Faster dieback of rainforests altering tropical carbon sinks under climate change

Faster dieback of rainforests altering tropical carbon sinks under climate change | npj Climate and Atmospheric Science

Abstract

Carbon sinks in the tropical rainforests are restricting the global warming to attain unprecedented heights. However, deforestation and climate change is switching them to a net carbon source at some of the deforested patches. Using machine learning algorithm we predict that more than 50% of the tropical rainforests will undergo rapid “Savannisation”/transformation by the end of 21st century under high emission scenarios. Climate change projects ‘El Niño-like’ warming condition, which decreases precipitation in the rainforests and favors atmospheric dryness. In Central Amazonia vegetation degradation saturates the carbon sink and more than 25% of the rainforests will transform into a net carbon source due to increase in soil microbial respiration. This transition will accelerate if Eastern Pacific/Global temperature warms beyond 1.5◦K/2.3◦K (by 2050’s) and will undergo a steeper transit by ~2075 (2.45◦K/3.8◦K warming). This alteration will exacerbate global warming and has consequences for policies that are intended to stabilize Earth’s climate.

Introduction

Tropical regions are home for 45% of the world’s forested landmasses, which comprises approximately 15% of global land areas. Among them tropical rainforest encompasses around 25% of all the ecological zones, which are spread largely in South America, Central America, Africa, and Southeast Asia 1. They are the home for two-thirds of the terrestrial biomass and act as natural carbon sinks by exchanging more carbon fluxes with the atmosphere compare to any other biomes in the world 2. In the last three decades deforestation and degradation destroyed more than 420 million ha of forest, which have posed a serious threat on biodiversity and weakens their resilience against climate change 3, 4. Deforestation not only decreases the local rainfall or increases surface warming, it affects continental scale climate by changing surface energy balance through excess carbon dioxide emissions 5, 6. Unprecedented land use changes, increasing frequency of drought and forest fire expedite the transition of Amazon rainforest to Savanna type of biome 7, 8. In the last two decades, Congo Basin of equatorial Africa has experienced a decrease in forest cover by 28%, primarily in the Democratic Republic of Congo (DRC), Cameroon, Equatorial Guinea and Republic of Congo. The ongoing deforestation and forest losses are mainly driven by socio-economic factors e.g. shifting agriculture, forestry services and rapid urbanization 9. About 50% of the tropical peatlands, which served as primary sink of carbon, are located in Southeast Asia (SEA). Nearly 80% of these landmasses are either deforested or drained for plantations and agriculture, that switches the SEA ecosystem from a net sink of carbon to its source 10, 11. The decrease in precipitation over SEA is due to forest loss and cropland expansion and these reductions are much higher than that of Amazon and Congo basins 12. It is noteworthy that the vegetation dynamics of tropical rainforests are more sensitive to climate change rather than deforestation and land use changes 13, e.g. extreme drought events have intensified the risk of Amazon forest fires during 2015–16 El Niño event 14, 15. A shift in Pacific Sea surface temperature (SST) pattern towards El Niño like warming condition in presence of warm North Atlantic Ocean will increase the risk of severe droughts and extensive occurrences of dry season. For instance, the Amazon forest was affected by four severe droughts over the last two decades: 2005, 2010, 2015-2016, and 2023-2024. The El Niño itself caused the 2015-2016 one and warm North Atlantic caused the 2005 drought. The other two droughts of 2010 and 2023-2024 were caused by the joint influence of El Niño and the warm North Atlantic Ocean. These factors will reduce moisture recycling and will increase vapor pressure deficit in the rainforests, which is affecting the critical plant physiology thresholds 16. It will apparently increase the water stress conditions of the forests and frequency of forest fires 17, 18.

Precipitation and temperature are the two key variables that keep the rainforests moist and warm. Any changes in these two factors could alter the rainforest ecosystem into Savanna or semi-desert like ecosystems within the deforested modules 19. It reduces the ability of the rainforests to absorb the excess carbon dioxide, which could serve as a precursor condition for rapid climate breakdown. During neutral El Niño Southern Oscillation (ENSO) years, the Walker circulation is characterized by strong upwelling of moist and warm air parcels from the maritime continents, eastern Africa and northern part of South America. These warm pockets coincide with the regions of deep convection and precipitation, thereby maintaining an abundant supply of moisture for the tropical rainforests to sustain in their natural landscape. In contrast, the downward and subsiding branches of the overturning circulation brought dry airmasses to the troposphere due to sinking motion over the eastern Pacific Ocean and the Arabian Sea, albeit the strength is weaker in the later sector. During an El Niño event, the lateral drift of the SST warming center towards the central and eastern equatorial Pacific Ocean eventually shifts the Walker Circulation further eastward. The rising branches of the Walker circulation weaken over the Maritime Continent and South America and are replaced by anomalous sinking motions. It decreases the average rainfall over tropical rainforests during the El Niño years, thereby, exacerbating the impacts of droughts and threatens the viability of rainforests in its normal functioning regime 20, 21. However, the decrease in precipitation has higher impact in the deforested patches of Amazon and South East Asia 12, while the impact is less in the Congo Basin 22.

Due to extreme climate change, tropical rainforests will experience more frequent and severe drought events in future, which will switch them out of their healthy functioning regime, despite they have the natural ability of resilience against climate stresses 23. The resilience is lowest in the South American rainforests compare to its African and Asian counterparts 24. Although, El Niño is the primary driver for the precipitation to decline in the rainforests, however, the severity of drought and strong drying trends are entirely due to climate change and warming of North Atlantic Ocean, which aggravated the probability of occurrences of meteorological and agricultural drought by a factor of 10 and 30, respectively 25. Furthermore, there are increasing evidences which suggested that the Southeast Asian peatlands 26 and Amazon forests 27 are vulnerable to the risk of forest fires due to warming, prolonged occurrences of extreme hydrological drought events 28, forest degradation 29, land use changes 30, changes in Hadley circulation 31 and transition of South American Monsoon system 32.

Increasing frequency of hydrological droughts constrain the soil moisture to decrease non-linearly as a function of precipitation 32, which dampens land evapotranspiration (ET) trend due to limited moisture supply 33. ET often serves as a dominant contributor of regional precipitation 18, which accounts ~ 41% and 50% of rainfall in the Amazon and Congo basins, respectively, in a positive feedback loop 34. Interannual variability of Amazonian carbon balance is sensitive to drought 35, which threatened the carbon sinks by increasing the CO2 emission through root autotrophic and soil microbial heterotrophic respiration; a primary contributor of global warming 36.

