# AI in climate

> Source: https://aiwiki.ai/wiki/ai_in_climate
> Updated: 2026-06-09
> Categories: AI Tools & Products, Artificial Intelligence
> From AI Wiki (https://aiwiki.ai), a free encyclopedia of artificial intelligence. Quote with attribution.

**AI in climate** refers to the application of [artificial intelligence](/wiki/artificial_intelligence) techniques to address climate change, both in mitigation (reducing greenhouse gas emissions) and adaptation (preparing for climate impacts that are already unavoidable). AI systems are being deployed across weather forecasting, energy optimization, carbon capture, climate modeling, satellite-based environmental monitoring, precision agriculture, and building efficiency. At the same time, AI itself carries a growing environmental cost: training and running large models consumes substantial electricity and water, raising questions about whether AI's climate benefits outweigh its own carbon footprint.

## AI for weather prediction

Weather forecasting has been one of the most successful applications of AI in climate science, with AI models now matching or exceeding the accuracy of traditional numerical weather prediction systems that have been developed over decades.

### GraphCast

[GraphCast](/wiki/graphcast) is a weather forecasting model developed by [Google DeepMind](/wiki/deepmind) that can make medium-range weather forecasts (up to 10 days in advance) more accurately and much faster than the European Centre for Medium-Range Weather Forecasts (ECMWF) system, which had been the industry gold standard. A 10-day forecast with GraphCast takes less than one minute on a single Google [TPU](/wiki/tensor_processing_unit_tpu) v4 machine, while the same forecast using conventional numerical methods takes hours of computation on a supercomputer with hundreds of machines. GraphCast uses a [graph neural network](/wiki/graph_neural_network) architecture trained on 39 years of historical weather data from the ERA5 reanalysis dataset [1].

### GenCast

GenCast, also from Google DeepMind, extends GraphCast's approach into probabilistic ensemble forecasting. Rather than producing a single forecast, GenCast generates 50 or more distinct scenarios by using a [diffusion model](/wiki/diffusion_model) framework, allowing it to provide percentage-based probability estimates for events like hurricane landfalls or heatwaves. GenCast outperforms the ECMWF's ensemble system (ENS) on 97.2% of tested targets for forecasts up to 15 days in advance. By early 2026, GenCast had established AI-driven probabilistic forecasting as a leading approach for predicting high-stakes weather events [2].

### WeatherNext and NOAA deployment

Google DeepMind followed GraphCast and GenCast with [WeatherNext 2](/wiki/weathernext_2), focused on energy trading applications. Meanwhile, the U.S. National Oceanic and Atmospheric Administration (NOAA) launched a suite of operational AI-driven global weather prediction models in 2025, providing forecasters with faster and more accurate guidance while using a fraction of the computational resources of traditional systems [3].

A limitation of these AI weather models is the "black box" problem. Unlike physics-based models where scientists can trace storm development back to specific physical laws, AI models are largely opaque. If a model predicts a catastrophic flood, forecasters may struggle to explain why the prediction was made, which complicates decision-making and public communication [4].

## Applications of AI in climate action

AI is being applied across a wide range of climate-related domains. The table below summarizes the major application areas, specific techniques, and real-world examples.

| Application area | AI techniques used | Examples and results |
|-----------------|-------------------|---------------------|
| Weather prediction | [Graph neural networks](/wiki/graph_neural_network), [diffusion models](/wiki/diffusion_model) | [GraphCast](/wiki/graphcast) (10-day forecasts in <1 minute), GenCast (probabilistic, 97.2% better than ENS) |
| Energy grid optimization | [Reinforcement learning](/wiki/reinforcement_learning), predictive analytics | AI-powered smart grids balance solar/wind variability, reduce blackouts during heat waves, forecast renewable output |
| Carbon capture optimization | [Machine learning](/wiki/machine_learning) for process control | 15-25% reduction in capture costs, 20% operational cost reduction in chemical absorption systems |
| Climate modeling | Deep learning, physics-informed [neural networks](/wiki/neural_network) | Faster simulation of climate scenarios, improved resolution of regional projections |
| Satellite monitoring | [Computer vision](/wiki/computer_vision), [CNNs](/wiki/convolutional_neural_network) | Tracking deforestation, greenhouse gas emissions, ocean heat absorption, and wildfire detection |
| [Precision agriculture](/wiki/precision_agriculture) | Predictive models, satellite image analysis | Projected 20% reduction in agricultural greenhouse gas emissions by 2025, optimized irrigation and fertilization |
| Building energy efficiency | Reinforcement learning, predictive HVAC control | 10-25% energy reductions in commercial buildings, [DeepMind](/wiki/deepmind) reduced Google data center cooling energy by 40% |
| Renewable energy forecasting | Time series models, weather prediction | Improved solar and wind output forecasting for grid stability |
| Ocean and marine monitoring | Acoustic analysis, satellite imagery, underwater sensors | Tracking ocean temperatures, marine biodiversity, coral reef health, and illegal fishing |
| Wildfire detection and prediction | Computer vision, satellite analysis, sensor networks | Early warning systems that detect fires within minutes of ignition using satellite and ground sensor data |

