Improving the Effectiveness of U.S. Climate Modeling (2001) / Chapter Skim
Currently Skimming:
Climate Models, Observations, and Computer Architechtures
Climate Models, Observations, and Computer Architechtures
Pages 13-28
The Chapter Skim interface presents what we've algorithmically identified as the most significant single chunk of text within every page in the chapter.
Select key terms on the right to highlight them within pages of the chapter.
From page 13... ...
, density, and sea ice in the oceans over time. Similarly, atmospheric scientists build models of the atmosphere incorporating surface geography and orography and the amount and distribution of gases in air (N2, O2, CO2, H2O and the more minor gases)
|
From page 14... ...
(Figure adapted from McGuffie and Henderson-Sellers, 1997. "A Climate Modelling Primer." Figure 1-3 John Wiley and Sons, New York.)
|
From page 15... ...
15 Land specialists build models containing factors influencing the distribution and runoff of water on the surface, soil moisture, the growth of vegetation (which in part determines land albedo, the amount of evapotranspiration, and the uptake of carbon dioxide) and include the geography and topography of the land surface.
|
From page 16... ...
Also, cost saving measures for weather observations can put long time series, of major value for climate, in jeopardy. It was a conclusion of NRC (1999b)
|
From page 17... ...
Because observations cost an order of magnitude more and the infrastructure to maintain sustained observations is again an order of magnitude more than any likely modeling infrastructure, the problem of producing and delivering climate information is, to first order, one of creating and maintaining a climate observing system. Improving Climate Models with Observations Climate models are built using our best scientific knowledge of the processes that operate in the atmosphere, ocean, land, and cryospheric systems, which in turn are based on our observations of these systems.
|
From page 18... ...
A similar method to improve the parameterizations of processes consists of confronting models with data taken from field programs designed to illuminate physical processes not adequately resolved by routine weather observations. Combinations of the two methods can be used.
|
From page 19... ...
Coupled climate models can be used to probe for predictability in the climate system on longer time scales when no observational data exists. A long simulation can be run and the model output treated as if it were true observational data.
|
From page 20... ...
An important recent use for models is to downscale information from the resolution at which the climate is simulated by global models, to a much smaller region at much higher resolution in order to capture the local characteristics of the specific region. Output from the global climate model is used as lateral boundary conditions for much higher resolution regional atmospheric models that then capture the local peculiarities of terrain and orography and, ideally, return details on local weather and weather changes under different climatic regimes.
|
From page 21... ...
4. Tightly integrated VPP systems with distributed shared memory using high-performance processors, memory, and interconnection network custom-designed to work together.
|
From page 22... ...
compared to custom-designed vector processors. This means, of course, that proportionately more commodity processors must be used to get the same
|
From page 23... ...
3. Finally, the large size and reduced performance of the interconnection network compounds the slowness of the memory chips, resulting in long delays (high latency)
|
From page 24... ...
systems have much higher peak performance per PE (~ 3 Gflops/PE for the NEC SX-4) than do cache-based distributed memory machines (~ 0.5 Gflops/PE on the SGI Origin 2000)
|
From page 26... ...
If the component models of a coupled model run in parallel, load imbalance can cause components to sit idle while waiting to receive information from another component.
|
From page 27... ...
3667 Mhz 39.6 Gflops peak 3667 Mhz 39.6 Gflops peak 41 PE 41 PE 4512 PEs 420 PE 5360 Mflops 52156 Mflops 545317 Mflops 527884 Mflops 627% 622% 66.6% 615% 76 716 The entries in the table are: 1manufacturer; 2model; 3processor characteristics; 4number of processing elements (PEs) ; 5sustained rate obtained with model (units are either "fdpd" = forecast days per day, or "Mflops")
|
From page 28... ...
We then compare them using the ratio of the sustained performance per processor between vector and microprocessor machines for both serial and parallel execution. The implications of the different performance characteristics will be explored in the following sections.
|
Key Terms
This material may be derived from roughly machine-read images, and so is provided only to facilitate research.
More
information on Chapter Skim is available.