Climate change accelerates global forest deadwood dynamics

Forests play a crucial role in the global carbon cycle, with carbon sequestration by live trees being recognized as one of the primary sinks for atmospheric carbon^1,2,3. Large quantities of carbon are stored in live trees, but also dead matter such as deadwood, litter, and soil organic material are important stocks in the forest carbon cycle³. In order to quantify the role of forests in the global climate system and to project potential future trajectories of the climate-regulating function of forests, the interplay between all of the components of the forest carbon cycle needs to be considered^4,5. However, while some of these components like live tree carbon are increasingly well understood¹, major knowledge gaps exist with regard to the role of deadwood⁶.

Globally, approximately 73 ± 6 Pg C are estimated to be currently stored in deadwood, representing 8% of the total carbon stored in forests³. Forest deadwood stocks are the result of two competing processes: deadwood formation—determined by tree mortality and the carbon stored in the woody compartments of trees when they die—and deadwood decomposition, i.e. the carbon released back to the atmosphere by heterotrophic organisms. Both of these processes are highly sensitive to changes in the climate system. Climate change is already accelerating forest growth dynamics via rising temperatures and extended growing seasons⁷. Simultaneously, tree mortality has been increasing in many regions of the globe, because of disturbances by agents such as fire, drought, wind, insects, and pathogens⁸, but also in the absence of major disturbance events^9,10. These ongoing changes are expected to increase the input of carbon into the forest deadwood pool in the future. At the same time, deadwood decomposition rates are also influenced by climate, with higher rates of decomposition at high temperatures⁶. Furthermore, a climate-induced change in tree species composition (e.g., from gymnosperms to angiosperms) could further increase or decrease the rate of deadwood decomposition^11,12. Consequently, the net change of global deadwood stocks in the face of climate change is uncertain; whether increasing deadwood formation will outweigh increasing deadwood decomposition or vice versa remains unresolved to date.

Dynamic Global Vegetation Models (DGVMs) are important tools for making inferences on the future carbon dynamics of global vegetation. These models have been successfully applied to simulate global forest vegetation dynamics and carbon uptake^{13,14,15,16,17}. They are important components in earth system modeling, and have contributed greatly to providing the foundation for global policy documents on climate change mitigation¹⁸. While all of the currently applied DGVMs are process-based models (i.e., representing global vegetation by simulating fundamental ecological processes, such as carbon uptake from photosynthesis, allocation of carbon to vegetation compartments, etc.), some processes are captured in more detail than others^19,20. Substantial progress has been made to improve DGVM process formulations²¹, e.g. with regard to the demography of forest ecosystems²² or the simulation of disturbance processes²³. However, deadwood dynamics have not been a focus of DGVM simulations to date, not least because of a lack of globally consistent data on deadwood stocks and decomposition rates. Recent advances in experimentally quantifying deadwood decomposition at the global scale⁶ now offer new avenues for understanding the future of global forest deadwood carbon by combining the strengths of DGVMs in estimating global deadwood formation with experimentally derived functions of global deadwood decomposition.

Here, our objective was to assess potential changes in global deadwood stocks under climate change by contrasting potential changes in deadwood formation and decomposition. Specifically, we investigated whether deadwood formation or decomposition responds more strongly to climate warming and thus, whether global deadwood stocks are likely to increase or decrease in the 21st century, assuming a continuation of current land-use practices (i.e., under the ceteris paribus condition of constant rates of wood extraction and salvage logging of dead and dying trees). Moreover, we quantified uncertainties related to climate pathways and model formulations by studying the effect of different models and climate scenarios on future global deadwood dynamics. We note that we here focus on potential climate-induced changes in the relationship between deadwood formation and decomposition for current forests; it was not our goal to make explicit projections of future global deadwood carbon, for which also deadwood extraction by humans, combustion of deadwood by wildfire as well as land-use change would have to be accounted for.

