Why use CSET?

Evaluation of weather and climate models requires an assessment of model characteristics, and the ability to understand what drives them. This typically involves comparing different models (whether physics-based or machine learned, or hybrid approaches) to each other and to a variety of observations across a range of model diagnostics. For model development, there is a requirement to assess the impacts of different model configurations and choices on results. Evaluation is an iterative process, and each step can unveil more questions that leads to a need for deeper investigation.

CSET aids in this by providing a flexible way to interrogate model and observational data, using diagnostics that can be quickly created by the combination of operators in operator recipes. Common operations such as reading, writing, and regridding are provided to reduce duplication of effort.

CSET provides a legacy for user-developed evaluation methods and diagnostics to be shared and documented, with well-defined working practices and review processes. It aims to embed and share best practice for evaluation and verification of weather and climate models using observational data. CSET primarily targets higher resolution atmosphere model processes (i.e. convective-scale with grid-spacing of order km’s and turbulence-scale with sub-km grid spacing) on regional to global domains across weather and climate timescales. CSET provides flexibility to provide consistent evaluation of physics-based or machine learning approaches, and to compare these directly.

CSET provides many benefits to users and developers, including:

  • Embeds and shares best practice for evaluation and verification of weather and climate models using observational data.

  • A legacy for user-developed evaluation methods and diagnostics to be shared and documented, with well-defined working practices and review processes.

  • A route for contributors to CSET to ensure legacy of newly developed diagnostics, to ensure benefit to wider weather and climate modelling communities.

  • Reproducibility, portability, accessibility, maintainability, and quality assurance of evaluation tools to support a range of applications.