Realistic parameterization of physical processes in a climate model is key for improving its simulation of both the historical climatic record and of potential future climate change. However, the origins of parameterization-based errors are usually not evident from inspection of model output statistics, since these reflect the nonlinear interactions of the dynamics with multiple parameterizations on climate timescales. Instead, parameterization deficiencies often are more clearly perceived at the shorter timescales that are characteristic of numerical weather prediction (NWP) models that usually are initialized using a data assimilation system (DAS).

Absent a DAS, it is still feasible to run an atmospheric climate model in NWP mode by:

1) initializing the model's atmospheric state variables realistically from a global NWP (re)analysis that is interpolated to the model's horizontal and vertical grids,

2) running the climate model in weather-forecast mode, outputting predicted variables on short timescales,

3) applying high-frequency observations to identify systematic errors in forecast variables that are strongly impacted by the parameterizations (e.g. heat/moisture/ momentum fluxes, clouds, precipitation, etc.).

Analysis of such systematic forecast errors then can guide efforts to improve selected climate model parameterizations. In turn, reductions in forecast errors resulting from parameterization improvements often are found to enhance model performance at climate timescales as well.

The U.S. Department of Energy (DOE) CAPT project is currently applying this NWP-inspired method to analyze weather forecasts made with two current-generation atmospheric climate models--the NCAR CAM3 and the GFDL AM2. High-frequency field observations recorded by the DOE's Atmospheric Radiation Measurement (ARM) program at sites representative of diverse climatic regimes (e.g. continental, maritime, or polar climates) also are key for identifying parameterization-based systematic errors in these model forecasts. Especially valuable for this purpose are “continuous forcing” datasets of land/atmosphere variables which are available for ARM's U.S. Southern Great Plains (SGP) site at hourly frequencies for the entire year 2000. For model grid points near the SGP site, it thus is possible to conduct unusually fine-grained evaluation of simulated land-atmosphere interactions over a wide range of synoptic conditions.

In this study, we focus on identifying systematic errors in model radiative and hydrological forcings, as well as land-surface response variables such as turbulent fluxes, temperatures, and humidities. We also identify systematic errors at different timescales, e.g. by considering both hourly samples and the daily averages of a surface variable, as well as its seasonal-mean diurnal cycle; further, we are able to quantitatively estimate the relative weights of bias vs. phase errors by using different statistical metrics.

Systematic errors are found to vary perceptibly according to process, season, and model. In the year-2000 summer season, for example, precipitation at the SGP site in the CAM3 model is much too excessive, while it is too weak in AM2, leading to qualitatively different errors in land-surface turbulent fluxes. Implications of such case studies for further parameterization development in these models also will be discussed.

Acknowledgments

This work was performed under the auspices of the U.S. Department of Energy (USDOE) at the University of California's Lawrence Livermore National Laboratory under contract W-7405-Eng-48. This work was supported through the USDOE's Atmospheric Radiation Measurement and Climate Change Prediction Programs which are directed by the Biological and Environmental Research program at the USDOE Office of Science.