To address our objectives we studied two ensembles of future projections of deadwood formation, whereof one contrasted five different DGVMs (LPJ-GUESS, ORCHIDEE, CABLE-POP, SEIB-DGVM and LPJmL) under the climate scenario RCP8.5 (henceforth referred to as the model ensemble), and the second compared 11 different climate projections (spanning four shared socioeconomic pathways from SSP126 to SSP585) as simulated with the DGVM LPJ-GUESS (the climate ensemble). Deadwood decomposition was estimated based on an experimentally derived statistical model of global wood decomposition, considering tree phylogenetic lineage (angiosperm or gymnosperm), precipitation, and temperature as driver variables⁶. We started our analysis from a map of global deadwood stocks in 2010⁶, and modeled changes in future deadwood stocks as the net result of deadwood formation (estimated from DGVMs) and deadwood decomposition (from the statistical model) at annual time steps. As we were mainly interested in the changes in the input/output ratio to global deadwood pools, and as we here combine different modeling approaches, we standardized all fluxes to the reference period 2000–2020 and calculated changes relative to this reference period. All analyses were conducted globally at a spatial resolution of 0.5°, spanning the time period from 2000 to 2099.

Future global deadwood dynamics

Across all models and climate scenarios, 21st-century deadwood formation exceeded deadwood decomposition by 5.0% (Fig.1). Both deadwood formation and decomposition increased in the 21st century compared to current values, with stronger increases in deadwood formation (median: 15.2%, range across all models and climate scenarios between 2.1% and 26.0%) than in decomposition (median: 8.4%, range between 1.5% and 19.0%). Climate change intensified both fluxes, with increasing carbon inputs to and outputs from global deadwood pools with increasing severity of climate change. Fluxes in SSP585 were 1.13 times higher than under SSP126 for both deadwood formation and decomposition. The uncertainty from different DGVMs and the resultant differences in deadwood formation are higher than the uncertainties in climate (Supplementary Fig. 1), with average deadwood formation fluxes spanning a range from 11.1 to 13.5 Pg C yr⁻¹, while deadwood formation in the climate ensemble ranged between 8.8 and 13.7 Pg C yr⁻¹.

Throughout the 21st century, both deadwood formation and decomposition fluxes followed an upward trend (Fig. 2), resulting in a relatively constant input-to-output (I/O) ratio over time. Deadwood formation increased from current values of 10.3 Pg C yr⁻¹ to 14.2 Pg C yr⁻¹ in the period 2080–2099 (mean across all simulations). Simultaneously, deadwood decomposition increased from 9.8 Pg C yr⁻¹ to 13.7 Pg C yr⁻¹. In addition to the main fluxes, their uncertainty (resulting from future climate uncertainty as well as model uncertainty) increased over time, with coefficient of variation (CV) values of 4.8%−18.7% for deadwood formation and 4.4%−14.6% for deadwood decomposition at the end of the century. This is also reflected in an increasing uncertainty in the I/O ratio over time (Fig. 2). In fact, 21.4% of all simulations (across both climate and model ensembles) had an I/O ratio of below 1.0 by the end of the century (2080–2099), suggesting that in roughly one out of five simulated trajectories, global deadwood pools are decreasing. When models were considered with equal weights in the analysis (i.e., model ensemble only), I/O ratios were still positive yet smaller than with the full dataset, and values approached 1.0 towards the end of the century (Supplementary Figs. 3 and 4).

The mean 21st century I/O-ratio was positive for all biomes, indicating higher deadwood formation than decomposition across the globe (Fig. 3). The arctic, boreal and temperate biomes showed the highest relative differences between deadwood formation and decomposition rates, with average I/O-ratios throughout the century of 1.69, 1.39 and 1.2. In tropical forests, on the other hand, deadwood formation only slightly exceeded decomposition, resulting in an I/O ratio of 1.05. However, the variation in I/O ratios was also high within biomes. While in coastal areas of temperate North America, deadwood decomposition increased more strongly than deadwood formation (I/O ratio < 1, indicating decreasing deadwood stocks), forests in the Rocky Mountains had I/O ratios > 1, suggesting potential increases in forest deadwood.

