Respiratory loss during late-growing season determines the net carbon dioxide sink in northern permafrost regions

We focused on the Northern High Latitudes (NHL, latitude > 50°N, excluding Greenland) due to their importance for carbon (CO₂-C, the same hereafter)-climate feedbacks in the Earth system. To minimize the potential human influence on the CO₂ cycle, we excluded areas under agricultural management (croplands, cropland/natural vegetation mosaic, and urban types), and considered only pixels of natural vegetation defined from the MODIS MCD12Q1 (v006) based IGBP land cover classification. Our main focus was the NHL permafrost region because permafrost plays a critical role in the ecology, environment, and society in the NHL. Permafrost, or permanently frozen ground, is defined as ground (soil, sediment, or rock) that remains at or below 0 °C for at least two consecutive years. The occurrence of permafrost is primarily controlled by temperature and has a strong effect on hydrology, soils, and vegetation composition and structure. Based on the categorical permafrost map from the International Permafrost Association⁵⁸, the permafrost region (excluding permanent snow/ice and barren land), including sporadic (10–50%), discontinuous (50–90%), and continuous (>90%) permafrost, encompasses about 15.7 × 10⁶ km², accounts for 57% of the NHL study dominion, and is dominated by tundra (shrubland and grass) and deciduous needleleaf (i.e., larch) forest that is regionally abundant in Siberia. The NHL non-permafrost region covers about 11.9 × 10⁶ km² and is dominated by mixed and evergreen needleleaf boreal forests (Fig. S1).

Atmospheric CO₂ inversions (ACIs)

ACIs provide regionally-integrated estimates of surface-to-atmosphere net ecosystem CO₂ exchange (NEE_ACI) fluxes by utilizing atmospheric CO₂ concentration measurements and atmospheric transport models⁵⁹. ACIs differ from each other mainly in their underlying atmospheric observations, transport models, spatial and temporal flux resolutions, land surface models used to predict prior fluxes, observation uncertainty and prior error assignment, and inversion methods. We used an ensemble mean of six different ACI products, each providing monthly gridded NEE_ACI at 1-degree spatial resolution, including Carbon‐Tracker 2019B (2000-2019, CT2019)⁶⁰, Carbon‐Tracker Europe 2020 (2000–2019, CTE2020)⁶¹, Copernicus Atmosphere Monitoring Service (1979–2019, CAMS)⁶², Jena CarboScope (versions s76_v4.2 1976–2017, and s85_v4.2 1985-2017)^63,64, and JAMSTEC (1996–2017)⁶⁵. The monthly gridded ensemble mean NEE_ACI at 1-degree spatial resolution was calculated using the available ACIs from 1980-2017. Monthly ACI ensemble mean NEE_ACI data were summed to seasonal and annual values, and used to calculate the spatial and temporal trends of net CO₂ uptake, and to investigate its relationship to climate and environmental controls.

Productivity dataset

Direct observations of vegetation productivity do not exist at a circumpolar scale. We therefore used two long-term gridded satellite-based estimates of vegetation productivity, including gross primary production (GPP) derived using a light use efficiency (LUE) approach (LUE GPP, 1982–1985)^21,66 and satellite observations of Normalized Difference Vegetation Index (NDVI) from the Global Inventory Modeling and Mapping Studies (GIMMS NDVI, 1982–1985)⁶⁷. LUE GPP (monthly, 0.5° spatial resolution, 1982–2015) is calculated from satellite observations of NDVI from the Advanced Very High-Resolution Radiometer (AVHRR; 1982 to 2015) combined with meteorological data, using the MOD17 LUE approach. LUE GPP has been extensively validated with a global array of eddy-flux tower sites^68,69,70 and tends to provide better estimates in ecosystems with greater seasonal variability at high latitudes. Following^66,71, we used the ensemble mean of GPP estimates from three of the most commonly used meteorological data sets: National Centers for Environmental Prediction/National Center for Atmospheric Research (NCEP/NCAR) reanalysis; NASA Global Modeling and Assimilation Office (GMAO) Modern-Era Retrospective analysis for Research and Applications, Version 2 (MERRA-2); and European Center for Medium-Range Weather Forecasting (ECMWF). GIMMS NDVI (bimonthly, 1/12 spatial resolution, 1982–2015) provides the longest satellite observations of vegetation “greenness”, and is widely used in studies of phenology, productivity, biomass, and disturbance monitoring as it has proven to be an effective surrogate of vegetation photosynthetic activity⁷².

