Are Earth’s natural carbon sinks collapsing?
Atmospheric CO2 levels jumping a record 3.5ppm in 2024, while human emissions remained largely constant. This suggests reduced carbon absorption by land and ocean sinks which will accelerate warming.
Continuous measurements of the carbon dioxide level in the atmosphere have been carried out since 1958 at the Mauna Loa Observatory on the island of Hawaii, and more recently from other stations around the world. The famous Keeling Curve shown below is named after Charles David Keeling who pioneered the work.
The concentration of the greenhouse gas has steadily risen over the years as more and more emissions are generated and released through the burning of fossil fuels and through the actions of land use change.
Looking deeper into the curve, you can see that each year there is a wobble linked to the seasons. The reason is quite simple, during the northern hemisphere summer, CO2 is absorbed by plants through photosynthesis, but released back through decomposition and respiration during winter. Since the southern hemisphere has much less land surface area, the southern summer has less of an effect and not enough to cancel out the northern cycle. Without human emissions adding to the concentration, the curve would be a generally level but wobbly line.
When the cycle is on the up-stroke, the natural system is a net source of CO2, increasing the concentration (October through April) and during the down-stroke, the natural system is a net sink of CO2, reducing the concentration (May through August).
There is a fair amount of natural variability in the annual curve depending on regional weather patterns, severity of the wildfire seasons and other natural emissions from volcanos for example. However it is possible to extract a trend in the annual swing which indicates the health of the natural sinks.
In a paper published earlier this year1, James and Samual Curran performed an analysis based on comparing the trends of the annual minimum and maximum concentrations. This should show if the amount of CO2 drop each year due to natural sinks is changing, and if so, in what direction. Rather than providing a stable linear result, which would suggest a stable source, sink relationship, the results showed an initial rise in CO2 uptake, peaking in 2008 and then starting to decline.
This analysis shows that until 2008, as the CO2 rose in the atmosphere, plants responded by increasing their rate of growth and storing some of the additional carbon away in the soils and woody matter. This is known as CO2 fertilisation and increased the annual natural drawdown of CO2, helping to sequester a portion of the human emissions. After 2008, it shows that this sequestration started to falter as a result of a combination of wildfires which release carbon from woody matter and increased plant stress through extreme weather events including heat, drought and floods. Since this is a global analysis, changes in the ocean carbon sinks may also be playing a part.
2024
The record jump in the annual average CO2 concentration in 2024 requires a closer look to see if the trend identified above is continuing and indeed accelerating.
3.5 parts per million may not sound a lot but it takes 7.8 billion tonnes of CO2 to raise the atmospheric concentration by 1 ppm, so the 2024 rise was 27.3 billion tonnes.
According to the Global Climate Project’s Carbon Budget 20242, human emissions were as follows:
Fossil fuel emissions were 37.4 billion tonnes
Land use was 4.2 billion tonnes
Cement, which is both a source and sink was a net 0.75 billion tonnes
Making total human emissions 42.35 billion tonnes.
As described above, those emissions don’t all stay in the atmosphere. The natural land and ocean sinks absorb over half of them. From 2013 to 2023 they averaged 26% absorbed by the oceans and 29% absorbed by the land sinks. So the ocean should have absorbed 11 billion tonnes and the land 12.3 billion tonnes of 2024’s human emissions.
That would have left 19.05 billion tonnes in the atmosphere, enough to raise the level by only 2.4 ppm. The gap between 2.4ppm and 3.5 ppm is 8.25 billion tonnes of CO2, almost 20% of the human emissions, which is a huge proportion to remain in the atmosphere.
The only ways this could happen are that other sources emitted large quantities of CO2, this would include forest fires, ecosystem degradation, permafrost melt, soil heat stresses etc. or land and ocean sinks failed to take up their normal share.
Both are extremely worrying since increasing natural sources and failed sinks will accelerate accumulation in the atmosphere, accelerating the rate of warming and bringing forward extreme events linked to climate change.
Looking at the annual curves can point us in the right direction. The plot below shows all the annual curves since 1959 overlayed on top of each other. The values for each year are relative to the previous December’s concentration to eliminate the human growth trend. The spread shows up the natural variation and the changes noted by the Currans over the decades. The red line is 2024.
January starts fairly normally, but through the rest of the spring, levels are high, but not record breaking. The recovery through the northern hemisphere summer is slow to start and ends with a record high in September and October which is never recovered by the end of the year. The main anomalies appear to be a late start to the drawdown in July and an early termination in September and October.
Machine Learning Analysis
This small section deals with the details of a neural network analysis which gets a little technical. You’re safe to skip to the next section if you like, where I cover the results.
