access_PCMDIMetrics

PCMDI Metrics

PCMDI Metrics have been installed, tested and configured for routine application to ACCESS model outputs.

The attachment PMs.pdf was given at a recent workshop and summarises the implementation of the tool for the needs of the ACCCESS modelling community. Since that presentation several changes have been made to the location of the tool and its inputs and outputs to more generally serve the modelling community at NCI.

The PCMDI metrics code is now maintained under subversion and checked out into the ~access account on raijin. Now the local implementation of the PCMDI metrics expects inputs from a common area and writes out metrics to a common area, both of which found under raijin:/g/data/access.

Two examples of the outputs are attached over a range of ACCESS models and experiments from the CMIP5 set as well as ACCESS1.4. The first figure (vars_global_abs.pdf) specifies an absolute error, and the second one (vars_global_norm.pdf) specifies a normalised error. This later results is analogous to Figure 9.7 from the WG1 AR5 report, chapter 9. The former one requires the specification of levels for each variable (and level for the case of 3D ones), which can be quite time consuming; the use of an normalised error reduces this effort but both options have their advantages and disadvantages in terms of interpretation of the results.

Further work needs to be done to allow the harvesting of data from the common area so that arbitrary plots and comparisons can be made without having to compute the full set of metrics each time. In addition, it is envisaged that eventually a standard set of metrics, including figures, will be automatically generated throughout ACCESS model runs to allow modellers an objective insight into how the experiment is going.

For further information on this tool contact mark.collier@csiro.au and/or refer to the github

https://github.com/PCMDI/pcmdi_metrics# Attachments

• PMs.pdf
• vars_global_abs.pdf
• vars_global_norm.pdf