Warming alters tundra ecosystems through a suite of interacting biotic and abiotic pathways, including changes in vegetation composition (Bjorkman et al., 2020; Collins et al., 2021; García Criado et al., 2025; Wilson and Nilsson, 2009), plant productivity (Hollesen et al., 2015; Myers-Smith et al., 2019), microbial activity (Frossard et al., 2021), decomposition rates (Sarneel et al., 2020; Schwieger et al., 2025), nutrient cycling (Weedon et al., 2012), growing season length (Barichivich et al., 2013; Collins et al., 2021; Myers-Smith et al., 2019; Oberbauer et al., 1998), and snow-mediated microclimate conditions (Morgner et al., 2010; Pattison and Welker, 2014; Rixen et al., 2022), as well as increased thermokarst activity and permafrost degradation, which can strongly alter soil structure and carbon availability and thereby contribute to ecosystem respiration (Reco) (Abbott and Jones, 2015; Lewkowicz and Way, 2019; Olefeldt et al., 2016; Vogel et al., 2009). These processes jointly regulate Reco, making it an integrative indicator of tundra ecosystem responses to climate warming.

As these responses occur at multiple spatial and temporal scales and are significantly influenced by the local environmental context, a robust evaluation of the effects of warming on Reco requires long-term, spatially diverse datasets that link flux measurements with detailed site and plot-level metadata. The TundraFlux Database was developed to address this issue, systematically compiling chamber-derived Reco measurements from experimental warming studies in Arctic and alpine tundra ecosystems. Fin these efforts by specifically compiling Reco observations from both ambient and experimentally manipulated tundra ecosystems, as well as including extensive metadata on vegetation (e.g., plant community composition, functional traits, biomass), soils (e.g., pH, organic carbon and nitrogen content, soil organic matter), and microclimate (e.g., air and soil temperature, soil moisture). This structure enables users to categorise analyses according to ecosystem properties, environmental factors or experimental design. This supports hypothesis testing, model evaluation and the large-scale synthesis of tundra carbon–climate feedback.

The TundraFlux Database compiles in-situ measurements of daily-aggregated terrestrial ecosystem-level CO2 fluxes (gC–CO2m-2d-1) from Arctic and alpine tundra ecosystems to assess warming effects on ecosystem CO2 respiration (Reco). We compiled data from experiments that used open-top chambers (OTCs) from across the Arctic and alpine tundra biome.

To achieve this, we contacted potential data contributors through established research networks like ITEX, WaRM, INTERACT and the Permafrost Carbon Network, and identified relevant contact information from authors of previously published meta-analyses on Reco responses in warming experiments (Table S2).

We included data from experiments situated within the Arctic and alpine tundra biome (i.e., treeless regions beyond the climatic limit for tree growth) that reported in-situ measurements of ecosystem respiration (Reco) in both warmed open-top chamber (OTC) and unmanipulated control plots. Based on these criteria, we compiled 40 160 individual Reco measurements, encompassing repeated observations across multiple plots, years, days, and, in some cases, multiple measurement times within a day at each site.

To maximize data usability and harmonize temporal resolution across studies, we provide two interlinked versions of the Reco dataset. The raw dataset retains all individual measurements at their original temporal resolution, including multiple measurements per day and associated quality-control information (in total 40 160 observations), allowing users full flexibility in data filtering and aggregation. In addition, we provide a daily-aggregated dataset in which Reco values were averaged within each site (site_id), treatment (OTC or control), plot (plot_id), year (flux_year), and day of year (flux_doy) when multiple measurements occurred per day. This aggregation reduces short-term variability and preprocessing effort for synthesis applications, resulting in a total of 24 951 daily mean observations (OTC, n = 11 046; control, n = 13 905; Fig. 1E).

We define a dataset as all Reco measurements from a given site and year (site_id × flux_year), including both OTC and control treatments, all replicate plots, and one or more measurement dates (Table S3 in the Supplement). Plot-level data were retained by design to preserve within-site replication and enable flexible, user-defined aggregation and statistical modeling approaches, including treatment-wise averaging or hierarchical analyses.

