Estimating Earth’s climate sensitivity, the global average warming from doubling CO2 above pre-industrial levels, is an urgent task as it governs how much the world will warm from fossil-fuel emissions.
Recent work has essentially ruled out low values of climate sensitivity below about 2C, but how much do we know about high values of climate sensitivity?
There is an active debate about the likely upper bound of climate sensitivity, which informs the worst-case projections of warming over the coming decades.
One way to narrow this down is to study the major climate changes of Earth’s past seen in palaeoclimate “proxy data”. These records are scattered across the planet, found in everything from ice cores and ocean sediments to tree rings and coral reefs.
Combining proxy data with climate models helps scientists understand what past changes were caused by natural fluctuations in CO2 and what changes we can expect from human-caused CO2 increases in the future.
In our new study, published in Science Advances, we find that the Last Glacial Maximum – a period of extensive ice cover about 21,000 years ago – provides even stronger evidence for modern-day climate sensitivity than previously thought.
Our estimates suggest a central estimate of “equilibrium” climate sensitivity of 2.9C, with a very likely range of 2.1-4.1C.
This is around a 30% narrowing of the 2.0-5.0C range in the latest assessment from the Intergovernmental Panel on Climate Change (IPCC).
Our findings give us more confidence that very high climate sensitivity is unlikely. They also support the IPCC’s central estimate for climate sensitivity of around 3.0C and lower estimate of 2.0C, confirming that warming would still be severe if we do not reduce fossil-fuel emissions.
Progress in climate sensitivity
For many years, the estimated range for equilibrium climate sensitivity (ECS) remained stubbornly wide.
After the landmark Charney report established a “consensus” position from a study group convened by the US National Research Council back in 1979, many IPCC assessments over several decades were unable to narrow it further.
This is illustrated in the chart below, which shows the Charney report estimated range for ECS (grey bar) on the left, followed by each IPCC report in sequence, from the first in 1990 (“FAR”, light blue) through to the most recent AR6 range on the right (red). Central estimates are shown with a dot, where available. The coloured bars indicate the estimated likely range for ECS and very likely ranges are marked with whiskers.
Estimates of ECS published in successive IPCC assessments since the Charney report in 1979. Dots show central estimates. The coloured bars show the likely range and the very likely range is given by whiskers. Chart by Carbon Brief
Ahead of the IPCC’s sixth assessment report (AR6), a four-year project from an international team of scientists made a significant advance.
The resulting paper by Prof Steven Sherwood, Dr Mark Webb and colleagues, published in the Review of Geophysics in 2020, formally combined multiple lines of evidence from observed data to improve understanding of the climate “feedbacks” that can amplify or dampen surface warming. The paper played a key role in the narrower ECS range in AR6 (red bar in chart above) – particularly in raising the lower bound to 2.0C and setting the upper bound at 5.0C.
However, observed data from recent warming are not as useful for constraining the upper end of ECS estimates. This is because climate feedbacks – especially how warming affects cloud properties that either enhance or reduce their overall cooling effect – depend on spatial patterns of temperature change. Climate scientists have dubbed this phenomenon the “pattern effect”.
Pattern effects
Temperature patterns over recent decades differ substantially from what we expect in the long-term. This means that climate feedbacks are likely to change in the future too, making observed warming a relatively poor predictor of how high climate sensitivity could be.
This is where palaeoclimate data comes in. The Last Glacial Maximum (LGM) has been touted as the best evidence for or against high values of climate sensitivity. By estimating how much colder the Earth was when CO2 levels were so much lower during the LGM, scientists can predict how much warmer the Earth would be with higher CO2 levels decades from now.
However, doing so requires isolating how much of the LGM cooling came from lower CO2 and correcting for how the climate’s sensitivity to CO2 differs between the cold LGM and the warm modern-day climate. Prior to our study, no one had quantified the impact of temperature patterns on climate sensitivity in the palaeoclimate record.
Traditionally, the additional cooling effect of the vast LGM ice sheets, which covered much of Canada and northern Eurasia, has been estimated from the amount of sunlight they directly reflect. But the ice sheets also caused the nearby northern Pacific and Atlantic oceans to cool and produce more low clouds, amplifying global cooling by reflecting even more sunlight. This ocean cooling is illustrated by the dark blue shading in the left-hand map below.
