The four study sites are Hisåsen, Finse and Iškoras on mainland Norway and Adventdalen on Svalbard. The sites are located along gradients in latitude, temperature and precipitation, covering a latitudinal gradient from 60–78° N, a precipitation gradient from 218–968 mm per year and a gradient in mean temperature from -3.9–2.7 °C (see Fig.  and Table ). Compared to northern latitude (above 60° N) sites available in the FLUXNET2015 dataset , our study sites span approximately the full range of latitudes and annual precipitation rates (including one site with higher precipitation), whereas they are somewhat in the mid-range with respect to mean annual temperature (Fig. b). The ecosystem types at the three mainland sites have been classified according to the “Nature in Norway” ecosystem and landscape diversity framework . Photos of the study sites are available in Fig. .

The Hisåsen site is a boreal peatland site, located south of the hill Hisåsen in the Regnåsen-Hisåsen nature reserve in eastern Norway. The area is undulating, slightly sloping towards north and covered by forest and peatlands. The climate is continental subarctic (Dfc), according to Köppen's classification, with a mean annual temperature of 2.7 °C and a mean annual precipitation of 857 mm. The measurement tower is located on a drained peatland with organic soils surrounded by forest on glacial till . The footprint ecosystem types are predominantly strongly lime-poor fen and lime-poor drained fen, according to the Nature in Norway system (maps are publicly available at http://github.com/geco-nhm/NiN_Hisaasen, last access: 28 November 2023). The vegetation is low and dominated by sedges, mosses and shrubs, such as Salix herbacea and Empetrum nigrum. A few trees, such as Pinus sylvestris and Betula pubescens are present.

Finse is a sub-alpine tundra and wetland site, located in the valley Finsedalen north of the Hardangerjøkulen glacier. The valley runs towards east-southeast, and wind directions at the site are controlled by the valley. The climate is tundra (ET), with mean annual temperature of -1.1 °C and mean annual precipitation of 967 mm. The climate has an oceanic influence, located approximately 140 km from the coastline in the west and exposed to the mild and moisture-baring westerlies. The instrument tower sits on a ridge running southwest-northeast. Southeast, the ridge slopes down towards the river Ustekveikja draining the lake Finsevatnet, located approximately 1 km east of the tower. The footprint ecosystem types are predominantly lime-poor open fens, arctic-alpine heath and lee side as well as snowbeds . The soil is thin and consisting of discontinuous glacial till with thickness less than 0.5 m and glacifluvial deposits . The vegetation is dominated by shrubs, such as Salix herbacea and Empetrum nigrum, and mosses and lichens in the drier areas and sedges and mosses in wetter areas.

Iškoras is a palsa mire site located north of the mountain Iškoras in the plateau area of Finnmarksvidda in northern Norway. The climate is continental subarctic (Dfc), with a mean annual temperature of -1.4 °C and a mean annual precipitation of 417 mm. The tower is located on a peat plateau surrounded by mires and ponds , with partly organic soil and partly glacial till . Shrubs and lichens dominate dry, elevated palsas, whereas sedges and mosses dominate wetter areas near unvegetated ponds . The shrubs are species such as Betula nana and Empetrum nigrum. Bog and open fen are the dominating ecosystem types (; Anders Bryn, personal communication, 2023). The peat plateau lies north of the mountain range Iškoras, and the terrain slopes gently towards north.

Adventdalen (NO-Adv) is an Arctic site featuring polygonal tundra, located in the valley Adventdalen on Spitsbergen, Svalbard. The climate is tundra (ET), with a mean annual temperature of -3.9 °C and a mean annual precipitation of 218 mm. The climate has an oceanic influence, located approximately 6 km from the coast. The tower is located on a river terrace on the flat part of a large alluvial fan, and the soil consists of a few decimetres of fine-grained eolian deposits on top of coarser-grained alluvial deposits . The vegetation is very low, dominated by dwarf shrubs like Salix polaris at dry places and mosses and sedges in wet depressions .