Climate sensitivity and uncertainty

Across all simulations, the deadwood I/O ratio was not sensitive to the magnitude of temperature increase. In other words, deadwood formation and decomposition responded similarly to climate warming, with the effects canceling each other out (Fig. 4a). However, upon closer examination, we found that the climate sensitivity of the deadwood I/O ratio was strongly contingent on the DGVM used in the analysis. LPJ-GUESS (Fig. 4b), for instance, projected an increase in the deadwood I/O ratio with warming, with deadwood formation responding more strongly positively to an increase in temperature compared to decomposition. In LPJ-GUESS, the I/O ratio increases by 0.007 per degree of global warming. In contrast, the models LPJmL, ORCHIDEE, CABLE-POP and SEIB (Fig. 4c–f) projected a decreasing I/O ratio with warming, with deadwood decomposition responding more strongly positively to increasing temperatures than tree growth and mortality. For these models, the I/O ratio decreased by between 0.002 and 0.017 per degree of warming. These findings illustrate the considerable uncertainties that remain regarding future global deadwood carbon dynamics. In general, the uncertainty introduced by differences between DGVMs was in the same order of magnitude as the uncertainty introduced by different SSPs and climate scenarios until the end of the 21st century (cf. Supplementary Fig. 1).

Here, we showed that both global deadwood formation and decomposition are likely to increase in the future. Forest productivity has increased in the past⁷, and is expected to increase further in the coming decades, especially in northern regions^24,25,26. Furthermore, tree mortality is one of the most climate-sensitive processes in forest ecosystems²⁷, and mortality rates are already increasing in many parts of the world^28,29. These trends will ultimately lead to the formation of more deadwood. At the same time, decomposing communities consisting of fungi and insects are highly sensitive to changes in the climate system^6,30,31,32 with warmer temperatures increasing their activity, and thus deadwood decomposition. It is important to note that not the full amount of carbon released from deadwood in the course of decomposition is going back to the atmosphere; a significant portion is also sequestered by organisms and soil organic matter and hence remains in the system^33,34. While our results suggest a potential increase in global forest deadwood stocks, the implications of elevated deadwood formation and decomposition could go considerably beyond the analyses presented here, affecting forest carbon and nutrient cycling more broadly^35,36,37,38.

We found high spatial variability in the response of deadwood dynamics to global warming. Specifically, our results suggest increases in deadwood stocks in temperate and boreal forests due to their considerably higher deadwood I/O ratio than tropical forests. This indicates that the current imbalance of the global deadwood distribution – with 65% of the global deadwood carbon being found in the tropics⁶ – could develop towards a more equal distribution of deadwood across biomes and latitudes with ongoing climate change. More broadly, our results suggest that the importance of deadwood for global terrestrial carbon storage will increase in the future. While deadwood contributes 8% to the carbon stored in forest ecosystems currently³, this value could increase in the future, as deadwood increases (indicated by I/O ratios > 1) and live tree carbon decreases (as a result of increasing tree mortality^16,39).