I applied a neural network approach to this data set called a Kohonen Self Organising Map (SOM). The SOM is a method of learning optimised representations of a complex multi-dimensional pattern. In this case, a year’s monthly relative CO2 concentration is represented as a point in 12 dimensional space, with each dimension taken as a month’s relative CO2 concentration. The SOM works like a clustering algorithm in that it finds a smaller number of cluster centres, or nodes, that represent the whole data set. The difference is that the nodes of a SOM are linked to each other through a grid or map structure, so that neighbouring nodes in the map represent proximal points in the data space, suggesting similar cluster characteristics. In this case I used a 4x4 SOM which provides 16 nodes. Once trained the data can be stepped through to see which nodes are closest to each year and how far away the years are from their closest node. This provides both a cluster classification which can track the evolution of the shape of the curves and the distance from the nearest node which provides a measure of uniqueness for a particular year.
Having trained the SOM on the 1959 to 2023 data (the green lines on the figure above), it was shown the 2024 data. Two outputs are created, firstly the closest node is identified allowing us to see which other years were similar, and secondly the vector distance between the year and the node provides a level of uniqueness.
Machine Learning Results
Initial analysis of the trained network showed that there was very little node overlap between the pre-2008 data and the post 2008 data which confirms the Currans’ analysis that a transition in CO2 curve shape happened around that point. The pre-1990 data was also well separated from the post-2000 data on the 4x4 map indicating a steady change in shape over the whole period. This also provides reassurance that the neural network has successfully separated the curve characteristics through the whole period and can therefore provide useful insights into the causes of the 2024 anomaly.
When shown the 2024 data, the network associated it with a node that also represented 1988, 1998, and 2023. Both 1988 and 1998 were exceptional years in that they were strong La Niña years which followed strong El Niños, just like 2024. We can conclude from this that part of the CO2 source and sink characteristics of 2024 is attributable to the ENSO cycle. The fact that 2023 also associated with this node suggests that whatever else influenced 2024, was underway, though to a lesser extent, during 2023.
The uniqueness value for 2024 was also very high. It was the second highest of all years, after 1991, with a score almost double the average for all the other years. This shows us that other new factors, beyond ENSO, are at play in 2024’s source and sink behaviour.
Breaking it down by month, the left hand chart above shows the monthly difference between 2024 and the winning network node. The right hand chart shows the accumulation over the year. They confirm the eyeballing we did above, showing that February and March had unusually strong CO2 accumulation, but that this was cancelled out by strong sink activity in April and May. There was some more source activity in July which was countered by sinking in August, then it was unusually high again from September through to the end of the year. The September high is of particular interest as it set the scene for the end of year record.
Natural Sources and Sinks
Assuming that human emission trends in terms of seasonal variation are fairly stable from year to year, the variations in the months identified above must be natural in origin and therefore linked to natural carbon source and sink behaviours.
ENSO
The El Niño Southern Oscillation is a phenomena in the Pacific Ocean which sees sea surface temperature extremes of hot (El Niño) and cold (La Niña) periods. They influence weather conditions around the world including rainfall and even hurricane suppression3.
Years with a warm anomaly in the tropical Pacific show a faster CO2 rise due to weaker land carbon sinks, particularly in the tropics, with a partial offset by stronger net uptake by the oceans. The opposite happens in years with cool Pacific sea surface temperature anomalies. These differences are mainly driven by physiological processes such as photosynthesis and respiration rates, with a smaller contribution from wildfires brought about by lower rainfall.
In the oceans, El Niño conditions involve decreased upwelling of carbon in the equatorial Pacific due to a weakening of the trade winds, causing this region to become a weaker sink of CO2, or even near neutral if the El Niño event is strong.
The neural network has however taken this into account as demonstrated by the low uniqueness output from 1988 and 1998. The 2023 El Niño was also not as strong as previous events, so the anomalies shown in the chart above are in addition to the ENSO effect.
Wildfires
Wildfires are a high natural emission source of CO2. Data from the Global Wildfire Information System4 shows that in 2024, 6.17 billion tonnes of CO2 was emitted through wildfires. However this was not as high as the 6.67 billion tonnes emitted in 2023. In fact 2024 ranks only 14th over the last 20 years. Interestingly though, the Brazilian emissions for 2024 were the highest since 2010 with peaks in February-April and September-November which aligns with the observed anomaly. Wildfires therefore made a contribution, however they don’t explain the whole of 2024’s jump in concentration. This is especially the case in the spring since the atmospheric levels returned to normal through the early summer.
Wildfires are something to look out for in the future however as climate change is making them more likely. According to the State of Wildfires 2024-2025 report5, anthropometric climate change made the fire weather conditions 2.1x more likely in Northeastern Amazonia, 3.3x more likely in the Pantanal-Chiquitano, 2.3x more likely in Southern California and 1.6x more likely in the Congo Basin.