For each site, the 3–4 years with the most complete data collection was chosen as the site measurement period. Thus, the measurement period varies slightly between sites (Table ). Figure  shows monthly mean temperature and cumulative precipitation at the nearest climate station of The Norwegian Meteorological Institute (MET Norway) for the reference period (1990–2020) and for each year in the respective measurement periods. Precipitation is relatively evenly distributed over the year at all sites, with the wettest month having a mean precipitation of 2.7 to 3.6 times the precipitation of the driest month. The wettest month at each site occurs in summer or early autumn (July at Iškoras, August at Hisåsen and September at Finse and Adventdalen). The driest month occurs in spring or early summer (March at Hisåsen and Iškoras, April at Finse and May at Adventdalen). Monthly mean temperature typically peaks in July and is below zero from November–March at Hisåsen, October–April at Finse and Iškoras and October–May at Adventdalen.

Overall, the measurement years were warmer than the reference period at all sites (Fig. ). At Hisåsen, the annual mean temperature was warmer than normal for all years (0.1–0.2 °C), with 2020 being the warmest year. Annual precipitation was both lower and higher than normal (ranging from 88 %–115 % of the long term mean) with 2020 being the wettest year. A similar pattern was found at Finse, with all years being warmer than normal (0.1–1.0 °C), and annual precipitation both higher and lower than normal (ranging from 78 %–133 % of the long term mean), with 2020 being the wettest year. Iškoras was the only site where annual temperature was both higher and lower than normal (varying from 0.4 °C lower in 2019 and 2021, to 0.7 °C higher in 2020). However, all years had temperatures 0.2–4.6 °C above normal in the warm season (May–September). Annual precipitation ranged from 101 %–118 % of the long term mean, with 2019 and 2020 being the wettest years. At Adventdalen, all years were warmer (0.4–3.8 °C) and wetter (114 %–143 %), than the long term mean, with 2016 being the warmest and wettest year as well as the warmest ever recorded at the station. Annual precipitation ranged from 114 % of the long term mean in 2013 to 143 % in 2016. The temperature deviations were typically larger in autumn and winter (September–March) than in spring and summer (April–August). On average, 40 %–44 % of the yearly precipitation fell as snow at Finse, Iškoras and Adventdalen, while 11 % fell as sleet and 45 %–48 % fell as snow. At Hisåsen, the distribution was 20 % snow, 13 % sleet and 67 % rain (Table ).

To study the controls on evaporation on a sub-daily timescale, we used hourly observations of evaporation from the eddy covariance data. Only observed data were used in this part of the study (i.e. gap-filled values were not included). We first calculated Pearson correlation coefficients between hourly evaporation and local meteorological and surface variables for the snow-free and snow-covered season separately, and tested whether the correlation coefficients were significant at significance levels p<0.05, p<0.01 and p<0.001. Available energy (Rn-G) is included in the analysis of correlations as it is a forcing variable in the Penman-Monteith equation (described in Sect. ), but is not included as a predictor variable in the gap-filling regression model or included in the factor analysis. To avoid spurious predictor importances due to the large degree of covariance between the controls of evaporation, we performed a Factor analysis (similar to the analysis in and ) to group variables with a large degree of common variance. To evaluate how the partitioning between sensible and latent heat flux changed with atmospheric and surface controls, we calculated mean Bowen ratio, i.e. the ratio of sensible to latent heat flux, for bins of vapour pressure deficit and soil water content. Finally, we modelled hourly evaporation using the Penman-Monteith equation to test how a widely used model is able to capture the sensitivity of evaporation to climatic and surface controls. We focus on controls in the snow-free season, as 68 % to 86 % of the annual evaporation occurred in the snow-free season.

To evaluate how the evaporation at the sites in this study compares to that of other northern latitude sites, we compared mean annual evaporation of the four sites to those in the FLUXNET2015 dataset . Only FLUXNET sites located above 60° N latitude with Creative Commons (CC-BY-4.0) licence were included in the comparison (see a list of the sites in Table ). To test the annual evaporation dependency on the site mean temperature, we fitted a linear regression with annual evaporation as dependent variable and mean annual temperature as independent variable (including both our four study sites and the fourteen FLUXNET sites) and checked if the regression slope was significant at p<0.05. Furthermore, we tested if a better fit could be obtained by using warm season (May–September) mean temperature only instead of annual mean temperature.

