A large amount of carbon is stored in global forests. However, the fraction of carbon stored as plant biomass versus soil organic carbon (SOC) varies among forest types, and potential changes over the 21st century are uncertain. Here, we used extensive data derived from inventories and remote sensing and Coupled Model Intercomparison Project Phase 6 (CMIP6) models to examine the current and 21st century dynamics in the proportion of biomass and SOC across global forests. We found that precipitation, elevation, soil, and wildfire were the primary controls of these differences in carbon pools. Under the SSP5–8.5 climate scenario, CMIP6 models project that the ratio of biomass to ecosystem carbon in global forests will increase across the 21st century, with the largest increases in boreal forests (95 ± 37%) compared to moist tropical forests (16 ± 15%). Changes in forest carbon pools resulting in greater biomass fraction will affect disturbance, and ecosystem carbon and energy balances, all of which interact with the climate system.

The North American (NA) terrestrial biosphere has been a long-term carbon sink but impacts of climate extremes such as drought on ecosystem carbon exchange remain largely uncertain. Here, changes in biospheric carbon fluxes with recent climate change and impacts of the major droughts of the past 30 years on continental carbon cycle across NA were studied using a comprehensive mathematical process model, ecosys. In test of these model responses at continental scale, the spatial anomalies in modeled leaf area indices, fully prognostic in the model, from long-term (1980–2010) means during major drought events in 1988 and 2002 agreed well with those in AVHRR NDVI (R2=0.84 in 1988, 0.71 in 2002). Net ecosystem productivity (NEP) modeled across NA declined by 92% (0.50PgC yr−1) and 90% (0.49PgC yr−1) from the long-term mean (+0.54PgC yr−1), in 1988 and 2002 respectively. These significant drops in NEP offset 28% of the carbon gains modeled over the last three decades. Although the long-term average modeled terrestrial carbon sink was estimated to offset ∼30% of the fossil fuel emissions of NA, only 0.03% and 3.2% were offset in 1988 and 2002 leaving almost all fossil fuel emissions to the atmosphere. These major drought events controlled much of the continental-scale interannual variability and mainly occurred in parts of the Great Plains, southwest US and northern Mexico. Although warming in northern ecosystems caused increasing carbon sinks to be modeled as a result of greater gross primary productivity with longer growing seasons, elsewhere in the continent frequent drought events of the past 30 years reduced carbon uptake and hence net carbon sinks of the NA.

An ecosystem model, ecosys, has been used to examine the effects of recent warming on carbon exchange in higher latitudes of North America. Model results indicated that gradual warming during the past 30 years has increased net ecosystem productivity (NEP) and leaf area index (LAI). Spring increases in LAI advanced by 2.3 days decade−1 and decreases in autumn were delayed by 5.0 days decade−1 from 1982 to 2006. These advances and delays were corroborated by similar trends observed in the normalized difference vegetation index. NEP modelled during this period increased at an average rate of 17.6 Tg C decade−1. Increasing carbon losses modelled with soil warming in autumn, when thaw depth was greatest, offset 34% of increasing carbon gains modelled in spring. If autumn warming continues, carbon losses in this season may further offset enhanced carbon sequestration in spring.

Ecosystem responses to the increasing warming in recent decades across North America (NA) are spatially heterogeneous and partly uncertain. Here we examined the spatial and temporal variability of warming across different eco-regions of NA using long-term (1979–2010) climate data (North America Regional Reanalysis (NARR)) with 3-hourly time-step and 0.25°×0.25° spatial resolution and run a comprehensive mathematical process model, ecosys to study the impacts of this variability in warming on gross primary productivity (GPP). In a site scale test of model results, annual GPP modeled for pixels which corresponded to the locations of 20 eddy covariance flux towers correlated well (R2=0.76) with annual GPP derived from the towers in 2005. At continental scale, long-term annual average modeled GPP correlated well (geographically weighed regression R2=0.8) with MODIS GPP. GPP modeled in eastern temperate forests and most areas with lower mean annual air temperature (Ta), such as those in northern forests and Taiga, increased due to early spring and late autumn warming observed in NARR and these eco-regions contributed 92% of the increases in NA GPP over the last three decades. However, modeled GPP declined in most southwestern regions of NA (accounting >50% of the ecosystems with declining GPP), due to water stress from rising Ta and declining precipitation, implying that further warming and projected dryness in this region could further reduce NA carbon uptake. Overall, NA modeled GPP increased by 5.8% in the last 30 years, with a positive trend of +0.012±0.01PgCyr−1 and a range of −1.16 to +0.87PgCyr−1 caused by interannual variability of GPP from the long-term (1980–2010) mean. This variability was the greatest in southwest of US and part of the Great Plains, which could be as a result of frequent El Niño–Southern Oscillation’ events that led to major droughts.

