Climate
34 gains9 Traces33 problems
Every measured gain and problem on the record for this topic, newest first. Updated .
AI and the climate
One square per record, darker for stronger evidence:Evidence: 3 high55 moderate18 emerging
Observational studies across countries associate more AI with lower carbon emissions and lower physical climate risk. TRV-2026-0377TRV-2026-1231 Against that, a model projects US AI servers adding water use and emissions through 2030, and the biggest tech companies' reported emissions have risen with datacentre construction. TRV-2026-0474TRV-2026-0296
A systematic review of more than 35 studies shows wearable and related sensors detected lameness in grazing sheep with up to 85% accuracy. TRV-2026-0951 A validated comparison found a machine learning correction roughly doubled the skill of ECMWF's AI forecasting model at two to six weeks and beat dynamical models from six operational centres in a 2025 real time competition. TRV-2026-1465 In a study of 1,163 advanced wheat breeding lines, cutting 570,441 genetic markers to a few hundred per trait raised prediction accuracy 14.4% over random markers. TRV-2026-2110 Gap filled satellite data matched ground monitors for hourly PM2.5 in the 50 most populous US counties (daily R2 0.85). TRV-2026-1611
Microsoft, Amazon and Google's combined emissions rose from about 101m to 119m tonnes of CO2e in the year to March 2026, and company reports link the rise to datacentre construction for AI and cloud. TRV-2026-0296 Global Energy Monitor reports US gas capacity under construction jumped 76% in early 2026, with about half of new capacity tied to AI datacentres. TRV-2026-0884 A study of China's 2000 to 2024 data found AI robotics did not help lower CO2 emissions. TRV-2026-2379 Tested on countries they had not seen, emission classification models reached only 0.637 to 0.676 accuracy, against 0.91 on a random split. TRV-2026-1743
Many tools were tested only on datasets or simulations: an apple leaf system reported 99.95% accuracy but not on farms, and a retrofit planner for UK public buildings ran 30.03 times faster only in simulated cases. TRV-2026-0262TRV-2026-1729 Estimates of AI's water and carbon cost are modelled, not metered: one paper puts GPT-3 training at 5.4 million liters of water, and data center reports do not separate AI from other workloads. TRV-2026-2359TRV-2026-0433 Claims of local harm from planned datacentres rest mostly on residents' accounts; in East Fishkill the project is at an exploratory stage and impacts on rates, water and jobs remain unmeasured. TRV-2026-1401TRV-2026-1555
Panel data from 67 countries associate greater AI development with significantly lower carbon emissions, yet a study of China's 2000 to 2024 data found AI robotics did not help lower CO2 emissions. TRV-2026-0377TRV-2026-2379 A review of 146 studies found that in four countries technical performance and investment alone did not ensure sustainable outcomes. TRV-2026-1905 Among Chinese listed firms, real AI investment was associated with lower pollution intensity, while AI talk beyond actual investment was associated with higher intensity. TRV-2026-2116
Projections and company reports point to rising emissions and water use, though most of the figures are modelled estimates rather than metered measurements. TRV-2026-0474TRV-2026-0296TRV-2026-0433
See these 9 recordsResidents, campaigners and voters in the US, Europe and India are contesting AI datacentres, and a Marquette survey found 73% of Americans think the costs outweigh the benefits. TRV-2026-1553TRV-2026-1401TRV-2026-2292
See these 13 recordsCrop and livestock tools mostly score well in dataset tests, with few field trials behind them. TRV-2026-0262TRV-2026-2052TRV-2026-2242
See these 12 recordsA Trace is one record with both a gain and a problem, so it counts on both sides: 43 records show a gain and 42 show a problem. 38 context records are in the record list but not counted, because they report no measured gain or problem.
| Evidence | Gain side | Problem side |
|---|---|---|
| HighStrong study design, backed by at least two independent sources | 3 | 0 |
| ModerateA measured study or an official source | 35 | 26 |
| EmergingEarly, small, unreviewed or reported second-hand | 5 | 16 |
By type of evidence: 26 benchmark or lab tests; 14 investigations by reporters; 11 observational studies; 7 news reports; 7 commentaries or narrative reviews; 7 controlled comparisons; 2 systematic reviews or meta-analyses; 2 surveys. How evidence is scored
Across 51 countries from 1996 to 2023, greater AI use was associated with lower physical climate risks, with a stronger effect where those risks were higher.
