Can Regional Models Be Used To Obtain Skillful Higher Spatial Resolution Climate Forecasts Decades Into The Future?

The answer is certainly NO.

An article on-line from the National Geographic entitled “No winter by 2105? New study offers grim forecasts for U.S.”
is of considerable relevance to this question. This news report is based on the Proceedings of the National Academy of Sciences (PNAS) November 1, 2005 paper by Diffenbaugh et al. entitled Fine-scale processes regulate the response of extreme events to global climate change”. This paper uses a regional model forced by a global climate model to produce “fine-scaleâ€? forecasts over the next century.

Should we accept the predictions from this study as skillful, given that to test against reality we have to wait 100 years?

As we have posted on our weblogs of July 15 and 22, 2005 (“What are Climate Models? What do they do?” and “Are Multi-decadal Climate Forecasts Skillful?” ) a necessary condition for skillful long-term climate forecasts is that all first-order climate forcings and feedbacks be included in the global climate model. This is not the case with the PNAS article. Thus, as skillful predictions, the global model is inadequate for this purpose, even before downscaling using a regional climate model.

There is another serious problem with this modeling approach, which is true even if we had accurate large-scale information. Not only is the global model deficient in its inclusion of climate forcings and feedbacks, such that information that is passed into the regional model through the lateral boundaries is deficient, but the regional model itself will not be able to retain the large-scale structure of the global model unless the regional model domain is small such that the lateral boundary conditions are close to the center of the regional model. The only value-added, therefore, of the regional model is its improved spatial resolution of surface forcing including terrain. However, to the extent these smaller-scale features are dependent on the lateral boundary conditions, they will degrade in accuracy.

In our paper,

Castro, C.L., R.A. Pielke Sr., and G. Leoncini, 2005: Dynamical downscaling: Assessment of value restored and added using the Regional Atmospheric Modeling System (RAMS). J. Geophys. Res. – Atmospheres, 110, No. D5, D05108, doi:10.1029/2004JD004721,

we concluded that regional climate models (in which the initial conditions are forgotten; Type 2 through Type 4 regional model applications as presented in Table 1 in Castro et al.) do not retain the larger-scale atmospheric structure except for very small regional domains, even when real-world observed larger-scale analyses (a Type 2 application) are inserted as lateral boundary conditions into the regional model.

The utility of the regional climate model is to resolve smaller-scale features which have a greater dependency on the surface boundary, but skill degrades if this surface forcing is significantly dependent on the lateral boundaries. In contrast, a Type 1 regional model application has observed atmospheric initial conditions which provide skillful fine-scale forecasts until the initial conditions are forgotten ( a numerical weather prediction application). Type 1 forecasts must, therefore, be more skillful than Type 2-4 regional model applications.

Thus downscaling from a global climate multi-decadal forecast using a regional climate model does not add appropriate insight to be used by policymakers. While presenting the appearance of more accuacy, since higher spatial resolution features are shown in model plots, there is no added skill over what is achieved from the global model. Unfortunately, the authors of this paper neglected to refer to (or seek to refute) our peer-reviewed paper in presenting their results and conclusions. This oversight by the climate community in the limitations of using regional climate models to downscale from multi-decadal global climate model predictions needs to be remedied.

Comments Off

Filed under Climate Models

Comments are closed.