iMIRACLI hackathon helps to create new benchmark dataset

iMIRACLI recently posed a hackathon challenge as part of the recent 3rd NOAA AI Workshop on Leveraging AI in Environmental Sciences.

Using the new 'ClimateBench' dataset, this challenge required the contestants to build machine learning models that were able to predict the climate response of a modern Earth system model (NorESM2) to a variety of forcers. The aim was to predict annual mean global distributions of temperature, diurnal temperature range and precipitation (including extreme precipitation) given emissions and concentrations of carbon dioxide, methane and aerosol out to the year 2100. This challenging problem required the teams to harness data from a variety of CMIP6 experiments and think about the complex spatio-temporal climate response to these varied forcing agents.

iMIRACLI ESRs joined researchers from other institutions to tackle the problem, and performed brilliantly - the top two teams 'benched cloudies' and the 'emulatean' contained four iMIRACLI ESRs between them. The winning approaches are currently being refined in advance of publication of the dataset and these models, which we hope will become a baseline for future efforts.

The completed article can now be read at https://www.essoar.org/doi/10.1002/essoar.10509765.2