With an international agreement to limit greenhouse gases remaining ever elusive, future impacts of climate change on human and natural systems are all but assured. The eventual magnitude of these impacts increases with each year that successful mitigation measures are not in place. Thus, the need for robust quantification of the risks posed to these systems due to climate change is becoming increasingly more urgent. Fundamental questions concerning the magnitude of the impacts and the time frames over which they occur need to be answered. With this knowledge, one can then ask: What adaptation options are available? And, when would an intervention need to occur to be successful?
All climate change impacts and adaptation research relies on climate projection information supplied by the climate research community. The required spatial and temporal scales and climate variables vary substantially, depending on the particular impact and system of interest. Quantifying impacts on large-scale water supplies requires knowledge of rainfall and evaporation at daily (or monthly) time scales over large catchments (thousands of square kilometers), while urban water infrastructure requires rainfall data derived over minutes to hours, over comparatively small areas (Figure 1).

Figure 2: End of the road due to coastal erosion. Skipsea Sands, U.K. Photo by Paul Glazzard.

Coastal erosion (Figure 2) studies require information on winds (and waves) over large expanses of ocean (100 km+) for days, with related storm damage requiring knowledge of extreme wind gusts over minutes. Understanding agricultural impacts requires knowledge of rainfall, temperature, solar radiation, and winds on time scales from hours (e.g. overnight frosts) to seasons, and spatial scales down to fields. Crops also are often sensitive to particular extreme events during parts of the growth cycle, so capturing a wide variety of these often temperature-related extremes also is needed. Urban air pollution is often related to the presence of a low-level atmospheric temperature inversion which traps the produced pollution, thus these impacts require knowledge of the vertical structure of the low-level atmosphere on time scales of hours. Natural systems cover a similarly broad range of variables and spatial and temporal scales.
Climate projection information originates from Global Climate Models (GCMs), and early impacts studies used direct climate data. The GCMs have several limitations when it comes to using them to study climate change impacts. The new CMIP5 (Coupled Model Intercomparison Project Phase 5) GCMs provide a step forward in terms of scale. Even so, they use spatial resolutions with grid cells of more than 100 kilometers per side. And while some variables are saved every three hours, the rest are available at daily and longer time scales. Even these new higher-resolution GCMs operate at scales too coarse to study systems of interest in impacts and adaptation research. This scale discord is addressed by downscaling using either a statistical or dynamical approach (Figure 3).

Figure 3: An example of improved climate information from a Regional v. a Global climate model over south-east Australia. The regional model is able to capture the effects of the coast, the mountains in the south-east and inland from the east coast.

Many different statistical downscaling approaches can be found in the literature. They range from simple interpolation and scaling of observations approaches to complex statistical approaches and weather generators. These statistical approaches rely on the existence of a long historical observational record from which statistical relationships can be calculated. In practice, this limits the variables that can be downscaled to generally temperature and rainfall, and the location of observation station determines where the downscaling can be applied. These observational constraints on statistical methods cannot be changed in the short to medium term.
Dynamical downscaling, or the use of Regional Climate Models (RCMs), is often seen as the alternative to statistical downscaling though most statistical approaches also can be applied to RCM output to obtain point location data. RCMs have some advantages over statistical techniques: They simulate the entire climate system so that all climate variables of interest are available, rather than being limited to the well observed variables; and they simulate the climate across the landscape regardless of whether observations exist.
RCMs do however, have some significant limitations. One major limitation being that they are computationally expensive which places limits on their resolution, often to 10s of kilometers, and on the number of simulations that can be performed to characterize the uncertainty. They also contain biases, both inherited from the driving GCM and generated within the RCM itself.
The first of these limitations is being addressed through improving computer technology which has seen, and will continue to see, the capacity of large computer centers increase dramatically. These improvements facilitate higher spatial resolutions and more simulations. The need for more simulations to characterize the uncertainty is being further addressed through international initiatives to have multiple groups contribute simulations to the same ensembles.
The GCM community has been doing this for many years through the CMIP experiments and the RCM community is building on previous projects with the CORDEX (COordinated Regional climate Downscaling EXperiment) initiative. New research into model independence also is pointing toward ways to create more statistically robust ensembles. Through a combination of further increases in computational power and growing international cooperation, the scale issues that exist between climate projections and impacts and adaptation research will be addressed with RCMs.
This will not solve all the problems with RCMs, however, particularly when it comes to dealing with the inherit bias. Here, a role for statistical downscaling techniques will continue, and will remain a vital step between the RCM output and the impacts analysis.
The future of climate projections for impacts and adaptation research is with RCMs. The challenge now is to build a dialogue between regional climate modellers and impacts and adaptation researchers. The regional climate modellers need to focus some model development toward producing particular climate variables at spatial and temporal resolutions appropriate for impacts and adaptation research. The impacts and adaptation researchers need to take advantage of the full spectrum of climate variables available from RCMs rather than continuing to limit themselves to the well-observed variables. This dialogue has already begun to some extent in CORDEX, and was a major part in the development of a regional climate modelling project in Australia, NARCliM (New South Wales and Australian Capital Territory Regional Climate Modelling project). With this dialogue developing, and cooperation between these groups increasing, the coming decade promises great leaps forward for climate change impacts and adaptation research. Perhaps just in time.
Jason P. Evans is an Australian Research Council Future Fellow, the climate projection lead for the NARCliM project, and the AustralAsia domain coordinator for the CORDEX project. He completed his Ph.D. in regional climate and hydrology at Australian National University before continuing his work in regional climate modelling at Yale University (USA). He is currently at the University of New South Wales in Sydney, Australia. Evans has worked with regional climate models for 15 years, with a particular interest in water resources.