Hourly evaporation was significantly correlated (p<0.001) with most of the climatic and surface controls, both in the snow-free and snow-covered season (Table ), however, the relation was typically weak (r<0.5) except for a few controls related to atmospheric evaporative demand. In the snow-free season, evaporation had strong linear relations (r>0.7) to vapour pressure deficit and incoming shortwave radiation. Most stations had a strong or moderate relation (r>0.5) to air and soil temperature and outgoing longwave radiation, while the relation to other variables such as soil water content and wind speed was weak. In the snow-covered season, the correlation coefficients between evaporation and its controls were overall weaker than in the snow-free season, however the relation with incoming shortwave radiation and vapour pressure deficit was still strong or moderate (r>0.5) for all sites except at Adventdalen where only vapour pressure deficit had r>0.5 in the snow-covered season. For other controls, the relation to evaporation was weak (r<0.5).

The factor analysis showed that evaporation, vapour pressure deficit and incoming shortwave radiation had a high degree of shared variance both in the snow-free and snow-covered season. In the snow-free season, evaporation loaded highest on the first factor at Hisåsen, Finse and Iškoras (Fig. ), together with vapour pressure deficit, incoming shortwave and outgoing longwave radiation (all with loadings >0.72). Air temperature, soil temperature and soil heat flux also loaded relatively high (>0.55) on the first factor. At Adventdalen, evaporation loaded highest on the second factor, together with incoming shortwave radiation and vapour pressure deficit (>0.62), whereas variables such as air temperature, surface temperature, and longwave outgoing radiation loaded high on the first factor. In the snow-covered season, evaporation had a lower loading on the first and second factor compared to the snow-free season, indicating a lower degree of common variance with other variables in the dataset (Fig. ). At Finse and Adventdalen, the loading for evaporation was less than 0.51. At Hisåsen and Iškoras however, the evaporation loadings in the snow-covered season was more comparable to the snow-free season (>0.7). At Hisåsen, evaporation had a relatively high loading (0.76) on the first factor, together with shortwave incoming radiation (0.79) and vapour pressure deficit (0.90). At Iškoras, evaporation had a relatively high loading on the second factor (0.84) together with incoming shortwave radiation (0.76) and vapour pressure deficit (0.71).

Comparing observed evaporation to Penman-Monteith estimates in the snow-free season, and looking at the distribution for bins of available energy and vapour pressure deficit (Fig. , upper row), we found that the Penman-Monteith equation reproduced the observed pattern of high evaporation when both vapour pressure deficit and available energy were high, and low evaporation when both controlling factors were low. The mean relative error (Fig. , lower row) showed a similar pattern across the sites, and showed a tendency towards overestimation (red hexagons) when vapour pressure deficit was low, and underestimation (blue hexagons) when high vapour pressure deficit was combined with low available energy. Furthermore, evaporation was underestimated by the Penman-Monteith equation when available energy was negative.

We evaluated the sensitivity of the observed and estimated evaporation, and the relative error of the estimates, to surface controls that can affect the surface water availability, such as the soil water content and time since rain (Figs. –), and growing degree day which is related to phenology (Fig. ). Overall, the sensitivity of the Penman-Monteith estimates corresponded to the observed sensitivity, and there was no tendency for the relative error to increase or decrease with changes in the surface controls. However, there was a tendency towards higher relative errors at high values of soil water content at Finse (Fig. ).

The Bowen ratio, i.e. the ratio of sensible to latent heat flux, decreased with vapour pressure deficit (Fig. ), meaning that latent heat flux was increasingly favoured over sensible heat flux with increasing vapour pressure deficit. At vapour pressure deficit over 0.4 to 1.0 kPa, the Bowen ratio was under 1 and the latent heat flux dominated over sensible heat. At Hisåsen, the mean Bowen ratio stabilised at just below 1 for vapour pressure deficit exceeding 1.0 kPa, while at the other sites, it continued to decrease for the whole range of observed values of vapour pressure deficit.

