## Regional Climate Models and Downscaling

In some situations, the spatial resolution of global climate model (GCM) outputs is too coarse for informing regional or local adaptation. In addition, GCMs may not be able to adequately represent the climate of a specific region with a varied and complex climate.

In these cases, it is necessary to transform GCM simulations into finer resolution climate simulations. This process is called downscaling. Descriptions of four downscaling methods follow.

Dynamically and statistically downscaled model data can be used to develop climate projections. Data downscaled using the scaling factor or bias correction methods can then be used in impact models (e.g. hydrological models, which can estimate future runoff).

**Dynamical downscaling**

This method, also known as regional climate modelling (RCM) simulates local climate using output from a GCM as input to a high-resolution climate model. Examples include CSIRO’s Conformal Cubic Atmospheric Model (CCAM[1]) and the ICTP RCM (RegCM4[2]).

**Statistical downscaling **

This method also simulates local climate using output from a GCM as input but through a statistical model. This is usually a two-step process. First, an empirical relationship between local climate variables such as rainfall and large-scale predictors such as the mean sea-level pressure is developed. The relationship is then applied to GCM simulation data to simulate local climate variables. Examples include the Generalised Linear Modelling for Daily Climate Series (GLIMCLIM[3]) and the nonhomogeneous hidden Markov model (NHMM[4]*)*.

**Scaling factor**

This method, also known as the delta method, combines the projected change from GCM or RCM outputs with observed climate data. The simple scaling technique[5] is an example.

**Bias correction **

This method is used to ‘adjust’ the mean, variance and/or distribution of GCM or RCM climate data.

All models have inadequacies due to such factors as resolution, differing internal dynamics, and model parameterizations, so different climate models can respond differently to the same inputs. Bias correction addresses this. Generally, this method is applied to generate modelled time series data required for impact modelling. Techniques include Distribution Mapping using Gamma Distribution (DM2G[6]) and the Empirical Quantile Mapping (QM[7]).

In some cases, bias correction needs observed climate data input. This is especially the case when the subsequent impact model (e.g. hydrological model) needs modelled climate time series such as rainfall. This is because GCMs and RCMs are not intended to produce exact values of climate variables on a specific day or in a given sequence, and climate model simulations data will never be exactly the same as in the observation.

[1] McGregor JL and Dix MR. 2008. An updated description of the conformal-cubic atmospheric model. *High-resolution numerical modelling of the atmosphere and ocean*, Springer, pp 51-76.

[2] Giorgi F, Coppola E, Solmon F et al. 2012. RegCM4: model description and preliminary tests over multiple CORDEX domains, *Climate Research*, 52, 7 – 29.

[3] Chandler RE. 2002. GLIMCLIM: Generalised Linear Modelling for Daily Climate Series (Software and User Guide). Department of Statistical Science, University of London.

[4] Bates BC, Charles SP. Hughes JP. 1998. Stochastic downscaling of numerical climate model simulations. *Environmental Modelling Software*, 13, 325-331.

[5] Chiew FHS, Kirono DGC, Kent DM et al. 2010. Comparison of runoff modelled using rainfall from different downscaling methods for historical and future climates, *Journal of Hydrology*, 387, 10-23.

[6] Smith A, Freer J, Bates P et al. 2014. Comparing ensemble projections of flooding against flood estimation by continuous simulation, *Journal of Hydrology*, 511, 205-219

[7] Bennett JC, Ling FLN, Post DA et al. 2012. High-resolution projections of surface water availability for Tasmania, Australia. *Hydrology and Earth System Sciences*, 16, 1287-1303.