The study was carried out at a 12 ha unmanaged drained grassland site at the Lækur Farm in Western Iceland (64.39° N, 21.89° W), hereafter termed the LD site. The site was artificially drained in 1961 through a network of ca. 2 m deep ditches spaced roughly 52 m apart and by putting in subterranean channels at ca. 60 cm depth between the ditches. Despite the drainage effort, the land has never been cultivated or fertilized, allowing natural vegetation succession under altered hydrological conditions. This has resulted in a dense, highly productive grass cover. There was some livestock grazing at the site prior to 1997, but since then it has remained protected. The terrain is predominantly flat, with subtle hummocks and the site drains southeast and southwest toward the ocean, which is > 120 m away from the lowest edge of the LD site (Fig. 1). The peat soil of the site is classified as Histosols, with a bulk density of 0.2 gcm-3, a pH of 4.6 and a C/N ratio of 20.36 in the 0–50 cm layer.

The climate is cool maritime, with a mean annual temperature (MAT) of +4.8 °C and mean annual precipitation (MAP) of 1011 mm, based on 2023 and 2024 data from the nearest weather station to the site (Icelandic Meteorological office, 2025). Seasonal minimum and maximum temperatures ranged from -14 °C in winter to 20 °C in summer during the same period. Prevailing winds originate mainly from the east and northeast, averaging 4 ms-1 (Table 1). The uniform microtopography of the site likely supports a relatively consistent microclimate.

The current vegetation at the site is classified as a Boreo-subalpine Agrostis grassland habitat type (Ottósson et al., 2017), with common bentgrass (Agrostis capillaris) and Arctic fescue (Festuca richardsonii) as dominating species. The peak season's plant canopy height is 20–50 cm and maximum annual aboveground biomass is ∼ 720 gm-2 (Table 1).

EC data were processed using EddyPro software (v7.0.9, LI-COR Biosciences, Lincoln, NE, USA), following standard protocols for open-path sensor configurations. The processing workflow included angle-of-attack correction, double coordinate rotation, and time lag compensation. Spectral losses were compensated following Moncrieff et al. (1997). To adjust for fluctuations in air temperature and water vapor, Webb–Pearman–Leuning (WPL) density correction was applied (Webb et al., 1980). To account for instrument-related density effects, the Burba et al. (2008) correction was applied to the entire 2023–2025 dataset. Owing to the high-latitude, cool-climate setting of the Icelandic site, a temperature gradient between the ambient air and the LI-7500DS sensor head was expected to persist for much of the year. Applying the correction consistently across all seasons ensured a uniform processing workflow and avoided introducing artificial discontinuities at seasonal transitions. The correction magnitude scaled with the observed air-instrument temperature gradient and naturally became negligible during warmer periods.

To account for CO2 accumulation below the measurement height, a storage flux correction was applied following the approach of Aubinet et al. (2001). Although the EC system height was increased from 2.15 to 3.5 m during the study period, data comparability was maintained through consistent storage corrections and the site's high degree of topographical and vegetative homogeneity.

The EC system data was integrated with the micro-meteorological sensors data (Table 2) and automatically time-aligned via the EddyPro software, enabling the computation of half-hourly averages for NEE, H, and LE. This integration also provided the metadata and diagnostic quality flags required for subsequent quality assurance and data analysis.

Following initial processing, a multi-step quality control (QC) procedure was implemented. Outliers were removed through a combination of statistical screening and visual inspection. The dataset was filtered using the turbulence and stationarity classification scheme of Mauder and Foken (2011); only data with high (0) or moderate (1) quality flags were retained. Data periods affected by instrument malfunction or signal loss were also excluded. u∗ thresholds were determined using the moving point test implemented in the REddyProc R package (Wutzler et al., 2018). Thresholds of 0.26 and 0.29 ms-1 were applied to the 2023 and 2024 datasets, respectively, to minimize underestimation of nocturnal fluxes (Papale et al., 2006). An additional filtering step was applied specifically to CO2 fluxes. Data collected during precipitation events, dew formation, or low vapor pressure deficit (VPD) were excluded, as these conditions interfere with open-path analyser performance and can yield biologically unrealistic values (Wohlfahrt et al., 2005; Aubinet et al., 2012).