Here we investigate the changes in precipitation in response to the El Niño like warming condition in the Pacific Ocean, which is altering the morphology of tropical rainforests and temperate forests in the subtropics. Using WaterGAP2-2e 37 based hydrological model, CMIP6 38 based GCMs and CLASSIC based land surface and terrestrial ecosystem model 39(Table 1 in method section) we unleashed a sequence of events that links the SST warming trend over Eastern Pacific Ocean with the gradual drying trend of the rainforests. Furthermore, using ISIMIP3B 40 based vegetation products we nailed down the eminent threats of rainforest degradation, due to irreversible alteration of photosynthesis, respiration and carbon uptake rate, a state that is conducive for global warming to exacerbate in a negative feedback loop.

Machine learning (ML) based regression models are employed to predict the vegetation transformation in the rainforests under different emission scenarios by feeding the models with meteorological, hydrological and vegetation parameters as input to envisage better accuracy in the prediction.ML techniques have several advantages over traditional CMIP6 based climate models for generating the vegetation data 41, 42. Although, the CMIP6 models includes the vegetation-related variables such as Gross Primary Production (GPP), Net Primary Production (NPP), and vegetation cover, it lacks the variable NDVI, which often been considered as the best representation of vegetation condition of a region. Secondly, ML models can downscale the coarse-resolution climate model outputs to much finer resolutions, allowing for more detailed and localized vegetation predictions. This is crucial for applications requiring high spatial resolution. Thirdly, the ML excels at capturing complex non-linear relationships between environmental variables and vegetation that linear assumptions in CMIP6 models struggle with. ML can integrate diverse data sources like remote sensing, meteorological, and ecological data to improve predictions and fill data gaps. Studies show ML models can outperform CMIP6 ensembles in predicting vegetation indices like NDVI. Once trained, ML models make predictions much faster than computationally expensive CMIP6 models. Additionally, ML models can adapt to new data, enabling continuous improvement in vegetation predictions as more observational data becomes available. This adaptability is the key in a changing climate. While CMIP6 models provide valuable insights, ML offers a complementary data-driven approach to enhance our understanding of vegetation dynamics 41, 42.

Results

Changes in tropical circulations and drying trend in rainforests

Historical (1971–2014) changes and future (2015–2100) projection (SSPs) of precipitation under different greenhouse gas emission scenarios (Fig. 1a–d) display a multi-model mean drying trend (standard deviation is shown in Supplementary Fig. 12. At higher emission scenarios, the drier zones will shift more northward and positioned over the central Amazonian rainforests (–3 to –5 Kg m2/dec). It is noteworthy that the proportion of rainfall that decreases over the Oceans (stretching from equator to subtropics), in South Africa, Southern tip of South America, Madagascar, western Australia and European temperate deciduous forests are linked with the descending branches of the Hadley cell. The subtropical drying trend 43 is consistent with the widening and poleward expansion 44 of tropics and thereby slowing down the regional Hadley circulation 45. On the other hand, at higher SSPs, rainfall increases over East Africa, Arabian Sea, South and East Asia, while the narrow strip of intensified rainfall (>5 Kg m2/dec) over central-eastern Pacific Ocean is the consequence of an eastward shift in Pacific SST warming pattern; a state that reflects more ‘El Niño-like’ conditions in future 46(Supplementary Fig. 2). The narrow strip of intensified rainfall collocates on the ascending branch of the altered Walker circulation 47, which is in response to the El Niño-like warming condition over central–eastern Pacific Ocean (Supplementary Fig. 2). The downward counterpart of the altered Walker circulation subsides over the Amazonia, Central America and western edge of tropical Africa, which strongly underpins the atmospheric dryness (–3 to –4 Kg m2/dec) and makes the rainforests deficit of rainfall. Similarly, the African branch of the Walker circulation favor wetness (2–3 Kg m2/dec) in East Africa, south of middle east Asia and Arabian Sea, while it’s subsidence over SEA rainforests will dampen the local precipitation trend (–4 to –5 Kg m2/dec) and will pose enormous threat for the rainforest ecosystem to thrive under excessive water stress.

The panels represent decadal trend in precipitation a–d, surface temperature e–h, evapotranspiration i–l, soil moisture m–p, surface runoff index probability/hydrological drought q–t, groundwater recharge u–x and groundwater storage y–z2 for the historical period (1971–2014, 1st panel), SSP126 scenario (2015–2100, 2nd panel), SSP370 scenario (2015–2100, 3rd panel) and SSP585 scenario (2015–2100, 4th panel). The subplots display multi-model mean trend unless otherwise stated and trends less than 95% significance level are masked.

El Niño-like warming condition and atmospheric dryness

To elucidate the physical mechanism that underlines the changes in precipitation in connection with Pacific Ocean warming, we plotted the latitudinal mean (5N–5S) SST and Sea level pressure (SLP) anomaly as a function of longitude and time in Fig. 2a, d and g for SSP126, SSP370 and SSP585 scenarios, respectively. The anomalies are calculated with respect to the 1981–2010 period. Irrespective of any scenarios, in post-2000 period the warming trends are distributed all along the tropics, however the eastern Pacific Ocean contribute maximum to this trend at higher SSPs. This is evident from area averaged (5N–5S, 170–110W) ensemble mean SST anomaly, which displays a monotonic warming (1–4 K, Fig. 2b) over the eastern Pacific Ocean. Furthermore, the area averaged Eastern Pacific warming is higher than tropical mean SST warming under SSP126 and SSP370 scenarios, while under SSP585 scenario the Eastern Pacific warming contributes almost whole of the warming over tropics (Supplementary Fig. 3). To validate the model based outputs we computed the multimodel mean SST and standard deviation and compare it with area averaged observed SST anomaly data. The interannual variability of the observed SST hovers between the upper and lower bounds of standard deviation, which reflects an absence of any systematic bias between observation and modeled SST. It inadvertently ensures the reliability of model projections that display an ‘El Niño-like’ warming condition over Eastern Pacific Ocean.

a,d and g represents SST anomaly (K, color shaded) and SLP anomaly (bold black line indicates SLP>0.5 hPa and dashed black line indicates SLP<−0.5 hPa) under SSP126, SSP370 and SSP585, scenarios, respectively. b, e and h represent area averaged SST anomaly (5N–5S, 170–110W), SLP anomaly over 5N–5S, 100–180E and 5N–5S, 75–50W, respectively. The anomalies are calculated with respect to the 1981–2010 period. Bold lines represent the mean and vertical bars represent the standard deviation. Blue, black and red lines represents CMIP6 model output data for SSP126, SSP370 and SSP585 scenarios, respectively, while magenta line represents HadleySST data (observed). c,f,i and j represent correlation coefficients between area averaged SST anomaly and precipitation anomaly for the historical period, SSP126, SSP370 and SSP585 scenarios, respectively. Coefficients less than 95% significance are masked.