By linking these adjacent cloud changes to the ice sheets, we found that relatively more of the LGM cooling was caused by the ice sheets and relatively less came from the lowered CO2.
Importantly, these amplifying cloud feedbacks are unique to the climate during the ice age and, therefore, do not apply to modern CO2-driven warming that will not have similarly large ice-sheet changes.
As a result, the amount of cooling from lower CO2 at the LGM is not consistent with very high values of modern-day climate sensitivity above around 4C.
Accounting for differences between the LGM and modern climates allows us to lower the upper end of climate sensitivity estimates, representing a major shift in how palaeoclimate data is used to inform our future.
Combining palaeoclimate data with climate models
Quantifying the climate differences mentioned above for the LGM requires estimating temperature patterns and climate feedbacks from 21,000 years ago.
An obvious challenge is that our instrumental temperature records span only centuries and our satellite observations of clouds span only decades. Fortunately, there have been two major advances in recent years that allowed us to quantify the pattern effects from the distant past.
First, palaeoclimate data assimilation is a recently developed method that combines climate model simulations with proxy data, producing globally complete reconstructions of surface temperatures during the LGM. However, there are still major uncertainties in this method. Various groups have recently reconstructed the ice age, finding somewhat different temperature patterns. To assess this uncertainty, we use four available reconstructions.
Second, we use these reconstructed surface temperatures in global atmosphere models to estimate climate feedbacks during the LGM and compare them to feedbacks expected from CO2 doubling. Because different models produce different climate feedbacks – even when given the same reconstructed surface temperatures – we assess this feedback uncertainty using five different atmosphere models.
Despite the uncertainties, a consistent story emerged. Climate feedbacks strongly amplified LGM temperature changes, much more so than climate feedbacks under modern-day warming from CO2 alone. We traced these differences primarily to cloud feedbacks in the North Pacific and North Atlantic Oceans where the patterns of temperature change strongly differ between the LGM – due to the presence of large ice sheets – and future warming from CO2.
At first glance, the extra amplifying feedbacks of the LGM could appear to suggest that climate sensitivity is higher than we thought. A recent study led by Dr James Hansen, former director of the NASA Goddard Institute for Space Studies, reported that some of the same reconstructions used here are consistent with a modern-day climate sensitivity of around 4.8C.
However, that interpretation is reversed when we recognise that those amplifying feedbacks are unique to the ice age climate and do not apply to the modern-day climate.
The result is that modern-day climate sensitivity, as estimated from the LGM, is actually lower than has been reported by studies such as Hansen’s that have not accounted for the unique effect of ice sheets on climate feedbacks.
Our findings also challenge a common assumption that warmer climates are more sensitive than colder climates. While that assumption appears likely to be true for climates warmer than today, it appears to be false for climates colder than today when the effect of ice sheets on temperature patterns and feedbacks is included.
Implications for future warming
The LGM, after accounting for how temperature patterns impact climate feedbacks, ends up being an even stronger constraint on modern-day climate sensitivity – especially the upper bound.
To the best of our knowledge, this is the first time that considering pattern effects has helped constrain climate sensitivity instead of adding to its uncertainty for modern-day climate.
Given the importance of combining multiple lines of evidence, we use the community framework from the Sherwood, Webb et al (2020) study, updated to include our estimate of how climate feedbacks differ between the LGM and modern-day warming.
As the chart below shows, our revised best estimate of climate sensitivity (blue) becomes 2.9C with a very likely range of 2.1-4.1C, a substantial narrowing of uncertainty by reducing the upper bound.
In contrast, the IPCC AR6 very likely range was 2.0-5.0C with a best estimate of 3C (red), while the Sherwood, Webb et al (2020) study reported 2.3-4.7C and central estimate of 3.1C (yellow).
Modern-day ECS estimates from recent assessments, including the Sherwood, Webb et al (2020 study (yellow), IPCC AR6 (red) and this study (blue). Dots show central estimates. The coloured bars show the likely range and the very likely range is given by whiskers. Chart by Carbon Brief based on data provided by V Cooper and K Armour
While the lower bound and central estimates are mostly unchanged, our findings reduce the upper bound by approximately 1C.
