Vegetation type is an important predictor of the arctic summer land surface energy budget

The Arctic is warming faster than the global average, triggering widespread changes in precipitation, permafrost, ice and snow cover, and vegetation composition, structure, and function. These changes feed back to climate via the surface energy budget (SEB), comprising net radiation, sensible and latent heat fluxes, ground heat flux, and melt terms. While multiple drivers influence the magnitude and seasonality of SEB components—such as vegetation type, snow cover, soil and permafrost conditions, topography, and meteorology—there has been no quantitative, circumpolar assessment of how important vegetation type is relative to other drivers. Earth system models typically simplify Arctic vegetation into a few plant functional types, potentially missing key variability. This study aims to quantify the SEB across the treeless land north of 60°N (1994–2021) and to compare the predictive skill of vegetation type versus other drivers (climate, topography, snow, permafrost, clouds, latitude). The authors harmonize in situ observations from 64 sites (652 site-years) and classify sites according to the Circumpolar Arctic Vegetation Map (CAVM) classes plus boreal peat bogs, to evaluate the magnitude and seasonality of surface energy fluxes with a focus on summer (June–August).

Prior research documents rapid Arctic environmental changes and highlights the SEB as a key mediator of land-atmosphere feedbacks. However, uncertainties in SEB projections remain, particularly for turbulent fluxes (sensible and latent heat). Arctic vegetation is diverse, yet land models often represent it with few plant functional types. Observational studies suggest vegetation types affect SEB components, including turbulent fluxes and albedo, but many are qualitative or geographically limited. Drivers such as cloud cover, snow regime, soil moisture, permafrost extent and ice content, and topography are known to influence SEB. The CAVM classes differ in SEB-relevant traits (height, productivity, albedo), implying potential causal vegetation effects on SEB. This study builds on these findings by providing a quantitative, pan-Arctic comparison of vegetation type against a suite of environmental drivers.

Study domain: treeless, terrestrial Arctic north of 60°N, aligned with CAVM extent. Data sources: half-hourly/hourly in situ energy fluxes and meteorology from FLUXNET (FLUXNET2015), AmeriFlux, AON, ICOS, GEM, GC-Net, PROMICE, covering 64 tundra and glacier sites and 652 site-years (1994–2021). Data processing: excluded gap-filled data; outlier filtering; time conversion to local standard time; daily mean/min/max computed when ≥65% of subdaily data available and gaps ≤4.8 h; harmonized units and flux sign convention (H, LE, G positive away from surface); derived Rnet, SWnet, LWnet, albedo from components; computed normalized fluxes (n.Rnet, n.SWnet, n.LWnet, n.H, n.LE, n.G) as percentage of daily maximum potential incoming shortwave radiation based on location and topography. Vegetation classification: site-level vegetation type assigned using literature descriptions and decision chain aligned to CAVM classes: barren complexes, graminoid tundra, prostrate dwarf-shrub tundra, erect-shrub tundra, wetland complexes; additional categories included glacier and boreal peat bog. Landscape-scale dominant vegetation (CAVM type) was extracted from raster CAVM (1 km resolution). Environmental drivers (SEB-drivers): climate variables (mean annual air temperature, annual precipitation, snow amount) from CHELSA V2.1 (1979–2018); Continentality (Conrad Index) and Summer Warmth Index; bioclimatic subzone (CAVM A–E); snow metrics (median snow duration, snow-free and snow-onset dates, 2000–2020) from MODIS MOD10C1; cloud cover and cloud-top temperature (1984–2016) from ISCCP H-series; permafrost extent and ground-ice content from NSIDC Circum-Arctic maps; topography (altitude, slope, aspect) from ArcticDEM. Analyses: (1) Variance partitioning to assess relative importance of 15 SEB-drivers in explaining summer (JJA) mean magnitudes of Rnet, H, LE, G (and additional fluxes including normalized fluxes), using all pairwise 2-predictor linear models with orderings to quantify independent and joint contributions; analysis on vegetation-only subset (31 sites; 234 site-years). (2) Linear mixed-models estimating mean ± 95% CI of Rnet, SWnet, LWnet, H, LE, G by vegetation type for summer (JJA) and annually (Y) for full dataset (64 sites; 652 site-years); post hoc pairwise comparisons among vegetation types; additional monthly analyses with normalized fluxes. (3) Seasonality analysis for 2000–2021 subset: constructed smoothed daily seasonal cycles per vegetation type; defined summer-regime periods as daily mean >0 W m−2 (Rnet, H, G) or Tsurf >0 °C, and albedo below mean of annual minimum and maximum; compared start/end of summer regimes relative to MODIS snow-free and snow-onset dates using Welch’s t-tests.

