Abstract

Assessments of the effects of climate change typically require information at scales of 10 km or less. In regions with complex terrain, much of the spatial variability in climate (temperature, precipitation, and snow water) occurs on scales below 10 km. Since the typical global climate model simulation’s grid size is more than 200 km, it is necessary to develop models with much higher resolution. Unfortunately, no datasets currently produced are both highly accurate and provide data at a sufficiently high resolution. As a result, current global climate models are forced to ignore the important climate variations that occur below the 200 km scale. This predicament prompted the creation of a global hybrid dataset with information for precipitation, temperature, and relative humidity. The resulting dataset illustrated the importance of having high-resolution datasets and gives clear proof that regions with complex terrain require a fine resolution grid to give an accurate representation of their climatology. For example, the Andes Mountains in Chile cause a temperature shift of more than 25° C within the same area as a single 2.5° grid cell from the NCEP dataset. Fortunately the CRU, U.D., GPCP, and NCEP datasets, when hybridized, are able to provide both precision and satisfactory resolution with global coverage. This composite will enable the development of both high-resolution models and quality empirical downscaling methods — both of which are necessary for scientists to more accurately predict the effects of global climate change. Without accurate long-term forecasts, climatologists and policy makers will not have the tools they need to effectively reduce the negative effects human activity have on the earth.

Introduction

Assessments of the effects of climate change typically require information at scales of 10 km or less. In regions with complex terrain, much of the spatial variability in climate (temperature, precipitation, and snow water) occurs on scales below 10 km (Daly et al. 1994; Ghan et al. 2000; Gyalistras et al. 1998; Leung et al. 1996). Current global climate model simulations typically use grid sizes over 200 km due to limits in computational power (Ghan et al. 2000; Delworth et al. 2000; Flato et al. 2000; Emori et al. 1999; Gordon et al. 2000; Russell et al. 1999; Zhang et al. 2000; Washington et al. 2000). Since halving the grid size requires eight times more computing power, and Moore’s Law indicates this large increase in processing speed will take six years per halving, reducing the grid size to 10 km would take about 25 years. This computational limitation has prompted climatologists to develop downscaling techniques, such as “empirical downscaling,” that can produce high-resolution climatological predictions without requiring such a massive increase in computational power. One method of downscaling global climate models adds highly detailed information about local geography (mountains are the most important added feature) to the model. As demonstrated later in this paper, mountainous geography has a very significant effect on weather conditions such as temperature and precipitation. Since these two items are among the most essential features to accurately predict, downscaling techniques may provide great insight into future climatology. However, in order to create and validate models that use these techniques, it is necessary to first have a high-resolution, global dataset. Using this dataset, work in this project will continue by creating a model that generates its data via an empirical downscaling method.

Materials and Methods

In order to facilitate ease of distribution and simplicity of use of the hybrid data set, the information was published in Network Common Data Form. NetCDF is an interface for array-oriented data access and a library that provides an implementation of the interface. The netCDF library also defines a machine-independent format for representing scientific data. Together, the interface, library, and format support the creation, access, and sharing of scientific data. The netCDF software was developed at the Unidata Program Center in Boulder, Colorado and is freely available from the Unidata website (unidata.ucar.edu).

The first dataset used is a global 0.5° (about 55 km) precipitation and air temperature monthly 1950-1999 dataset (land data only) developed at the University of Delaware by Cort Willmott, Kenji Matsuura, and David Legates from an analysis of station data. This analysis only adjusts air temperature to account for the influence of topography (Willmott and Matsurra 1995; Willmott et al. 1985). The second global land dataset is the Climatic Research Unit 10-minute (about 18km) 1961-1990 climatology, also developed from station data (New et al. 2002). The CRU has more topographic sensitivity than the U.D. method and provides information for precipitation, surface air temperature, relative humidity, and water vapor pressure, but only one climatological value for each month.

Results

Plots of the hybrid data best show the success of the new dataset. The plots illustrate the importance of high-resolution data, and the close correlation of the source datasets to each other. Figure 1 illustrates just how large is the difference in resolution between older models and the current one. Figure 2 shows that the datasets used are indeed quite compatible. Figures 3, 4, and 5 illustrate that fine resolution datasets are essential to truly capture the details of weather patterns.

Discussion

As stated previously, it is essential that the size of the grid cell is small enough to accurately represent the underlying climate conditions. Without sufficiently fine resolution, many important details will be omitted, and that can definitely introduce inaccuracies into any prediction models based upon such a dataset. Additionally, in the case of empirical downscaling methods, a highly detailed and highly accurate dataset is a must. Regrettably, none of the currently available datasets can provide both suitably detailed resolution and dependable accuracy. Fortunately the CRU, U.D., GPCP, and NCEP datasets, when hybridized, are able to provide both precision and satisfactory resolution with global coverage. This composite will enable the development of both high-resolution models and quality empirical downscaling methods.

The hybridization process was designed specifically to make the resulting data as accurate as possible. Using multiple data sources enhances the reliability of the data since each data source provides independent confirmation that the other data sources are accurate. While rounding errors are a potential concern for many computationally generated datasets, the hybridization was done with floating point numbers and algorithms that minimized the effects of round off errors by paying careful attention to the order of the calculations. Additionally, after the hybrid was completed, it was compared to the original data sources and no significant differences were observed. Data for the oceanic regions were not available in high-resolution, but climatological differences over water take place over much greater areas than on land, so the oceanic resolution is sufficient for our purposes. It is important to note that translating low-resolution data into its high-resolution equivalent loses no information. In fact, the accuracy is increased because this new data is scaled by an independent high-resolution data source. All of these factors combined to yield a very accurate, high-resolution, global, hybrid dataset.

Acknowledgements

I would like to thank the U.S. Department of Energy and the National Science Foundation for giving me the opportunity to participate in the Student Undergraduate Laboratory Internships program and the chance to have an incredible learning experience at Pacific Northwest National Laboratory. I am giving special thanks to my mentor, Dr. Steven Ghan for giving me such significant work to do and for providing positive feedback. Thanks to everyone for letting me work on such an enjoyable and meaningful project.