The gridded GPP data were resampled to 1-degree resolution at monthly time scales, to be consistent with NEE_ACI, and used to test (H1) whether greater temperature sensitivity of vegetation productivity explains the different trends in net CO₂ uptake across the NHL. LUE GPP was also used to calculate monthly total ecosystem respiration (TER) as the difference between GPP and NEE_ACI (i.e., TER_residual =  GPP– NEE_ACI) from 1982-2015, as global observations of respiration do not exist. The NEE_ACI, GPP and TER_residual were used as observation-constrained top-down CO₂ fluxes to investigate mechanisms underlying the seasonal CO₂ dynamics in the structural equation modeling and additional decision tree-based analysis.

Eddy Covariance (EC) measurements of bottom-up CO₂ fluxes

A total of 48 sites with at least three years of data representing the major NHL ecosystems were obtained from the FLUXNET2015 database (Table S1 and Fig. S1). EC measurements provide direct observations of net ecosystem CO₂ exchange (NEE) and estimate the GPP and TER flux components of NEE using other climate variables. Daily GPP and TER were estimated as the mean value from both the nighttime partitioning method⁷³ and the light response curve method⁷⁴. More details on the flux partitioning and gap-filling methods used are provided by⁷⁵. Daily fluxes were summed into seasonal and annual values and used to compare with trends from ACIs (Fig. S7), to estimate the climate and environmental controls on the CO₂ cycle in the pathway analysis (Fig. 5), and to calculate the net CO₂ uptake sensitivity to spring temperature (Fig. S14).

Ensemble of dynamic global vegetation models (TRENDY simulations)

The TRENDY intercomparison project compiles simulations from state-of-the-art dynamic global vegetation models (DGVMs) to evaluate terrestrial energy, water, and net CO₂ exchanges⁷⁶. The DGVMs provide a bottom-up approach to evaluate terrestrial CO₂ fluxes (e.g., net biome production [NBP]) and allow deeper insight into the mechanisms driving changes in carbon stocks and fluxes. We used monthly NBP, GPP, and TER (autotrophic + heterotrophic respiration; Ra + Rh) from ten TRENDY v7 DGVMs⁷⁶, including CABLE-POP, CLM5.0, OCN, ORCHIDEE, ORCHIDEE-CNP, VISIT, DLEM, LPJ, LPJ-GUESS, and LPX. We analyzed the “S3” simulations that include time-varying atmospheric CO₂ concentrations, climate, and land use. All simulations were based on climate forcing from the CRU-NCEPv4 climate variables at 6-hour resolution. CO₂ flux outputs were summarized monthly at 1-degree spatial resolution from 1980 to 2017. Monthly ensemble mean NBP, GPP, and TER were summed to seasonal and annual values, and then used to compare with observation-constrained ACI top-down CO₂ fluxes (Figs. 4 and 5).

Satellite data-driven carbon flux estimates (SMAP L4C)

We also used a much finer spatio-temporal simulation of carbon fluxes from the NASA Soil Moisture Active Passive (SMAP) mission Level 4 Carbon product (L4C) to quantify the temperature and moisture sensitivity of NHL CO₂ exchange⁷⁷. The SMAP L4C provides global operational daily estimates of NEE and component CO₂ fluxes for GPP and TER at 9 km resolution since 2015; whereas, an offline version of the L4C model provides a similar Nature Run (NR) carbon flux record over a longer period (2000-present), but without the influence of SMAP observational inputs. The L4C model has been calibrated against FLUXNET tower CO₂ flux measurements and shows favorable performance and accuracy in high latitude regions^4,77. In this analysis, daily gridded CO₂ fluxes at 9-km resolution from the L4C NR record were summed to seasonal and annual values, and used to calculate the sensitivity of net C uptake in response to spring temperature (Fig. S14).

CO₂ fluxes in this analysis are defined with respect to the biosphere so that a positive value indicates the biosphere is a net sink of CO₂ absorbed from the atmosphere. The different data products described above use different terminology (e.g., NEE, NBP) with slightly different meanings; however, they all provide estimates of net land-atmosphere CO₂ exchange⁷⁸.

Climate, tree cover, permafrost, and soil moisture data

Monthly gridded air temperatures at 0.5-degree spatial resolution from 1980 to 2017 were obtained from the Climate Research Unit (CRU TS v4.02) at the University of East Anglia⁷⁹. Air temperature was summarized at seasonal and annual scales to calculate temperature sensitivities of net CO₂ uptake and to investigate the mechanism underlying the seasonal CO₂ dynamics.

