|
The current decade (2000-2010) may be regarded as the decade of the coming-of-age of Earth System models. Such models are coming into routine use in both research and operational settings: for understanding the planetary climate in terms of feedbacks and balances between its many components; for translating such understanding into projections that inform policy to address anthropogenic climate change; and increasingly for medium- and long-term forecasts that require coupled models as well.
These activities manifest themselves in aspects of current scientific methodology. Earth System science is becoming “big science” where experiments systematically involve large international modeling campaigns, matching in scale the observational campaigns that are responsible for producing the climate record. A key example of such a modeling campaign is the activity surrounding the Inter-Governmental Panel on Climate Change (IPCC) Assessment Reports. These reports, issued every 6 years, are a culmination of systematic and coordinated modeling experiments run at multiple institutions around the world. Figure 1 shows a list of participating IPCC institutions from the recently concluded Fourth Assessment Report (IPCC AR4) (missing ref: ). A comparative study of results from multiple models run under the same external forcings remains our best tool for understanding the climate system, and for generating consensus and uncertainty estimates of climate change. Several key papers based on the IPCC AR4 data archive at PCMDI document recent leaps in understanding of aspects of the climate system in stable and warming climates, such as ENSO (Guilyardi 2006; van Oldenborgh et al. 2001), the tropical circulation (e.g Vecchi et al. 2006), Southern ocean circulation (Russell et al. 2006), and others (missing ref: ). Other similar campaigns underway include the Aqua-Planet Experiment (APE) (missing ref: ), the ENSEMBLES project (Hewitt and Griggs 2004) as well as several older ones.
It has also become apparent that a similar molt-model ensemble approach is of utility in seasonal and interannual forecasting as well. An example of such a modeling campaign is the DEMETER project (Palmer et al. 2004). Studies (e.g Hagedorn et al. 2005) show that such operational ensemble forecasts have demonstrably better forecast skill than any individual ensemble member.
A third trend in current modeling studies is the increased use of downscaling, reviewed in Wilby and Wigley (1997). Where fine-scale simulation over some domain is sought, and it is either useless (because there is limited impact of fine-scale structure on larger scales) or impractical (for computational reasons) to extend the high resolution over the entire domain, one often creates model chains, where models over larger domains at coarser resolution are used to force finer-scale models nested within. The use of model chains is also a sort of multi-model study, where output data from one model serves as input to another. In all the approaches above, the need for data standards to enable ready access to data from diverse models is apparent.
As Earth System science increasingly comes to depend on models created from multiple components, and on comparative studies of output from such models, standardization has become a serious issue as we grapple with the practicalities of carrying out such studies. Emerging efforts at standardization of model component interfaces include the Earth System Modeling Framework (ESMF) (ESMF: Hill et al. 2004; Collins et al. 2005) and the PRISM project (missing ref: eric,sophie).
Model output data in the Earth System Science community increasingly converges on the netCDF format, and, to a lesser degree, the HDF5 format. In the weather forecasting domain, the WMO-mandated GRIB and BUFR formats (missing ref: ) continue to be used. While the data formats themselves are relatively mature, recent efforts in this domain focus on developing consistent and comprehensive metadata, data descriptors that provide human- and machine-readable information about the data necessary in interpreting its contents. Metadata vocabularies are intended eventually to enable the inclusion of data into a semantic web (Berners-Lee 1999; Berners-Lee and Hendler 2001) which human and other reasoning agents will be able to use to make useful inferences about found entities. In the climate and weather modeling domain, efforts at developing a common vocabulary for metadata have converged on the Climate and Forecasting (CF) conventions. Similar initiatives for observational data (e.g the Marine Metadata Initiative (MMI)) abound, and there are attempts underway to align the CF vocabularies with the observational ones. The Open Geospatial Consortium (OGC) is a possible mechanism to shepherd the CF conventions toward a formal standard.
This paper focuses on a key element of the metadata under development: the grids on which model data is discretized. Experience from the international modeling campaigns cited above in Section 1.1 indicates that there is a wide diversity in the model grids used; and further, it appears that this diversity is only increasing. However, in the absence of a standard representation of grids, it has been rather difficult to perform comparative analyses of data from disparate model grids. Rather, the lead institutions in these campaigns insist upon having data delivered on very simple grids, on the credible argument that the sites running the models are best placed to perform regridding operations of appropriate quality, meeting the relevant scientific criteria of conservation, and so on.
This approach was followed in the IPCC AR4 campaign, and while the resulting data archive was an extraordinary boon to data consumers (analysts of model output), the burden it placed on data producers (modeling centres) was considerable. Further, the issues surrounding regridding are common to most modeling centres, capable of being abstracted to common software. We believe a suite of common regridding methods and tools is now possible, given a grid standard.
The grid standard becomes even more necessary in considering the other sorts of uses outlined in Section 1.1, such as in model chains where gridded data from one model becomes input to another. And last but not least, multiple model grids and data transformations between them are intrinsic to modern Earth System models themselves, and are the basis for coupled model development from components developed across the entire community.
This paper proposes a grid standard: a convention for describing model grids. We have described so far its general features and purposes:
An outline of such a grid standard is the topic of this paper.
The paper is structured as follows. In Section 2 we survey the types of grids currently in use, and potentially to be used in emerging models, that the standard must cover. This includes the issue of vector fields and staggered grids. In Section 2.7 we develop the key abstractions of mosaics, required for handling nested grids and other “non-standard” tilings of the sphere. In Section 2.9 we cover the issue of masks and exchange grids, required for transformations of data between grids. In Section 3 we develop a vocabulary for describing grids in the context of the CF conventions.
created by v. balaji (balaji
princeton.edu) in emacs using Tex4HT.