## Case studies

### Google DeepMind: data center cooling

In 2016, DeepMind applied [machine learning](/wiki/machine_learning) to Google's data centers and achieved a 40% reduction in the energy used for cooling, which translated to a 15% reduction in overall power usage effectiveness (PUE) overhead. The system trained self-learning algorithms to predict how hot data centers would get within the next hour, then used that prediction to supply only as much cooling as necessary. The algorithm also produced the lowest PUE the site had ever recorded [5].

Google has stated that the framework is general-purpose and can be applied to other domains, including improving power plant conversion efficiency, reducing semiconductor manufacturing energy and water usage, and helping manufacturing facilities increase throughput.

### Google: broader climate and sustainability initiatives

Beyond data center cooling, Google has deployed AI across several climate-related applications:

- **Wildfire detection:** Google's AI-powered wildfire tracking system uses satellite imagery to detect and track wildfires, providing early warnings to communities and emergency responders.
- **Flood forecasting:** Google's flood prediction models cover rivers in over 80 countries, providing alerts to millions of people in flood-prone regions.
- **Traffic optimization:** Google Maps uses AI to suggest fuel-efficient routes, which the company estimates has helped prevent more than 2.4 million metric tons of CO2 emissions since launch.
- **Carbon-free energy matching:** Google is advancing its goal of operating on 100% 24/7 carbon-free energy across all grids by 2030. In 2025, the company achieved an average of 64% carbon-free energy usage globally, with several regions in Europe and the Americas exceeding 90% [18].

However, Google's carbon emissions rose 48% over the past five years, largely due to AI infrastructure expansion. In 2025, Google reported a 13% increase in total greenhouse gas emissions compared to the previous year, driven primarily by data center growth [18].

### Microsoft: carbon negative pledge and AI tensions

In 2020, Microsoft announced its commitment to become carbon negative, water positive, and zero waste by 2030. The company has used AI through its AI for Good Lab to accelerate materials discovery for clean energy and improve climate forecasting models.

However, Microsoft's massive investments in AI infrastructure have created a significant tension with its climate goals. The company plans to spend $80 billion on data center infrastructure in fiscal year 2025, and its total planet-warming impact is approximately 30% higher in 2025 than it was when the pledge was made in 2020. Microsoft has acknowledged that it will not meet its original 2030 carbon-negative timeline, attributing the delay to AI energy demands. The company has pivoted toward longer-term, higher-impact carbon removal projects and large-scale procurement of carbon-free electricity, achieving a 29.9% reduction in Scope 1 and Scope 2 emissions from the 2020 baseline [6][7].

Microsoft's specific climate AI projects include:

- **Project Guacamaya:** A joint effort using AI models to monitor deforestation and protect biodiversity in rainforest ecosystems, combining satellite imagery, camera traps, and bioacoustics analysis [19].
- **Planetary Computer:** A platform that aggregates petabytes of environmental data (satellite imagery, weather data, land use records) and provides AI tools for analyzing environmental change at global scale.
- **Materials discovery:** AI models accelerating the search for new materials for solar cells, batteries, and carbon capture systems.

### AI for carbon capture

AI is being used to optimize carbon capture and storage (CCS) operations. Machine learning algorithms analyze data from capture plants to find optimal settings for temperature, pressure, flow rates, and chemical reactions. Studies have found that process-optimization algorithms can reduce capture costs by 15-25%, while AI-guided energy management in chemical absorption systems has achieved approximately 20% operational cost reductions. Projects at the Technology Centre Mongstad in Norway and the Boundary Dam facility in Saskatchewan have reported 10-20% cost reductions through AI optimization of solvent-based and membrane-based capture systems [8].

