Edinburgh Research Explorer Complex controls on nitrous oxide flux across a large elevation gradient in the tropical Peruvian Andes

Current bottom-up process models suggest that montane tropical ecosystems are weak 18 atmospheric sources of N 2 O, although recent empirical studies from the southern Peruvian 19 Andes have challenged this idea. Here we report N 2 O flux from combined field and 20 laboratory experiments that investigated the process-based controls on N 2 O flux from 21 montane ecosystems across a large elevation gradient (600-3700 m a.s.l.) in the southern 22 Peruvian Andes. Nitrous oxide flux and environmental variables were quantified in four 23 major habitats (premontane forest, lower montane forest, upper montane forest and 24 montane grassland) at monthly intervals over a 30-month period from January 2011 to June 25 2013. The role of soil moisture content in regulating N 2 O flux was investigated through a 26 manipulative, laboratory-based 15 N-tracer experiment. The role of substrate availability 27 (labile organic matter, NO 3- ) in regulating N 2 O flux was examined through a field-based litter- 28 fall manipulation experiment and a laboratory-based 15 N-NO 3- addition study, respectively. 29 Ecosystems in this region were net atmospheric sources of N 2 O, with an unweighted mean 30 flux of 0.27 ± 0.07 mg N-N 2 O m -2 d -1 . Weighted extrapolations, which accounted for 31 differences in land surface area among habitats and variations in flux between seasons, 32 predicted a mean annual flux of 1.27 ± 0.33 kg N 2 O-N ha -1 year -1 . Nitrous oxide flux was 33 greatest from premontane forest, with an unweighted mean flux of 0.75 ± 0.18 mg N-N 2 O m - 34 2 d -1 , translating to a weighted annual flux of 0.66 ± 0.16 kg N 2 O-N ha -1 year -1 . In contrast, 35 N 2 O flux was significantly lower in other habitats. The unweighted mean fluxes for lower 36 montane forest, montane grasslands, and upper montane forest were 0.46 ± 0.24 mg N-N 2 O 37 m -2 d -1 , 0.07 ± 0.08 mg N-N 2 O m -2 d -1 , and 0.04 ± 0.07 mg N-N 2 O m -2 d 30 % WFPS. This trend in N 2 O flux suggests a complex relationship between WFPS and nitrate-reducing processes (i.e. denitrification, dissimilatory nitrate reduction to ammonium). Changes in labile organic matter inputs, through the manipulation of leaf litter- fall, did not alter N 2 O flux. Comprehensive analysis of field and laboratory data demonstrated that variations in NO 3- availability strongly constrained N 2 O flux. Habitat – a 53 proxy for NO 3- availability under field conditions – was the best predictor for N 2 O flux, with 54 N-rich habitats (premontane forest, lower montane forest) showing significantly higher N 2 O 55 flux than N-poor habitats (upper montane forest, montane grassland). Yet, N 2 O flux did not 56 respond to short-term changes in NO 3- concentration.


Litter-fall manipulation experiments 294
We conducted a field-based litter-fall manipulation experiment to test for the effects of 295 variations in labile organic matter availability on trace gas flux. This study took place over a 296 14-month period (April 2012 to June 2013), and consisted of 4 experimental treatments 297 (control, +50 % litter addition, +100 % litter addition, litter removal) implemented across 3 298 habitats (premontane forest, lower montane forest, upper montane forest), with 6 replicate 299 plots per treatment per habitat (each treatment plot was 0.5 x 0.5 m in size; n = 24 300 observations per habitat; n = 72 observations per sampling increment). Leaf litter addition 301 rates for the +50 % and +100 % litter addition treatments were determined based on prior research from this study site, and fell within the natural range of variability observed across 303 this elevational gradient (Girardin et al., 2010). 304 inputs simply reflected natural background litter-fall rates. For the +50 % and +100 % litter 309 addition treatments, background litter inputs were supplemented with additional litter 310 taken from the litter baskets. Briefly, wet litter was weighed in the field using portable scale, 311 gently mixed (homogenized), and then re-distributed to the +50 % and +100 % litter addition 312 plots in amounts proportional to the average amount of wet litter that fell into the litter 313 baskets over the course of the month. As a consequence, the amount of litter added in the 314 two litter addition treatments was not fixed but varied according to the natural background 315 rate of litter-fall. For the litter removal treatment, leaf litter was removed from the forest 316 floor at the start of the experiment, and 3mm nylon mesh was placed over the surface of the 317 treatment plot to prevent further litter ingress to the soil surface. Any debris accumulating 318 on the mesh was removed at monthly intervals. 319 320 Trace gas flux and environmental variables were determined at 7 time points over the 321 course of the 14-month experiment using the methods described in section 4.2. In addition, 322 soil moisture (WFPS from the 0-10 cm depth), soil temperature (0-10 cm depth), air 323 temperature, soil gas concentrations (O 2 , CH 4 , N 2 O, CO 2 ) from the 0-10 cm and 20-30 cm 324 depths, litter C, and litter N were determined concomitantly. Litter C and N content was 325 determined on a Carlo-Erba NA 2500 elemental analyser (CE Instruments Ltd, Wigan, UK) at 326 the University of Aberdeen. 327 328