On the other hand, area averaged ensemble mean SLP will portray an anomalous high pressure (~1 hPa, Fig. 2h) center over the western–central Pacific Ocean and a low pressure (~ –0.6 hPa, Fig. 2e) center over the eastern Pacific Ocean–South American landmasses. Since east–west differences in the Pacific SLP pattern is the key aspect of Walker circulation, its deviations from climatological mean behavior in response to the ‘El Niño-like’ warming condition is indicative of a strengthened (due to stronger upwelling but weakened compare to the climatology) but shifted Walker circulation at higher SSPs. It not only reflects a stronger upwelling of moist and warm air parcels from eastern Pacific Ocean but it projects a stronger subsidence of drier airmasses over the Amazon rainforest, SEA and western Africa, albeit weakly on the latter domain. Fig. 2e, h represents an area averaged SLP anomaly over 5N–5S, 100–180E (western–central Pacific Ocean) and 5N–5S, 75–50W (eastern Pacific Ocean–South American landmasses), respectively; the regions that collocate with the ascending and descending branches of the altered Walker circulation are displaying a monotonic increasing and decreasing trend in SLP from the early 20th century, respectively. Even under SSP126 scenario, the warming trend is not projected to slow down atleast before 2050, which inadvertently demands a fast track action to cut down global emissions to its lowest possible extent (Fig. 2b). To further ascertain the covariability between SST and precipitation, we perform correlation coefficient between area averaged SST anomaly and precipitation (Fig. 2c, f, i and j) for historical period, SSP126, SSP370 and SSP585 scenarios. Higher the emission and surface warming is (Fig. 1e–h), higher is the degree of coupling or covariability between SST and precipitation. SST anomaly accounted for approximately 80–90% of the explained variances in precipitation changes in tropics and subtropics by modulating the position and intensity of the Pacific Walker and Hadley circulation, respectively.

Since high amount of precipitation is the key for tropical rainforests to survive under climate stress, ET acts as a prime determinant for their moisture abundance by returning a considerable amount of land precipitation back to the atmosphere in a positive feedback loop 48. However, changes in ET are usually governed by the complex interplay between surface temperature and terrestrial moisture supply 49. ET decreases (Figs. 1i–l, –2 to –3 Kg m2/dec) drastically due to increasing soil moisture limitations (Figs. 1m–p, –3 to –5 Kg m2/dec) in the drier (Ptrend<0) and hotter (Ttrend>>0) patches of northern Amazonia, Central America, Western and Southern Africa, temperate deciduous forest, SEA (Java, Sumatra and Southwest Borneo islands) and coastal Australia. While in the less hot (Ttrend>0) and wetter (Ptrend>0) pockets excess precipitation surpasses the soil moisture deficit (–2 to –3 Kg m2/dec), leading to an ET surplus in North America, South and East Asia. The deficit is more critical in places that are stretching from northeastern (Guyana, Suriname, French Guiana and Northeast Amazonia) to the southeastern part of Amazon rainforest, which in turn would disrupt the moisture feedback loop to the atmosphere and will further exacerbate the regional dryness and aridity. The multi-model standard deviation of ET, soil moisture and T are shown in Supplementary fig. 1.

Persistent decrease in rainfall that propagate over time will aggravate the occurrences of agricultural (soil moisture) and hydrological (runoff) droughts, based on where in the water cycle moisture deficit occurs 50. Although the probability of occurrences of hydrological drought (Standardized Runoff index i.e. SRI (12) <–1) will not vary much (0.15–0.16, Fig. 1q–t) in future, its spatial extent will encroach wider areas of Amazon rainforest, central America, western, central and southern Africa, SEA (Java, Sumatra and Southwest Borneo islands) and temperate deciduous forest of Europe, that are projected to face acute water stress in future. Regions that are exposed under persistent and prolonged droughts, groundwater storage (GWS) often act as a natural reserve against erratic rainfall variability 51. In response to the changing rainfall patterns GW recharge will decrease significantly due to warming in the areas of increasing atmospheric and surface dryness (Figs. 1u–x, –1 to –2 Kg m2/dec). It will destabilize the GWS (Figs. 1y–z2, –3 to –4 Kg m2/dec) by disrupting the water inflow deep underneath of Amazon rainforest, western Africa, SEA and temperate deciduous forest of Europe. It is noteworthy that during 1971–2014, GWS exhibits a decreasing trend (Fig. 1y, ~ –4 Kg m2/dec) in the Southern part Brazil, which is consistent with satellite based GRACE GWS trend (Supplementary Fig. 4, ~ –5 Kg m2/dec), albeit for a different time period (2003–2022). The pattern of decreasing GWS trend is projected to shift northwestward over Central Amazonia and Central America, which reduces the strategic reserve for the rainforest to survive under water stress. Under high emission scenarios, the combined influence of rainfall, ET and soil moisture deficit, recurrent droughts and gradual depletion of the natural buffer (GWS) will make the tropical rainforests and subtropical forests more vulnerable to the climate change 3, 4. The multi-model standard deviation of GW recharge and GWS are shown in Supplementary Fig. 1.

Vegetation degradation and tropical Carbon sinks

Projected warming and drying trends across Amazonia, Western Congo Basin, parts of the Central America, SEA and temperate deciduous forests could transform the forest biome to Savanna or grassland vegetation type in response to the deforestation 4, dry season intensification 52, climate change 4, 53, increasing respiration 54, 55 and decreasing photosynthesis 56, 57. Such transformations could saturate the tropical carbon sinks 58 and alter the ability of the forests to absorb or sequester atmospheric Carbon 59. However, it is still unclear how the tropical and subtropical vegetation dynamics will transform under increasing dryness and water stress that are instrumented by Pacific SST warming and changes in tropical circulation patterns.

Machine learning (ML) tools are often used to predict the variables by training the models with appropriate and sufficient input parameters as predictors 60. Furthermore, ML techniques can also be employed to estimate the nonlinear contributions of predictors on the predictand variables 61. Here we selected six spatial domains e.g. Amazon and Central American rainforests, Western and Central African rainforests, SEA rainforests, South African forest, temperate deciduous forest of Europe and Southern tip of South America that will experience a decline in precipitation (Fig. 1a–d) under high emission scenarios. Within these spatial clusters we simulate the annual mean NDVI (2015–2100) as a proxy of vegetation coverage and type under each of the emission scenarios (SSP126, SSP370 and SSP585) by training four ML based regression models (Tree, Support Vector Machines (SVM) Gaussian, Bagged ensemble and Neural network). Normalized Difference Vegetation Index (NDVI)>0.8 corresponds forest vegetation, while 0.8>NDVI>0.4 represents Savanna type/semi-arid vegetation 62. More details on the results are available in the method section and Supplementary Table 1.