### Smart grid optimization

Smart grids powered by AI are becoming essential for managing the increasing share of renewable energy in the power mix. Because solar and wind generation are inherently intermittent, grid operators must constantly balance supply and demand. AI enhances smart grid operations in several ways:

| Smart grid AI application | Description | Impact |
|---|---|---|
| Renewable output forecasting | AI predicts solar and wind generation 24-72 hours ahead using weather data, historical patterns, and satellite imagery | Reduces the need for backup fossil fuel generation by improving prediction accuracy |
| Demand response optimization | AI predicts electricity demand patterns and coordinates demand-side management (shifting loads to times of high renewable generation) | Reduces peak demand and associated emissions |
| Grid fault detection | AI monitors grid infrastructure for anomalies that indicate potential failures | Detects outages instantly, reducing downtime and transmission losses |
| Energy storage management | AI optimizes charging and discharging of battery storage systems based on price signals, demand forecasts, and renewable generation | Maximizes the value of stored renewable energy |
| Electric vehicle integration | AI manages the charging patterns of electric vehicle fleets to avoid grid stress and take advantage of renewable energy peaks | Enables V2G (vehicle-to-grid) and smart charging at scale |

Smart grids automatically balance supply and demand, detect outages instantly, and reduce transmission losses. As renewable energy capacity continues to grow globally, AI-powered grid management is becoming critical infrastructure [20].

### Precision agriculture

AI-driven precision agriculture uses data from satellite imagery, soil sensors, and climate models to optimize crop management. The market for AI in precision agriculture is expected to grow from $0.78 billion in 2024 to $0.94 billion in 2025, with a CAGR of 20.3% [21].

Key AI applications in precision agriculture include:

- **Satellite-based crop monitoring:** AI analyzes NDVI (Normalized Difference Vegetation Index), multispectral, and hyperspectral satellite imagery to assess crop health, detect disease outbreaks, and estimate yields.
- **[Precision](/wiki/precision) irrigation:** Predictive models forecast crop water requirements based on soil moisture, weather forecasts, and plant growth stage, enabling irrigation that reduces water waste by 20-30%.
- **Targeted fertilizer application:** AI-guided variable-rate application of fertilizers minimizes nitrous oxide emissions (a greenhouse gas approximately 300 times more potent than CO2 per unit) while maintaining or improving yields.
- **Pest and disease prediction:** Machine learning models predict pest outbreaks and disease spread based on weather patterns, enabling targeted pesticide application rather than blanket spraying.
- **Carbon farming:** AI systems monitor soil carbon sequestration and recommend practices that increase carbon storage in agricultural soils.

Platforms like Farmonaut and Cropin analyze satellite data to provide crop health assessments and yield predictions. By 2025, AI climate solutions in agriculture are projected to reduce the sector's greenhouse gas emissions by approximately 20% globally [9].

Climate-smart farming relies on real-time monitoring of carbon emissions, soil health, water use, and other environmental indicators. AI-infused systems collect and analyze this data to recommend sustainable land management practices, supporting regenerative agriculture approaches.

### Satellite monitoring of deforestation and emissions

AI-powered satellite monitoring systems use [computer vision](/wiki/computer_vision) and [convolutional neural networks](/wiki/convolutional_neural_network) to analyze satellite imagery for tracking deforestation, land degradation, wildfires, greenhouse gas emissions, and ocean heat content. These systems can process vast amounts of satellite data in near real-time, identifying changes that would take human analysts far longer to detect.

