Content area
Full text
1. Introduction
Climate change impact assessment can use regional climate models (RCMs) to provide higher-resolution projections than available from global climate models (GCMs). RCMs are driven by large-scale circulation patterns from the GCM being transferred through lateral boundary conditions (LBCs), as well as specified initial conditions and lower boundaries. As these models are driven by coarse-scale GCM data, they are dependent on the provision of realistic simulations of the global climate to serve as their lateral boundaries.
There is a range of literature associated with biases in GCMs, including bias in the sea surface temperature (Bruyère et al. 2014), wind fields (Capps and Zender 2008), specific humidity (John and Soden 2007), atmospheric variables in CMIP3 (van Ulden and van Oldenborgh 2006; Vial and Osborn 2012) and CMIP5 (Brands et al. 2013; Jury et al. 2015), and the low-frequency variability of precipitation (Rocheta et al. 2014a). These GCM biases result from poorly understood physical processes, model resolution, and numerical parameterizations within the GCM. While reduced when aggregated over continental spatial or yearly time scales, they result in biases in the RCM simulations to which they provide input. Many studies have shown that improper boundary conditions will affect the entire limited-area RCM domain (Caldwell et al. 2009; Rojas and Seth 2003; Warner et al. 1997; Wu et al. 2005).
A common approach to deriving useful atmospheric variables for impact studies involves performing bias correction on RCM output (Casanueva et al. 2016; Kim et al. 2015, 2016; Ruiz-Ramos et al. 2016; Teutschbein and Seibert 2012). This requires observational datasets at the spatial and temporal resolution of the RCM output, which is often not available. Most bias correction techniques do not maintain intervariable relationships that may be important for driving subsequent impact models (see Rocheta et al. 2014b; Ehret et al. 2012). However, some new approaches are attempting more sophisticated multivariable bias correction methods such as the Inter-Sectoral Impact Model Intercomparison (ISI-MIP) method (Hempel et al. 2013) and the empirical copula-bias correction method (Vrac and Friederichs 2015).
The question of whether correcting these biases in the RCM input improves the quality of the RCM simulated outputs is one that has only started to be investigated in detail. For example, Holland et al. (2010) show improvement of the RCM simulation...