Energy balance closure (EBC) was evaluated by comparing the sum of the turbulent energy fluxes to the available energy (Foken, 2008). To account for energy temporarily stored within the ecosystem, storage terms were included in the energy balance equation: 2H+LE=Rn-G-S,

Where H is the sensible heat flux (Wm-1), LE is the latent heat flux (Wm-2), Rn is the net radiation (Wm-2), G is the ground heat flux (Wm-2), and S represents the sum of heat storage terms, including storage in the air column (Sa) and the soil layer above the heat flux plates (Sg).

The spatial representativeness of the EC fluxes was assessed using the Flux Footprint Prediction (FFP) model of Kljun et al. (2015) using Tovi software (v2.9.1, LI-COR Biosciences, Lincoln, NE, USA). Footprints were computed at 30 min resolution using measured u∗, Obukhov length (L), wind direction, and lateral wind variability (σv), with constant surface parameters of roughness length (z0 = 0.035 m) and displacement height (d = 0.23 m).

The high-resolution DSM with 7 cm ground sampling distance shows that the site is predominantly flat, with most of the area situated around 7 ma.s.l. Elevation ranged from 1 to 11 m, with the lowest point along the eastern boundary and the highest in the northwest, though these extremes fell largely outside the main flux footprint (Fig. 2a).

In 2023, when the measurement height was 2.15 m, the 80 % footprint averaged 50–70 m around the tower, which was smaller than expected due to the rough, hummocky surface (Fig. 2a). After the measurement height was increased to 3.5 m in 2024, the 80 % EC flux footprint extended on average 100–120 m around the tower (Fig. 2a). The 80 % footprint remained within the main study area for both years and covered relatively uniform terrain with less than 2 m elevation variation and similar vegetation composition. The site was characterized by predominantly easterly winds and high u∗ (mostly above 0.2 ms-1), indicating that mechanical turbulence dominated and atmospheric conditions were mostly neutral.

Within the footprint, subtle microtopographic features were present, including hummocks and drainage ditches with their associated spoil ridges. Surface classification indicated that ∼ 3.5 % of the footprint consisted of ditches, 9.5 % of excavated peat soil, and the remaining ∼ 87 % was the main drained unmanaged peatland grassland field (Fig. 2a).

Measured peat depths across the 46 sampling points ranged from 120 to 387 cm. Spatially, deeper peat deposits were primarily concentrated in the central and eastern parts of the site. In contrast, shallower regions were typically located near the northern and southern site boundaries. The average depth recorded from field measurements was 235 cm, while the mean depth derived from the generated peat depth map was 237 cm (Fig. 2b).

Temporal trends in NDVI followed a clear seasonal pattern, with values peaking in mid-summer (July–August), reflecting maximum canopy development and light interception. NDVI values were lowest in early spring, late autumn and winter corresponding to the dormant phases and the deciduous character of the vegetation and relative lack of moss cover in the fertile grassland (Fig. 2c).

The seasonality of key environmental drivers, including Ta, Ts (at 10 cm below the surface), PAR, VPD, Rain, SWC and WL, was analyzed to provide essential context for interpreting the ecosystem CO2 fluxes (Fig. 3).

The GS began in early May and extended into early October in 2023 (3 May–5 October), reflecting a relatively long period of active plant growth. In 2024 the GS was noticeably shorter; it started later and ended earlier, covering the period from 19 May to 22 September (Fig. 3; vertical red dashed lines). This corresponds to the length of the GS being 45 % of the year in 2023 and 35 % in 2024.

Mean Ta during GS was slightly higher in 2023 (9.8 °C) than in 2024 (9.5 °C), with similar patterns during the NGS with mean Ta of 1.7 °C in 2023 and 1.6 °C in 2024. Ts followed similar trends, remaining low during NGS (mostly close to 0 °C due to freezing of surface peat layers) and reaching GS means of 10.6 °C (2023) and 10.4 °C (2024) (Fig. 3a).

PAR showed the strongest seasonal signal (Fig. 3b), with GS means of 302 µmolm-2s-1 in 2023 and 281 in 2024, while NGS values remained low. VPD remained relatively low in both years (with GS mean of 5.8 hPa in 2023 and 5.1 hPa in 2024 and even lower values during NGS), reflecting the humid maritime climate, but showed slightly stronger seasonality in 2023 (Fig. 3c).