Percent tree cover (%TC) at 0.05-degree spatial resolution was averaged over a 35-year (1982-2016) period using annual %TC layers derived from the Advanced Very High-Resolution Radiometer (AVHRR) (Fig. 1a)⁴². %TC was binned using 5% TC intervals to assess its relation to net CO₂ uptake, or aggregated at a regional scale (e.g., TC > 50% or TC < 50%) to contrast variation of net CO₂ uptake or used to explore the mechanism underlying the seasonal CO₂ dynamics.

We used two permafrost maps, including a continuous permafrost extent map from European Space Agency’s (ESA) Climate Change Initiative (CCI) Permafrost project (Permafrost_CCI, 1-km spatial resolution)⁸⁰ and a categorical permafrost zone map from IPA (International Permafrost Association, Permafrost_IPA, vector data)⁵⁸. The Permafrost_CCI product was derived from a thermal model driven and constrained by satellite observations of land surface temperature. The percent of permafrost extent (%P) for each year was calculated as the yearly fraction of permafrost-underlain (ground temperatures at 2 m depth < 0) and permafrost-free area within a pixel. The %P was averaged over 1997–2017 and aggregated to 0.05-degree spatial resolution (Fig. 1b). The %P metric corresponds well with Permafrost_IPA permafrost zones, which distinguish isolated (0–10%), sporadic (10–50%), discontinuous (50–90%) and continuous permafrost (90–100%). In this analysis, the %P map was binned at 5% intervals to show its relation to net CO₂ uptake or used to explore the mechanism underlying the seasonal CO₂ dynamics. The permafrost map was also aggregated into continuous permafrost (ConP, P > 90%), discontinuous permafrost (DisconP, 10% < P < 90%), and non-permafrost (NoP, P < 10%) zones for regional analysis.

We used the ESA CCI soil moisture product (SM, v4.5) produced from combined satellite active and passive microwave remote sensing observations⁸¹ at daily, 0.25-degree spatial resolution from 1980 to 2017. The ESA CCI SM product was developed using data derived from C-band scatterometers suitable for SM retrieval, such as European Remote Sensing Satellites (ERS-½) and METOP, as well as the use of data from multi-frequency microwave radiometers such as the Scanning Multichannel Microwave Radiometer (SMMR), Special Sensor Microwave Imager (SSM/I), Microwave Imager (TMI), Advanced Microwave Scanning Radiometer (AMSR-E), and Windsat⁸¹.

The ESA CCI SM product characterizes surface (0–5 cm depth) SM conditions that are highly variable at daily time scales. We first aggregated the daily gridded SM estimates into monthly mean gridded SM to reduce the temporal variability and spatial gaps. Monthly gridded SM estimates were then aggregated into seasonal and annual averages and used to explore their relationships with the land-atmosphere CO₂ fluxes.

Net CO₂ trends along vegetation, climate, and permafrost gradients

We binned %TC and %P into 5% intervals, and annual mean air temperature into 1-degree intervals. The net CO₂ uptake for the early-growing season (EGS: May–August), late-growing season (LGS: September–October), winter (November–April), and annual periods from 1980 to 2017 was summarized using the ensemble mean of the ACIs (NEE_ACI) at each binned interval. The seasonal and annual mean net CO₂ uptake for each interval of %TC, %P, and temperature was then regressed against years using linear regression. The slope of the regression was interpreted as the net CO₂ uptake trend (gC m² yr⁻²). Finally, trends of net CO₂ uptake at seasonal and annual scales were plotted against %TC, %P, and air temperature to understand the trend and seasonality of net CO₂ uptake along the vegetation, climate, and permafrost gradient (Figs. 1 and 3).

Regional analysis

To reduce the pixel-scale uncertainties of net CO₂ uptake using ACIs data, we also calculated the trends of net CO₂ uptake at regional scales, which were classified by %TC and %P (Fig. S4). Using %TC, the NHL was divided into low (< 30%), intermediate (30–50%), and high (>50%) tree cover regions. Using %P, the NHL was divided into continuous (ConP, P > 90%), discontinuous (DisconP, 10% < P < 90%), and non-permafrost (NoP, P < 10%) regions. %TC and %P are highly correlated, such that short-vegetated regions (TC < 50%) are primarily overlying permafrost (ConP and DisconP), whereas the tree-dominated regions (TC > 50%) are primarily in non-permafrost (Fig. S4). Spatial and temporal patterns for the net CO₂ uptake trend were calculated seasonally and annually from 1980 to 2017 using the ensemble mean of the ACIs (NEE_ACI) over different NHL regions. Seasonal and annual mean net CO₂ uptake for each region was regressed against years using linear regression. The slope of the regression was interpreted as the net CO₂ uptake trend (gC m² yr⁻², Fig. S4).