Model uncertainty was considerable in our analysis of future deadwood dynamics. The uncertainty of forest mortality processes is well recognized^40,41,42,43 and underlines the complexity of predicting ecosystem dynamics. However, the potential for increased deadwood stocks, as identified in our study, may offset some of the anticipated future net carbon losses in forests^18,28,44,45. This highlights the importance of accurately modeling deadwood decomposition in ecosystem models to capture its offsetting effect on total ecosystem carbon fluxes. Additionally, model assumptions and structural differences regarding representation of plant-functional types, their distributions, and controls are likely contributing to uncertainty⁴⁶. The model CABLE-POP, for instance, does not incorporate changes in plant-functional types with climate, while other models like LPJ-GUESS do consider these dynamic changes, which influence the simulated deadwood formation as well as its estimated decomposition rates. Furthermore, not all models are equally able to simulate climate-driven mortality pulses, given that disturbance modeling is currently an active field of research in the community^21,23. A further uncertainty pertains to acclimation processes, which are currently not considered in the suite of DGVMs applied here. CO2 fertilization effects that are strongly driving the increase in productivity (and subsequently deadwood formation) in current simulations might subside in the future, as forests adapt to higher levels of CO2 in the atmosphere^47,48. But also, statistical models like the one used here to estimate deadwood decomposition rates are associated with uncertainty. Our approach, for instance, ignores that decomposer communities might change dynamically over time due to, e.g., changes in climate^32,49, tree species composition⁵⁰ or increasing deadwood stocks^51,52 as well as interactions between resource availability and climate⁵³. More broadly, statistical models assume that relationships estimated in the past are also applicable in the future, which might not be the case in a world that is strongly changing⁵⁴. Additionally, our decomposition model does not account for the impact of wildfires on deadwood decay. Wildfires are a major source of carbon release from forest ecosystems^55,56, and also modify deadwood decay processes. While we acknowledge the complexity of the interactions between wildfires and deadwood decay, we assume that the direct effects on decay rates are relatively minor⁵⁷. Yet, as fires are expected to increase under climate change^58,59, the effects of wildfire on deadwood via combustion and altered decomposition will likely increase in importance in the future.

In this study, we were interested in how climate change affects deadwood formation and decomposition, yet human land use—assumed to be time-invariant here—is one of the main drivers of global deadwood stocks³. In fact, analyses of recent changes in global deadwood report moderate decreases since 1990, as increases in the boreal and temperate biome were overcompensated by decreases in the tropics, resulting from land-use change⁶⁰. Changes in human land use could also drastically alter future global deadwood stocks: An increase in protected areas to 30% of the global land surface, as envisioned by the Kunming-Montreal Global Biodiversity Framework⁶¹, could distinctly increase the amount of deadwood retained in ecosystems. Conversely, a further increase in timber harvesting²⁴ in response to the resource needs of a growing global population could reduce global deadwood stocks. Similarly, forest policies recommend changes in tree species composition in managed forests to adapt to a changing climate⁶², which will influence deadwood formation and decomposition. While the overall state of global deadwood is thus strongly contingent on future land-use decisions, the climate-driven processes analyzed here will act to further modulate the outcomes of human land use.

Our finding that climate change could lead to an increase in forest deadwood stocks has important implications for forest functions and services. Deadwood is a long-lived carbon compartment in forests³⁵, and an increasing deadwood carbon pool can thus contribute to increasing the carbon density in forests, supporting their climate-regulating function³. Furthermore, deadwood also affects microclimatic conditions by retaining moisture³⁷, and can protect against gravitational natural hazards such as rockfall⁶³. Importantly, deadwood is habitat to about 25% of all forest species and thus crucial for biodiversity conservation^64,65. Our results suggest that deadwood-dependent insects⁵³ —many of which are currently threatened with extinction^52,66 —could benefit from the combined effects of increasing temperature and deadwood availability in the future. We conclude that climate change will profoundly accelerate deadwood dynamics in the forests of the world. As deadwood is at the heart of the nexus of forest carbon and biodiversity, and thus a key element for tackling the ongoing climate and biodiversity crises, deadwood should receive increasing attention in forest policy and management.

To assess future deadwood dynamics at the global scale we combined three elements: 1.) Estimates of future deadwood formation under climate change, derived from five dynamic global vegetation models, 2.) an analysis on climate-mediated deadwood decomposition, based on empirical relationships derived from a global experiment of deadwood decomposition, and 3.) a map of current global deadwood stocks, generated by combining inventory data with remote sensing information, serving as starting point for the analysis. In the following, these three elements are described in more detail.