Deforestation and Degradation
Fire is not the only disturbance that can cause a tropical forest to become a carbon source.
Two recent studies of the Amazon show the impact of continued deforestation on the regions ability to act as a carbon sink and how they are becoming a carbon source. Deforestation has significantly increased surface air temperatures and reduced rainfall during the Amazonian dry season6. Over the past 35 years, deforestation has accounted for approximately 74% of the ~ 21 mm per year dry season decline and 16.5% of the 2°C rise in maximum surface air temperature. This is pushing the Amazon towards a tipping point where it can no longer sustain a tropical rainforest habitat and becomes a savanna instead.
The 2023-2024 drought in the Amazon surpassed previous records. This would have led to increased temperatures and water shortages affecting plant growth and therefore carbon uptake. Coupled with forest fragmentation this has increased fire risk but also degradation rates. Clément Bourgoin et al.7 found a 152% surge in forest disturbances from deforestation and degradation in 2024, reaching a 2-decade peak of 6.64 Mha (million hectares).
The Amazon is not the only forest in the world however. A recent study from Australia has shown that its forests’ ability to sink carbon is also in decline8. They analysed the carbon content of above ground woody biomass in Australian moist tropical forests.
They found that a transition from sinking 0.62 Mg C ha−1 yr−1 (Megagrams of carbon per hectare per year) between 1971 and 2000 to a source of 0.93 Mg C ha−1 yr−1 between 2010 and 2019 has occurred for the aboveground woody biomass of these forests. The transition was driven by increasingly extreme temperature and other climate anomalies, which have increased tree mortality and associated biomass losses, with no evidence of the hoped for stimulation of woody growth through carbon fertilisation. Their work suggests similar trends in other tropical regions such as the Congo Basin.
Boreal forests are also experiencing die offs caused by drought, insect attacks and disease. This would tend to affect the sink of CO2 during the summer months though, so does not explain the Autumn rise in concentrations.
Peatland & Permafrost
Staying on land, the other potential sources of natural carbon are peatlands and permafrost. Peatlands only account for 3% of the land surface but hold 30% of the world’s soil carbon. They are susceptible to drought and elevated CO2 levels which causes them to dry out, reduce productivity and emit carbon back into the atmosphere. The role of CO2 in this action has been studied recently and surprisingly found that while it helped productivity at lower temperatures, it actually caused higher carbon emissions from the soils as temperatures rose9.
Taken with the Australian woody biomass study, this also provides some explanation for the Currans’ detection of a peak in sink activity in 2008.
Permafrost too has the potential to release significant amounts of CO2 during the melt season.
Looking at the surface air temperature anomaly for September 2024, the point at which the trend veered off the normal pattern, signifiant parts of the northern hemisphere where peat and permafrost exists, were under extremely high temperatures for the time of year.
The Oceans as a source and sink
Standing on a beach, the ocean looks like a single body of water, but there is a complex layering structure beneath the surface with currents going in different directions at different depths and vast areas of interchange where waters rise from the deep and others sink to the abyss.
Deep water currents collect carbon from the depths as organic matter rains down from above. Shallow water interacts with the air to exchange carbon both biologically through plankton growth at the base of the food web and chemically. When deep waters upwell they release some of their stored carbon into the air, when shallow waters overturn to the depths they take away carbon from that zone.
Water temperature plays a large part in the surface processes. Warmer water absorbs less carbon dioxide from the air. You can try this at home by warming apparently flat coke. More CO2 is released as the liquid warms. Temperature, oxygen levels and alkalinity also affect the ability of plankton to grow. To make matters more complex the three primary plankton species responsible for converting the majority of bio-utilised carbon dioxide into calcium carbonate behave differently under different conditions. One is particularly susceptible to pH changes and is under pressure through ocean acidification.10
Looking at the sea surface temperature anomaly for September 2024, shows huge marine heatwaves in the Pacific, North Atlantic, Caribbean and Mediterranean. This would have suppressed carbon absorption, even leading to regional net emissions.
The Southern Ocean is of key importance here. It is responsible for 40% of the global ocean’s uptake of our carbon emissions. However it is also the site of deep carbon rich upwelling which until recently has been capped by strong layers of different densities. Two recent studies are relevant here. The first by Léa Olivier and Alexander Naumann11 shows how the carbon rich deep water has been kept contained by the cap of fresh water at the surface from the melting glaciers around Antarctica. They have detected a shallowing over the last several decades and suggest that when released, the Southern Ocean would suffer a significant loss of sink potential.
In the second paper by Alessandro Silvano et al.12 they note that since the abrupt change in sea ice conditions since 2016, the surface salinity has been rising as more deep water makes it to the surface exposing it to high winds over lower ice covered seas. With September being the peak sea ice season in the Southern Hemisphere, this could have made a major contribution to the September 2024 rise in atmospheric CO2 since the Southern Ocean would have been limited in its sink potential having less ice to protect it from the winter winds.
