Analyse observations and produce an environmental prediction
See the requirement-level assessment and its sources.
SCIENCE UNDER AUTOMATION
AI in Earth & environmental science: capabilities and evidence
Earth science combines remote observations, physical samples, models and tests of processes that unfold in space and time.
Autonomy · up to 4.4 / 10
Model interval: 2.1–4.4. Assessed 21 October 2026.
Two equally weighted workflow families: a bounded operational task and an investigation or deployment beyond that task. Inherited subfields use this parent scenario.
Calculation and work mix
Each requirement uses the operational scale from 0 (independent execution unestablished) to 4 (reliable independent execution in the stated setting). A workflow takes its lowest requirement level. The two workflows receive equal weight; the headline is 1 + 9 × weighted level / 4.
These levels are source-linked model assignments, not measurements reported by the papers. “Up to” is the upper model bound, not a probability or a forecast date.
Analyse observations and produce an environmental prediction: 1–2 / 4.
Test a new process with field observations: 0–1 / 4.
Upper bound with 75:25 workflow weights: 4.9. With 25:75 weights: 3.8.
What has AI demonstrated?
Nine models were tested over 103 geospatial tasks. Claude Sonnet 4 scored 61.7% ± 0.7% under strict all-checks scoring.
- Test
- Primary preprint; October extended evaluation
- Comparison
- Comparators vary by study; see the linked sources.
- System and version
- GeoNatureAgent extended benchmark · 2026-10-06
- Evidence checked
- 2026-10-21
Tasks use a fixed 16-tool API for three indicators in Spain and Portugal. API task success does not establish field sampling or general Earth-science autonomy.
What does the field require?
Analyse observations and produce an environmental prediction
- Retrieve and combine the appropriate geospatial observations
Autonomy level: 1–2 / 4.
GNA’s fixed API evaluation exposes incomplete reliability in choosing and combining operations.
GeoNatureAgent extended benchmark - Produce a forecast from available state information
Autonomy level: 2–3 / 4.
Aurora demonstrates several forecasting applications with established data infrastructure.
A foundation model for the Earth system - Check spatial validity and model outputs
Autonomy level: 1–2 / 4.
Tool-task scoring and forecast comparisons support checks within their evaluated domains.
GeoNatureAgent extended benchmark · A foundation model for the Earth system
Test a new process with field observations
- Design and carry out physical sampling
Autonomy level: 0–1 / 4.
The evaluated systems consume existing observations and API data; field collection remains external to those systems.
A foundation model for the Earth system · GeoNatureAgent extended benchmark - Choose measurements that distinguish causal processes
Autonomy level: 1–2 / 4.
Geospatial analysis and prediction assist planning, without establishing an independent field investigation.
GeoNatureAgent extended benchmark - Validate across sites, instruments and conditions
Autonomy level: 1–2 / 4.
Transfer requires checks beyond the particular forecasting tasks and regional API coverage.
A foundation model for the Earth system · GeoNatureAgent extended benchmark
Research tasks
Test a new process with field observations
See the requirement-level assessment and its sources.
Can AI replace the whole workflow?
The results above measure specific tasks. The field assessment also requires completion of the connected workflow, including the requirements listed here. A field-wide replacement percentage has not been established.
Read the assessment method