Sample Models Like Observations
Scientific Motivation
Model geophysical fields are not always directly comparable to observations. Observations have limited spatial sampling (e.g., we do not have surface temperature observations over much of the ocean) and may have detection limits (e.g., spaceborne lidars cannot see through thick clouds).
Despite known differences between model fields and observed products, they are frequently compared. Such observational limitations can be important when drawing conclusions about model performance. For example, Cowtan et al (2015) found that by accounting for differences in HadCRUT4 and “climate model temperature fields accounts for 38% of the discrepancy in trend between models and observations over the period 1975–2014.” This finding helped to reconcile model-observational differences in the inferred transient climate response (Richardson et al., 2016).
In order to compare models and observations, post-processing is required to transform model fields into synthetic observed fields.
Proposed Hacking
During the CMIP6 Hackathon we will:
- produce a publicly available code base that can be used to transform model fields into synthetic observed fields
- make model-based synthetic observations publicly available for CMIP6 models
- an obvious synthetic observation to include is surface air temperature, but the code could produce other synthetic observations as well (e.g., MSU temperature fields)
Anticipated Data Needs
Historical and future scenario surface fields including tas, tos, ts, sftof/sftlf, and sic.
Anticipated Software Tools
CDAT would work (NetCDF IO, regridding, masking), but other tools (e.g., xarray, ESMF) could also work. The goal would be to produce a library (e.g., in GitHub) that would be easy to use (e.g., with a conda environment) along with CMIP6 synthetic observation datasets.
Desired Collaborators
I proposed this idea because it seems like this is a project that would be generally useful to the climate community. This project could benefit from scientists who have experience in software development (to write a solid, reliable library) or scientists who see value in creating software to create their own synthetic observational dataset (from model fields).