Based on the modelled relationship between climate sensitivity and future warming, lowering the upper bound on climate sensitivity from 5C to 4.1C translates to a reduction of 0.4-0.8C in the upper end of global warming estimates by 2100 (the range reflects uncertainty in future emissions). In other words, this is a substantial reduction in the likelihood of extremely high warming.
Our results for the LGM show that temperature patterns and their effects on climate feedbacks must be accounted for when using past climates to estimate modern-day climate sensitivity.
Spatial reconstructions of past climates are a necessity and there are opportunities for further progress in constraining climate sensitivity by analysing temperature patterns and feedbacks in other past climates.
The post Guest post: Ice-age analysis suggests worst-case global warming is less likely appeared first on Carbon Brief.
Guest post: Ice-age analysis suggests worst-case global warming is less likely
Climate Change
‘Ride the wave of momentum’: Australia announces once-in-a-decade Marine Parks Network review
In response to the federal government announcing its once-in-a-decade review of Australia’s Marine Parks Network, the following lines can be attributed to Elle Lawless, Senior Campaigner at Greenpeace Australia Pacific:
“Greenpeace Australia Pacific welcomes today’s announcement that the Albanese Government will review Australia’s Commonwealth Marine Parks Network. This is a rare, once-in-a-decade opportunity to strengthen our marine parks and ban industrial fishing in Australia’s marine protected areas.
“Australians would be appalled to know that more than half of Australia’s Marine Parks Network currently allows for extractive industries, like longlining, bottom trawling and oil and gas mining. These so-called ‘protected’ areas were designed to safeguard our beloved ocean wildlife and underwater ecosystems – that is what Australians expect. Damaging industrial industries should not be given a free pass to trawl, fish, drill or extract from our marine parks.”
“With the first Ocean COP just around the corner, and off the back of Australia’s move to ratify the Global Ocean Treaty earlier this year, the Australian government has a unique opportunity to ride the wave of this momentum and solidify itself as a true global ocean leader.
“Greenpeace Australia Pacific is calling for industrial activities to be banned from our protected waters and for at least 30% of Australia’s ocean to be protected as ocean sanctuaries. This review presents a rare opportunity to create more ocean sanctuaries, true blue havens where ocean life can recover, thrive and repopulate the surrounding waters.”
—ENDS—
‘Ride the wave of momentum’: Australia announces once-in-a-decade Marine Parks Network review
In recent days, prominent climate sceptics and rightwing commentators have shared charts on social media incorrectly implying that Europe is having its “quietest” year for wildfires in 2026.
These include Dr Matthew Wielicki, a former University of Alabama geochemist and self-described “professor in exile”, who was recently appointed by the Trump administration to lead the US Global Change Research Program.
However, these charts paint a misleading picture as they are skewed by encompassing the entirety of Russia in the data – including the vast plains of Siberia.
These charts also use data that include fires that are deliberately lit to manage cropland, which is a declining practice across much of Europe.
In this factcheck, Carbon Brief shows that the area burned by wildfires across the European Union in 2026 is second only to 2022 for this time of year.
The latest data from the European Forest Fire Information System (EFFIS) also shows that France has set a new modern record for area burned and Spain’s wildfire season is among the worst on record.
The fires have displaced more than a third of a million people across south-western Europe, while an impending heatwave has also raised fears of the fires worsening in the coming days.
‘Quietest year’
On 27 July, as wildfires raged across multiple European countries, former Conservative peer and climate-sceptic commentator Matt Ridley posted on Twitter that “2026 is the quietest year for wildfires in Europe by some distance”.
Ridley, who sits on the academic advisory council of the Global Warming Policy Foundation (GWPF), a UK-based climate-sceptic lobby group that refuses to reveal the sources of its funding, was responding to an article by Daily Telegraph columnist Tim Stanley.
Stanley’s column, headlined: “Climate change is real – and the right needs to get serious about it”, warned:
“This is no longer a matter of speculation: the wildfires of Europe, pitiless and persistent, are the way we live now.”
Ridley included a chart from Our World In Data, showing the cumulative area burned by wildfires by week for Europe. The chart puts 2026 as having the smallest area for this time of year in a dataset going back to 2012.
Ridley’s post was widely shared by prominent rightwing figures – including Richard Tice, deputy leader of the hard-right, climate-sceptic Reform UK party, former Conservative cabinet minister Jacob Rees-Mogg and multiple commentators.