A study of G20 economies from 2015 to 2023 found that higher AI readiness was linked to better national sustainability performance, with a larger association than FinTech or Blockchain.
In panel data from 67 countries, greater AI development was associated with significantly lower ecological footprints and carbon emissions and faster energy transitions, with the strongest link for energy transitions.
A review of 146 studies found that AI's expansion in energy systems also raises electricity and infrastructure demand, and that in four countries, technical performance and investment alone did not ensure sustainable outcomes.
A study of 432 coastal Ghana wells found modelled health risk, driven mainly by fluoride, averaged above the safety threshold for children (HI 1.04) and infants (1.10) but was low for adults (0.38).
Models detect microplastics, soil metals and wastewater sources well in lab or regional tests, but they miss sites and often overstate accuracy outside their training data. TRV-2026-1989TRV-2026-1744TRV-2026-0643TRV-2026-0932
See these 14 recordsAI weather and wave forecasts matched or beat operational systems in benchmarks, and a first 10 meter Arctic wetland map reports 89% overall accuracy. TRV-2026-0528TRV-2026-1465TRV-2026-1437
See these 7 recordsCountry and firm studies link AI to lower emissions, while lab and plant tests show energy cuts such as 26% on a packaging line. TRV-2026-0377TRV-2026-2054TRV-2026-1849
See these 16 recordsEach record sits in one thread at most. A record with both a gain and a problem counts on both sides. 4 records fit none of these threads.
A study found the model missed 2 of 5 paint-contaminated field sites and could not predict particle concentration accurately enough.
34 records · page 2 of 3
"Climate change - environment" by RecondOil, CC BY 2.0.Researchers built an explainable machine learning system that links how biochar is prepared, its physical and chemical properties, and how well it remediates saline-alkali soil. SHAP analysis was used to derive recommended parameter ranges, which guided a phosphoric acid-chitosan composite modification.
Environmental Research
This peer-reviewed review from September 2026 synthesized more than 35 studies on precision livestock farming for extensive sheep production, where large grazing areas and limited supervision make…
Researchers present probabilistic bias correction, a machine learning framework trained on historical probabilistic forecasts to remove systematic errors.
Researchers built the Pan-Arctic Wetlands Inventory, a 10 m resolution map covering the circumpolar Arctic. It was produced with a Random Forest classifier trained on Sentinel-1, Sentinel-2 and ALOS PALSAR-2 imagery,…
Researchers tested four machine learning algorithms on 57 site-years of data from rainfed Oklahoma wheat environments. Inputs included proximal-sensing NDVI at several growth stages, pre-emergence nitrogen rate, soil…
By October 2026, researchers had developed and validated genomic selection models to address phenotyping bottlenecks in breeding watermelon for resistance to Phytophthora capsici fruit rot.
A peer-reviewed study published October 1, 2026 analyzed 51 countries between 1996 and 2023 to test whether AI helps manage physical climate risks.
Real-time and accurate monitoring of heavy metal concentrations in rice is essential for ensuring food safety and supporting the safe utilization of contaminated agricultural land.
Researchers proposed a framework that integrates biological and non-biological indicators for soil heavy metal risk assessment.
Researchers compared a domain-specific EfficientNet-B0 multi-task deep learning classifier trained on about 700 annotated optical microscope images against a zero-shot Claude Vision API augmented with…
Researchers tested a field-scale targeted sampling strategy inside a regional hybrid model that combines machine learning and geostatistics for soil organic carbon mapping.
Researchers built EcoSchedAI, a carbon-aware scheduling framework that places deep learning training jobs in periods when grid carbon intensity is low.
Using genomic data from 220 UK Holstein cows, researchers tested Random Forest and Multi-Layer Perceptron models against conventional gBLUP for predicting residual feed intake, a feed-efficiency…
On 2026-08-06, a peer-reviewed study described an integrated assessment for a mountainous area in northern Iran covering flood, avalanche, rockfall and landslide.
Researchers developed a hybrid framework that couples the STICS biophysical crop model with machine learning to predict apple fruit maturity dates across China.
ace