DATeS is not a new dating website, nor is it a health twist on the popular wrinkly fruit. It is, instead, a new suite that allows researchers to compare different data assimilation methodologies and understand their performance in various settings.
Data assimilation involves the integration of information from diverse sources, with the aim of accurately describing the state of a physical system. It is important in a wide range of scientific fields, including geoscience and atmospheric research, with new data assimilation algorithms an ongoing research effort.
Essential to the testing of such algorithms are numerical experiments, and therein lies the problem. Currently available testing environments are either too simplistic or too general, and many are tied to a specific model or a specific language.
To address this problem, researchers at Argonne National Laboratory and Virginia Polytechnic Institute and State University have developed a highly extensible and flexible data assimilation testing suite called DATeS.
“Our intention behind DATeS is to provide a suite that is unified, is easy to use and can be extended with minimal effort,” said Ahmed Attia, an assistant computational mathematician in the Mathematics and Computer Science division at Argonne National Laboratory.
Unified: DATeS includes the most common elements found in data assimilation applications – numerical models, data assimilation algorithms, linear algebra solvers and time discretization routines. The important point here is that these elements are independent of each other.
Application developers can use DATeS as an experimental testing pad to try out new ideas, without being overly concerned about the other components.
Easy to use: DATeS allows for easy interfacing with third-party code written in various languages – for instance, linear algebra routines in Fortran and forecast models In C.
Extensible: Users can add components by following a few rules in order to guarantee interoperability with the other pieces in the suite. In particular, an appropriate model class has to be created that inherits the corresponding base class. To help in this regard, DATeS provides base classes with definitions of all the necessary methods.
The researchers have tested DATeS successfully in various numerical experiments, demonstrating the utility of the suite through visualizations such as rank histograms.
“The design of DATeS makes it easy to generate benchmarks for a new experiment. For example, a scientist can write short scripts to iterate over a combination of settings of a filter to find the best possible results,” Attia said.
He noted that the current version of DATeS functions somewhere between professional data assimilation packages and simple research implementations.
“We still need to add support for parallel algorithms and for interfacing with large-scale models,” Attia said. “But application developers can access numerous external solvers and configure new data assimilation applications. And students can use DATeS as an interactive learning tool for experimenting with data assimilation techniques.”
For further information about DATeS, see the paper “DATeS: a highly extensible data assimilation testing suite v1.0,” A. Attia and A. Sandu, Geoscientific Model Development 12, 629-649. 2019