Figure 3a displays the NDVI climatology (1982–2014), which reflects the mean state of the vegetation dynamics and greenness. Values higher than 0.8 corresponds dense rainforests of Amazonia, central America, Congo basin and SEA, while NDVI between 0.5 and 0.8 stands for tropical Savanna type biome of western and central Africa and values below 0.5 represents tropical dry or temperate deciduous forests of Europe. At higher emissions, the simulated NDVIs will deviate alarmingly from its climatological mean value (Fig. 3c, f and i for SSP126, SSP370 and SSP585, respectively), a state that indicates a rapid transformation of the rainforest biome towards Savanna or tropical dry vegetation type. In South America, degradation of the rainforest (NDVI decreases sharply below 0.8) stretches northward from Brasilia, Para, Rondonia and Amazonas in Brazil, to Guyana, Suriname, and French Guiana and to the Peruvian and Bolivian sector; following a pattern that is consistent with the northwestward shift in the center of rainfall, soil moisture and GWS deficit (Fig. 1). Central American rainforest usually receives less attention in studies that are linked with climate change and tropical rainforest dieback will experience a gradual transformation into a tropical dry vegetation state (NDVI decreases below 0.5), in particular the strongest trends are observed in Costa Rica, Nicaragua and Honduras. Figure 4 represents the same as Fig. 3, but is zoomed in for Amazon and Central American rainforests, African rainforests and Southeast Asian rainforests.

a NDVI climatology (1982–2014) from NASA ARC ECOCAST GIMMS NDVI3g, c, f and i represent Machine learned predicted (2015–2100) NDVI under SSP126, SSP370 and SSP585 scenarios, respectively. NDVI>0.8 represents tropical rainforest, b, d, g and j represent probability of occurrences of –NEP i.e. potential carbon sources (probability>0.5) for historical period (1971–2014) and SSP126, SSP370 and SSP585 scenarios, respectively. e, h and k are the bivariate risk map (–NDVI i.e. vegetation degradation and –NEP i.e. net Carbon uptake by atmosphere).

It represents the same as Fig. 3, but is zoomed in for Amazon and Central American rainforests, African rainforests and Southeast Asian rainforests.

On the contrary, the African rainforests are more resistant to the climatic extremities, because of slower warming rate, higher elevations 52, abundance of drought-adapted species and their natural ability to sustain in the dry conditions 24, however, unprecedented hydrological droughts (Fig. 1q–t) and decreasing rainfall (Fig. 1) will underscore the threat of “Savannisation” on the eastern edge of the Congo basin (DR Congo, Rwanda and South Sudan). In SEA rainforests, Java, Sumatra, Borneo and New Guinea islands will suffer greater share of the transformations in vegetation dynamics amid rapid climate change, while the subtropical arid (Southern Africa) and temperate (deciduous forest of Europe) biomes are more robust and sustainable in response to the atmospheric dryness.

Therefore, in unabated emission and warming scenarios, climate change will implicate a rapid transformation/degradation of the rainforests from its tropical moist form to Savanna or tropical dry state, which inadvertently will cease their natural ability of sequestering ~50% of the global terrestrial carbon uptake 63, 64. As evident from the Fig. 4, 58% of the Amazon and Central American forests (left panel), 43% of the African forests (middle panel) and 61% of the SEA forests (right panel) will experience a rapid transformation under climate change. The percentages are calculated by accumulating the number of grid points that are projected to transform under climate change in their respective forest biomes. Any more degradation in the structurally intact tropical forests will alter the rate at which net carbon is exchanging with the atmosphere (net ecosystem production (NEP)) either by decreasing photosynthesis (gross primary production (GPP)) and fixing less CO2 or by increasing respiration (RE). In the tropical forest, carbon balance is often defined as the difference between GPP and RE i.e. NEP=GPP–RE, where, +NEP corresponds a net carbon sink, while –NEP corresponds a net carbon source to the atmosphere 53.

The carbon content of soil is about three times higher than that of atmosphere or vegetation 36, which makes soil respiration as the largest contributor of carbon efflux into the atmosphere through root (autotrophic) and microbial (heterotrophic) respiration 65, 66. Recent studies have revealed an increase in soil heterotrophic respiration 54, 55 and highlighted the role of hyperactive soil microbes in weakening the ability of land surface to act as a carbon sink. Figure 3b,d,g and j illustrate the probability of occurrences of NEP<0 for the historical period, SSP126, SSP370 and SSP585 scenarios, respectively, which portrays the regions of net carbon uptake that are transitioning to the regions of net carbon release. The patches where probabilities are higher than 0.5 correspond to the sections of degraded forests that no longer will serve as net carbon sink, rather they will act as a net source with greater release of carbon to the atmosphere. Forest degradation due to climate change and atmospheric dryness will project the carbon sink to weaken across the wider swath of Central Amazonia (probability >0.4) which are in the verge of rapid transformation, while the reversal (probability >0.5) is unbound along the Amazon river basin that stretches from Guyana, Suriname, and French Guiana to the Peruvian sector of the rainforest.

In Central America the transformation is visible in Costa Rica, Honduras, Nicaragua, Guatemala and Mexican sector of the rainforests. While in Africa and SEA, coastal Ghana, Benin, Togo, Nigeria and Greater Sunda islands (Sumatra and Java) will emerge as the potential sources of carbon in future. Ceasing ability of the rainforests to sequester anthropogenic emissions and their reversal to a net source will ensure a two-fold rise in atmospheric carbon budget that will have worrying implications for future warming trends. To examine the risk of the transformation due to rapid “Savannisation” of the rainforests we plotted bivariate choropleth map of probability of occurrences of –NPE and –NDVI in Fig. 3e,h and k for SSP126, SSP370 and SSP585, respectively. Across Central Amazonia, Central America (Nicaragua, Honduras, Mexico), Western Africa (coastal Ghana, Benin, Togo, Nigeria) and SEA (Greater Sunda islands, particularly lower Sumatra and upper Java) rainforests, an increasing convergence of –NPE at higher values of –NDVI indicate a gradual transformation of traditional carbon sinks to a net carbon source in response to the vegetation degradation. At higher emission scenarios, about 25% of the degraded Amazon and Central American rainforests are transforming into a net carbon source and thereby looming threat on its faster dieback that will further exacerbate the global warming trend in a negative feedback loop.