Robustness analysis

Since trends of net CO₂ uptake are not statistically different between low tree cover (< 30% tree cover in ConP regions) and intermediate tree cover (30-50% tree cover in DisconP regions) (Fig. S4), we aggregated these two regions into a short-vegetated (TC < 50%) permafrost region. We contrasted the net CO₂ uptake between short-vegetated (TC < 50%) permafrost and tree-dominated (TC > 50%) non-permafrost regions. To confirm the net CO₂ uptake trend between permafrost and non-permafrost regions obtained by the ACIs, we performed a series of additional analyses as follows.

Individual ACI analysis

We repeated the trend analysis on permafrost and non-permafrost regions for each individual ACI. All ACIs showed that net CO₂ uptake is increasing faster in the permafrost region than in the non-permafrost region, although only three ACIs (CT2019B, CTE2020, CAMS) showed significantly faster trends over the available data periods (Fig. S5).

Random years and length analysis

We randomly selected starting years and length of years (>= 10 years) for the ensemble mean ACIs, and then repeated the trend analysis. Results consistently showed that net CO₂ uptake is increasing at a much faster rate in permafrost regions than in non-permafrost regions (Fig. S6).

Site-level analysis

We conducted a site-level comparison between EC observations and the ACI ensemble at NHL tower site locations since the 1990s. Both the EC measurements and ACI ensembles confirmed that the net CO₂ uptake is increasing with decreasing tree cover over and increasing permafrost extent, such that the short-vegetated permafrost region (%TC < 50%) had a higher rate of net CO₂ uptake increase than the tree-dominated non-permafrost region (%TC > 50%) during the 1990s – 2010s (Fig. S7). This result is also consistent with an independent EC-based analysis showing that forest sties had a slower increasing rate of net CO₂ uptake than other ecosystem types (e.g., tundra, wetland) in northern permafrost regions¹³. Expectedly, the net CO₂ uptake trends calculated by the EC observations are much higher than the ACIs because (1) the ACIs represent a much larger spatial footprint than the ecosystem-level EC measurements, and therefore average out the local-scale variability, and (2) some episodic ecosystem processes, such as fire disturbance, were not accounted for by the EC observations.

ACI uncertainty analysis

The uncertainties among different ACIs may also affect trend estimates. Uncertainty in ACIs estimates may be due to (i) spread across inversions, (ii) differences among inversions in partitioning of fluxes between permafrost and non-permafrost regions, and (iii) time-dependent differences in inversion spread. Therefore, we used a Generalized Linear Mixed effects Model (GLMM) to estimate trends, considering (i)-(iii) as random effects. The GLMM showed that even after accounting for the uncertainty due to inversion spread, the rate of net CO₂ uptake in permafrost regions is still significantly faster than non-permafrost regions (see supplementary text for full details).

Spatial trend agreement analysis

to assess the spatial consistency of trends of net CO₂ uptake derived from individual ACIs, we calculated the number of ACIs that showed similar trends (either decreasing or increasing). If the majority of ACIs (i.e., >= 5 of 6 ACIs) showed similar trends, we considered the trend in these areas to be true. We reported the trends within these areas here. Generally, we found more areas in the EGS showing similar trends than the other seasons, reflecting larger uncertainties in understanding the carbon cycle in LGS and winter season. In the EGS, 76% of the non-permafrost region and 77% of the permafrost region showed increasing net CO₂ uptake. Annually, only 46% of the non-permafrost region and 51% of the permafrost region showed increasing net CO₂ uptake because of high uncertainties in the LGS and winter (Fig. S8).

Time-series analysis

to see if the trend analysis was affected by including more ACIs in the record after year 2000, we compared the trends calculated from the two longest ACIs (i.e., CAMS, and Jena CarboScope (s76)) with trends calculated from the 3 shortest ACIs (i.e., CT2020B, CTE2020, and JAMSTEC). Results showed that permafrost regions still have a significantly higher rate of net CO₂ uptake using the two longest continuous ACIs since 1980 (CAMS and Jena CarboScope (s76)). The trend of net CO₂ uptake is significantly higher after 2000, which may elevate trends in the permafrost region since the 1980s (Fig. S9). Therefore, including more ACIs after 2000 may change the magnitude of trends in permafrost regions, but does not alter our finding that net CO₂ uptake increased faster in permafrost than non-permafrost regions.