Total annual precipitation was higher in 2024 (950 mm) than in 2023 (823 mm), an increase of roughly 15 % (Fig. 3d). During the GS, rainfall increased from 328 mm in 2023 to 391 mm in 2024, with 2023 characterized by a pronounced dry period in July–August, whereas precipitation in 2024 was more evenly distributed.

SWC reflected the precipitation pattern, with reduced values during the dry period of 2023 and consistently higher, more stable levels (∼ 0.4 m3m-3) during GS 2024. Lower winter SWC in both years was associated with soil freezing, consistent with Ts values near 0 °C.

The WL fluctuated between approximately -0.5 and -0.9 m through most of the study period, close to the subterranean channel depth of -0.6 m (Fig. 3f). During GS 2023 the mean WL was -0.88 m, while GS 2024 showed a shallower mean of -0.73 m, indicating that the GS WL was on average about 17 % higher in 2024. The deepest drawdowns occurred in late GS of 2023, reaching approximately -1.1 m, while the deepest drawdowns in 2024 was only -0.86 m.

Overall, 2023 was warmer and drier compared to 2024, which was colder and had more precipitation especially during GS.

In EBC evaluation, linear regression of half-hourly H + LE against Rn - G yielded a slope of 0.79 and an intercept of 2.68 Wm-2 in 2023 and a slope of 0.74 and an intercept of 2.25 Wm-2 in 2024, indicating reasonable EBC within the expected range for peatland and wetland ecosystems (Wilson et al., 2002).

The final gap-filled time series (Fig. 4) combined accepted MDS estimates with XGBoost predictions, producing a smooth and internally consistent NEE dataset. Model evaluation demonstrated strong predictive performance, with R2 values of 0.8 for MDS model and 0.91 for XGBoost, and RMSE below 1.36 µmolCO2m-2s-1, highlighting the high level of agreement between predicted and observed fluxes. Based on SHAP values, radiation and thermal variables emerged as the primary contributors, while seasonal and diurnal temporal indicators were additionally used by the model to capture systematic patterns in ecosystem exchange not fully resolved by meteorological drivers alone. A detailed breakdown of XGBoost feature attribution based on SHAP values is provided in Appendix B.

The gap-filled NEE was partitioned into flux components (Reco and GPP). Evaluation metrics showed an acceptable model fit, yielding an R2 of 0.59 and an RMSE of 0.913 CO2m-2s-1 (see Appendix B for the complete SHAP value analysis). These continuous NEE, GPP and Reco data provided a robust foundation for analyzing seasonal dynamics and annual carbon budgets.

As expected, the average diurnal cycle of CO2 fluxes showed a clear seasonal contrast, reflecting the interplay between seasonal light availability, temperature, and phenological state (Fig. 5). During GS, the ecosystem acted as a strong daytime carbon sink, with NEE and GPP exhibiting synchronized diurnal patterns. Peak GPP uptake and maximum NEE sequestration occurred at 13:00 GMT, reaching -12.1 and -6.9 µmolm-2s-1, respectively, close to local solar noon. In contrast, Reco peaked later at 16:00 GMT (5.4 µmolm-2s-1), reflecting thermal inertia and the delayed response to peak air and soil temperatures. The broad “shoulders” of the GPP curve indicate extended photosynthetic activity during long twilight periods at this high-latitude site.

In contrast, during the NGS, photosynthetic activity was strongly suppressed, and the ecosystem functioned as a persistent CO2 source (Fig. 5). GPP remained low, with a maximum uptake of -1.4 µmolm-2s-1 (≈ 10 % of GS peak), while Reco exceeded GPP throughout the day, resulting in consistently positive NEE values, peaking at 1.2 µmolm-2s-1 around midday. Although the diurnal amplitude of Reco was reduced compared to the GS, it still showed a cycle, with maximum values of 2.7 µmolm-2s-1 (∼ 50 % of GS peak). The sustained positive NEE during the extended NGS highlights the significant contribution of winter CO2 emissions to the annual carbon balance.

The daily NEE time series showed that the unmanaged drained grassland acted as a net CO2 source during most of the year (Fig. 6a). During the NGS, NEE remained positive, fluctuating around 2 µmolm-2s-1, indicating continuous CO2 release under the long, cold and light-limited conditions. With the onset of the GS, increasing temperature and PAR drove NEE toward negative values, reflecting net carbon uptake. The strongest sink strength was observed in July (∼ -4 µmolm-2s-1), corresponding to peak photosynthetic activity.