Separately, Wielicki also shared a chart on Twitter to imply that wildfires in Europe are declining. Wielicki has previously claimed that the “science is not settled on climate change”.
The charts posted by Ridley and Wielicki both use data from the Global Wildfire Information System (GWIS). The GWIS category for “Europe” encompasses all the countries on the continent and includes the whole of Russia.
As a result, Russia accounts for about 74% of the area included in the GWIS definition of “Europe”.
Wildfires in Russia typically account for 80-90% of the burned area in the GWIS Europe dataset. In 2026, fires in Russia are substantially below average. Therefore, including Russia in this comparison creates the false impression that wildfire activity across Europe is unusually low.
Dr Calum Cunningham, a research fellow at the University of Tasmania’s Fire Centre, says that such claims are “highly misleading”, noting that “they rely on aggregating fire activity across an enormous and climatically diverse region”. He tells Carbon Brief:
“A relatively quiet season in Russia can easily mask an exceptionally active season in France or Spain. If the analysis is focused on the regions actually experiencing the current fires, the picture is very different.
“The reality is that western Europe has experienced an extraordinary sequence of climate conditions this year.”
In contrast, the EFFIS provides a subset of wildfire data specifically for the area covered by the 27 nations of the EU, which, therefore, excludes Russia.
Another difference between the two datasets is that GWIS monitors all fires – including those on agricultural land that are intentionally set alight. The burned area as measured by GWIS contains significant cropland area.
By contrast, EFFIS uses land-cover data and other information to filter specifically for forest fires.
Looking at the EU-only data from EFFIS reveals that Europe is far from having its “quietest” year. The bloc’s burned area, as of 29 July, is almost 435,000 hectares (ha) – second only to 2022 for this time of year.
Notably, Wielicki has actually continued to post charts based on GWIS data, even after acknowledging that “includ[ing] all of Russia, including vast areas of Siberia…isn’t a good proxy for Europe”.
French fires
Even looking at EU-wide data misses the scale of this year’s wildfires for some individual countries.
The chart below shows the surge in burned area in France since mid-July.
For much of the first half of the year, the country was having a wildfire season that was only slightly above average in terms of total burned area. However, a notable uptick began in the first week of July.
The third week of the month saw France break its previous cumulative annual record by more than 19,000ha. That gap has widened as the fires continue to burn; as of 29 July, the cumulative burned area in France during 2026 was nearly 24,700ha above the previous record.
The fires in France follow a record-breaking June heatwave that “dried out vegetation across the region, allowing fires to spread quickly”, wrote the New York Times.
On 27 July, French president Emmanuel Macron called a “crisis cabinet meeting” in order to address the fires “ravaging several areas of south-west France”, said France 24.
More than 220,000 people have been evacuated due to the Gironde fire, west of Bordeaux, in “what may be France’s largest peacetime evacuation”, reported the Associated Press.
In the Conversation, Cunningham and two other University of Tasmania researchers write that evacuation orders “protec[t] human lives, but makes it more likely houses and other structures will burn if there’s no one to defend them”. They add:
“There is little doubt climate change has made France and Spain’s wildfires worse. They represent yet another reason to redouble our efforts to tackle climate change and stabilise our climate.”
Central Spain scorched
While Spain’s fire season has not broken records in the same way that France’s has, it is on track to be among the worst since EFFIS began reporting data in 2006.
The chart below shows the rapid increase in burned area in Spain since 8 July. The latest data from EFFIS reveal that, as of 29 July, Spain has almost matched its previous record at this point in the year. It is also nearly five times the average area burned for this time of year.
In Spain, the wildfires have been concentrated in the central part of the country, near Madrid.
BBC News reported that the fires outside the capital have burned “an area more than twice as large as the city itself”.
Nearly 90,000 people were forced from their homes in central Spain by the fires, said the Associated Press.
Pedro Sánchez, Spain’s prime minister, called the fires a “painful expression” of climate change.
Meanwhile, the UK, French and Spanish governments have issued joint statements this week in response to the fires. The UK/Spain statement begins:
“This summer’s wildfires demonstrated that climate change was now a national security emergency facing Europe and threatening our way of life.”
Related
The post Factcheck: No, Europe is not having its ‘quietest’ year for wildfires appeared first on Carbon Brief.