Daily evaporation rates showed a clear seasonal pattern, with lower values in the snow-covered season and higher values in the snow-free season. Across the sites, mean daily evaporation (averaged over season) ranged from 0.0 to 0.1 mmd-1 in the snow-covered season, from 0.2 to 0.4 mmd-1 in the shoulder season and from 0.5 to 1.0 mmd-1 in the snow-free season. In the shoulder and snow-free season, the magnitudes of mean daily evaporation followed the gradient in mean annual temperature, with the highest evaporation at Hisåsen and lowest at Adventdalen, while in the snow-covered season there were only minor differences between the sites. Looking at the distribution of daily evaporation per month of the year (Fig. , upper row), we found that the mean daily evaporation was highest in the summer months, June–August, at all sites. However, the month of the highest mean daily rates were slightly different between the sites. The mean daily evaporation peaked in June at Hisåsen and Adventdalen, June/July at Iškoras and July/August at Finse. Comparing across sites, evaporation was highest at Hisåsen, with mean daily evaporation of 1.9 mmd-1 (June), compared to 0.8 mmd-1 at Finse (July/August), 1.2 mmd-1 at Iškoras (June/July), and 0.8 mmd-1 at Adventdalen (June).

The seasonal pattern of available energy and vapour pressure deficit followed a similar seasonal pattern, with lower values in winter and higher during summer (Fig. , middle and lower row). Mean daily available energy ranged from -1.6 to 0.6 MJd-1 in the snow-covered season, from 1.2 to 5.0 MJd-1 in the shoulder season, and from 6.1 to 7.0 MJd-1 in the snow-free season. Mean daily vapour pressure deficit ranged from 0.04 to 0.12 kPa in the snow-covered season, from 0.03 to 0.15 kPa in the shoulder season, and from 0.14 to 0.35 kPa in the snow-free season. Available energy peaked in June for all sites, while vapour pressure deficit peaked in either June, July, or August depending on site.

Across the sites, we found a large variation in the role of evaporation in the vertical water balance, due to a higher variation in precipitation (Table ) than in evaporation. The evaporation ratio, i.e. the ratio of evaporation to precipitation, was highest at Adventdalen, the site with the lowest precipitation, and lowest at Finse, the site with the highest precipitation. For individual months, the evaporation ratio was occasionally over 100 % at all sites, and up to 400 % at Adventdalen (Fig. ). At Finse, the monthly evaporation ratio was generally below 40 %, except in August 2021 when evaporation was 124 % of the precipitation. At Adventdalen, the monthly evaporation ratio was typically higher than 100 % in May, June, and July. Considering the warm season (May–September), the total evaporation was up to 72 % of the precipitation in the same months (at Adventdalen in 2015). The mean warm season evaporation ratio was lowest at Finse (20 %), intermediate at Hisåsen (48 %) and Iškoras (47 %) and highest at Adventdalen (58 %). The mean annual evaporation ratio ranged from 9 % to 30 % of MAP (Fig. ), and the evaporation ratio increased with decreasing MAP across sites.

Overall, annual evaporation followed the gradient in mean annual temperature, with evaporation increasing with increasing temperature (Fig. ). The highest annual evaporation was found at the Hisåsen site, with mean annual evaporation amounting to 208 mm, followed by 107 mm at Iškoras, 81 mm at Finse and 80 mm at Adventdalen. Interannual variability in evaporation was generally low. At Hisåsen, Iškoras and Adventdalen, annual evaporation deviated less than 10 % from the mean. However, at Finse, the interannual variability was larger, with evaporation in 2020 being 34 % lower and in 2021 27 % higher than the mean of all four years.

We found that some site-specific interannual variability in evaporation corresponded with the interannual variability of the end of the snow-covered season. At Hisåsen, Finse and Iškoras the year with the lowest annual evaporation corresponded to the year with the longest lasting snow cover in spring (Fig. ). Accordingly, at Hisåsen and Finse, the year with the highest annual evaporation corresponded to the year with the earliest snow cover melt-out. To assess whether the lower annual evaporation in years with a long-lasting snow-cover was mainly caused by a shorter evaporative season or if years with a longer lasting snow-cover had lower evaporative demand in the snow-free season, we compared weekly values of evaporation, incoming solar radiation and vapour pressure deficit for overlapping weeks in the snow-free season of each year. The comparison was set up as an ANOVA-analysis for each variable, using a significance level of p=0.05. We found that weekly evaporation rates did not differ significantly between years, nor were there significant differences in incoming solar radiation or vapour pressure deficit. Comparing across sites, a larger interannual variability in snow-cover duration in spring corresponded with a larger interannual variability in evaporation, with Finse showing the highest interannual variation.