In accordance to the atmospheric carbon balance equation, for NEP to decrease in future, either GPP should decrease or RE should increase under high emission scenarios. In the Amazon basin and Central America, higher occurrences of –GPP (decreasing photosynthesis) coincide with the higher values of –NDVI (increasing degradation) in smaller patches of northeast Amazonia, Nicaragua, Honduras and Mexican sector of the rainforests (Fig. 3a, c and e), which perhaps linked with the direct or indirect effects of increasing temperature or decreasing soil moisture on photosynthesis 65. The bivariate maps (Fig. 3b, d and f) illustrate that the areas of the degraded forests (with higher –NDVI’s ~1) which are projected to experience precipitation deficits and hydrological stress are falling at higher probabilities of +RE (increasing respiration ~1). Therefore, due to rapid “Savannisation” an increasing soil microbial respiration has the strongest effect on NEP to decrease across the wider swath of South America, Central America, Africa and SEA rainforests (Supplementary Fig. 5). It will eventually disrupt the existing carbon balance and will transform the traditional carbon sinks to a net carbon source.

In addition to the human caused deforestation and land use changes, vegetation degradation in the rainforests is an obvious outcome of El Niño-like warming condition in the Eastern Pacific Ocean. It will eventually transform the Amazonian and Central American carbon sinks to a net source of carbon for the atmosphere to uptake. To ascertain the variability in vegetation transformation in context of SST warming, we plotted the area averaged (South and Central America) simulated NDVI as a function of area averaged (central-eastern Pacific) SST in Fig. 5a. The vegetation dynamics follows an exponentially decaying curve i.e. at low–medium warming (<1.5 ◦K) and at low emission scenario (SSP126) rainforest vegetation will recover slowly, while at higher warming (>1.5 ◦K) and higher emission scenarios (SSP370 and SSP585) vegetation degradation will acquire a path of faster dieback at 2.45◦K by ~2075. Figures 5b,c displays an increasing trend in total respiration as a function of Eastern Pacific warming and vegetation degradation, respectively. SST warming accounts approximately 83% of the explained variances in carbon emission rise from Amazon rainforest due to root and soil microbial respiration. 1.5 ◦K and 2.45 ◦K of SST warming corresponds 2.3 ◦K and 3.8 ◦K in global mean warming scale, respectively, which is shown in Fig. 6. Therefore, any level of warming beyond 2 ◦K will be disastrous for the very existence of the Amazon and Central American rainforest.

a Changes in area averaged (South and Central America) simulated NDVI as a function of area averaged (Central-Eastern Pacific Ocean) SST warming. The black empty circles represent NDVI under all scenarios, while blue, black and red bold circles represents NDVI under SSP1–2.6, SSP3–7.0 and SSP5–8.5 scenarios, respectively. The black line is the polynomial fitted profile of order 3, indicates an exponential decay in vegetation at warming beyond 1.5 ◦K and transformation at 2.45 ◦K. b It represents the same as (a) but for total respiration (autotrophic & heterotrophic) as a function of SST. The black line indicates the linear trend. c Changes in total respiration as a function of simulated NDVI over South and Central America.

It represents the same as Fig. 5a but as a function of global mean temperature (GMT) rise. The black empty circles represent NDVI under all scenarios, while blue, black and red bold circles represents NDVI under SSP1–2.6, SSP3–7.0 and SSP5–8.5 scenarios, respectively. The black line is the polynomial fitted profile of order 3, indicates an exponential decay in vegetation at global warming beyond 2.3 ◦K and transformation at 3.8 ◦K.

Discussions

Global oceans play a crucial role in controlling the global warming by absorbing a significant amount of the excess heat trapped by greenhouse gases in the Earth's atmosphere. Oceans absorbed over 90% of the excess heat trapped by human-caused greenhouse gas emissions since the 1950s, which has moderated the rate of atmospheric warming 67. In response, oceanic heat content has increased alarmingly in the recent decades over the Pacific Ocean (0.1 ◦K/dec, Supplementary Fig. 2a) and a multi-fold (2–4 times) increase in warming is projected under SSP1–2.6, SSP3–7.0 and SSP5–8.5 scenarios (supplementary fig. 2c, e and g, respectively). This warming will be strongest in tropical central–eastern Pacific Ocean, a pattern that resembles more of an ‘El Niño-like’ warming condition in the Pacific Ocean. It will strengthen the air-sea interaction mechanisms by transferring warm SST anomalies of eastern Pacific origin up into the atmosphere, which will have serious ramifications on tropical and subtropical precipitation trends via changes in Pacific Walker and Hadley circulations, respectively. In El Niño-like conditions, an eastward shift in the Walker cell towards Eastern Pacific Ocean will subsequently alter the position of its descending branches and will impose anomalous dryness over the tropical rainforests; in particular, the Amazon and Central American sectors will be widely affected. Poleward extension and slowdown of the Hadley cell will bring severe rainfall deficit over subtropical temperate deciduous forests in Europe and South Africa.

Precipitation deficit and regional warming trend will reduce the soil moisture content in the rainforests, which vehemently will disrupt the moisture feedback loop by reducing ET to recycle the percipitable moisture back to the atmosphere. This will increase the probability of hydrological droughts to encroach wider landmasses of the tropical and subtropical forests. In absence of abundant moisture supply, increasing dryness will deplete the GWS by hindering its recharge, which will further complicate the sustainability of the forests and their normal functioning during severe water stress conditions.

Here we used machine learning algorithms to predict the future changes in vegetation patterns under different emission scenarios, which indicate that ~50% of the tropical rainforests will undergo a rapid transformation to Savanna or tropical dry vegetation type due to projected increase in atmospheric dryness and climate change. The situation is more critical in the Amazon rainforest. The tropical biomes south of the Amazon forest is a biome with the highest biodiversity and much soil carbon storage. Once the Amazon tropical forest reaches the tipping point of "Savannisation", the highly degraded ecosystem will not have the same biodiversity as the tropical savannas south of the Amazon basin and will store very little soil carbon. In sum, it will become an open-canopy, highly degraded ecosystem.

These transformations in vegetation morphology underpins the higher probability of occurrences of NEP<0, due to increasing microbial activity in the degraded patches of Central Amazonia, Central America (Nicaragua, Honduras, Mexico), Western Africa (coastal Ghana, Benin, Togo, Nigeria) and SEA (Greater Sunda islands) rainforests, and thereby transform the traditional carbon sinks into a net source of carbon to the atmosphere. These transformations are more acute in the Amazonian and Central American sectors of the rainforests, which is expected to accelerate by the middle of 21st century due to faster degradation of rainforests if SST/GMT warms beyond 1.5 ◦K/2.3 ◦K (by late 2050’s) and will undergo total transformation if warming exceeds 2.45 ◦K/3.8 ◦K (by ~2075).