Factcheck: No, Europe is not having its ‘quietest’ year for wildfires
The general public often underestimate support for climate action, while overestimating the real-world actions taken by other people to address the problem, according to new research.
The study, published in Nature Climate Change, explores the differences between people’s support for climate change, their behaviour and their assumptions about other people’s behaviour.
It is based on multiple surveys of more than 5,000 people across Germany and the US.
The study expands on previous research on how the general public systematically underestimate the climate commitment of their peers.
The difference between actual and perceived support for climate action among the public is sometimes known as a “perception gap”.
The surveys tested how people’s perceptions of climate attitudes and behaviours relate to their own willingness to contribute and undertake “climate-friendly” actions.
One of the authors tells Carbon Brief that this perception gap is not due to “ignorance or bias”, but because “people are just not good at making good estimations”.
The research also reveals that people’s opinions and behaviours are more “nuanced than previously assumed” and suggests that simply “correcting misperceptions” does not automatically lead to greater climate action.
Measuring climate actions
The study notes that correcting the perception gap is often seen as a “cost-effective” way to promote public engagement and drive action to reduce the intensification and impacts of climate change.
Most studies that explore the perception gap have primarily focused on surveys that have asked people to report their willingness to support climate change.
In other words, researchers have relied upon people saying they would support efforts to tackle climate change, rather than measuring people’s real-world actions, such as financial donations, attending protests or changing their behaviour.
To fill this gap, the researchers behind the new study surveyed a total of more than 5,000 people in Germany and the US over 2024-25. Surveys were split across five different experiments, each focused on public perceptions of climate attitudes and how they relate to individuals’ actual behaviour:
| Experiment | What they did |
| Survey one | Participants were asked if they were willing to donate 1% of their household income to WWF – and then were given the chance to do so. Follow-up questions asked participants to predict how many of their peers said “yes” and how many actually donated to the charity. |
| Survey two | Participants read a constitutional complaint against the German government, led by Greenpeace, which demands for stricter climate policies. They were asked if they were willing to participate as a claimant and/or donate to the cause – and then were given the chance to do so. Follow-up questions asked participants to predict how many of their peers said “yes” and how many went on to support the complaint. |
| Survey three | Participants were requested to complete an online “work for environmental protection task” where the more “pages” they completed resulted in more donations to WWF. They then predicted how many pages their peers completed. Participants also rated their individual behaviours and support for eight climate policies and then estimated the same for other people. |
| Surveys four and five | Participants were split into three groups that were either informed that 4% of participants had donated 1% of their household income to WWF, that “68% were willing to contribute” or given no information. They then had to state whether they were willing to support WWF and then were given the opportunity to do so. |
The authors note that Germany and the US are two of the “top 10 CO2 emitters” and are places where climate action is “especially necessary”. However, they add that the two countries are not reflective of “diverse cultural contexts” and further research is needed across the world.
The perception gap
The researchers find that most of their participants supported climate action, but much fewer actually performed verifiable behaviours.
For example, survey one finds that 37% of participants said they were willing to donate to WWF, yet just 4% did when given the opportunity.
Participants generally overestimated the climate actions of their peers, predicting that 23% of other people donated. Willingness, on the other hand, was slightly underestimated with respondents averaging around 34%.
The results from survey three suggest that this perception gap is likely due to general cognitive processes within the human brain that make accurate estimations about large groups difficult, say the authors.
The chart below shows the actual percentage of people who supported different environmental policies and performed climate-friendly behaviours (blue dots) compared to average predictions from the surveys (red dots).
They reveal a “consistent pattern” where “small proportions were overestimated and large ones were underestimated”, the authors say, driving predictions towards the middle. This phenomenon is known as “regression to the mean”.
In other words, where public support for a policy was high, participants in the survey estimated it was lower than it was. When the support was lower, estimates would be higher.
The study finds that individual and environmental factors played a role in shaping people’s perceptions of their peers’ climate actions, which were distinct from general misestimations.
For example, people who were already involved in climate action, had more frequent climate discussions and consumed more climate-focused news and media predicted a higher proportion of climate support “across the board”.
The results from the fourth and fifth surveys show that knowing the context of other people’s beliefs and behaviour in surveys can impact the attitudes of participants.
Participants that were told that 68% of people were willing to donate 1% of their household income to the WWF were more willing to donate.