The quality of gridded weather and soil datasets is one of the main sources of uncertainty in modeling regional scale land-atmosphere carbon exchange. This uncertainty may be assessed by comparing net ecosystem productivity (NEP) modeled for selected grid cells with these datasets against NEP modeled at representative sites within that grid cell for which detailed site scale measurements of weather and soil are available. In this study, a comprehensive mathematical ecosystem model, ecosys, was used to simulate differences in NEP at six eddy covariance (EC) flux tower sites across North America caused by using inputs for weather and soil extracted from 0.25°×0.25° gridded datasets for the grid cells within which the towers were located vs. those measured at each tower site during years with contrasting weather (cool vs. warm and wet vs. dry). NEP differences attributed to gridded vs. measured model inputs varied among sites when tested against EC-derived values. At some sites (e.g. coastal coniferous site in British Columbia), adverse impacts of summer warming on NEP under contrasting weather in 2001 (cooler) vs. 2004 (warmer) were accurately modeled with both inputs, giving R2 with EC NEP values >0.80. At other sites, reduced accuracy in NEP modeled with gridded vs. measured inputs was attributed to shallower soil depth in the gridded soil database. This shallower depth caused site NEP in a boreal deciduous stand in Saskatchewan not to be well simulated in 2001 and 2003, the first and third years of a major drought in central North America, due to early soil drying. At yet other sites, reduced accuracy was attributed to overestimates of soil organic nitrogen (SON) calculated from general soil organic carbon (SOC):SON relationships used in the absence of SON values in the gridded soil database. These overestimates caused excessive productivity to be modeled with gridded inputs at a boreal black spruce site in Quebec with unusually large SOC:SON. Gridded weather and soil inputs that caused such differences in NEP would certainly affect regional and global carbon budget model estimates, and so need to be improved in future large scale model input datasets.

Recent changes in species composition, and increases in shrub abundance in particular, have been reported as a result of warming in Arctic tundra. Despite these changes, the driving factors that control shrubification and its future trajectory remain uncertain. Here we used an ecosystem model, ecosys, to mechanistically represent the processes controlling recent and 21st century changes in plant functional type using RCP8.5 climate forcing across North American Arctic tundra. Recent and projected warming was modeled to deepen the active layer (spatially averaged by ~0.35 m by 2100) and thereby increase nutrient availability. Shrub productivity was modeled to increase across much of the tundra, particularly in Alaska and tundra-boreal ecotones. Deciduous and evergreen shrubs increased from ~45% of total tundra ecosystem net primary productivity (NPP) in recent decades to ~70% by 2100. The increased canopy cover of shrubs reduced incoming shortwave radiation for low-lying plants, causing declines in graminoids NPP from a current 35% of tundra NPP to 18%, and declines in nonvascular plants from 20% to 12%. The faster-growing deciduous shrubs modeled with less efficient nutrient conservation dominated much of the low Arctic by 2100 where nutrient cycling became more rapid, while the slower-growing evergreen shrubs modeled with more efficient nutrient conservation dominated a wider latitudinal range that extended to the high Arctic where nutrient cycling remained slower. We conclude that high-latitude vegetation dynamics over the 21st century will depend strongly on soil nutrient dynamics, diversity in plant traits controlling nutrient uptake and conservation, and light competition.

Recent observed shifts in Arctic tundra shrub cover have uncertain impacts on 21st century net ecosystem carbon exchanges. Here we applied a well-tested ecosystem model, ecosys, to examine the effects of North America Arctic tundra plant dynamics on ecosystem carbon balances from 1980-2100 under the RCP8.5 scenario. Tundra productivity was modeled to increase from enhanced carbon fixation and N mineralization under recent and future climates. Between 1982 and 2100 and averaged across the region, predicted increases in relative dominance of woody versus non-woody plants increased ecosystem annual net primary productivity by 244 g C m-2 that offset concurrent increases in annual heterotrophic respiration (139 g C m-2), resulting in an increasing net carbon sink over the 21st century. However, smaller increases in seasonal carbon uptake during winter (1 g C m-2) and autumn (22 g C m-2) and greater increases in ecosystem respiration (winter (23 g C m-2) and autumn (47 g C m-2)) by 2100 versus 1982 resulted in larger carbon losses during these seasons that completely offset the gains in spring (13 g C m-2) and 25% of the gains in summer (140 g C m-2). Modeled soil temperatures were predicted to increase more slowly than air temperatures (∼0.6 °C for every 1 °C increase in air temperature over the 21st century). This slower soil versus air warming, and thus greater increases in CO2 fixation versus soil respiration rates, also contributed to the tundra remaining a carbon sink through 2100. However, these higher gains versus losses of carbon may be a transient response and not sustainable under further soil warming beyond 2100. Our modeling analysis allows us to extend beyond results from short-term warming experiments, which cannot characterize effects associated with decadal-scale changes in plant communities.