Methods

CMIP6 simulations

The historical simulation and the SSP1–2.6, SSP3–7.0 and SSP5–8.5 experiments from CMIP6 models (Table 1 in method section) are used to represent the present-day and future climates, respectively 38. We analyzed the monthly mean precipitation, surface temperature, surface wind, sea surface temperature (SST) and sea level pressure (SLP) data to calculate the multi-model mean trend. The data can be downloaded from https://pcmdi.llnl.gov/CMIP6 or https://climate-scenarios.canada.ca. All the monthly model data are interpolated onto a 0.5◦ × 0.5◦ grid before analysis. This study presents the results as the multi-model mean unless stated otherwise.

ISIMIP3B simulations

Inter-Sectoral Impact Model Intercomparison Project (ISIMIP) provides climate and socioeconomic forcing datasets for cross-Sectoral climate impact model output data. ISIMIP3b is the latest version which incorporates GCM-based quantification of impacts at different levels of climate change 40. It provides bias-corrected CMIP6 climate forcing for pre-industrial, historical, SSP1–2.6, SSP3–7.0 and SSP5–8.5 conditions. The historical simulations are forced by observational climate and socioeconomic information, while the future projections are designed with fixed socioeconomic condition, which include the possible trends in agriculture, land use and deforestation. Here we used WaterGAP2-2e 37 based global water availability and water use model data for analysis. Historical and simulated atmospheric datasets from MRI-ESM2-0, IPSL-CM6A-LR, MPI-ESM1-2-HR, UKESM1-0-LL, GFDL-ESM4 are used in the WaterGAP2-2e model to generate 5 sets of soil moisture, total runoff, total discharge, groundwater storage, groundwater recharge and groundwater discharge for historical period (1901–2014), SSP1–2.6, SSP3–7.0 and SSP5–8.5 scenarios (2015-2099). For gross primary production (GPP), autotrophic and heterotrophic respiration we used the Canadian Land Surface Scheme Including Biogeochemical Cycles (CLASSIC) based land surface and terrestrial ecosystem model which simulate the biomes data. CLASSIC is developed by coupling the Canadian Land Surface Scheme (CLASS) and the Canadian Terrestrial Ecosystem Model (CTEM). CLASS simulates the fluxes of energy, water, and momentum, while CTEM simulates biogeochemical cycles including fluxes of carbon. All the monthly model datasets are available at 0.5◦ × 0.5◦ resolution. More details are available in Table 1, method section. This study presents the results as the multi-model mean unless stated otherwise. The data can be downloaded from https://data.isimip.org/search/tree/ISIMIP3b/OutputData/water_global/wa… and https://data.isimip.org/search/tree/ISIMIP3b/OutputData/biomes/classic.

NDVI data

The Normalized Difference Vegetation Index-3rd generation: NASA/GFSC GIMMS, version number 3g.v1 data are generated using the Global Inventory Monitoring and Modeling System (GIMMS) 68, 69. The vegetation indices are measured radiometrically by absorbing photosynthetically active radiation by chlorophyll in the green leaves of vegetation canopies and therefore serve as a good measure of surface greenness level of a region. The data is available from 1981–2015 at a resolution of 0.83◦ × 0.83◦, which are later rescaled at 0.5◦ × 0.5◦ resolution by interpolation. This product can be downloaded directly from https://ecocast.arc.nasa.gov.

Machine learning (ML) algorithm

Within each of the spatial domains we simulate the annual mean NDVI as a proxy of vegetation coverage and type under each of the emission scenarios (SSP126, SSP370 and SSP585) by training four ML based regression models (Tree, SVM Gaussian, Bagged ensemble and neural network). The satellite based NDVI data (1982–2014) is regressed with CMIP6 based ensemble mean precipitation, surface temperature and surface wind data, WaterGAP2-2e model based ensemble mean GWS, Soil moisture, Evapotranspiration, Potential evapotranspiration, surface runoff, total runoff, GW recharge, total discharge, GW runoff and CLASSIC model based GPP, root autotrophic respiration and soil heterotrophic respiration data as input parameters to predict the future changes in NDVI in the selected domains using the aforementioned supervised ML (more details in Table 1 below). The inputs for each of the four ML based models are the ensemble mean of the variables mentioned above, which are derived by forcing the WaterGAP2-2e and CLASSIC model with atmospheric data sets of the 5 GCMs (MRI-ESM2-0, IPSL-CM6A-LR, MPI-ESM1-2-HR, UKESM1-0-LL, GFDL-ESM4). For each run we performed 10 cross-validations to examine the predictive accuracy of the fitted models and to protect the models against overfitting. The results of ML algorithms at different geographical sectors are tabulated at Supplementary Table

GRACE and GLDAS groundwater storage (GWS)

The gravity recovery and climate experiment (GRACE) total water storage (TWS) data is available from the JPL NASA during the period from 2002 to 2023, in the form of monthly gravity mascon solutions. The GRACE TWS is the anomaly data that are available with the resolution of 0.5°. The GRACE data can be available from https://grace.jpl.nasa.gov/data/get-data/. In many cases, the GRACE observation datasets are combined with the global land data assimilation (GLDAS) system to estimate the groundwater storage (GWS) by subtracting GLDAS soil moisture (SM), snow water equivalent (SWE), and the canopy (CA) from GRACE based TWS data 67. The GLDAS data is available from http://ldas.gsfc.nasa.gov/gldas/gldas-get-data/.

Carbon sink

Carbon balance is defined as the difference between GPP and RE i.e. net ecosystem production NEP=GPP–RE, where, +NEP corresponds a net carbon sink, while –NEP corresponds a net carbon source to the atmosphere.

Standardized Runoff Index (SRI)

SRI correspond the hydrological drought index, which is derived using a nonparametric approach 70. The marginal probability distribution of surface runoff is computed using the empirical Gringorten plotting position. The empirical probabilities are then standardized as SRI=ϕ−1(pr) where ϕ is the standard normal distribution (probability density function), and ps denotes the empirical probability of surface runoff.

HadSST

The Met Office Hadley Centre's sea surface temperature (HadSST) data set, Version 4.0.1.0 is a monthly global SST data on 5° latitude by 5° longitude grid from 1850 to 2024. The bias adjusted data is produced by taking in situ measurements of SST from ships and buoys and the anomalies are calculated by subtracting climatological values from the measurements. The data is downloaded from https://www.metoffice.gov.uk/hadobs/hadsst4/.

