Explicit Support of DCPP Output for CMIP6 in climpred
Decadal climate prediction is an emerging field that initializes Earth System Models (ESMs) and integrates forward an ensemble forecast for months to decades. This borrows numerical weather prediction (NWP) principles and applies it to the full Earth system for longer lead times. Recent modeling centers (e.g., MPI and NCAR) have begun to include ocean and terrestrial biogeochemistry in these predictions, allowing scientists to focus on near-term predictability in ecosystems and the carbon cycle.
However, initialized ESMs are quite daunting to work with. They tend to be 5-dimensional or higher datasets. These ensembles necessitate an initialization dimension, ensemble member dimension, lead time dimension, and then spatial dimensions (x, y, z). This generates single-variable output on the order of 10s to 100s of gigabytes, requiring the user to either pre-process with something like NCO or use out-of-memory computations with tools such as dask.
We recently released a python package (“climpred”) that wraps xarray to help make this analysis process much easier on the user. Our goal is to make climpred the standard package for everything from subseasonal to multidecadal climate prediction.
Documentation here: https://climpred.readthedocs.io/en/latest/
DCPP information: https://www.geosci-model-dev.net/9/3751/2016/
climpred has been developed by testing our functions on MPI and CESM output, since the two lead developers are from those institutes. We also currently only support annual averages, which is a huge limitation.
The goal here would be to develop climpred to explicitly support DCPP CMIP6 output. This could include making notebooks that use xarray and dask to post-process raw DCPP output into climpred-compliant xarray datasets, but might also extend to core development on climpred to support sub-annual predictions.
Anticipated Data Needs
DCPP output from the CMIP6 project.
Anticipated Software Tools
python (xarray, dask, jupyter)
Anyone interested in initialized climate prediction and open-source python development. Ideally you are comfortable with git, contributing to python packages, etc., but this would be a great opportunity to learn and practice these skills!