CliMA, a research coalition led by scientists, engineers and applied mathematicians from Caltech, MIT and NASA’s Jet Propulsion Laboratory, announced an effort to build a new Earth system model that learns directly from observations. The group said the model will draw on satellite and ground data, targeted high resolution simulations and modern computational methods to improve predictions of droughts, heat waves and extreme rainfall.
The project will combine machine learning and data assimilation with physics based simulation. CliMA described plans to train components of the model using observations and on demand high resolution simulations of phenomena such as clouds and ocean turbulence. The aim is to reduce and quantify uncertainties in climate projections so planners can better assess hazard risk and resilient infrastructure needs.
Engineers on the project said the platform will be scalable, designed to run on the fastest supercomputers and on cloud systems, and to grow to ever finer global resolution. The team committed to open science. They will release code openly and provide interfaces so external developers can build front end applications for flood risk, extreme heat, crop yield and other local impact tools. Funding comes from a consortium of private foundations and federal agencies, led by the generosity of Eric and Wendy Schmidt by recommendation of the Schmidt Futures program and the National Science Foundation, the coalition said.
Context And Local Implications
Climate is commonly defined as the long term average and variability of weather in a region, typically measured over a thirty year baseline. The climate system includes atmosphere, hydrosphere, cryosphere, lithosphere and biosphere and their interactions. Models simulate energy exchanges among these components to track Earth’s energy balance and its changes.
Climate models range from simple energy balance tools to coupled atmosphere ocean sea ice simulations and vary in horizontal resolution from greater than one hundred kilometres down to around one kilometre. High resolution datasets and regional downscaling are used to translate global projections into local risk information. Examples of modelling efforts named by climate scientists include ICON and mechanistic downscaling products such as CHELSA.
Authorities report rising global temperatures and note the need to monitor Earth’s energy imbalance. At the same time, operational forecasts show how local conditions vary. A two week forecast for New York provided with these files shows temperatures mainly in the high teens to mid twenties Celsius, bouts of rain, high humidity and gusty winds with gusts reported up to forty seven kilometres per hour. That combination illustrates why the CliMA team wants models that link global signals to city scale impacts.