Comparison with Global Carbon Budget 2020 (GCB2020)

The global land CO₂ sink and trend between the ensemble of ACIs and the Global Carbon Budget 2020 (GCB2020), which was estimated from the multi-model mean of 16 DGVMs¹⁸, were also compared from 1980 to 2017 (Fig. S2). We note that this comparison is not a fully independent representation of the land sink, as the ACIs and GCB2020 land-sink estimates used similar fossil fuel emissions, ocean fluxes, and land-use change emissions, and some TRENDY model calculations served as a-priori land-sink for individual ACIs.

Comparison with TRENDY NBP simulations

To see if current state-of-art land surface simulation models reproduce the trend and seasonality of net C uptake relative to the ACIs, the same analysis of net CO₂ uptake spatial and temporal patterns and trends was applied to the NBP simulations from TRENDY (Fig. 4).

Trend and seasonality of productivity

To test H1(whether greater temperature sensitivity of vegetation productivity in the NHL explains the different trends in net CO₂ uptake along the gradient of tree cover), we first calculated the trend and seasonality of productivity using the two productivity products described above, following the same trend analysis procedure used for the ACIs (Fig. S10). The analysis showed that the trend and seasonality are similar using different productivity proxies despite different algorithms, observations, and driving variables among products. Both annual NDVI and GPP data showed positive trends in the NHL (Fig. S10), consistent with greening trends reported from other studies. Furthermore, trends in productivity generally increased with tree cover (Fig. S10), in direct contrast to the decreasing trends of net CO₂ uptake with increase in tree cover.

To further test H1, we calculated the correlations between the trends in productivity and net CO₂ uptake along the tree cover gradient at both pixel level and 5% intervals (Fig. 2a, c). We also calculated the correlation between pixel-level time-series net CO₂ uptake and productivity from 1982 to 2015 (Fig. 2b, d).

Path analysis to explore the mechanism underlying the seasonal CO₂ cycles

Structural equation modeling (SEM) is a powerful multivariate technique to evaluate direct and indirect effects between pre-assumed causal relationships within multivariate data⁸². SEM combines two statistical methods, confirmatory factor analysis and path analysis, and aims to find the causal relationship among variables by creating a path diagram. SEM is well suited to test the complex interaction among CO₂ cycles and controls.

Based on a priori expectations, we constructed one structural equation model (SEM) for each season (EGS and LGS) to test the relative influence of component CO₂ fluxes (i.e., productivity [GPP] and respiration [TER]) on the net CO₂ flux (NEE), and the climate and environmental controls on the NEE component CO₂ fluxes. Our goal was not to precisely predict the spatial and temporal CO₂ cycle variability, but rather to illuminate and quantify the relative influence of the major climatic and environmental controls on CO₂ fluxes. We constructed different models to explain EGS and LGS CO₂ cycles because of the varied climate and environmental controls across seasons over the NHL⁴. For EGS, we tested the relative influences of air temperature (Air T), photosynthetic active radiation (PAR), soil moisture (SM), percent tree cover (%TC), and percent permafrost extent (%P) on vegetation photosynthetic activity and respiration, and the resultant influence on net CO₂ uptake^4,24. For LGS, we tested the relative influences of Air T, PAR, SM, %TC, %P, and labile carbon carried over from the early growing season (PreGPP) on vegetation photosynthetic activity and respiration, and the resultant influence on the net CO₂ uptake results^4,10,26,52 (Fig. 5).

SEMs were fit using the sem function of the lavaan package in R⁸³. The performances of the SEMs were evaluated using a combination of the chi-square statistic (where χ2 ≤ 2 and p > 0.05 indicate a good fitting model), Bentler’s comparative fit index (CFI, where CFI ≈ 1 indicates a good fitting model), and the root mean square error of approximation (RMSEA; where RMSEA ≤ 0.05 and p > 0.1 indicate a good fitting model). The standardized regression coefficient can be interpreted as the relative influences of exogenous (independent) variables. The R² indicates the total variation in an endogenous (dependent) variable explained by all exogenous (independent) variables.