The mean annual evaporation at our sites (80–208 mm), was in the lower range of the mean annual evaporation at the selected FLUXNET2015 sites (45–385 mm), despite being in the mid-range of mean annual temperature. Furthermore, we found no significant linear relationship between annual evaporation with mean annual temperature when including the 14 FLUXNET2015 sites (Fig. ). However, we found a significant increase in annual evaporation with mean temperature in the warm season (Fig. ), with a slope of 16.4 mm°C-1 (p<0.05). Still, all our four sites were below the trendline. We were unable to detect any pattern in the deviation from the regression line based on ecosystem type, however, the ecosystem types were not evenly distributed over the temperature range. For example, all forest sites were in the upper end of the temperature range.

Our results show that evaporation from northern latitude wetlands is mainly controlled by atmospheric evaporative demand, and furthermore, that the evaporative demand depends mainly on incoming solar radiation and vapour pressure deficit. On the sub-daily (hourly) timescale, vapour pressure deficit and incoming solar radiation had a strong correlation with evaporation, loaded high in the factor analysis on the same factors as evaporation, and had a high relative importance in the random forest model. Though vapour pressure deficit and incoming solar radiation had a high degree of shared variance, we found that evaporation was occasionally limited by low vapour pressure deficit despite available energy being high (Fig. ). Especially at Finse and Adventdalen, where the climatic setting and oceanic influence lead to lower warm season temperatures and higher air humidity, the vapour pressure deficit was typically low, leading to constrained evaporation. At low vapour pressure deficit, more of the available energy was partitioned into sensible heat flux, as shown by the higher Bowen ratio for low vapour pressure deficit (Fig. ). Additionally, we found that evaporation at Adventdalen in the snow-covered season was mainly controlled by vapour pressure deficit. The correlation to other controls was weak. This is likely due to the long polar night at Adventdalen with low variation in solar energy.

Other studies such as and have discussed the role of vapour pressure deficit in controlling northern latitude evaporation. found latent heat flux to persistently exceed sensible heat flux when vapour pressure deficit was above 0.3 kPa for wet soils, and above 1.2 kPa for dry soils. We found a similar threshold at around 0.8–1.0 kPa (Fig. ) with no clear effect of soil moisture content (Fig. ). Additionally, we found that the Bowen ratio continued to decrease with vapour pressure deficit at Iškoras and Finse while the midday mean Bowen ratio at Hisåsen levelled off at values just below 1 at high vapour pressure deficits. The higher Bowen ratio at Hisåsen during high vapour pressure deficit might be caused by a higher cover of vascular plants. Vascular plants close their stomata during periods of high vapour pressure deficit to prevent excessive water loss , resulting in reduced transpiration rates. At Finse, Iškoras, the footprints of the eddy covariance measurements have a higher percentage than Hisåsen of open water surfaces and non-vascular vegetation where the soil remains saturated most of the year (Fig. ), and evaporation may continue to increase in response to the increased atmospheric evaporative demand as vapour pressure deficit increases. found a varying response to increasing evaporative demand based on vegetation type, and showed that evaporation from boreal peatlands increased more than evaporation from boreal forests with increasing vapour pressure deficit. At Hisåsen, a few trees are present in the footprint, however it is unclear how they contribute to the total evaporation measured. found that overstory transpiration contributed less than 1 % to total evaporation of a boreal bog with sparse tree density in north-western Canada.