Deadwood formation

To estimate the annual rate of deadwood formation, i.e., the carbon input to the deadwood pool, we used Dynamic Global Vegetation Models (DGVM). Simulations were performed on a spatial resolution of 0.5° grid cells. We analyzed two ensembles of deadwood formation: One designed to quantify model uncertainty, contrasting five different DGVMs under one common climate scenario (referred to as model ensemble), and the second assessing the climate sensitivity of deadwood formation by comparing projections of a single model across 11 different climate scenarios (climate ensemble). For the model ensemble, fluxes of woody carbon resulting from tree mortality and thus representing deadwood formation were obtained from the DGVMs LPJ-GUESS⁶⁷, LPJmL⁶⁸, SEIB-DGVM⁶⁹, ORCHIDEE⁷⁰, and CABLE-POP⁷¹. Previous model runs for the IPCC-RCP 8.5 climate scenario⁷² for these DGVMs were available from Pugh et al. (2020)⁷³. The time series for this analysis spanned the period 2000 to 2099. Non-forest areas such as shrubs and grasslands were excluded, focusing exclusively on deadwood dynamics in forests. With the exception of CABLE-POP, all models simulated dynamic changes in the cover and occurrence of plant-functional types in response to climate. Simulated information on the emergent global state of forests (e.g., the distribution of angiosperms and gymnosperms) was used as input for the calculation of deadwood decomposition (see below).

For the climate ensemble, new LPJ-GUESS simulations were conducted with climate forcing derived from the Earth System Models (ESMs) EC-Earth3-Veg⁷⁴, MRI-ESM2-0⁷⁵, and NorESM2-LM⁷⁶ for historical climate and the SSP scenarios 126, 245, 370, and 585⁷⁷. All ESMs shared the same initial (historical) values for all SSPs from 2000 to 2014. The resultant 11 simulations represent different climate futures for the Earth, and the respective simulated forest conditions were again used to drive deadwood decomposition. For the climate ensemble, three global climate models that represented a wide spread of the likely climate sensitivity range were selected. For this study, the climate data was generated to explore a broad spectrum of potential climate change scenarios. To achieve this, three Global Circulation Models (GCMs) were carefully selected to represent a diverse range within the likely climate sensitivity spectrum. Additionally, the above-mentioned SSPs (126, 245, 370, and 585) were chosen to capture a wide range of plausible future trajectories. The selection of carbon dioxide and nitrogen deposition was aligned with each RCP scenario. Soil data utilized in this study were sourced from the WISE dataset by ref. ⁷⁸. The climate data generated was on a daily scale and underwent bias adjustment using the methodology mentioned above. The key climate variables included in the dataset are temperature (minimum, maximum, and mean), total precipitation, net shortwave radiation, wind, and relative humidity. These variables are essential inputs for the LPJ-GUESS model. GCM data was bias-corrected with the ISIMIP approach⁷⁹ against the WFDEI dataset⁸⁰.

Deadwood decomposition

To calculate the annual output from the deadwood pool, a generalized additive model (GAM) was used. The model was based on a global decomposition experiment, spanning 55 sites distributed across six continents and representing the global distribution of forest ecosystems in climate space well⁶ (see also Supplementary Fig. 2). The GAM used tree functional type (Angiosperm, Gymnosperm), mean annual temperature (in °C), and mean annual precipitation (in dm) as independent variables to determine the output fraction (k) of deadwood for each 0.5° grid cell. The model structure is as described in Seibold et al.⁶, including the effect of both insect and fungal decomposer communities. The climate data used to drive the GAM was the same as the one used in DGVM simulations, i.e. deadwood formation and decomposition analyses used consistent climate data. Tree functional type was directly determined from DGVM output and thus changed dynamically in time. We here decided to use an empirical function to determine deadwood decomposition rather than using the fluxes simulated by DGVMs. Our rationale was that the representation of deadwood processes in models varies strongly, contingent on how trees are represented in models (individuals, cohorts, or carbon pools without structure). Furthermore, deadwood processes have received considerably less attention in the global modeling community compared to processes of tree growth and mortality^17,22, and the climate sensitivity of deadwood processes (a central element of our analysis here) remains underexplored for many models. Hence, we assumed that the most consistent estimate of climate-dependent deadwood decomposition currently available comes from global experimental data⁶, as captured in the statistical model used here. In the decomposition model, it was assumed that decomposition rates for standing deadwood vs. downed deadwood in temperate and boreal forests would be reduced by 50%, while in tropical forests, no reduction was applied. Based on extensive inventory data, it was estimated that standing deadwood constitutes approximately 25% of total deadwood in boreal forests and 30% in temperate forests⁶.

