Interrogating empirical-statistical downscaling

Abstract

The delivery of downscaled climate information is increasingly seen as a vehicle of climate services, a driver for impacts studies and adaptation decisions, and for informing policy development. Empirical-statistical downscaling (ESD) is widely used; however, the accompanying responsibility is significant, and predicated on effective understanding of the limitations and capabilities of ESD methods. There remain substantial contradictions, uncertainties, and sensitivity to assumptions between the different methods commonly used. Yet providing decision-relevant downscaled climate projections to help support national and local adaptation is core to the growing global momentum seeking to operationalize what is, in effect, still foundational research. We argue that any downscaled climate information must address the criteria of being plausible, defensible and actionable. Climate scientists cannot absolve themselves of their ethical responsibility when informing adaptation and must, therefore, be diligent in ensuring any information provided adequately addresses these three criteria. Frameworks for supporting such assessment are not well developed. We interrogate the conceptual foundations of statistical downscaling methodologies and their assumptions, and articulate a framework for evaluating and integrating downscaling output into the wider landscape of climate information. For ESD there are key criteria that need to be satisfied to underpin the credibility of the derived product. Assessing these criteria requires the use of appropriate metrics to test the comprehensive treatment of local climate response to large-scale forcing, and to compare across methods. We illustrate the potential consequences of methodological choices on the interpretation of downscaling results and explore the purposes, benefits and limitations of using statistical downscaling.

On using principal components to represent stations in empirical–statistical downscaling

“…While there are various RCMs with different abilities to simulate local climate (Linden and Mitchell, 2009), ESD too may involve many different approaches and techniques (Wilby et al, 1998a(Wilby et al, , 1998bEasterling, 1999;Zorita and Storch, 1999;Huth, 2002;Salath, 2005;Benestad et al, 2008;Brands et al, 2013;Gutman et al, 2012). Indeed, the results of downscaling may vary depending on the chosen strategy or method, and hence the sensitivity of the results to different options must be examined (Hewitson et al, 2014). A relevant question for the projection of local climate change is how results from downscaling depend on assumptions and chosen strategies.…”

Providing future climate projections using multiple models and methods: insights from the Philippines

“…no lateral boundaries), bias-corrected SSTs from GCMs can introduce errors due to assumptions in the bias correction process. & Statistical downscaling methods vary in complexity (Hewitson et al 2014) but typically use empirical-statistical relationships between large-scale predictors (i.e. GCM-derived atmospheric parameters) and local predictands (e.g.…”

“…Whilst global climate models (GCMs) can provide large-scale future climate change projections (Stocker et al 2013), dynamical downscaling using regional climate models (RCMs; Rummukainen 2010) and statistical downscaling methods (Hewitson et al 2014) can provide more spatial detail to better inform local adaptation to climate change.…”

Evaluating the stationarity assumption in statistically downscaled climate projections: is past performance an indicator of future results?

“…There are a broad range of ESD methods of varying levels of complexity (Benestad et al 2008;Hewitson et al 2014). At their core, they generally use as input GCM simulations and observation-based datasets to determine statistical relationships that in turn are applied to transform GCM outputs into downscaled products.…”

“…In practice, ESD outputs typically are viewed as value-added products -deemed to be more credible and suitable for downstream applications than the raw GCM results from which they are derived. However, assumptions that may limit the suitability of downscaled projections for some applications often are not conveyed to or appreciated by end users (Hall 2014, Hewitson et al 2014. For past time periods, ESD performance characteristics can be determined by comparing observational datasets with ESD-generated products representing the same time period (e.g., via cross-validation (Bishop 2006;Wilks 2011)).…”