~/aixsci
weekly updates

fields/climate-weather

AI in weather, climate and oceans

How artificial intelligence is used to study weather, climate and the oceans: filling gaps in patchy observations, turning satellite pictures into physical measurements, and forecasting storms and seasonal swings.

How AI is used

Weather, climate and ocean science works from measurements that are always incomplete. Satellites see the surface but not the depths. Towers and gauges measure one spot well and say nothing about the next valley. Physical models fill the gaps by solving equations for air and water motion, but they are slow and their small-scale detail is approximate. Machine learning, in which a model is tuned on many examples until it reproduces known answers, is used here to bridge between these sources. In the examples on this site it takes satellite imagery and predicts the ocean currents underneath, combines rain radar and cloud data with ground gauge readings to estimate rainfall, and separates the carbon that plants absorb from the carbon the soil releases in tower flux measurements.

The models used follow the shape of the data. Images and maps are handled by convolutional networks, which scan for local patterns: one example trains a U-Net, an encoder-decoder network with shortcut links between its layers, on simulated ocean output so that sequences of temperature gradient images yield surface velocities. Time sequences spread over space are handled by transformers, which weigh how every point and moment relates to every other: one example builds such a model to forecast the El Niño cycle. Simpler tree-based methods such as random forests are used to rank which inputs matter, or to relate laboratory measurements to chemical outcomes. Much of the work is preparation: regridding, filtering, aligning in time, and correcting for rare events being rare.

Why researchers use it

Researchers reach for these methods when the relationship they need exists in the data but not in a usable equation. Rainfall intensity is tied to cloud properties, yet no tidy formula converts one into the other across a coastline's varied terrain. Surface currents leave traces in the swirls and filaments of sea surface temperature, but reading them back requires inverting a tangle of physics. A trained model stands in for that missing inverse: it learns the mapping from paired examples, often from a physical simulation where both sides are known, and then applies it to real observations.

It also stands in for cost. Running a high-resolution ocean or atmosphere model is expensive, so one can be run once and a network trained on its output used thereafter. Seasonal forecasting has a long-standing weak point in spring, when signals from the tropical Pacific are faint, and a model that can draw on patterns across several ocean basins at once is attractive where a narrower method stalls. Flux measurements give a single combined number where two opposing processes are at work, and a model constrained by physical relationships can split them apart.

In the record

Limits

A model trained on simulation output inherits that simulation's errors, and real observations carry noise and gaps the training data never showed. Gauges and towers cluster where people live, so coverage is uneven and performance away from them is hard to check. Heavy rain and strong storms are rare, so the cases that matter most are the ones with fewest examples, and the fixes for that imbalance can distort the rest. These models reproduce relationships seen in past data; under a shifting climate, those relationships may not hold. None of them explain the physics they imitate.

Written by AIxSci: the background from general knowledge, the examples from checked records.

What the AI was for

Model families

Named models

Every article

aix-00306

Benchmarking the cost of ensemble Kalman methods for calibrating chaotic models

simulation surrogate

climate-weather · ai-result

aix-00303

Neural network stands in for a weather model inside a forecast-correction scheme

simulation surrogate

climate-weather · ai-result · Quarterly Journal of the Royal Meteorological Society · 2026

aix-00283

Deep learning forecast of El Niño suggests Indian and Atlantic oceans extend warning time

forecastingsimulation surrogate

climate-weather · ai-result · Science Advances · 2026

aix-00284

Neural network turns hourly satellite heat images into Gulf Stream current maps

structure determination

climate-weather · ai-result · Nature Geoscience · 2026

aix-00285

Neural network forecasts freak waves minutes ahead from a single buoy's record

forecastingclassification

climate-weather · ai-result · Scientific Reports · 2024

aix-00286

AI fills gaps in river temperature records to track overlapping heatwaves

property predictionforecasting

climate-weather · ai-result · Nature Geoscience · 2026

aix-00288

Machine learning turns satellite cloud images into hourly rainfall maps for central Vietnam

classificationproperty prediction

climate-weather · ai-result · PLoS ONE · 2026

aix-00282

Neural network trained on a decade of records forecasts typhoon strength and position

forecasting

climate-weather · ai-result · PLoS ONE · 2024

aix-00287

Satellite image dataset extended to four channels for typhoon strength prediction

property prediction

climate-weather · ai-result · arXiv · 2026

aix-00289

Machine learning points to filter loading as main driver of levoglucosan desorption temperature

property predictionattribution

climate-weather · ai-result · Environmental Science & Technology · 2026

aix-00281

Bacteria-derived aerosol matter made to produce oxidants, then modelled from its optical fingerprints

property prediction

climate-weather · ai-supporting · Environmental Science & Technology · 2025

aix-00290

Machine learning separates weather from policy in China's winter heating air pollution

property predictiondenoisingattribution

climate-weather · ai-result · Environmental Science & Technology · 2023

aix-00291

Statistical model estimates what Makkah's air would have held without lockdown

forecasting

climate-weather · ai-supporting · Toxics · 2022

aix-00292

Neural network forecasts daily sea surface temperature up to 90 days ahead

forecasting

climate-weather · ai-result · PLoS ONE · 2026

aix-00293

Machine learning used to predict fine particle emissions at a Greek regional airport

property prediction

climate-weather · ai-supporting · Toxics · 2025

aix-00294

Neural network trained on simple equations spots tipping points in real data

classificationforecasting

climate-weather · ai-result · Royal Society Open Science · 2022

aix-00295

Diffusion model trained on past weather produces 15-day global ensemble forecasts

forecasting

climate-weather · ai-result · Nature · 2024

aix-00296

Neural networks trained on 39 years of weather data produce week-ahead global forecasts

forecasting

climate-weather · ai-result · Nature · 2023

aix-00297

Weather and climate model built from physics equations plus a trained neural network

forecastingsimulation surrogate

climate-weather · ai-result · Nature · 2024

aix-00298

Generative AI model forecasts UK rainfall 90 minutes ahead from radar images

forecasting

climate-weather · ai-result · Nature · 2021

aix-00299

Neural network forecasts weather straight from raw satellite and station observations

forecastingstructure determination

climate-weather · ai-result · Nature · 2025

aix-00280

Neural network splits forest carbon measurements into uptake and release

property prediction

climate-weather · ai-result · Global Change Biology · 2026