Our results indicate that evaporation during the snow-free season has a low sensitivity to surface conditions, apart from the effect of the surface on available energy. Overall, the evaporation dynamic in the snow-free season was well represented by the Penman-Monteith equation with a fixed site-specific surface conductance. We found no tendency for the relative error of the Penman-Monteith estimates or Bowen ratio to change with decreasing soil moisture content or time since rain (Figs. , and ), indicating that evaporation was not constrained by low soil moisture content at our sites. The tendency of larger relative error (model overestimation) at high soil water content at Finse was likely because the highest levels of soil moisture were only reached during or shortly after rain events, when vapour pressure deficit was very low. The Penman-Monteith model tended to overestimate evaporation during periods of low vapour pressure deficit (Fig. ). Furthermore, we found no change in relative error with growing degree day (Fig. ), indicating that the seasonal vegetation development has limited influence on the total evaporation. However, the transpiration response to leaf phenology and soil moisture may be hard to detect due to a larger soil and free water evaporation in the total evaporation measured. Other studies have found that vegetation type is an important predictor of high-latitude evaporation . The low sensitivity to soil moisture and phenology at our sites agrees with the results of , who studied evaporation in 65 boreal and arctic eddy covariance sites and found that surface conditions exert strong control on the latent heat flux in mature forest, but has less influence in ecosystems with shorter vegetation such as grassland, wetlands, and tundra.

The four sites in this study can all be described as having well-watered soils throughout the year, and the measurement periods did not include substantial dry periods. Therefore, it is likely that the soil moisture content did not decrease to a level where it would restrict evaporation. However, as the surfaces at the sites included are heterogenous, evaporation may have been restricted by lower soil moisture in the drier parts of the footprint although not detectable in the observations due to averaging out by higher evaporation from wetter parts. found low Bowen ratios in wet years after large snowmelts in the Canadian shield subarctic terrain and increased over the growing season. Although this study did not specifically examine how the Bowen ratio changed with time since snowmelt, point measurements of soil water content (available at Hisåsen, Finse and Iškoras) indicated that soil moisture during the snow-free season was influenced more by individual rainfall events than by the time elapsed since snowmelt (Fig. ). At Finse and Iškoras, the highest levels of soil moisture were found during and right after rain events, while at Hisåsen the relation of soil moisture and time since rain was unclear. We also found no clear relationship between years with greater snow accumulation (Fig. ) and level of soil moisture at the start of the snow-free season. Other studies have found contrasting results regarding the sensitivity of evaporation to soil moisture content in northern latitude ecosystems. and found that the Bowen ratio in a high-arctic tundra site and at an Arctic coastal wetland, respectively, decreased with higher soil water content, i.e., latent heat flux was favoured over sensible heat flux when soil water content was high. found soil moisture to be a strong control on interannual variation in evaporation from a deciduous needle leaf forest in eastern Siberia, while found that temporal changes in soil moisture did not affect evaporative fluxes in subarctic deciduous woodland. Transpiration and soil evaporation typically decrease with soil moisture content below a certain threshold .

We found that the presence or absence of snow on the surface had a large effect on evaporation rates, especially during spring, when the atmospheric evaporative demand was high. The amount of accumulated snow during the winter can affect the warm season flux budget. investigated the effect of snow cover duration on annual evaporation, in a study of water and carbon fluxes at Finse. In 2020, a year with one month delay in snow cover melt-out, the annual evaporation was reduced by 50 % compared to 2021 – a year with normal snow cover duration. found that, in a year with above average snow accumulation, accumulated growing season latent heat flux was reduced by 33 % in a high arctic wet fen in Zackenberg, Northeast Greenland, while it increased by 24 % in a nearby dry heath at the expense of sensible heat. We found that that sites with lower interannual variation in snow-cover duration have lower interannual variation in evaporation, and that a longer lasting snow-cover is typically associated with lower annual evaporation.

The dynamics of evaporation in the snow-free season was well represented by the Penman-Monteith equation, despite using a fixed surface conductance parameter. Predicting the magnitude, however, depends on finding a suitable value for surface conductance. The Penman-Monteith equation assumes surface energy balance, i.e. available energy is partitioned into either sensible or latent heat flux, whereas the measured fluxes do not amount to the measured available energy at the sites. Accordingly, the surface conductance parameter acts to compensate for the lack of for energy balance closure. By optimizing the surface conductance parameter using the sum of observed sensible and latent heat as available energy (i.e. forcing energy balance closure), we found higher values of surface conductance, with less variation between sites (0.0028–0.0044 ms-1 compared to 0.000–0.0028 ms-1 when the difference between net radiation and soil heat flux was used as available energy). Lack of energy balance closure is not uncommon for eddy covariance sites, and can be caused by e.g. unmeasured storage terms, large scale exchange processes, and landscape heterogeneity . In the snow-free season, energy balance closure varied from 0.7 at Hisåsen to around 0.5 at Finse, Iškoras and Adventdalen (Tables  and ). When including data for the whole year, the degree of energy balance closure was lower at Finse and Iškoras (Fig. ). This might be partly due the energy used for snow melt, which is not accounted for in the energy balance calculation. As Finse on average accumulated more than the double amount of snow (in mm water equivalent) as the other sites (Fig. ), energy spent on snow melt will have a lower effect on the full year energy balance of the other sites. Additionally, weather conditions associated with low energy balance closure may be more common in winter.

