Canopy airspace of riparian forest mitigates soil N2O emission during hot moments

Riparian forests are known as hot spots of N cycling in landscapes and climate warming speeds up the cycle. Here we present results from the rst multi-annual high temporal-frequency study of soil, stem and ecosystem (eddy covariance) uxes of N 2 O from a typical riparian forest in Europe. Hot moments (extreme events of N 2 O emission) last a quarter of the study period but contribute more than a half of soil uxes. For the rst time we demonstrate that high soil emissions of N 2 O do not reach the ecosystem level. During the drought onset, soil N 2 O emission peaks at intermediate soil water content. Rapid water content change is the main determinant of the emissions. The freeze–thaw period is another hot moment. However, according to the eddy covariance measurements the riparian forest is a modest source of N 2 O. We propose photochemical reactions and dissolution in canopy-space water as consumption mechanisms.


Introduction
Forests are important regulators of carbon dioxide (CO 2 ) uxes 1 but their role in regulating other greenhouse gas (GHG) budgets, in particular for nitrous oxide (N 2 O), is still largely unknown 2 . The accelerated increase in atmospheric N 2 O concentrations (from a pre-industrial concentration of 270 ppbv to 328 ppbv in 2016 3 is of concern not only because N 2 O is responsible for approximately 6% of global radiative forcing from anthropogenic GHGs. Its ozone-depletion potential outweighs the sum of emissions from all other ozone-depleting substances controlled by the Montreal Protocol 4 . Thus, N 2 O is the most dangerous stratospheric ozone-layer depleting agent in the 21 st century 5 , and the third most important GHG having a global warming potential 296 times (100-yr lifetime adjustment, with feedbacks) that of CO 2 3 .
Riparian forests provide important ecosystem services 6 . They regulate runoff ow, prevent nonpoint source pollutants from entering water bodies, enhance the in-stream processing of both nonpoint and point source pollutants, create habitats for many species, support landscape connectivity, and serve as recreational and cultural-educational areas 7 . Their multifunctional role reinforces current policy in several countries endorsing riparian forest buffers as best management practice subsidizing riparian reforestation for stream restoration and water quality 8 . One of the main functions of riparian forests in agricultural landscapes is removal of excess nitrate (NO 3 − ) via complete denitri cation 9 , converting NO 3 − to N 2 gas. However, incomplete denitri cation can result in the production of N 2 O, a powerful greenhouse gas. Generally, a variety of nitrogen cycle processes can produce N 2 O 10 but in riparian zones, denitri cation has been found the most important source of N 2 O 11,9 .
Grey alder (Alnus incana (L.) Moench) is a fast-growing tree species typically found in riparian zones, with great potential for short-rotation bioenergy forestry 12 . Their symbiotic dinitrogen (N 2 ) xation ability makes alders important for the regulation of the nitrogen (N) cycle 13,12 . Grey alder forests are widely distributed in Europe and North America 14 ( Supplementary Fig. S1) whereas in Europe they are often dominating communities in riparian zones 15 . In Europe there are 15,000 km 2 of Alnus incana subsp.
N 2 O uxes measured in forest soils [16][17][18][19][20] varied from 0.00054 mg N 2 O-N m -2 h -1 in a birch plantation in China 20 to 0.082 mg N 2 O-N m -2 h -1 in a spruce forest under high N deposition in Germany 18 . In general, temperate forest soils emit 1.57 kg N 2 O ha −1 yr −1 whereas in boreal forests the net emission is four times lower and in tropical forests three times higher than that in temperate forests 21 .
For most of soil N 2 O ux studies, manual chambers have been used, and few investigations are based on automated chambers 19,28 whereas Pihlatie et al 16,29 compared uxes measured by chambers and the eddy covariance (EC) technique.
Hot spots and hot moments (extreme events of emissions) largely determine spatio-temporal variation of N 2 O emissions from soils 30 , and soil water content (SWC) is a leading factor controlling all the hot moments. A SWC value of 50-80% has been shown to be optimal for soil N 2 O emissions in forests on both mineral soils 31 and organic soils 32 . Therefore, depending on the initial moisture, both ooding and drought can induce hot moments in tropical forests 33,34 . Drought has been observed to decrease soil N 2 O emissions and account for soil N 2 O consumption 35,36 . The majority of these studies have been conducted in relatively dry mineral soils and only few focused on wetter conditions of organic soils 17 (Fig. 1). Since during the Freeze-Thaw the soil temperature at 0-10cm was almost constant, near-ground air temperature was more signi cant determinant.
Anomalies from the mean of each hot moment period illustrate the pattern of uxes during the hot moments ( Fig. 2). At the end of the Freeze-Thaw period, the rising SWC driven by snow melt became a leading determinant (Fig. 2). Except the Wet period, during all hot moments only soil N 2 O ux showed signi cant peaks. During the Wet period remarkable increase was observed also in the stem uxes and EC-based ecosystem uxes.   Table 1). We observed a distinctive diurnal pattern with small negative uxes during the morning hours (8)(9)(10)(11)(12) in summer months of both years. We observed no diurnal pattern neither in simultaneous soil uxes nor in EC uxes of autumn, winter or spring months.

