CISL Science

Comparison of compression methods
A comparison of compression methods for reducing the size of climate model output. This figure is an example of applying different methods of data compression to one of the primary output fields from the Community Earth System Model (CESM). In this case the variable being compressed is geopotential height and each method is effecting a lossy compression. A useful benchmark is comparing the maximum error that is incurred by the compression to natural variability among ensemble members from the same simulation The red horizontal line on the plot indicates a factor of 10 decrease in variability relative to the ensemble. Besides the image-based method supported by NCEP (grib2) there are two other compression algorithms that have a small error. Of note is the performance of fpzip (24) an open source tool that does not require additional information about the field’s geometry. Initial findings for CESM output suggest that the amount of compression will vary among variables but overall one can achieve an average compression ratio of 1:5 using fpzip while controlling the amount of error. This kind of reduction in the output size could be important to maximize the amount of storage available for archiving climate model experiments.

CISL research activities support scientific computation, numerical methods, geophysical modeling, and the analysis of geophysical data and model experiments. These activities are chosen to lead the geophysics community in adopting new computational methods and mathematical tools to improve research.

Diverse scientific disciplines often share common tools and numerical methods. The kind of mathematical, computational, and physical sciences housed in CISL focus on general mathematical tools, models, and algorithms that have broad application across NCAR. Hallmarks of this research are innovative and standout contributions that not only have relevance for the overall NCAR scientific program, but also are significant in their specific area of mathematical, physical, or computational science.

The figures shown here illustrate some of the diversity of scientific research in CISL and also suggest some unifying themes. The compression of model output (top figure) not only draws on computer science but must also address the particular requirements of the geophysical application. Any solution should not compromise the scientific value of the model output and also should fit into the standard workflow of model analysis used by the community.

The development of HOMME (second figure) is part of the next generation of numerical methods for CESM. Anticipating future computing architectures, these algorithms must be designed to be highly scalable but also must limit communication between cores. Moreover, to be scientifically relevant, the algorithms must also give an accurate solution of the dynamical equations. Thus the numerical development is tied to the end goals of the communities who will use CESM.

Parallel communication improvement
Improvements in parallel communication for large numerical codes. The left figure shows the connectivity in two versions of the HOMME, a dynamical core of the NCAR Community Atmospheric Model. The main overhead in communication for this algorithm is between tiles (spectral elements) in this plot and is indicated by the color blocks around the central tile. The grey lines are additional connectivity that is required by the spectral element method but not the discontinuous Galerkin (DG) method. Using this connectivity and also implementing a strategy where additional computation is done during periods of communication latency the run time for DG has improved. The figure on the right shows the speedup in run time for a standard test case and the DG version of HOMME. Note that, as expected, the improvement becomes more pronounced for larger number of processors (cores).

The final two figures indicate some of the breadth involved in CISL research. Here a statistical model is applied to observational data not directly related to the time series of interest. These kinds of reconstructions will have benefit for forcing model simulations, but will also give insight into the natural variability of CO2 before industrialization. The hierarchical statistical models used for this application are general and have wide applicability to interpreting climate model experiments, data assimilation, and the impacts of climate change.

In summary, the seemingly diverse nature of CISL research is motivated and has impact on the core mission of NCAR to support research with both Earth System models and observations.

Atmospheric CO2 from Antarctica
A reconstruction of atmospheric carbon dioxide concentrations using ice core measurements from Law Dome, Antarctica. This figure plots the CO2 concentrations measured in bubbles trapped at different depths in a core sample from Southern Antarctica. Although deeper ice layers tend to represent older air samples, the exact relationship is more complicated due to diffusion of air within the layer as snow from the surface becomes consolidated into ice. Here a combination of a physical model for the ice core trapping process and a statistical model to handle measurement errors and to constrain the CO2 time series is used to reconstruct concentrations of atmospheric CO2 for the past 2,000 years.
Reconstruction of atmospheric CO2 concentrations
This figure further illuminates the reconstruction of atmospheric carbon dioxide concentrations using ice core measurements from Law Dome, Antarctica. It reports the reconstructed concentration time series (black) with estimates of the uncertainty (blue envelope). The solid points are the depth measurements placed at the average times associated with depth, and the open circles are direct measurements of the atmosphere from recent time. The gray points are based on a statistical sampling method to indicate the uncertainty in where the minimum value of CO2 occurs during this period. Although uniformly distributed up to approximately 1790, there is no evidence for a minimum occurring beyond 1795. This is a useful, objective measure of determining the onset for the increase in CO2 due industrialization. Besides being a contribution to paleoclimate, this example is also a testbed for developing statistical methods for inverse problems that arise for many kinds of remotely sensed data.

Some notable highlights in CISL research during FY2014 include:

  • Data centric research that extends tools for data assimilation and data analysis of complex observational data, processing of large simulations, and visualization of model output.

  • Algorithm developments that accelerate the simulation of geophysical processes and make better use of computational and storage resources.

  • Developing and evaluating computational strategies for new architectures to anticipate how codes and workflow may have to adapt to future systems.

This work is funded as specified in the following individual reports.