdfs) of simulated changes in temperature and precipitation were constructed and compared to the original statistically downscaled data. The methodology and results presented represent a significant contribution to the traditional approach of statistical downscaling through the development of associated likelihoods, rather than just a change in the mean value. While the methodology presented should enable the rapid development of probabilistic based climate projections, based on a limited availability of downscaled climate scenarios, caution needs to be exercised in the interpretation of the results. While they provide a basis for risk or policy assessment, estimates of the level of risk are not independent of the method employed.