Specific deforestation monitoring tools and achievements include:

| Tool / Initiative | Developer | Approach | Results |
|---|---|---|---|
| Global Forest Watch | World Resources Institute | Satellite imagery combined with ML algorithms for near-real-time deforestation alerts | Covers global tropical forests; alerts issued within days of forest loss |
| Starling | Earthworm Foundation / Airbus | Geospatial solution combining satellite imagery, AI, and supply chain expertise with field verification | Used by major corporations to monitor deforestation risk in commodity supply chains |
| Project Guacamaya | Microsoft / partners | AI models combining satellite imagery, camera traps, and bioacoustics | Monitors deforestation and biodiversity in rainforest ecosystems [19] |
| YOLO-based detection | Academic research | YOLOv8 object detection with LangChain-based agentic AI for real-time anomaly detection | Detects tree stumps, logging machinery, and unauthorized human presence in protected areas |

Unilever uses satellite monitoring combined with high-resolution (1.5-meter) optical imagery and radar data for continuous monitoring of its supply chains, achieving 95.7% deforestation-free sourcing across 20 million hectares. AI has increased detection rates and enforcement efficiency by up to 70% in monitored regions by 2025 [22].

### Ocean and marine monitoring

AI is increasingly applied to ocean monitoring and marine conservation:

- **Ocean temperature tracking:** AI models analyze satellite sea surface temperature data and autonomous underwater vehicle readings to track ocean warming patterns and predict marine heatwave events.
- **Coral reef monitoring:** Computer vision systems analyze underwater imagery to assess coral bleaching, biodiversity, and reef health at scales that would be impossible with manual surveys.
- **Illegal fishing detection:** AI analyzes vessel tracking data (AIS transponder signals), radar imagery, and satellite data to identify vessels engaged in illegal, unreported, and unregulated (IUU) fishing. Global Fishing Watch, supported by Google, uses AI to monitor fishing activity across the world's oceans.
- **Methane emissions tracking:** AI-powered satellite analysis detects methane emissions from offshore oil and gas facilities, landfills, and agricultural operations, providing data for regulatory enforcement and emissions reduction.

### Building energy efficiency

Buildings account for roughly 40% of global energy consumption. AI-driven building energy management and control systems (BEMCS) use predictive modeling and automation to optimize HVAC (heating, ventilation, and air conditioning), lighting, and other systems. Research has shown that [reinforcement learning](/wiki/reinforcement_learning) approaches achieve average energy savings of 22.3%, while hybrid AI methods can reach savings of 28.1%. These systems identify patterns and anomalies in building data that traditional control systems miss, allowing dynamic responses to weather changes, occupancy patterns, and electricity pricing [11].

## AI's own carbon footprint

The environmental cost of AI itself is a growing concern, and it creates a paradox: the technology being deployed to fight climate change is simultaneously contributing to the problem.

### Training energy consumption

Training large [language models](/wiki/large_language_model) requires enormous amounts of electricity. Training [GPT-3](/wiki/gpt-3) consumed approximately 1,287 megawatt-hours of electricity, enough to power about 120 average U.S. homes for a year, and generated roughly 552 tons of carbon dioxide. Larger models released since then have even greater energy requirements, though exact figures are often not disclosed by companies [12].

### Data center energy growth

Data centers have been doubling their electricity consumption, driven largely by AI-intensive hardware. Global electricity demand from data centers is projected to more than double by 2030, reaching approximately 945 terawatt-hours. Google alone expects to spend $75 billion on AI infrastructure in 2025 [13]. Many new AI data centers require 100 to 1,000 megawatts of power, equivalent to the demands of a medium-sized city, while grid operators face connection lead times of over two years to connect to clean energy supplies [23].

### Carbon and water projections

Researchers at Cornell University and MIT have projected that at the current rate of AI growth, the technology would annually emit 24 to 44 million metric tons of CO2 by 2030, equivalent to the emissions of 5 to 10 million additional cars on U.S. roads. AI operations would also consume 731 to 1,125 million cubic meters of water per year, equal to the household water usage of 6 to 10 million Americans. For 2025 specifically, estimates place AI's carbon footprint between 32.6 and 79.7 million tons of CO2 emissions, with a water footprint of 312.5 to 764.6 billion liters [14][15].

| Metric | Value | Context |
|--------|-------|---------|
| [GPT-3](/wiki/gpt-3) training energy | ~1,287 MWh | Equivalent to powering 120 U.S. homes for a year |
| GPT-3 training CO2 | ~552 tons | Equivalent to ~120 round-trip flights from NYC to London |
| Projected data center electricity by 2030 | ~945 TWh | More than double current consumption |
| Projected AI CO2 emissions by 2030 | 24-44 million metric tons/year | Equivalent to 5-10 million cars |
| Projected AI water consumption by 2030 | 731-1,125 million m3/year | Equal to 6-10 million Americans' household water use |
| Google AI infrastructure spending (2025) | $75 billion | Single year |
| Microsoft AI data center spending (FY 2025) | $80 billion | Single year |
| Google carbon emissions increase (5-year) | 48% | Largely attributed to AI infrastructure expansion |
| Microsoft emissions increase since 2020 | ~30% | Despite carbon-negative pledge |
| Typical new AI data center power demand | 100-1,000 MW | Equivalent to a medium-sized city |

## The net impact debate

Whether AI's benefits for climate action outweigh its environmental costs is an active area of debate.