In contrast, participants that were told that 4% of people actually donated did not report more willingness to “discuss climate change, sign petitions or donate” than the control group.
However, there was no obvious impact on actual donations for any of the three groups, the study notes.
Lead study author Dr Kevin Tiede, scientific managing director of the Institute for Planetary Health Behaviour at the University of Erfurt, tells Carbon Brief that the findings suggest that “just telling people how many people support climate action is likely not enough to really change something”.
However, Tiede adds that “direct comparability” between people saying they would donate and actually donating is “limited” and that giving people more time to answer and autonomy over where to donate might result in more people taking action.
‘Pluralistic ignorance’
Tiede explains that the study findings demonstrate the existence of “pluralistic ignorance”, where a person believes their own views differ from the majority.
For climate change, this means that the “vast majority of people around the world support climate action, but people considerably underestimate the extent of this support”, the study says.
However, the surveys reveal that pluralistic ignorance “in the climate domain” is more nuanced than previously thought, say the authors.
Prof Madalina Vascleanu, an assistant professor at Stanford University’s Doerr School of Sustainability, who was not involved in the study, tells Carbon Brief that encouraging climate action is complex.
It may take multiple and repeated “attempts” at effective communication, or for people to directly “experience” the “norm” that climate change is widely supported, she says, rather than simply being told.
“Observable” behaviours, such as “identity signalling” – which could involve anything from protesting to vegetarianism – might have more of an impact on encouraging climate action among peers than “private behaviours like donations”, she adds.
The study is a “great addition to the literature”, Vascleanu says, because “correcting” the perception gap did not have an effect on climate-friendly behaviour, as “scholars had previously assumed”. She adds that it has “sparked several new hypotheses” that her “lab is now working on”.
Prof Mauro Bertolotti, associate professor of social psychology at the Università Cattolica del Sacro Cuore, explains that the “attitude-behaviour gap” revealed by the research is a “rather common finding”.
However, he is “sceptical” of the “simplified and abstract” measures, warning that experiment environments often come with “assumptions and expectations” that are different from real life.
As a result, they might not “replicate” the process people go through when choosing to “make a donation to an environmental cause”, he says.
‘Targeted’ communication strategies
The researchers argue that it is more effective to focus on “targeted” communication strategies – encouraging climate-friendly behaviours that aim to reach the majority who already support climate action, rather than trying to convert climate sceptics.
They call for attention to be paid to the attitude-behaviour gap between people saying they support efforts to tackle climate change and following up with real-world climate actions.
The study suggests strategies for decision-makers to reduce the attitude-behaviour gap, such as “facilitating climate-friendly behaviour” with “convenience and subsidies”. They also recommend ensuring environmental policy prioritises fairness to gain visible and widespread public support.
They add that the public would benefit from understanding the “effectiveness and co-benefits” of climate action.
Tiede, K.E. et al. (2026) People systematically under- and overestimate public engagement in climate action, Nature Climate Change, https://doi.org/10.1038/s41558-026-02668-z
Related
Scientists are ‘most trusted’ source of climate information in global-south survey
Survey: ‘Very few’ Africans place responsibility for climate action on ‘rich nations’
Guest post: How public attitudes towards ‘CO2 removal’ differ in the UK and US
Guest post: How ‘discourses of delay’ are used to slow climate action
The post Correcting climate ‘misperceptions’ may not boost climate action appeared first on Carbon Brief.
Correcting climate ‘misperceptions’ may not boost climate action
-
Greenhouse Gases12 months ago
Guest post: Why China is still building new coal – and when it might stop
-
Climate Change12 months ago
Guest post: Why China is still building new coal – and when it might stop
-
Greenhouse Gases2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago嘉宾来稿：满足中国增长的用电需求 光伏加储能“比新建煤电更实惠”
-
Climate Change2 years ago
Bill Discounting Climate Change in Florida’s Energy Policy Awaits DeSantis’ Approval
-
Renewable Energy9 months agoSending Progressive Philanthropist George Soros to Prison?
-
Carbon Footprint2 years agoUS SEC’s Climate Disclosure Rules Spur Renewed Interest in Carbon Credits
-
Greenhouse Gases1 year ago
嘉宾来稿：探究火山喷发如何影响气候预测