Initial deadwood stocks

The initial global deadwood carbon stock was taken from Seibold et al. (2021), who combined country-level information derived from forest inventories³ with remote sensing-based information on the distribution of forest biomass⁸¹ to generate a forest deadwood map for the year 2010 at 5 arcminute resolution. The estimated global deadwood stock was 86.0 Pg C, considering all woody materials above 2 cm in diameter. For the analysis here, data were aggregated to the common 0.5° resolution of DGVM model simulations.

Analyses

Current global deadwood pools strongly reflect the prevailing stand structure as well as the management and disturbance history of the recent past. The DGVMs used here, however, simulate potential vegetation development⁴⁶. It was thus necessary to scale potential deadwood formation rates from DGVMs to current levels, accounting for wood extraction rates and other land-use effects. Erb et al.⁸² showed that the rate between potential and actual carbon stocks is relatively conservative across a wide range of different forests. We thus scaled DGVM-derived deadwood formation rates (potential vegetation) to current values to match them with current deadwood stocks and decomposition estimates. Specifically, we derived scaling factors SF for each tree functional type (i.e., angiosperm or gymnosperm) tft, DGVM m, and biome b, equating DGVM-derived deadwood formation rates (DF) with GAM-estimated decomposition rates (DD) for the period 2000–2020 (Eq. 1 and Supplementary Table 1).

with DD calculated from average temperature ((hat{T})) and precipitation ((hat{P})) over the period 2000–2020, and n number of years within the period. These scaling factors were subsequently applied to scale DGVM-derived deadwood input values for the full simulation period. The factors contain all effects of land use, but also ensure that all models are standardized to the same starting point, controlling for initial differences in DGVMs. In other words, we set the I/O ratio for all models and scenarios to 1.0 for the reference period 2000–2020, and only studied relative changes in inputs and outputs to deadwood pools until the end of the 21st century. This approach of standardization (at the level of individual models) thus allows us to consistently analyze the deadwood responses to climate over the 21st century, while factoring out initial differences between models. We note, however, that structural legacy effects (such as differences in forest age distribution and their implications on deadwood formation) are not considered in this approach. Rather, our analysis solely focuses on the effects of climate change on deadwood formation and decomposition. Furthermore, in cases where the formation value for a given biome and tree functional type was equal to zero or not existent during the 2000–2020 time period, the corresponding carbon stock was adjusted to 0 (see Supplementary Table 2).

The main analysis variable was the deadwood input/output (I/O) ratio, contrasting the input to the deadwood pool (deadwood formation, from scaled DGVM results) to the output via decomposition (from the statistical model). The I/O ratio gives an indication of potential changes in deadwood stocks in response to climate, with values > 1 indicating climate-induced increases in deadwood stocks, and values < 1 indicating climate-induced losses in deadwood. We assessed the deadwood I/O ratio over time (using 20-year bins) as well as in space to glean insights into the spatio-temporal patterns of future deadwood dynamics. We furthermore analyzed deadwood I/O ratios within climate and model ensembles to better understand uncertainties in projected future deadwood dynamics. In addition to reporting results across both ensembles (i.e., across all available climate scenarios and model runs, but with models not equally weighted) we also analyzed results for scenario RCP 8.5 only, giving equal weight to all five DGVMs (model ensemble, see Supplementary Figs. 3 and 4).

All analyses were limited to the common surface area of forests in 2010, the available climate data, and the initial carbon fluxes for each RCP and SSP climate scenario. This spatial restriction was consistently applied to all years, ensuring comparability of deadwood stock dynamics within the specified area of analysis. Additionally, an ESI CCA land cover (version 2.0.7, Defourny, 2017) based forest mask with a threshold of 1% minimum forest area within a 0.5° grid cell was applied. Ecozones were separated according to FAO (2012) and binned into the arctic, boreal, temperate, subtropical, and tropical zones following Seibold et al. (2021)⁸³.