GPP showed particularly strong seasonality (Fig. 6b). Photosynthetic uptake started to increase sharply in May and peaked around -10 µmolm-2s-1 during July. The decline in late September was also rapid, marking the end of active vegetation growth under sharply shortening day length at high latitude.

Reco followed a similarly clear seasonal pattern (Fig. 6c). Reco increased rapidly in late spring (May), reached peak values of 6–7 µmolm-2s-1 around August, and declined gradually during autumn (late September). Winter respiration generally remained relatively high, (∼ 2 µmolm-2s-1), occasionally falling below 1 µmolm-2s-1, but never ceasing entirely.

Despite a mid-season reduction in cumulative NEE reflecting temporary carbon uptake, emissions during the NGS dominated the annual balance (Fig. 6d). The consistently increasing cumulative NEE in both years confirms the system as a persistent net CO2 source, with slightly higher total emissions observed in 2024. Overall, these flux patterns revealed an ecosystem characterized by intense but short summer CO2 uptake on one hand, and continuous and substantial winter respiration on the other hand.

The annual net C-budget (NEE) was surprisingly similar between the 2 years (Table 3), showing the drained unmanaged peatland being a C source to the atmosphere of 4.1 and 4.4 tCO2-Cha-1yr-1 in 2023 and 2024, respectively (approximately an 8 % difference). All the net seasonal CO2-C loss from the unmanaged drained peatland occurred during the NGS, as the GSs were small sinks in both years.

When comparing the seasonal NEE budget between the 2 years, it should be noted that the NGS was shorter in 2023 than in 2024 (55 % versus 65 % of the year, respectively). The NGS NEE-sum was correspondingly lower in 2023 (Table 3). The GS NEE-sum was also less negative in 2023, when the seasonal temperature variation included a larger part of the late season within the defined GS.

The component fluxes (GPP and Reco) provided further insight into these annual patterns. The annual GPP was 11 % higher in 2023, which benefited from a longer GS, but the annual Reco was only 5 % higher in 2023 compared to 2024. The higher GPP/Reco ratio in 2023 than in 2024 resulted in the site being a smaller net CO2 source in 2023 (Table 3).

While most of the GPP fluxes occurred during the defined GS, the Reco showed a more modest intra-annual variability, with an average of 39 % and 50 % of seasonal emissions occurring during the NGS in 2023 and 2024, respectively (Table 3). Consequently, in both years nearly half of the annual respiratory CO2 release occurred outside the GS. The average annual Reco-sum (12.6 tCO2-Cha-1yr-1) was approximately 3 times larger than the average NEE-sum (4.2 tCO2-Cha-1yr-1), showing the relative importance of the annual GPP in the site's C-balance (Table 3).

To explain the relatively small interannual differences in NEE despite contrasting weather conditions (2023 being warmer and drier while 2024 was colder and wetter), weekly aggregated Reco and GPP values were related to measured environmental drivers (Fig. 7). For Ts and WL, the analysis was restricted to observed GS data to ensure comparability between years, as winter data were unavailable for 2023. Conversely, for PAR and NDVI, the full annual dataset was used to capture the complete seasonal development of the radiation-use efficiency of the canopy.

Weekly Reco consistently followed a similar exponential temperature-response curve in both years relative to Ts (Fig. 7a; Q10 values of 2.37 in 2023 and 2.23 in 2024). While the range of weekly temperatures was also relatively similar across both years, 2024 experienced more weeks with lower Ts and fewer peak-temperature weeks compared to 2023 (Fig. 7a). This difference in frequency of high-temperature weeks primarily explained why Reco was significantly higher in 2023 (annual median Reco of 4.087 µmolm-2s-1 in 2023 and 2.751 µmolm-2s-1 in 2024; Mann–Whitney U test, p < 0.0001). Additionally, a significant negative relationship between WL and Reco was observed in 2023 (R2 = 0.40, p = 0.009), whereas no such relationship existed in 2024 (R2 = 0.00, p = 0.934) (Fig. 7b). Consequently, the higher cumulative Reco in 2023 was also driven, to a lesser extent, by the lower WL, which increased the peat volume exposed to aerobic microbial decomposition.