• Vegetation type is a strong predictor of summer turbulent fluxes: variance partitioning shows vegetation type explains on average 56.3% (range 53.8–58.8%) of variance in H and 71.7% (61.4–81.9%) in LE.
• For ground heat flux (G), vegetation type is among top predictors (average explained variance 31.7%) after landscape-scale dominant vegetation (CAVM type, 40.6%) and bioclimatic subzone (39.1%).
• For net radiation (Rnet), latitude and snow cover duration are most important; when using normalized fluxes, vegetation type ranks among the top three for n.Rnet and is the most important predictor for n.LWnet; vegetation type also predicts albedo and surface temperature.
• Summer mean flux magnitudes differ markedly among vegetation types, often comparable to differences between glaciers and vegetated surfaces. Examples (summer means): • LE: barren tundra 0 W m−2 vs boreal peat bog 75 W m−2 (P < 0.001). • H: barren −18 W m−2 vs erect-shrub 38 W m−2 and prostrate-shrub 39 W m−2 (P < 0.05). • LWnet: barren −71 W m−2 vs erect-shrub −29 W m−2 (P < 0.01).
• Bowen ratio (H/LE) in summer: >1 for prostrate-shrub (1.6), erect-shrub (1.5), wetlands (1.1); <1 for graminoid (0.6), boreal peat bog (0.1), barren (<0).
• Seasonality: starts of summer-regimes occur significantly before the spring snow-free date across vegetation types: Rnet −56 ± 27 days, H −33 ± 18 days, G −39 ± 11 days (P < 0.05). Ends are close to snow-onset, except G which ends earlier (−19 ± 20 days, P < 0.05).
• Vegetation types differ in timing: start of H summer-regime varies by 41 days (earliest erect-shrub 60 days before snow-free; latest prostrate-shrub 20 days before). Start of Rnet summer-regime varies by 74 days (earliest erect-shrub 108 days before; latest boreal peat bog 34 days before). For G, start varies by 19 days; end varies by 50 days (earliest in boreal peat bogs 56 days before snow-onset; latest in graminoid 6 days before).

Vegetation type has high predictive power for summer SEB, particularly turbulent fluxes, likely because CAVM classes integrate multiple environmental conditions and differ in SEB-relevant traits (height, productivity, albedo), implying both proxy and causal roles. Cloud cover shows low predictive ability across sites due to high and spatially persistent cloudiness in Arctic summer despite its strong radiative effect. Differences among vegetation types in turbulent fluxes can equal or exceed those between glacier and vegetation, underscoring the need to represent Arctic vegetation diversity in land models. Boreal peat bogs exhibit very high summer latent heat fluxes (likely due to Sphagnum moss lacking stomatal control), implying that northward expansion of peatlands could shift energy partitioning toward latent heat with hydrological and climatic implications. Barren tundra shows negative summer sensible heat flux, acting as a heat sink; with its areal extent projected to decline, this may reduce a current cooling influence, but data are sparse. Vegetation-snow interactions likely modulate SEB seasonality: erect shrubs may protrude earlier through snow, advancing H and Rnet onset relative to snow-free dates, affecting cumulative sensible heating. The seasonality of fluxes, not only their mean magnitude, is critical for understanding land-atmosphere coupling, soil thermal regimes, and permafrost dynamics. Overall, results support refining vegetation representation in Earth system models to reduce SEB uncertainties.

This circumpolar synthesis demonstrates that vegetation type is a key predictor of the Arctic summer surface energy budget, especially for sensible and latent heat fluxes, with differences among vegetation types comparable to those between glacier and vegetated surfaces. Vegetation also influences the timing of SEB summer regimes relative to snow phenology. Incorporating a refined set of Arctic vegetation types and associated functional traits into land surface components of Earth system models should improve simulation of high-latitude energy fluxes and feedbacks. Future research should identify specific vegetation traits driving SEB variability, improve year-round SEB observations across all vegetation types (especially barren tundra), and further resolve vegetation–snow–permafrost interactions and their implications for SEB seasonality and cumulative energy exchange.

SEB observations remain sparse outside summer, with many sites lacking autumn and winter data and incomplete turbulent flux records, limiting annual assessments and SEB closure. Seasonality analyses could only be resolved at vegetation-type level, not per site. Barren tundra is severely underrepresented (only one site), introducing uncertainty in inferred differences. Potential misclassification or limited trait coverage in CAVM classes could obscure specific trait–SEB linkages.