The winter sea ice maximum around Antarctica is reached in September each year. 2024 was the second lowest on record, only slightly behind 2023. The plot from NOAA below shows the extend compared to the 1981-2010 average.
Conclusions
The 2024 rise in atmospheric CO2 was record breaking and since human emissions were stable, must have been due to significant changes in the Earth’s natural carbon sources and sinks. 20% of human emissions that would have been expected to be absorbed by the natural system remains in the atmosphere.
A neural network analysis of the year’s monthly concentration trend showed that part of the underlying pattern for the year was due to the ENSO cycle, in particular the La Niña which followed a strong El Niño in 2023. This was not enough to explain the whole anomaly though.
The Amazon, among other areas suffered from widespread and damaging wildfires which together with continued deforestation and degradation helped drive the peak levels early in the year, but these were largely recovered through the middle of the year.
Widespread drought and extreme temperatures in the early Autumn could have contributed to a rise in natural emissions from boreal forests, peatlands and permafrost in the northern hemisphere, together with moist forest emissions from Australia during it’s summer.
The ocean sink is also a strong contender for the high natural sources of carbon from September through the rest of the year, especially the Southern Ocean where increased surface salinity and low sea ice cover would have enabled deep carbon rich water upwelling, reducing the ocean’s ability to absorb atmospheric CO2.
All of these systems which contributed to the 2024 CO2 rise continue to be under increasing pressure suggesting that the Earth’s ability to absorb our emissions is indeed diminishing and will continue to do so in the coming years and decades. This will have a significant effect on the rising greenhouse gas concentration and accelerating global heating and climate change.
Given this fact, there has never been a more urgent time to dramatically cut human emissions.
James C. Curran, Samuel A. Curran, Natural sequestration of carbon dioxide is in decline: climate change will accelerate, 2025, https://doi.org/10.1002/wea.7668
Friedlingstein et al. Global Carbon Budget 2024, Earth Syst. Sci. Data, 17, 965–1039, https://doi.org/10.5194/essd-17-965-2025, 2025.
El Niño Southern Oscillation in a Changing Climate - Chapter 20 ENSO and the Carbon Cycle, Book editors: Michael J. McPhaden, Agus Santoso, Wenju Cai, Chapter authors: Richard A Betts et al. 2020 Print ISBN:9781119548126 Online ISBN:9781119548164 DOI:10.1002/9781119548164
Douglas I. Kelley et al. (2025) State of Wildfires 2024-25, Earth Syst. Sci. Data, 17, 5377, https://doi.org/10.5194/essd-17-5377-2025
Franco, M.A., Rizzo, L.V., Teixeira, M.J. et al. How climate change and deforestation interact in the transformation of the Amazon rainforest. Nat Commun 16, 7944 (2025). https://doi.org/10.1038/s41467-025-63156-0
Bourgoin, C. et al. Extensive fire-driven degradation in 2024 marks worst Amazon forest disturbance in over 2 decades, Biogeosciences, 22, 5247–5256, https://doi.org/10.5194/bg-22-5247-2025, 2025.
Carle, H., Bauman, D., Evans, M.N. et al. Aboveground biomass in Australian tropical forests now a net carbon source. Nature 646, 611–618 (2025). https://doi.org/10.1038/s41586-025-09497-8
Quan Quan et al. ,Drought-induced peatland carbon loss exacerbated by elevated CO2 and warming.Science390,367-370(2025).DOI:10.1126/science.adv7104
Patrizia Ziveri et al. ,Calcifying plankton: From biomineralization to global change.Science390,eadq8520(2025).DOI:10.1126/science.adq8520
Olivier, L., Haumann, F.A. Southern Ocean freshening stalls deep ocean CO2 release in a changing climate. Nat. Clim. Chang. (2025). https://doi.org/10.1038/s41558-025-02446-3
Silvano, A.et al. Rising surface salinity and declining sea ice: A new Southern Ocean state revealed by satellites, Proc. Natl. Acad. Sci. U.S.A. 122 (27) e2500440122, https://doi.org/10.1073/pnas.2500440122 (2025).
Good analysis and a stark reminder that nature’s buffering systems are reaching their limits. Outside of reducing emissions, investing in the natural systems that regulate our climate to build resilience is key. Rewilding, soil health, ocean protection must be climate priorities.
This preprint could also be important - microbial respiration went up by 3.5Gt C in 2024 due to abnormal wet and warm conditions mostly over grasslands - but not sure if their results will hold, so lets see:
"Dramatic increase in ecosystem respiration causes record-breaking atmospheric CO2 growth rate in 2024"; https://www.researchsquare.com/article/rs-6956425/v1