Weekly GPP reached higher peak values in 2023 compared to 2024 (Fig. 7c). PARacc alone explained 72 % of the variation in GPP during 2023 but only 38 % in 2024, where several weeks exhibited a low GPP:PARacc ratio following a delayed start to the GS. However, when GPP was plotted against the PARacc × NDVI product, the relationships for both years converged and explanatory power increased to 86 % and 64 % for 2023 and 2024, respectively (Fig. 7d). 2023 included several weeks with higher PARacc × NDVI values than any observed in 2024. Further relationships between the Reco and GPP with other environmental variables are provided in Appendix C.

The present study site is broadly representative of unmanaged drained peatlands in Iceland, where drainage commonly leads to a shift from wetland vegetation toward graminoid-dominated grassland communities, often influenced by land use such as grazing (Arnalds et al., 2016a). These systems are classified as “Grassland on organic soil” in the Icelandic National Inventory Report (Umhverfisstofnun, 2023) and cover approximately 4200 km2.

The results show that the drained unmanaged peatland remained a consistent net CO2 source despite different environmental conditions between 2023 (warmer and drier) and 2024 (colder and wetter), with a mean annual loss of 4.2 tCO2-Cha-1yr-1 (Table 3). This value was 26 % lower than the IPCC (2014) default boreal EF of 5.7 tCO2-Cha-1yr-1 that is currently used in Iceland's national reporting for unmanaged drained peatlands (Umhverfisstofnun, 2025). It is, however, notably similar to the 3.8 ± 0.65 tCO2-Cha-1yr-1 that was obtained by chamber measurements on unmanaged drained peatland in western Iceland (Ólafsdóttir, 2015). Further, it is a similar value to estimated mean soil organic matter losses of approximately 4 tCO2-Cha-1yr-1 in a regional study using chamber point measurements and NDVI-driven upscaling on similar seven unmanaged drained sites in W-Iceland (Guðmundsson et al., 2026). The obtained CO2-C loss in the present study is, however, higher than the mean annual loss estimated from the peat-stock change method in eight paired undrained and drained site-pairs in southern Iceland, which found on average annual topsoil loss of 1.7 tCO2-Cha-1 during 15–50 years of draining (Gunnarsdóttir, 2017). However, in that study a 1500 AD ash layer at 17–30 cm depth in the peat profile was used for the stock-change estimates, which may have underestimated the total CO2 loss from the whole drained peat profile as the WLs were generally lower.

Despite the similar NEE observed in this study compared to other measurements on drained peatlands in Iceland, the annual GPP (-7.85 to -8.8 tCO2-Cha-1) and Reco (12.2 to 12.7 tCO2-Cha-1) observed here were somewhat higher than most chamber-based estimates from other Icelandic drained sites (GPP: -4.3 to -7.4 tCO2-Cha-1 and Reco: 7.19 to 12.3 tCO2-Cha-1) (Guðmundsson et al., 2026; Ólafsdóttir, 2015), likely reflecting the broader spatial footprint and continuous temporal coverage of the EC method.

One pertinent question is whether the 60-year duration since initial drainage contributes to the lower CO2 emissions observed compared to the IPCC (2014) default EF. There is currently a general lack of long-term monitoring studies evaluating how long high CO2-C emissions are sustained post-drainage (Qiu et al., 2021). Truskavetskii (2014) demonstrated that the loss of peat organic matter in croplands was linear during the first 45 years following drainage, excluding enhanced losses during the initial 7-years period. Studies extending beyond 50 years have primarily focused on how decadal peat subsidence may reduce effective drainage (WL) over time, leading to decreasing rates of organic matter loss as the peat column compacts (Rojstaczer and Deverel, 1993; Liu et al., 2026). Evans et al. (2021) analyzed CO2 flux measurements from 16 EC towers on drained peatlands across the UK and concluded that mean annual effective WL represents the overwhelmingly dominant control on CO2 fluxes.

Given that the subsidence at the present study site was only -0.16 m since the onset of drainage (Streeper, 2026), and the present WL generally remained deep (mainly between -0.5 and -0.7 m), the age of the drainage (60 years) is unlikely to be the primary explanation for the lower CO2-C losses relative to the IPCC (2014) default EF for the boreal zone. Furthermore, both Gunnarsdóttir (2017) and Guðmundsson et al. (2026) investigated multiple drained peatland sites in Iceland using stock change and repeated chamber flux methods, respectively. They observed considerable variation in annual CO2-C losses among sites that was not correlated with the age of drainage. Consequently, it is recommended that for future assessments of unmanaged drained peatlands, sites should be classified by their effective annual WL rather than the time that has elapsed since drainage.

