Thealleninstitute The Research Scientist will develop an AI-powered physics-informed hybrid climate model for multidecadal forecasts. The role requires expertise in AI, earth science, and machine learning application development.
Responsibilities
You'll get to work on:
Designing, implementing and testing a strategy for coupling NeuralGCM with the chosen ocean model. Note that NeuralGCM is written in JAX and implemented on TPUs.
Designing, implementing and testing a strategy for decomposing physical parameterizations and constraints into parts encoded based on physical equations vs. parts that can be learned (even for unseen climates) from historical observations.
Presenting results at major relevant AI and climate modeling conferences and in peer-reviewed publications.
Open-source code development as part of a tightly connected team, including daily meetings, code review, design documents and interaction with external collaborators.
Mentoring postdocs and interns.
Qualification
Your Next ChallengeYou must have a PhIt's A Bonus If YouFormal graduate-level ML coursework
Required
We seek one or more Research Scientists for a new three-year project to develop an AI-powered physics-informed hybrid climate model trained on historical data that is capable of multidecadal forecasts that are more accurate than the current state of the art. We’re looking for someone with experience in AI and earth science who is excited to push the frontier in climate model development.
This position is supported by a grant with a current period of performance ending September 22, 2029. Continuation of the position beyond that date depends on renewal, extension, or identification of additional funding. Ai2 will make reasonable efforts to notify the employee if funding for the position is expected to end.
Ai2’s Climate Modeling team is an international pioneer in using machine learning (ML) methodologies to improve on current physics-based climate models. Our fast, accurate autoregressive open-source emulator, ACE, stably reproduces the climate and weather extremes of an existing climate model or a reanalysis. We have coupled ACE to a ML ocean model, trained it to realistically account for changing CO2 concentrations, and downscaled (super-resolved) ACE outputs from their 100 km native scale to to km-scale detail. Our tight-knit team of scientists and software engineers works closely with leading physics-based climate modeling centers to obtain unique training and testing data and get expert feedback on our progress. We collaborate with other leading research groups doing related AI work.
Your Next Challenge:
Physically-based climate models discretize equations representing individual time-evolving processes like clouds, rain, wind, land and sea-ice, and ocean currents on a computational grid. Some processes (e.g. clouds) are less reliably encoded than others (e.g. winds). Hence such climate models have biases in representing present-day climate and produce an undesirably large range of projections of future climate change for a given human-caused change in CO2 or other climate forcings. Current AI-based climate models reduce present-day climate bias, but don’t generalize well to future climate change.
We are starting a 3-year project funded by google.org to develop an open source ‘AI-first’ climate model that can demonstrably project future climates more accurately than current physically-based climate models when trained on historical observations by judicious design of the AI to incorporate physical principles that reliably apply in any climate, seen or unseen. The model should be lightweight and easy for an ML-savvy climate scientist to train and deploy. Our starting point is the Google Research NeuralGCM model, a hybrid ML model that uses a conventional discretization of winds encoded in JAX and learns a column-local representation of all other atmospheric and land processes from historical data. Compared to other AI climate models, NeuralGCM uses ML in a conceptually simpler way and shows somewhat better future-climate generalization skill. It is an open-weight model but its training pipeline and documentation need to be adapted for a broader user community.
You must have substantial experience building and deploying a relevant ML application in a practical or academic research setting and be fluent in Python and collaborative coding practices.
You must have a Ph. D. in computer or computational science, applied mathematics/statistics, or atmospheric science or a related geophysical science, with at least one accepted conference paper or peer-reviewed first-authored publication making extensive use of ML.
Demonstrated excellent communications skills.
It's A Bonus If You:
Have formal training and/or practical experience with physically-based atmospheric, oceanic, or related model development.
Have previously held a professional or post-doctoral position using your machine learning skills.
Preferred
Formal training and/or practical experience with physically-based atmospheric, oceanic, or related model development
Previous professional or post-doctoral position using machine learning skills