Is it feasible to use global-scale general circulation models (GCMs) to assess climate impacts on regional and local scales? How reliable are these methods and how well do they estimate regional climate factors such as rainfall and stream flow? What can such estimates tell us about the regional-scale impacts of climate change?

INTRODUCTION:

OVERVIEW

Global-scale climate models can be successfully employed in examining the potential climate impacts of global warming on a regional scale, using a variety of recently developed techniques. Regional climate change results (assuming a doubling of the atmospheric CO2 concentration) derived from such techniques project, for example, that the northeastern U.S. will have higher wintertime precipitation while the southwestern U.S. is projected to be substantially drier during winter. In summer, warmer global conditions are predicted to lead to increased precipitation over the southern U.S. These results would suggest that the Susquehanna River Basin, which is being examined closely, would receive higher levels of precipitation during every season in a doubled CO2 world, with the largest increases being in spring and summer.

Global climate models, coupled with careful, regional modeling and analysis techniques, are the only tool available for providing long-term predictions of future climate and for assessing the climate implications of human activities. These comprehensive models require considerable computer resources, and consequently, they resolve the Earth's atmosphere and land surface only at very coarse spatial resolution (hundreds of miles). Using this spatial resolution, the ability of global models to produce simulations of the variables essential for assessing the regional impacts of global climate change on human or ecological systems is generally limited. For example, because precipitation is highly variable in time and geographic location, the prediction of this critical variable by global models tends to be inadequate for use in evaluating the regional consequences of precipitation changes for agriculture and/or water resources.

In order to address this fundamental dilemma, two unique approaches are being explored so that results from global-scale climate models can, in fact, be successfully transformed into information that is useful in examining regional-scale changes relating to economic or ecological interests in a particular area. The first technique is called Ònesting,Ó and involves embedding a high-resolution, limited-area climate model within a global-scale general circulation model (GCM) of the atmosphere. This is now being done for the United States. With this technique, the prediction of precipitation, particularly in the central U.S., is substantially improved compared to the global-scale model. The model results show a high correspondence with observations. Furthermore, the high-resolution precipitation prediction provides a firm foundation for predicting river flow in major regions of the U.S. such as the Susquehanna River Basin which feeds into Chesapeake Bay. This has been demonstrated by an ability to simulate precipitation over the Basin and to match observed measurements of precipitation and water flow when the mesoscale model is coupled to a hydrologic model. The reason for the improved prediction of precipitation in the nested model is directly related to achieving better representation of the precipitation physics and because of the improved incorporation of topography.

Statistical techniques also have significant potential as a method of ÒdownscalingÓ (scaling from a coarse resolution model to a high spatial resolution prediction for a region). As an example, a set of so-called Òneural net transfer functionsÓ (a set of mathematical expressions) are being used to derive high-resolution precipitation predictions for the Susquehanna River Basin based on global-scale GCM predictions of the circulation and humidity - an approach similar to what is used to derive local weather forecasts. The downscaled precipitation is, once again, a close match to the observed data.

The improved ability to simulate precipitation using both downscaling methods and nested models indicates potential for greatly improved estimates of the regional impacts of climate change. For this reason, both techniques are being used to produce precipitation predictions for the initial case of a warmer world resulting from a doubling of atmospheric carbon dioxide. The nested model domain includes the entire continental United States. In winter, the northeastern U.S. is predicted to have higher precipitation (rising from an average of 1-2 mm/day to 2-4 mm/day), and the southwestern U.S. is predicted to be substantially drier. In summer, the largest changes from a doubled CO2 concentration involve increased precipitation over the southern U.S. The neural net technique, which is centered on the Susquehanna River Basin, indicates higher precipitation during every season in a doubled CO2 world, with a substantial increase (32%) in spring and summer. The smallest increases occur in the southeastern part of the Basin. Such increases would have dramatic effects on river flow, on valley communities, and on the Chesapeake Bay.