Direct and legacy effects of temperature on seasonal net CO₂ uptake

Because landscape thawing and snow conditions regulate the onset of vegetation growth and influence the seasonal and annual CO₂ cycles in the NHL^24,84, we also analyzed the legacy effects of spring (May–Jun) temperature on seasonal net CO₂ uptake. We regressed seasonal and annual net CO₂ uptake from the site-level EC observations, regional-level ACI ensemble, and the TRENDY NBP ensemble against spring (May-June) air temperature. For EC observations, net CO₂ uptake (i.e., NEE) and air temperature were summarized from site-level measurements. For the ACIs and TRENDY ensemble, net CO₂ uptake (i.e., NEE_ACI and NBP) was summarized as regional means from the ACIs and TRENDY ensemble outputs, and air temperature was summarized as regional means from CRU temperature. The slope of the regression line was interpreted as the spring temperature sensitivity of the CO₂ cycle. Simple linear regression was used here mainly due to the strong influence of spring temperature on the seasonal and annual CO₂ cycle in NHL ecosystems³⁰. Temperature sensitivity (γ: g C m⁻² day⁻¹ K⁻¹) is the change in net CO₂ flux (g C m⁻² day⁻¹) in response to a 1-degree temperature change. The sensitivity of net CO₂ uptake to warm spring anomalies was calculated for different seasons (EGS, LGS, and annual) and regions (i.e., permafrost and non-permafrost), and the T-test was used to test for the difference in γ among different regions, seasons, and datasets. Similarly, direct effects of temperature on net CO₂ uptake were calculated using the same season data (Fig. S14).

Observationally-constrained estimates (EC and ACIs) showed that the sensitivity of net CO₂ uptake in the EGS to spring temperature is positive (γ > 0) and not statistically different (p > 0.05) between permafrost and non-permafrost regions (({gamma }_{{ACI}}^{{np}})=0.125 ± 0.020 gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{EC}}^{{np}}) = 0.052 ± 0.013 gC m⁻² d⁻¹ K⁻¹). In contrast, the sensitivity of net CO₂ uptake in LGS to spring temperature is negative (γ < 0) and significantly smaller in magnitude in permafrost regions (({gamma }_{{ACI}}^{p})= −0.054 ± 0.0064 gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{EC}}^{p})=−0.0157 ± 0.0084 gC m⁻² d⁻¹ K⁻¹) than non-permafrost regions (({gamma }_{{ACI}}^{{np}})= −0.093 ± 0.0070 gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{EC}}^{{np}}) = −0.061 ± 0.013 gC m⁻² d⁻¹ K⁻¹). Net CO₂ uptake in the LGS is also negatively correlated to LGS temperature in the non-permafrost region (({gamma }_{{ACI}}^{{np}})= −0.056 ± 0.0062 gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{EC}}^{{np}}) = −0.061 ± 0.011 gC m⁻² d⁻¹ K^–), but showed a varied response in permafrost regions (({gamma }_{{ACI}}^{p})= −0.027 ± 0.0046 -gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{EC}}^{{np}}) = 0.024 ± 0.011 gC m⁻² d⁻¹ K⁻¹_, Fig. S14). Overall, spring warming has stronger legacy effects on LGS net CO₂ release in the NHL non-permafrost regions, and therefore contributes to a larger seasonal amplitude of net CO₂ fluxes, smaller annual net CO₂ uptake sensitivity to spring temperature, and slower rate of increase in net CO₂ uptake compared to permafrost regions.

Consistent with regional estimates from ACIs, NBP simulated by DGVMs showed a positive sensitivity of EGS net CO₂ uptake to spring temperature (({gamma }_{{TRENDY}}^{{np}})= 0.044 ± 0.025 gC m⁻² d⁻¹ K⁻¹; ({gamma }_{{TRENDY}}^{p}) = 0.111 ± 0.0075 gC m⁻² d⁻¹ K⁻¹), contributing to an annual net CO₂ uptake in the NHL. However, simulated LGS NBP showed non-significant (({gamma }_{{TRENDY}}^{p}) = 0.0034 ± 0.0036 gC m⁻² d⁻¹ K⁻¹) or positive (({gamma }_{{TRENDY}}^{{np}}) = 0.0201 ± 0.0061 gC m⁻² d⁻¹ K⁻¹) sensitivities to spring warming, failed to capture the seasonal compensation in net CO₂ uptake in both permafrost and non-permafrost regions, which therefore partially explained the lack of trends in NBP along the NHL tree cover gradient (Fig. 4).

Source: Ecology - nature.com