Alongside Reco, we compiled in-situ metadata on microclimate conditions, geolocation, vegetation cover, plant traits, and soil characteristics. These ancillary variables were measured at both plot and site levels, with some recorded concurrently with flux measurements (and provided within the Carbon Flux data, Fig. 2), and others representing broader site-level environmental context that can be linked to the relevant flux measurements (Figs. 2 and 3, Table S4 in the Supplement).

Part of the database compiled through this effort was previously presented in Maes et al. (2024), which focused exclusively on growing season measurements (i.e., June to August) and included data up to the year 2020. This current published version of the TundraFlux Database extends the earlier synthesis by incorporating updated measurements through 2024 (Fig. 1B) and by including shoulder season fluxes from April to May and September to November (Fig. 1D).

The TundraFlux Database comprises 76 variables describing Reco from 64 tundra sites spanning 2000–2024 (Fig. 1A and B). A comprehensive data dictionary with a description of the variables of the site, vegetation, soil and method metadata, including data-type and unit is provided in the Supplements (Table S4).

Data were processed in R version 4.5.1 (R Core Team 2025). We used the R package tidyverse (v.2.0.0) for data handling. For each dataset, defined by a unique combination of site_id and flux_year, we calculated treatment- and plot-specific daily mean Reco values, whenever multiple measurements were made within the same day. We provide an R script on Zenodo (DOI: 10.5281/zenodo.17976235) that documents the full aggregation procedure, and the un-aggregated dataset (see the “Data Availability” section). All individual flux measurements were standardized to a common unit (gC–CO2m-2d-1) to ensure comparability across studies. We screened the data for inconsistencies in unit or sign conventions, contacting data contributors in cases of uncertainty. Spatial coordinates were validated by visualizing all sites on a map, and any imprecise or conflicting locations were corrected in consultation with site researchers.

All experiments measured daytime Reco using dark or opaque chambers, except for the CiPEHR site in Alaska (ALA_1), where automated chambers continuously measured CO2 respiration using clear chambers. To obtain comparable respiration estimates for this site, we extracted only night-time fluxes, defined by photosynthetically active radiation (PAR) values below 5 µmolm-2s-1, as these best approximated dark chamber conditions. These night-time measurements were therefore used as the site's Reco values.

The same quality-control procedures were applied to associated metadata to ensure consistency in spatial reference and measurement units across sites. Soil moisture data, reported at the plot level, were provided either as volumetric or gravimetric values and were subsequently converted to percentages. Bulk density was standardized to gdwtcm-3. Soil organic carbon (SOC) and soil organic nitrogen (SON) were standardized to gCkg-1 soil and gNkg-1 soil, respectively, while soil organic matter (SOM) was standardized to percentages.

The TundraFlux Database includes occasional negative and exact-zero Reco values (Table S3). These originate from the heterogeneity of measurement techniques and pre-processing procedures used in the studies that contribute to the database. Negative fluxes can occur in chamber-based measurements due to short-term CO2 concentration fluctuations, instrument noise, pressure-induced artefacts or brief periods of apparent net CO2 uptake when Reco is close to the detection limit. We retained such values exactly as reported in the original datasets in order to preserve data fidelity. Exact-zero values appear when the net change in chamber CO2 concentration falls within instrument noise, or when contributors apply local preprocessing (e.g. rounding small negative fluxes to zero or applying minimum detection thresholds) prior to submission. As preprocessing conventions differed between studies, both negative and zero values are present in the compiled dataset. Users may apply additional filtering or thresholding, in line with their research objectives, for instance by removing negative fluxes, setting minimum detection limits or utilizing the quality control flags provided in the non-aggregated dataset.

The TundraFlux Database is based on a global synthesis of how ecosystem respiration in Arctic and alpine tundra responds to experimental warming (Maes et al., 2024). Using 136 datasets from 56 OTC experiments at 28 tundra sites worldwide (60 % of the data presented here), the study quantified that a mean rise of 1.4 °C (confidence interval (CI) 0.9–2.0 °C) in air and 0.4 °C (CI 0.2–0.7 °C) in soil temperatures causes a 30 % (CI 22 %–38 %) increase in growing season Reco. It further showed that this stimulation persisted for up to 25 years of warming treatment, with evidence pointing to enhanced plant and microbial activity as the underlying drivers. By linking respiration responses to changes in local soil conditions (i.e., nitrogen concentration and carbon-to-nitrogen ratio), Maes et al. (2024) demonstrated that tundra sites with stronger nitrogen limitations and sites in which warming had stimulated plant and microbial nutrient turnover seemed particularly sensitive in their Reco response to warming. This example highlights the power of standardized, long-term experimental data to uncover generalizable patterns in climate responses. It also showcases how the TundraFlux Database enables large-scale syntheses that identify not only the direction but also the mechanism of possible climate change feedbacks on tundra ecosystems, which are essential to improve carbon–climate feedback projections in Earth System Models.

