Updating Available Patterns of Mean Field Behavior for CMIP6 Models
The Pangeo-Enabled ESM Pattern Scaling (PEEPS) dataset simplifies and updates the generation of spatial patterns following past efforts to provide an available library of spatial patterns (average global-to-local relationships) of multiple earth system model (ESM) outputs for Coupled Model Intercomparison Project (CMIP)5 data. Using well-established linear pattern scaling methodology and cloud-hosted CMIP6 data via Pangeo, PEEPS is a repository of trained annual and monthly patterns from CMIP6 outputs for temperature, precipitation, and relative humidity for each of 38 models under 6 scenarios (where data is available). Importantly, because all training data for the PEEPS data set is cloud-based, users do not need to download and house the ESM output data to reproduce the patterns in PEEPS or to use them to provide rapid estimates of mean fields as a function of global average temperature trajectories.
This work shows that linear pattern scaling is an effective means of obtaining global-to-local relationships for CMIP6 models, as it has been in past model eras. It successfully extends the application of linear pattern scaling to the relative humidity in addition to the traditional temperature and precipitation and to monthly values in addition to annual. Most importantly, by relying on cloud-hosted training data, calculating these patterns is significantly faster than in past eras and without the need for extensive local data storage.
This research introduces the Pangeo-Enabled ESM Pattern Scaling (PEEPS) dataset providing a repository of trained patterns from CMIP6 outputs for climate variables such as temperature, precipitation, and relative humidity. Average global-to-local relationships like these spatial patterns also offer the opportunity for understanding differences among long-term outcomes from CMIP6 models, so that multisectoral modelers can use these outcomes intentionally. While calculating these spatial patterns is new, doing so in past model eras has been tremendously labor and memory-intensive. PEEPS offers a significant advancement by eliminating the need for local data storage and processing, as it utilizes cloud-based training data. The patterns are validated against the CMIP6 archive and show effective performance for annual and monthly outputs across different future scenarios, with some limitations in regions with strong climate feedbacks. Mean field patterns like those in the PEEPS dataset also support rapid plotting of future scenarios, such as that performed in hectorUI, the HECTOR simple climate model’s graphical interface tool (https://jgcri.github.io/hectorui/).
The PEEPS dataset is accessible as NetCDF files and is fully reproducible using provided Python code, which is available via GitHub in Jupyter notebook and script formats. This approach greatly reduces the effort required to update and validate trained patterns, enhancing the reproducibility of climate impact studies.