The development and application of a pattern-scaled stochastic weather generator as a computationally efficient tool to study future climate variability

Pattern scaling is a common way to emulate the future climate projections of Earth System Models (ESMs) but in most applications they are limited to emulating changes in mean climate and do not represent changes in climate variability. We have extended the pattern scaling technique to apply it to the parameters of a stochastic weather generator. Stochastic weather generators are computationally inexpensive tools that can be used to produce a suite of weather variables with long time series. Weather generator parameters under future climates are obtained using spatial climate response patterns from ESMs scaled by any global temperature change. Combining a stochastic weather generator with pattern scaling of its parameters allows the study of a wide range of future scenarios and time periods at a local scale, whilst maintaining reduced computational complexity. Here we use a Markov chain-gamma model to generate daily precipitation, and multiple linear regression to generate daily temperature. We will present the development of the pattern-scaled weather generator and illustrate its application using three case studies. Spatial response patterns for weather generator input parameters have been derived for three ESMs. 300-years of daily precipitation and temperature time series have been produced using perturbed weather generator parameters for three levels of global warming relative to 1850-1900: 1.5, 2.0 and 4.0°C. These sequences of generated weather, under different future climates, will be used to show the implications for the precipitation and temperature distributions, and frequency of extreme events. This technique provides an alternative to traditional downscaling approaches, whilst reducing ESM bias and allowing a wide emulation of climate variability.