At the four Norwegian eddy covariance sites studied, we found that the mean annual evaporation ranged between 80–208 mm. This is at the lower end and below the range of pan evaporation rates measured at 42 sites in Norway between 1967 and 1972 . The pan evaporation sites were mostly located at lower altitudes with higher temperatures. For the five pan sites with a mean annual temperature in the reference period 1961–1990 below 3 °C (data from MET Norway), annual pan evaporation ranged from 250–315 mm (mean annual temperature was 1.0–2.4 °C).

The magnitudes of annual evaporation at our four eddy covariance sites were within the range found at northern latitude eddy covariance sites in FLUXNET2015 (45–385 mm). Including the FLUXNET2015 sites in the data sample, annual evaporation showed a strong relation to mean temperature in the warm season (May–September). However, all four Norwegian sites ended up below the regression line. Grouping the site ecosystem types into wetland and forest sites did not reveal any pattern that could explain the deviation seen. However, the ecosystem types were not evenly distributed in the temperature range. Another possible explanation for the deviations is the degree of continentality, where sites in more oceanic climates are influenced by maritime moist air masses and have lower vapour pressure deficit for the same temperatures compared to sites in more continental climates.

We found a large variation in the role that the evaporation plays in the vertical water balance. At Finse, the annual evaporation was less than 13 % of the annual precipitation, while at Adventdalen evaporation amounted to more than 28 % of the precipitation. The large variation in the evaporation ratio (evaporation as a fraction of precipitation) was caused by a high variation in annual precipitation across sites with evaporation varying less. The wettest site, Finse, is also the site with the lowest evaporation. The evaporation ratio is thus mainly controlled by precipitation. As the precipitation data used in this study have not been corrected for potential under-catch, the precipitation might be underestimated, especially in the months when precipitation falls as snow. Thus, the estimated evaporation ratios might be overestimated, especially for months in the cold season as well as annual estimates. The evaporation ratio for the warm season is less affected by precipitation under-catch when precipitation falls as rain .

Our results suggest that evaporation from northern latitude wetlands will increase in a warmer climate with the expected reduction in snow-cover duration and increased vapour pressure deficit . We found that site-level variation in annual evaporation was linked to spring snow-cover melt out timing, whereas we did not find any significant difference in the evaporative demand between years for overlapping periods (weeks) of the snow-free season. Although our dataset only included 3–4 years of data per site, it is reasonable to conclude that the snow-cover melt out date has a notable effect on annual evaporation. The mean snow melt-out dates across all sites varied from 23 May to 14 July, a period of the year when incoming solar radiation typically is high. An earlier snow cover melt out date implies more solar radiation will be absorbed by the surface (lower albedo as compared to a snow-covered surface), enhancing an early start of the growing season. The timing of the start of the snow covered season in autumn is less likely to affect the evaporation, as the evaporative demand is typically low this period of the year.

The study finds that low vapour pressure deficit is likely a constraining factor for evaporation at northern latitudes. Thus, it is likely that evaporation will increase with the expected increase in atmospheric demand through an increase in vapour pressure deficit in response to global warming.

The response of the ecosystem evaporation to increased evaporative demand will, however, depend on soil moisture availability . Though we did not observe any soil moisture constrains on evaporation, evaporation was regularly exceeding precipitation in May, June, and July at Adventdalen, which had the lowest precipitation, and was typically around 50 % of precipitation in the warm season. The soil moisture availability therefore depends on precipitation and/or melt water from preceding months. An earlier snow-cover melt-out and increased evaporation in spring, might lead to lowered soil moisture availability later in the season, potentially limiting evaporation in anomaly warm and dry years.