Relationships of N 2 O ux and environmental parameters
The main factors related to N 2 O soil uxes in this ecosystem were SWC and soil temperature (Fig. 6a).
Based on the full data set measured during the study period, there was an optimal SWC value of about 0.5 m 3 m -3 (50%) at which the soil ux was the highest (Fig. 6b). The relationship between the soil surface temperature and soil N 2 O ux was more complex showing two peaks: one at 0-4 o C and a second one at 13-14 o C (Fig. 6c). The rst peak corresponds to the Freeze-Thaw period and the second one represents the Dry and Dry-minor hot moments.
During the Dry hot moment, the correlation between the SWC and soil N 2 O emission was very strong showing a clear peak at SWC values between 0.35 and 0.5 % (Fig. 7). N 2 O stem uxes were in uenced by SWC, however, a positive relationship was found only during the Wet period and it was not statistically signi cant. We did not nd any signi cant relationships between ecosystem N 2 O uxes and environmental parameters (air and soil temperature, SWC, wind speed) or gross primary production (GPP) on half-hourly scale. However, a general pattern of weekly average uxes followed that of SWC with the modifying in uence of changes in air temperature (Fig. S4).
Discussion N 2 O emissions from terrestrial ecosystems are always irregular and highly variable at both temporal and spatial scale 10,19,57 . Therefore, consideration of hot moments and hot spots is essential to obtain an adequate long-term account of N 2 O uxes 30 and for inferring the mechanisms that drive these events under eld conditions.
Soil water content was the main factor associated with hot moments of N 2 O emission from soils, and it has been argued that drought and rewetting trigger such events by different mechanisms 58 . Likewise, we propose that hot moments in current study were caused by a variety of different mechanisms. Speci cally, we identi ed the Wet period associated with increasing SWC, Drought periods associated with decreasing SWC and the Freeze-Thaw period based on uctuating ground temperature around 0 o C and slight increases in SWC (Figs. 1 and 2). In all cases, SWC remained above 50%. We found that warmer conditions (Drought in our case) had a greater in uence on emissions than wetter conditions (Wet), however, the combined effect of wetter and warmer conditions would be more offsetting than synergetic ( Fig. 6) Analogous results were observed by Shrestha & Wang 37 . Likewise, a climate manipulation experiment in a post-harvest forest showed that wetting increased soil N 2 O ux but not when combined with heating 59 .
In our study, hot moments Dry and Freeze-Thaw contributed correspondingly 9 and 8-10 times higher soil N 2 O emission than the period average ( Fig. 2 end of Dry period) with short-term emission peaks caused by precipitation (rewetting; Dry-minor; Fig. 2). In all cases, the highest emissions occurred when the SWC level passed the optimum value around 50%.
Lesser uctuations of N 2 O were observed in the beginning of the Wet period and during the hot moment Dry-minor, when the SWC uctuated between 45 and 50% (Fig. 2).
During the Freeze-Thaw period a different complex of factors caused a substantial increase in N 2 O emissions from soil. Several hypotheses have been posed to explain this phenomenon, the most common of which are: i) freeze-thaw disrupts soil aggregates exposing physically protected organic matter to be rapidly mineralized by microorganisms 40 ; ii) large proportions of microorganisms, ne roots, and mycorrhiza die during the freeze 44 65 ). This was also the case in our investigations (Fig. 2). In addition, we could see that the SWC plays an important role in N 2 O uxes, especially at the end of the Freeze-Thaw period (Fig. 2) We observed the freeze-thaw effect only in February and March 2019 when the snow cover was sporadic and thin, and several open patches were frozen in the nighttime causing slight uctuations in diurnal pattern of N 2 O soil uxes (Fig. S3). In contrast, in February and March 2018, there was a continuous snow cover of 20-30 cm under which the SWC was decreasing from 70% to 45% (Fig. S4), however, no signi cant emission of N 2 O uxes from the soil was observed (Fig. 2 (Fig. S4). During February and March 2018, the decreasing EC N 2 O ux almost perfectly followed the decrease in SWC, and over the longer period from October 2018 until October 2020 the trends were similar (Fig. S4). Due to a lack of comparable studies on EC N 2 O measurements in forest ecosystems, conventional comparison is impossible. However, comparison of uxes from all compartments measured may provide valuable information. The cumulative emission from ecosystem (87.3 mg N 2 O-N m -2 ) was 5.3 times smaller than the emission from the soil (458.8 mg N 2 O-N m -2 ; Table 1), whereas the difference between these two sources was constantly increasing throughout the whole study period (Fig. 4). During the 1.5-year period the cumulative ux from alder stems was 3.53 mg N 2 O-N m -2 , which constituted only 0.77% of cumulative soil uxes (Table 1). This is about the same level with the ndings by Wen et al 2017 50 on A. glutinosa trees (1.1%).
Since the measurement frequency was su cient to capture all the uxes it is di cult to explain such a great difference between the soil and ecosystem level. During the Drought Onset episode in 2019, N 2 O concentrations increased at wind-still nights (Fig. S5) and uxes declined in the subsequent clear sunny days (Fig. S3b). The consumption could be explained by photochemical reactions, which have been observed in both boreal 68 and tropical forests 69 . The monthly sum of sunshine hours during our Drought Onset and Dry-minor hot moments was high (Fig. S6) 72 found that NOx uxes from the forest canopy were smaller than measured soil NOx emissions and referred to the phenomenon as a "canopy reduction factor" which they applied to soil NOx emissions in large-scale models. The interpretation of these differences was a chemical conversion of NOx to other nitrogen oxides within the forest canopy. Fulgham et al 73 report that wet surfaces of leaves/needles and branches in a mixed forest control the vertical exchange of gases (volatile organic acids). Since the exchange velocity of these gases was well correlated with dew point depression (DPD) we compared the monthly average soil and EC uxes of N 2 O with DPD in our forest. We found that during the autumn and winter (except the Freeze-Thaw period) differences between soil and EC uxes were lowest (Fig. S6).
We also checked whether the trees' photosynthetic activity could be related to N 2 O EC uxes but the correlation analysis with gross primary production (GPP; obtained from the EC tower by LiCor system) did not show signi cant correlation neither throughout the whole study period nor at a monthly basis.
Almost all of previous upscaled annual rates of N 2 O exchange between the atmosphere and forest ecosystem were based on soil emission values. Therefore, for the upscaling to annual and hectare level we used both a soil-and canopy-based approach. Based on the soil values, this riparian alder forest emitted on average 2.18 kg N 2 O-N ha -1 y -1 . According to canopy-based calculations, the emission is only 0.26 kg N 2 O-N ha -1 y -1 . Thus, both calculations show that this type of riparian forest is emitting several times less N 2 O than the agricultural areas 4 or drained N-rich peatlands 32 .
Upscaling these values to the whole Alnus incana subsp incana distribution area (15,000 km 2 ) 14 , we estimate the total annual emissions of 3,270 (soil-based) or 390 tons (canopy-based) of N 2 O-N a year.
Thus, in addition to several ecosystem services which riparian alder forests can provide, they are low emitters of N 2 O which make them attractive for riparian zone management