Data availability

All the original datasets used in this research are publicly available from their sources e.g. CMIP6- https://pcmdi.llnl.gov/CMIP6, https://climate-scenarios.canada.ca ISIMIP3B- https://data.isimip.org/search/tree/ISIMIP3b/OutputData/water_global/wa…, https://data.isimip.org/search/tree/ISIMIP3b/OutputData/biomes/classic, NDVI NASA/GSFC GIMMS- https://ecocast.arc.nasa.gov, GRACE- https://grace.jpl.nasa.gov/data/get-data/, GLDAS- http://ldas.gsfc.nasa.gov/gldas/gldas-get-data/. The machine learning based output datasets can be available on request from the corresponding authors.

Code availability

The software codes to generate the datasets will be available freely from the corresponding authors.

1.
Ometto, J. P. et al. Cross Chapter Paper 7: Tropical Forests. In: Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge University Press, Cambridge, UK and New York, NY, USA, pp. 2369–2410 (2022).
2.
Pan, Y. et al. The structure, distribution, and biomass of the world’s forests. Annu. Rev. Ecol., Evolution, Syst. 44, 593–622 (2013).
3.
Denning, S. Southeast Amazonia is no longer a carbon sink. Nature 595, 354–355 (2021).
4.
Gatti, L. V. et al. Amazonia as a carbon source linked to deforestation and climate change. Nature 595, 388–393 (2021).
5.
Bonan, G. B. Forests and climate change: forcings, feedbacks, and the climate benefits of forests. Science 320, 1444–1449 (2008).
6.
Hansen, M. C. et al. High-resolution global maps of 21st-century forest cover change. Science 342, 850–854 (2013).
7.
McKay, D. I. A. et al. Exceeding 1.5°C global warming could trigger multiple climate tipping points. Science 377, eabn7950 (2022).
8.
Parry, I. M. et al. Evidence of localised Amazon rainforest dieback in CMIP6 models. Earth Syst. Dynam. 13, 1667–1675 (2022).
9.
Barimalala, R. et al. Effects of the Congo Basin Rainforest on Rainfall Patterns. Report no. 12-2023, NORCE Climate and Environment (2023).
10.
Mishra, S. et al. Degradation of Southeast Asian tropical peatlands and integrated strategies for their better management and restoration. J. Appl. Ecol. 58, 1370–1387 (2021).
11.
Miettinen, J. et al. From carbon sink to carbon source: extensive peat oxidation in insular Southeast Asia since 1990. Environ. Res. Lett. 12, 024014 (2017).
12.
Smith, C. et al. Tropical deforestation causes large reductions in observed precipitation. Nature 615, 270–275 (2023).
13.
Dadap, N. C. et al. Climate change-induced peatland drying in Southeast Asia. Environ. Res. Lett. 17, 074026 (2022).
14.
Erfanian, A. et al. Unprecedented drought over tropical South America in 2016: significantly under-predicted by tropical SST. Sci. Rep. 7, 5811 (2017).
15.
Nobre, C. A. et al. Land-use and climate change risks in the Amazon and the need of a novel sustainable development paradigm. PNAS 113, 10759–10768 (2016).
16.
Kath, J. et al. Vapour pressure deficit determines critical thresholds for global coffee production under climate change. Nat. Food 3, 871–880 (2022).
17.
Xu, L. et al. Non-equilibrium early-warning signals for critical transitions in ecological systems. Proc. Natl Acad. Sci. USA 120, e2218663120 (2023).
18.
Staal, A. et al. Forest-rainfall cascades buffer against drought across the Amazon. Nat. Clim. Change 8, 539–543 (2018).
19.
Tang, K. H. D. Impacts of Climate Change on Tropical Rainforests' Adaptive Capacity and Ecological Plasticity. Clim. Change Facts, Impacts Solut. 1, 1–5 (2019).
20.
Cai et al. Climate impacts of the El Niño– Southern Oscillation on South America. Nature Reviews. Earth Environ. 1, 215 (2020).
21.
Zemp, D. C. et al. Self-amplified Amazon forest loss due to vegetation-atmosphere feedbacks. Nat. Commun. 8, 14681 (2017).
22.
Esquivel-Muelbert, A. et al. Compositional response of Amazon forests to climate change. Glob. Change Biol. 25, 39–56 (2019).
23.
Tao et al. Increasing and widespread vulnerability of intact tropical rainforests to repeated droughts. PNAS ecology sustainability. Science 119, e2116626119 (2022).
24.
Bennett et al. Resistance of African tropical forests to an extreme climate anomaly. PNAS Ecol. Sustainability Sci. 118, e2003169118 (2021).
25.
Clarke et al. Climate change, not El Niño, main driver of exceptional drought in highly vulnerable Amazon River Basin. World Weather Attribution Report (2024).
26.
Pan, X. et al. Connecting Indonesian fires and drought with the type of El Niño and phase of the Indian Ocean dipole during 1979–2016. J. Geophys. Res. Atmos. 123, 7974–7988 (2018).
27.
da Silva, S. S. et al. Dynamics of forest fires in the southwestern Amazon. Ecol. Manag 424, 312–322 (2018).
28.
Sullivan Martin, J. P. et al. Long-term thermal sensitivity of Earth’s tropical forests. Science 368, 869–874 (2020).
29.
Dwomoh, F. K. et al. Forest degradation promotes fire during drought in moist tropical forests of Ghana. Ecol. Manag. 440, 158–168 (2019).
30.
Sloan, S. et al. Fire activity in Borneo driven by industrial land conversion and drought during El Niño periods, 1982–2010. Glob. Environ. Change 47, 95–109 (2017).
31.
Rao, V. B. et al. In a changing climate Hadley cell induces a record flood in amazon and another recorded drought across South Brazil in 2021. Nat. Hazards 114, 1549–1561 (2022).
32.
Bochow, N. & Boers, N. The South American monsoon approaches a critical transition in response to deforestation. Sci. Adv. 9, eadd9973 (2023).
33.
Jung et al. Recent decline in the global land evapotranspiration trend due to limited moisture supply. Nature 467, 951–954 (2010).
34.
Baker, J. C. A. & Spracklen, D. V. Divergent representation of precipitation recycling in the Amazon and the Congo in CMIP6 models. Geophys. Res. Lett. 49, e2021GL095136 (2022).
35.
Liu, J. et al. Contrasting carbon cycle responses of the tropical continents to the 2015-2016 El Niño. Science 358, eaam5690 (2017).
36.
IPCC: Climate change 2021: the physical science basis. Contribution of working group I to the sixth assessment report of the inter-governmental panel on climate change. Cambridge University Press (2021).
37.
Schmied, H. M. et al. The global water resources and use model WaterGAP v2.2e: description and evaluation of modifications and new features. Geoscientific Model Development,