The expanded TundraFlux Database, which now includes spring and autumn (“shoulder-season”) measurements and updated observations through 2024, enables new high-impact research questions that could not be addressed in Maes et al. (2024). For example, users can quantify how experimental warming affects Reco beyond the peak growing season and evaluate whether the environmental controls of Reco differ between early-season, peak-season, and late-season periods. These additions substantially improve the capacity to analyze seasonal dynamics and long-term trajectories of tundra carbon cycling, thereby supporting more robust evaluations of climate–carbon feedbacks.

Accurately predicting carbon release from permafrost soils under warming scenarios remains a major challenge in climate science (Knoblauch et al., 2021; Schuur et al., 2015) due to the high variability in ecosystem responses and limited availability of long-term data (Swindles et al., 2015). The Warming Permafrost Model Intercomparison Project (WrPMIP) led by Woodwell Climate Research Center is using the TundraFlux Database to bridge the gap between experimental warming studies and large-scale carbon modeling (warmingpermafrost.nau.edu). In this project, multi-model simulations are being run at both regional and site scales to align with the spatial and temporal dimensions of experimental data (Wells et al., 2023). By aligning field-based warming measurements from the TundraFlux Database with model simulations, the project will enhance our ability to project the magnitude and timing of carbon release from permafrost regions under climate change (Schädel et al., 2018).

The TundraFlux Database inherently reflects a sampling imbalance in the field, particularly the underrepresentation of key Arctic regions such as the Canadian High Arctic archipelago and Siberia (López-Blanco et al., 2024; Metcalfe et al., 2018; Virkkala et al., 2019; Fig. 1A, Table S5 in the Supplement). Still it represents the most comprehensive effort that is currently available of Reco data from warming experiments in the tundra. In particular, high-latitude North America and northern Europe are the best-represented regions (Table 1). This geographical concentration reflects a well-documented spatial bias in Arctic field sampling toward long-established research hubs with good accessibility and infrastructure (Metcalfe et al., 2018).

There is substantial variability in the number of measurements across sites (median = 120, IQR = 133.5; total measurements = 24 951). In particular, the CiPEHR site (ALA_1) near Eight Mile Lake, Alaska, USA, with 13 572 daily-aggregated observations, contributed 52.5 % of the Reco data in the TundraFlux Database (Table S5). A large proportion of observations from a single, well-studied site increases temporal and treatment-level detail but may disproportionately influence cross-site or pan-Arctic analyses if not accounted for. How to address this imbalance depends on the user's research question and analytical framework, but several approaches can help prevent disproportionate influence of high-density sites. These include applying hierarchical or mixed-effects models that treat site as a random effect, using equal or inverse-effort site weighting, aggregating fluxes to common temporal resolutions, conducting sensitivity analyses (e.g., subsampling or excluding dominant sites), or using site-level bootstrapping or partial-pooling approaches (Choi and Kang, 2025). At the same time, these long-term, high-density datasets provide valuable opportunities for method development, uncertainty quantification, and benchmarking models at sites with robust metadata.

By systematically compiling data from OTC experiments across Alaska, Greenland, Svalbard, Iceland, Fennoscandia, Canada, and Russia, the TundraFlux Database is the most comprehensive resource currently available for evaluating warming effects on tundra carbon cycling. Importantly, tundra ecosystems exhibit substantial spatial variability in both abiotic and biotic conditions (Aalto et al., 2022; Magnússon et al., 2023), meaning that each site, regardless of data quantity, contributes distinct information about ecosystem responses under different environmental and vegetation contexts.