Conclusions
The outcomes of our long-term study support our hypothesis that this alder forest is a net source of N 2 O.
The second hypothesis on the role of hot moments in long-term pattern of N 2 O uxes was also supported -hot moments contributed about 56% of soil emissions throughout the whole study period. The third hypothesis was not supported -ecosystem (eddy covariance) ux was not coherent with the soil + stem uxes. The stem ux was almost close to zero showing only some increase during the Wet period. In comparison to high soil N 2 O emission, the ecosystem level emission was about 5.3 times lower.
Photochemical reactions and dissolution in atmospheric water may be the consumption mechanisms behind that.
As hypothesized, soil N 2 O ux peaked at 50% of SWC whereas during the Drought Onset the correlation was strong and N 2 O ux mainly depended on speed of SWC change. During Freeze-Thaw, near-surface air temperature was the main factor of N 2 O soil ux.
In the next decades we anticipate a global increase in frequency of disturbances causing hot moments of greenhouse gas emissions in terrestrial ecosystems. Our study reveals the importance of high-frequency eld measurements across the year. Full understanding of nitrogen budgets of riparian forests cannot rely on soil level measurements only but must be combined with tree-stem, canopy and ecosystem-level EC uxes. Identi cation of microorganisms and biogeochemical pathways associated with N 2 O production and consumption is another future challenge.