https://doi.org/10.5194/gmd-2023-213(2023).
38.
Eyring, V. et al. Overview of the coupled model intercomparison project phase 6 (CMIP6) experimental design and organization. Geosci. Model Dev. 9, 1937–1958 (2016).
39.
Melton, J. R. et al. CLASSIC v1.0: the open-source community successor to the Canadian Land Surface Scheme (CLASS) and the Canadian Terrestrial Ecosystem Model (CTEM) – Part 1: Model framework and site-level performance. Geoscientific Model Dev. 13, 2825–2850 (2020).
40.
Warszwaski, L. et al. The inter-sectoral impact model intercomparison project (ISI–MIP): project framework. PNAS 111, 3228–3232 (2013).
41.
de Burgh-Day, C. O. & Leeuwenburg, T. Machine learning for numerical weather and climate modelling: a review. Geosci. Model Dev. 16, 6433–6477,
https://doi.org/10.5194/gmd-16-6433-2023(2023).
42.
Gibson, P. B. et al. Training machine learning models on climate model output yields skillful interpretable seasonal precipitation forecasts. Commun. Earth Environ. 2, 159,
https://doi.org/10.1038/s43247-021-00225-4(2021).
43.
Shaw, T. A. & Voigt, A. Land dominates the regional response to CO2 direct radiative forcing. Geophys. Res. Lett. 43, 11383–11391 (2016).
44.
Grise, K. M. & Davis, S. M. Hadley cell expansion in CMIP6 models. Atmos. Chem. Phys. 20, 5249–5268 (2020).
45.
Douville, H. et al. Climate Change 2021: The Physical Science Basis, Contribution of Working Group I to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge: Cambridge University Press., 1055–1210 (2021).
46.
Bayr, T. et al. The eastward shift of the Walker circulation in response to global warming and its relationship to ENSO variability. Clim. Dyn. 43, 2747–2763 (2014).
47.
Zhang, L. & Li, T. Relative roles of differential SST warming, uniform SST warming and land surface warming in determining the Walker circulation changes under global warming. Clim. Dyn. 48, 1–11 (2016).
48.
Oki, T. & Kanae, S. Global hydrological cycles and world water resources. Science 313, 1068–1072 (2006).
49.
Jung, M. et al. Towards global empirical upscaling of FLUXNET eddy covariance observations: validation of a model tree ensemble approach using a biosphere model. Biogeosciences 6, 2001–2013 (2009).
50.
Grafton, R. Q. et al. The paradox of irrigation efficiency. Science 361, 748–750 (2018).
51.
Famiglietti, J. S. The global groundwater crisis. Geophys. Res. Lett. 38, L03403 (2011).
52.
Hubau, W. et al. Asynchronous carbon sink saturation in African and Amazonian tropical forests. Nature 579, 80–87 (2020).
53.
Caveleri et al. Tropical rainforest carbon sink declines during El Nino as a result ~ of reduced photosynthesis and increased respiration rates. N. Phytologist 216, 136–149 (2017).
54.
Lamberty, B. B. et al. Globally rising soil heterotrophic respiration over recent decades. Nature 560, 80–83 (2018).
55.
Nissan, A. et al. Global warming accelerate soil heterotrophic respiration. Nat. Comm. 14, 3452 (2023).
56.
Ogle, K. Hyperactive soil microbes might weaken the terrestrial carbon sink. Nature 560, 32–33 (2018).
57.
Lin, Y. S. Optimal stomatal behaviour around the world. Nat. Clim. Chang. 5, 459–464 (2015).
58.
Rammig, A. Tropical carbon sinks are out of sync. Nature 579, 38–39 (2020).
59.
Heinrich, V. H. A. et al. The carbon sink of secondary and degraded humid tropical forests. Nature 615, 436–442 (2023).
60.
Green, J. K. et al. Amazon rainforest photosynthesis increases in response to atmospheric dryness. Sci. Adv. 6, eabb7232 (2020).
61.
Paliwal, M. & Kumar, U. A. Assessing the contribution of variables in feed forward neural network. Appl. Soft Comput. 11, 3690–3696 (2011).
62.
Stachowicz, I. et al. Baseline for monitoring and habitat use of medium to large non-volant mammals in Gran Sabana, Venezuela. Therya 11,
https://doi.org/10.12933/therya-20-891(2020).
63.
Sitch, S. et al. Recent trends and drivers of regional sources and sinks of carbon dioxide. Biogeosciences 12, 653–679 (2015).
64.
Gaubert, B. et al. Global atmospheric CO2 inverse models converging on neutral tropical land exchange, but disagreeing on fossil fuel and atmospheric growth rate. Biogeosciences 16, 117–134 (2019).
65.
Kuzyakov, Y. V. & Larionova, A. A. Contribution of rhizomicrobial and root respiration to the CO2 emission from soil (a review). Eurasia. Soil Sci. 39, 753–764 (2006).
66.
Piao, S. L. et al. Evaluation of terrestrial carbon cycle models for their response to climate variability and to CO2 trends. Glob. Change Biol. 19, 2117–2132 (2013).
67.
Cheng, L. J. et al. New record ocean temperatures and related climate indicators in 2023. Adv. Atmos. Sci. 40, 963–974 (2024).
68.
Schneider, D. P. et al. Climate Data Guide Spurs Discovery and Understanding. Eos Trans. AGU 94, 121–122 (2013).
69.
Rzepecka, Z. & Birylo, M. Groundwater storage changes derived from GRACE and GLDAS on smaller river basins—a case study in Poland. Geosciences 10, 124 (2020).
70.
Hao, Z. et al. Global integrated drought monitoring and prediction system. Sci. Data 1, 140001 (2014).

Acknowledgements

The research work is supported by Yunnan University, China startup funding grant and National Natural Science Foundation of China (NSFC) project Grant No. 42250410320.

Ethics declarations

Competing interests

The authors declare no competing interests.

Additional information

Publisher’s note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Supplementary information

Rights and permissions

Open Access This article is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License, which permits any non-commercial use, sharing, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if you modified the licensed material. You do not have permission under this licence to share adapted material derived from this article or parts of it. The images or other third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by-nc-nd/4.0/.

About this article

Cite this article

Nath, D., Nath, R. & Chen, W. Faster dieback of rainforests altering tropical carbon sinks under climate change. npj Clim Atmos Sci 7, 235 (2024). https://doi.org/10.1038/s41612-024-00793-0

-
Received:
-
Accepted:
-
Published:
-
DOI: https://doi.org/10.1038/s41612-024-00793-0