Proponents argue that AI can accelerate decarbonization across entire sectors of the economy (energy, agriculture, transportation, manufacturing, buildings) in ways that more than compensate for the energy consumed by AI systems themselves. The DeepMind data center cooling example alone, if applied across all of Google's facilities and replicated by other companies, would save more energy than the AI systems consume. AI-optimized energy grids, more efficient carbon capture, and precision agriculture each offer significant emission reductions.

Critics point out that much of the current growth in AI compute is not directed at climate applications. The vast majority of AI training and inference supports commercial applications like chatbots, image generation, recommendation systems, and advertising optimization. The climate benefits of AI are real but represent a small fraction of total AI energy consumption. Furthermore, the rebound effect is a concern: as AI makes certain processes more efficient, it may also increase total demand for those processes [16].

A 2025 BCG and CO2 AI survey found that while 85% of companies recognize AI's potential for sustainability, only a fraction are systematically deploying AI for emissions reduction. The gap between potential and deployment remains wide [24].

MIT researchers have emphasized that the challenge is not whether AI can help with climate change (it clearly can) but whether the AI industry will grow in a way that prioritizes climate applications and minimizes its own environmental footprint. Policy measures such as carbon pricing for data centers, renewable energy requirements for AI infrastructure, and transparency mandates for reporting AI energy consumption are among the proposed solutions [17].

## Current state (2025-2026)

AI for climate applications is maturing rapidly across multiple fronts. NOAA's operational deployment of AI weather models in 2025 marked a turning point for institutional adoption. AI-driven optimization of energy grids, buildings, and industrial processes is moving from pilot projects to production systems. Carbon capture facilities are increasingly integrating AI-based process control.

Specific developments shaping the current landscape include:

**Institutional adoption of [AI weather forecasting](/wiki/ai_weather).** NOAA's 2025 deployment of AI-driven global weather models represents the first operational use of AI weather prediction by a major national meteorological agency. These models provide faster and more accurate guidance while consuming a fraction of the computational resources of traditional numerical weather prediction systems.

**Corporate sustainability tensions.** Both Google and Microsoft have seen their carbon emissions rise significantly due to AI infrastructure expansion, even as they deploy AI for climate applications. Google's emissions rose 48% over five years, and Microsoft acknowledged it will not meet its 2030 carbon-negative target. This tension highlights the gap between AI's climate potential and the industry's actual environmental trajectory.

**Precision agriculture scaling.** AI-powered precision agriculture is moving from pilot programs to widespread adoption, with satellite-based monitoring, AI-guided irrigation, and targeted fertilizer application reducing agricultural emissions while maintaining productivity.

**Deforestation monitoring improvements.** AI-powered satellite monitoring has improved deforestation detection rates by up to 70% in monitored regions, and major corporations are using these tools to verify deforestation-free supply chains.

**Smart grid expansion.** As renewable energy capacity grows, AI-powered smart grid management is becoming critical infrastructure, with AI systems balancing supply and demand, managing energy storage, and integrating electric vehicle charging.

At the same time, the environmental footprint of AI continues to grow. Major technology companies have acknowledged that their AI expansion has complicated their climate commitments. The tension between AI's promise as a climate tool and its reality as a growing source of energy demand is likely to intensify as AI models continue to scale.

The emerging consensus among researchers is that AI's net climate impact depends heavily on governance: what proportion of AI development is directed toward climate-beneficial applications, how AI infrastructure is powered, and whether the industry is held accountable for its environmental costs.