Study site and set-up
The studied hemiboreal riparian forest is a 40-year old Filipendula type grey alder (Alnus incana (L.) Moench) forest stand grown on a former agricultural agricultural land. It is situated in the Agali Village  (Table S2).
The long-term average annual precipitation of the region is 650 mm, and the average temperature is 17.0°C in July and -6.7 °C in January. The duration of the growing season is typically 175-180 days from mid-April to October 75 . The

Soil ux measurements
Soil uxes were measured using 12 automatic dynamic chambers located close to each studied tree and installed in June 2017. The chambers were made from polymethyl methacrylate (Plexiglas) covered with non-transparent plastic lm. Each soil chamber (volume of 0.032 m³) covered a 0.16 m² soil surface. To avoid strati cation of gas inside the chamber, air with a constant ow rate of 1.8 L min -1 was circulated within a closed loop between the chamber and gas analyzer unit during the measurements by a diaphragm pump. The air sample was taken from the top of the chamber headspace and pumped back by distributing it to each side of the chamber. For the measurements, the soil chambers were closed automatically for 9 minutes each. Flushing time of the whole system with ambient air between measurement periods was 1 minute. Thus, there were approximately 12 measurements per chamber per day. A Picarro G2508 (Picarro Inc., Santa Clara, CA, USA) gas analyzer using cavity ring-down spectroscopy (CRDS) technology was used to monitor N 2 O gas concentrations in the frequency of approximately 1.17 measurements per second. The chambers were connected to the gas analyzer using a multiplexer.
Since the 9 minutes of closing each soil chamber for measurements consisted of two minutes for stabilization the trend in the beginning and about two minutes unstable uctuations at the end, for soil ux calculations, only 5 minutes of the linear trend of N 2 O concentration change has been used for soil ux calculations.
After the quality checking 105,830 ux values (98.7% of total possible) of soil N 2 O uxes could be used during the whole study period.