The annual CO2-C balance of the unmanaged drained peatland was strongly influenced by NGS emissions, which accounted for 39 % and 50 % of the annual Reco in 2023 and 2024, respectively. Notably, positive NEE occurred only during the NGS in both years, which turned the study site into an annual C source. For comparison, Alm et al. (1999) found that winter (November-May) CO2-C losses accounted for 23 % and 21 % of annual totals in undisturbed and drained peatlands, respectively, at latitudes of 62–65°N in Finland using a closed chamber method. In another study Tikkasalo et al. (2025) found by EC measurements over a drained peatland forest in southern Finland (latitude of 61°N), that only 28 % of the positive NEE was released during the winter season. Although the exact timing of NGS defined in the present study (October–April) differed from these reference studies (November–May), the duration was the same. This comparison suggests that the proportion of NGS CO2 flux relative to the annual flux is higher in the present study than other studies on drained peatlands at similar latitudes in Fenno-Scandia. Climatic shifts in these continental regions are most evident during the winter, a season that has warmed markedly over the past 30 years and is projected to experience the greatest degree of warming through the end of the century (Lind et al., 2023). Recent evidence suggests that the C balance of drained peatlands in Finland has already shifted toward becoming a larger CO2 source over the last three decades, due to warmer (winter) climates (Alm et al., 2023). In this context, Iceland may serve as a pertinent case study, illustrating how increasingly mild winters are likely to impact the annual C balance of drained peatlands at similar latitudes globally.

The year 2023 was an extreme weather year at the field site, characterized by the coldest winter since 1995, followed by an unusually warm and dry summer (including the driest July since the 1860s) and a warm autumn (Icelandic Meteorological Office, 2025a). The 2024 winter was also among the coldest on record but was followed by an unusually wet and cold summer (the coldest since 1998) and autumn (Icelandic Meteorological Office, 2025b). Despite these contrasting meteorological conditions between 2023 (drier and warmer) and 2024 (wetter and colder), the annual CO2-C balance remained surprisingly similar across both years (Table 3). The stabilizing mechanism that appears to have balanced the enhanced CO2 fluxes during the warm 2023 was a strong positive response in the productivity of the unmanaged grassland vegetation. This was reflected in higher PARacc × NDVI (indicating enhanced canopy development and higher irradiance) and elevated GPP values during a 6-week window of the relatively short GS (Fig. 7d). Although Reco was significantly higher in 2023 due to higher temperatures and reduced precipitation, which led to lower WL and greater peat exposure to aerobic microbial decomposition, the enhanced GPP largely offset this increase. Consequently, the annual NEE remained consistent across both years. High covariance between Reco and GPP which led to small interannual differences in NEE has been observed in other studies (Wohlfahrt et al., 2008). Previous research has shown that the NEE of drained peatlands can be relatively stable (e.g. Wilson et al., 2016), while other studies have highlighted significant interannual variability in NEE (e.g. Bubier et al., 2003). These differences in results often depend on whether interannual driving factors consistently stimulate both C uptake and release, or primarily influence only one side of the C balance.

Unmanaged grasslands at high latitudes, particularly in Iceland, have shown a robust positive response to warming, characterized by significant increases in biomass production and NDVI (Elmendorf et al., 2012; Mortier et al., 2024). This trend has also been noted for the country as a whole as MAT has increased since the 1980s (Raynolds et al., 2015). Such a strong positive relationship between GPP (or NDVI) and temperature is not unexpected at the northern limits of the boreal/Subarctic zone, particularly where water availability or biotic factors, such as high grazing pressure, do not limit aboveground biomass.

The summer drought in 2023 resulted in a significant drawdown of WL; however, no apparent negative drought response was observed in either GPP or Reco. In fact, Reco showed a significantly enhanced response to the lower WL and higher Ts. This aligns with findings from 16 EC towers on drained peatlands across the UK (Evans et al., 2021), which concluded that mean annual effective water-table depth represents the overwhelmingly dominant control on CO2 fluxes in these ecosystems. In the present study, however, the effect of lower WL was relatively small compared to the direct Ts effect on Reco.