## References

1. "GraphCast: AI model for faster and more accurate global weather forecasting." Google DeepMind. https://deepmind.google/blog/graphcast-ai-model-for-faster-and-more-accurate-global-weather-forecasting/
2. "GenCast predicts weather and the risks of extreme conditions with state-of-the-art accuracy." Google DeepMind. https://deepmind.google/blog/gencast-predicts-weather-and-the-risks-of-extreme-conditions-with-sota-accuracy/
3. "NOAA deploys new generation of AI-driven global weather models." NOAA, 2025. https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models
4. "Can AI models reliably forecast extreme weather events?" Nature, 2026. https://www.nature.com/articles/d41586-026-00842-z
5. "DeepMind AI Reduces Google Data Centre Cooling Bill by 40%." Google DeepMind, 2016. https://deepmind.google/blog/deepmind-ai-reduces-google-data-centre-cooling-bill-by-40/
6. "Progress on the road to 2030." Microsoft, 2025. https://blogs.microsoft.com/on-the-issues/2025/02/13/progress-on-the-road-to-2030/
7. "AI Expansion Jeopardizes Microsoft's Carbon-Negative Pledge." Environment+Energy Leader. https://www.environmentenergyleader.com/stories/ai-expansion-jeopardizes-microsofts-carbon-negative-pledge,1355
8. "Revolutionizing carbon capture efficiency." IJSTRA, 2025. https://sciresjournals.com/ijstra/sites/default/files/IJSTRA-2025-0043.pdf
9. "AI-Driven Future Farming: Achieving Climate-Smart and Sustainable Agriculture." MDPI, 2025. https://www.mdpi.com/2624-7402/7/3/89
10. "Artificial Intelligence-Driven Analytics for Monitoring and Mitigating Climate Change Impacts." MDPI, 2025. https://www.mdpi.com/2673-4591/108/1/7
11. "Artificial intelligence for energy optimization in smart buildings: A systematic review." Springer Nature, 2025. https://link.springer.com/article/10.1186/s42162-025-00592-8
12. "Explained: [Generative AI](/wiki/generative_ai)'s environmental impact." MIT News, 2025. https://news.mit.edu/2025/explained-generative-ai-environmental-impact-0117
13. "AI: Five charts that put data-centre energy use and emissions into context." Carbon Brief. https://www.carbonbrief.org/ai-five-charts-that-put-data-centre-energy-use-and-emissions-into-context/
14. "'Roadmap' shows the environmental impact of AI data center boom." Cornell Chronicle, 2025. https://news.cornell.edu/stories/2025/11/roadmap-shows-environmental-impact-ai-data-center-boom
15. "The Real Environmental Footprint of Generative AI: What 2025 Data Tell Us." Online Learning Consortium, 2025. https://onlinelearningconsortium.org/olc-insights/2025/12/the-real-environmental-footprint-of-generative-ai/
16. "We did the math on AI's energy footprint. Here's the story you haven't heard." MIT Technology Review, 2025. https://www.technologyreview.com/2025/05/20/1116327/ai-energy-usage-climate-footprint-big-tech/
17. "Responding to the climate impact of generative AI." MIT News, 2025. https://news.mit.edu/2025/responding-to-generative-ai-climate-impact-0930
18. "Google 2025 Environmental Report." GreentechLead. https://greentechlead.com/sustainability/google-2025-environmental-report-ai-driven-growth-raises-emissions-as-company-accelerates-clean-energy-and-net-zero-strategy-52687
19. "Project Guacamaya uses satellites & AI to battle deforestation." Microsoft News. https://news.microsoft.com/source/latam/features/ai/project-guacamaya-rainforest-deforestation/?lang=en
20. "AI and Climate Change in 2026: How Artificial Intelligence Is Transforming Global Sustainability." Writac. https://writac.com/ai-and-climate-change/
21. "AI Applications In Precision Agriculture: 7 Key Uses 2025." Farmonaut. https://farmonaut.com/precision-farming/ai-applications-in-precision-agriculture-7-key-uses-2025
22. "Satellite AI Revolutionizes Deforestation Detection & Supply Chain Transparency." Fiegenbaum Solutions. https://www.fiegenbaum.solutions/en/blog/satellite-ai-deforestation-detection-supply-chain-transparency
23. "AI scale and climate commitments: A 2026 outlook." Carbon Direct. https://www.carbon-direct.com/insights/ai-scale-and-climate-commitments-a-2026-outlook
24. "Climate Survey 2025." CO2 AI / BCG. https://co2ai.com/climate-survey-2025