In the TundraFlux Database, 93 % of the averaged daily data points (n = 22 534) come from warming experiments that lasted 11 years or less, and over half (51 %, (n = 12 239) lasted fewer than 4 years, at the time of Reco measurements. At the time of publication, all warming experiments included in the database were still ongoing (Table S2 in the Supplement). Out of the 64 sites in total, 7 sites (11.1 %) represent long-term experiments with continuous measurements (≥ 5 years), while 56 sites (88.9 %) represent short-term experiments (<5 years) with often only single measurements (Table S5). Consequently, longer-term measurements (>11–24 years) are rare (n = 1573), and data from warming experiments that lasted longer than 24 years are absent (Fig. 1C). This clearly highlights the importance of maintaining long-term experiments, particularly given that changes in soil processes or vegetation composition driven by warming may unfold over decades in Arctic and alpine tundra (Hollister et al., 2005; Jónsdóttir et al., 2023; Wei et al., 2025). At the same time, both OTCs and manual chamber measurements may introduce experimental disturbance effects (e.g., trampling, vegetation damage, soil compaction) that can accumulate over time (Hollister et al., 2023). Such disturbance may increasingly influence ecosystem functioning the longer an experiment runs. Thus, while longer-term data would help capture slow ecological changes, extended experiment duration may simultaneously amplify disturbance-related artefacts, complicating the interpretation of long-term trends.

Our database contains 21 830 daily-aggregated Reco observations from the growing season (defined here as June-September in the Northern Hemisphere, and December-March in the Southern Hemisphere), and 3297 observations from outside the growing season, highlighting a gap in our understanding of carbon fluxes in the shoulder (i.e., April, May, October, and November, n = 3330) and winter seasons (i.e., December–March in the Northern Hemisphere, n=0, Australia excluded) (Figs. 1D and 4).

As Arctic soils are usually covered in snow during the winter months, the spring and autumn shoulder seasons are particularly vulnerable to the effects of global warming, as these transitional periods experience rapid changes in snow cover and soil temperatures (Hassol, 2004; Shukla et al., 2019). Despite its length and significance, winter remains the most understudied season in Arctic ecosystems, even though cold-season processes can contribute substantially to annual carbon budgets through continued microbial activity and ecosystem respiration beneath the snowpack (Natali et al., 2019; Zona et al., 2016). Climate warming is altering Arctic winters by increasing air and soil temperatures, changing snow accumulation and duration, and increasing the frequency of winter warming events and rain-on-snow events, all of which affect soil thermal conditions, microbial activity, and carbon fluxes (Cooper, 2014; Rixen et al., 2022). Similarly, alpine ecosystems are experiencing reduced snowpack, earlier snowmelt, and glacier retreat, with important consequences for soil temperature regimes, hydrology, and ecosystem functioning (Ernakovich et al., 2014). Warming effects and mechanisms identified during the growing season (Maes et al., 2024) may not apply to the winter season, when factors such as snow depth and duration exert great control over carbon fluxes (Björkman et al., 2010; Grogan, 2012; Morgner et al., 2010; Rixen et al., 2022; Semenchuk et al., 2016; Slatyer et al., 2022).

More research focusing on the effects of warming on carbon fluxes during the underrepresented winter and shoulder seasons while developing a mechanistic understanding of winter carbon dynamics is therefore essential to improve predictions of future CO2 emissions from Arctic soils. Overlooking these seasons risks underestimating both the extent and variability of carbon release from rapidly warming Arctic soils.

Reco responses to climate change depend on the dynamics of its two main components: autotrophic (plant-derived) and heterotrophic (microbial and faunal) respiration (Bond-Lamberty et al., 2004; Hicks Pries et al., 2013). Because these two components can respond differently to changes in climate (Borken et al., 2006; Muhr and Borken, 2009; Gomez-Casanovas et al., 2012), understanding their individual contributions is crucial to accurately predict ecosystem carbon dynamics and modeling carbon–climate feedback processes. However, partitioning Reco into these source fluxes remains methodologically challenging. As respiration partitioning measurements are usually destructive, altering ecosystem dynamics and making long-term measurements difficult, such data remain scarce. For example, in the meta-analysis on the drivers of Reco in Arctic and alpine tundra by Maes et al. (2024), only nine out of 136 Reco datasets included partitioning data on the autotrophic and heterotrophic respiration components.