Stem ux measurements
The tree stem uxes were measured manually with frequency 1-2 times per week from September 2017 until December 2018. Twelve representative mature grey alder trees were selected for stem ux measurements and equipped with static closed tree stem chamber systems for stem ux measurements 49 . Soil uxes were investigated close to each selected tree. The tree chambers were installed in June 2017 in following order: at the bottom part of the tree stem (approximately 10 cm above the soil) and at 80  Stem uxes were quanti ed on a linear approach according to change of N 2 O concentrations in the chamber headspace over time. A data quality control was applied based on R 2 values of linear t for CO 2 measurements. When the R 2 value for CO 2 e ux was above 0.9, the conditions inside the chamber were applicable, and the calculations for N 2 O gases were also accepted in spite of their R 2 values.
To compare the contribution of soil and stems, the stem uxes were upscaled to hectare of ground area based on average stem diameter, tree height, stem surface area, tree density, and stand basal area estimated for each period. A cylindric shape of tree stem was assumed. To estimate average stem emissions per tree, tted regression curves for different periods were made between the stem emissions and height of the measurements as previously done by Schindler et al 32 .

Eddy covariance instrumentation
Eddy analyser. Time lags were detected using covariance maximisation in a given time window (5±2s was chosen based on the tube length and ow rate). While WPL-correction is typically performed for the closed-path systems, we did not apply it as water correction was already performed by the Aerodyne and the software reported mixing ratios. Both low and high frequency spectral corrections were applied using fully analytic corrections 80,81 .
Calculated uxes were ltered out in case they were coming from the half-hour averaging periods with at least one of the following criteria: more than 1000 spikes, half-hourly averaged mixing ratio out of range (300-350 ppb), quality control (QC) ags higher than 7 8 2.
Footprint area was estimated using Kljun et al 83 implemented in TOVI software (Li-Cor Inc.). Footprint allocation tool was implemented to ag the non-forested areas within the 90% cumulative footprint and uxes appointed to these areas were removed from the further analysis.
Storage uxes were estimated using point concentration measurements from the eddy system, assuming the uniform change within the air column under the tower during every 30 min period (calculated in EddyPro software). In the absence of a better estimate or pro le measurements, these estimates were used to correct for storage change. Total ux values that were higher than eight times the standard deviation were additionally ltered out (following Wang et al., 2013 84 ). Overall, the quality control procedures resulted in 61% data coverage.
While friction velocity (u*) threshold is used to lter eddy uxes of CO 2 85 , visual inspection of the friction velocity in uence on N 2 O uxes demonstrated no effect. Thus, we decided not to apply it, taking into account that 1-9 QC ag system already marks the times when the turbulence is not su cient.
To obtain the continuous time-series and to enable the comparison to chamber estimates over hourly time scales, gap-lling of N 2 O uxes was performed using marginal distribution sampling method implemented in ReddyProcWeb online tool (https://www.bgcjena.mpg.de/bgi/index.php/Services/REddyProcWeb) (described in detail in Wutzler et al 86 . MATLAB (ver. 2018a-b, Mathworks Inc., Natick, MA, USA) was used for all the eddy uxes data analysis.

Ancillary measurements
Air temperature and relative humidity were measured within the canopy at 10m height using the HC2A-S3 -Standard Meteo Probe / RS24T (Rotronic AG, Bassersdorf, Switzerland) and Campbell CR100 data logger (Campbell Scienti c Inc., Logan, UT, USA). Based on these data, dew point depression was calculated to characterise chance of fog formation within the canopy. The incoming solar radiation data were obtained from the SMEAR Estonia station located at 2 km from the study site 87 using the Delta-T-SPN-1 sunshine pyranometer (Delta-T Devices Ltd., Cambridge, UK). The cloudiness ratio was calculated based on radiation data.
Near-ground air temperature, soil temperature (Campbell Scienti c Inc.) and soil water content sensors (ML3 ThetaProbe, Delta-T Devices, Burwell, Cambridge, UK) were installed directly on the ground and 0-10 cm soil depth close to the studied tree spots. During six campaigns from August to November 2017 composite topsoil samples were taken with a soil corer from a depth of 0-10 cm for physical and chemical analysis using standard methods 88