Seasonal variations in high-latitude terrestrial carbon (C) fluxes are predominantly driven by air temperature and radiation. At present, high-latitude net C uptake is largest during the summer. Recent observations and modeling studies have demonstrated that ongoing and projected climate change will increase plant productivity, microbial respiration, and growing season lengths at high-latitudes, but impacts on high-latitude C cycle seasonality (and potential feedbacks to the climate system) remain uncertain. Here we use ecosys, a well-tested and process-rich mechanistic ecosystem model that we evaluate further in this study, to explore how climate warming under an RCP8.5 scenario will shift C cycle seasonality in Alaska throughout the 21st century. The model successfully reproduced recently reported large high-latitude C losses during the fall and winter and yet still predicts a high-latitude C sink, pointing to a resolution of the current conflict between process-model and observation-based estimates of high-latitude C balance. We find that warming will result in surprisingly large changes in net ecosystem exchange (NEE; defined as negative for uptake) seasonality, with spring net C uptake overtaking summer net C uptake by year 2100. This shift is driven by a factor of 3 relaxation of spring temperature limitation to plant productivity that results in earlier C uptake and a corresponding increase in magnitude of spring NEE from -19 to -144 gC m-2 season-1 by the end of the century. Although a similar relaxation of temperature limitation will occur in the fall, radiation limitation during those months will limit increases in C fixation. Additionally, warmer soil temperatures and increased carbon inputs from plants lead to combined fall and winter C losses (163 gC m-2) that are larger than summer net uptake (123 gC m-2 season-1) by year 2100. However, this increase in microbial activity leads to more rapid N cycling and increased plant N uptake during the fall and winter months that supports large increases in spring NPP. Due to the large increases in spring net C uptake, the high-latitude atmospheric C sink is projected to sustain throughout this century. Our analysis disentangles the effects of key environmental drivers of high-latitude seasonal C balances as climate changes over the 21st century.

Plant growth and distribution in high-latitude tundra ecosystems is strongly limited by nutrient availability and is critical for quantifying centennial-scale carbon-climate interactions. However, land model representations of plant-nutrient interactions are uncertain, leading to poor comparisons with high-latitude observations. Although it has been recognized for decades in the observational community that plants continue to acquire nutrients well past when aboveground activity has ceased, most large-scale land models ignore this process. Here we address the role tundra plant nutrient acquisition during the non-growing season (NGS) has on centennial-scale vegetation growth and dynamics, with a focus on shrub expansion. We apply a well-tested mechanistic model of coupled plant, microbial, hydrological, and thermal dynamics that explicitly represents nutrient acquisition based on plant and microbial traits, thereby allowing a prognostic assessment of NGS nutrient uptake. We first show that the model accurately represents observed seasonality of NGS plant nutrient uptake in a northern Alaskan tundra site. Applying the model across the North America tundra indicates that NGS nutrient uptake is consistent with observations and ranges between ∼5% and 50% of annual uptake, with large spatial variability and dependence on plant functional type. We show that NGS plant nutrient acquisition strongly enhances modeled 21st century tundra shrub growth and expansion rates. Our results suggest that without NGS nutrient uptake, total shrub aboveground dominance would be ∼50% lower, limited primarily by their inability to grow tall enough to maximize their inherent capacity for light competition. Evergreen shrubs would be more strongly affected because of their relatively lower capacity for nutrient remobilization and acquisition compared to deciduous shrubs. Our results highlight the importance of NGS plant and soil processes on high-latitude biogeochemistry and vegetation dynamics and motivate new observations and model structures to represent these dynamics.

The high-latitude carbon (C) cycle is a key feedback to the global climate system, yet because of system complexity and data limitations, there is currently disagreement over whether the region is a source or sink of C. Recent advances in big data analytics and computing power have popularized the use of machine learning (ML) algorithms to upscale site measurements of ecosystem processes, and in some cases forecast the response of these processes to climate change. Due to data limitations, however, ML model predictions of these processes are almost never validated with independent datasets. To better understand and characterize the limitations of these methods, we develop an approach to independently evaluate ML upscaling and forecasting. We mimic data-driven upscaling and forecasting efforts by applying ML algorithms to different subsets of regional process-model simulation gridcells, and then test ML performance using the remaining gridcells. In this study, we simulate C fluxes and environmental data across Alaska using ecosys, a process-rich terrestrial ecosystem model, and then apply boosted regression tree ML algorithms to training data configurations that mirror and expand upon existing AmeriFLUX eddy-covariance data availability. We first show that a ML model trained using ecosys outputs from currently-available Alaska AmeriFLUX sites incorrectly predicts that Alaska is presently a modeled net C source. Increased spatial coverage of the training dataset improves ML predictions, halving the bias when 240 modeled sites are used instead of 15. However, even this more accurate ML model incorrectly predicts Alaska C fluxes under 21st century climate change because of changes in atmospheric CO2, litter inputs, and vegetation composition that have impacts on C fluxes which cannot be inferred from the training data. Our results provide key insights to future C flux upscaling efforts and expose the potential for inaccurate ML upscaling and forecasting of high-latitude C cycle dynamics.