Nitrate addition experiment 329
To quantify the effect of NO 3 availability on N 2 O flux, we conducted a 15 N-NO 3 addition 330 experiment. Background concentrations of NO 3 were determined prior to the start of 331 experiment using soil subsamples (n = 5 per elevation), after which the soils from each 332 habitat were divided into three treatment groups, and supplemented with surplus NO 3 -333 which raised these background levels by +50 %, +100 %, and +150 % ( Table 2). The NO 3 -from the organic layer (O horizon; n = 6) and the other from the mineral layer (A horizon; n = 341 6). Soil samples were then shipped to the University of Aberdeen and sampled within one 342 week of arrival. Transport times from Peru to the UK varied between one and two weeks. 343 Five of these soil cores, one for each replicate, were split into four equal parts (3 treatment 344 samples and one control sample) and distributed into 1 L screw top jars (Kilner, UK). A small 345 soil subsample from each core was used to determine WFPS, background NO 3 content 346 (extracted in 100ml 1M KCl for a 10g soil sample prior to the start of the experiment), as well 347 as total C and N content. If necessary, the samples were gravimetrically amended with water 348 until the cores reached 80% WFPS. Soil cores were kept under constant conditions for 3 days 349 before the start of the experiment to minimize the effects of changing water content on soil 350 processes. 351

352
At the start of the experiment, dissolved 15 N-labelled KNO 3 (30 atom %) was added 353 according to the measured NO 3 concentrations of each core to reach the required NO 3 concentration for each treatment (Table 2). Initial NO 3 concentration (prior to 15 N addition) 355 averaged (± standard error) 157 ± 12 µg N g soil -1 for pre-montane forest, 140 ± 12 µg N g 356 soil -1 for lower montane forest, 19 ± 7 µg N g soil -1 for upper montane forest organic layer 357 soil, 18 ± 5 µg N g soil -1 for upper montane forest mineral layer soil, and 6 ± 2 µg N g soil -1 for 358 montane grassland soil ( Table 2). The jars were then sealed with lids fitted with a two-way 359 stopcock to allow for gas sampling. Gas samples were taken with gas tight syringes, and 360 stored in pre-evacuated containers for determination of 15 N-N 2 , 15 N-N 2 O, N 2 O, CO 2 and CH 4 361 content. Isotope samples (150 ml) were stored in 100 mL serum bottles and gas 362 concentration samples (20 ml) were stored in 12 ml Exetainers® (Labco Ltd., Lampeter, UK). 363 After gas sampling, the stopcock was opened to allow the sampled air from the jar to be 364 replaced by lab air, and lab air was sampled to allow for correction of the gas concentrations 365 in the jars due to dilution. Samples were taken at 0, 6, 12, 24, 36, and 48 hours, after which the jars were opened and soil was sampled for determination of NO 3 -, NH 4 + and total C and 367 N. Gas flux, isotopic and elemental concentrations were determined according to the 368 methods described previously. 369

Variations in N 2 O flux among habitats and between seasons 391
The overall mean N 2 O flux for the entire dataset was 0.27 ± 0.07 mg N-N 2 O m -2 d -1 , with a 392 range from -8.40 to 75.0 mg N-N 2 O m -2 d -1 . We investigated the effect of habitat, season, 393 topography, and the interaction of habitat by season on N 2 O flux by using a three-way 394 ANOVA on plot-averaged data (F 10,307 = 3.28, P < 0.0005; Supplementary Online Materials 395 Table S1A). We found that there was a significant effect of habitat (P < 0.003) and an effect 396 of season at the borderline of statistical significance (P < 0.07). However, we found no effect varied significantly between seasons (t-Test, P < 0.05), with a mean dry season value of 52.1 447 ± 2.4 % compared to a mean wet season value of 59.5 ± 1.6 % ( Table 3). The significant 448 habitat by season interaction is due to the fact that some habitats showed seasonal trends in 449 WFPS whereas others did not. Whereas lower montane and upper montane forests all 450 showed a significant reduction in WFPS during the dry season, premontane forest and 451 montane grasslands showed no seasonal differences in WFPS (Table 3, Figure 3). For 452 topography, the main effect was that the basin landform had significantly higher WFPS than 453 the other landforms. The basin landform showed a mean WFPS of 89.3 ± 0.1 % whereas 454 WFPS in other landforms ranged from 51.7 ± 2.2 to 57.7 ± 2.7 %. 455 456 Soil oxygen in the 0-10 cm depth varied significantly as a function of habitat, habitat by 457 season, and topography (F 10,242 = 27.70, P < 0.0001; Table 3; Supplementary Online Materials 458 Table S1C). Habitat accounted for the largest proportion of variance in the model (66.9 % of 459 the total variance), followed by topography (8.4 %), habitat by season (3.5 %) 460 Table S1C). For habitat, multiple comparisons tests grassland showed a significant difference in O 2 content between wet and dry season, 469

Effects of environmental variables on N 2 O flux 522
For the whole dataset, the relationship between N 2 O flux and environmental variables was 523 examined using an ANCOVA on Box-Cox transformed data with habitat, season, topography, 524 and environmental variables as covariates. Environmental variables included WFPS, oxygen, 525 air temperature, soil temperature, and resin-extractable inorganic N flux (NH 4 + and NO 3 -).

Litter manipulation experiment 645
In order to investigate the relationship between leaf litter input rates and N 2 O flux, we used 646 a Generalized Linear Model (GLM) and an ANCOVA that included habitat, litter treatment, 647 season, WFPS, litter input rate, litter C input rate, litter N input rate, soil temperature and air 648 temperature as independent variables. The analysis was also repeated using ANCOVA on 649 Box-Cox transformed data. Both analyses revealed no significant statistical relationship 650 between N 2 O flux and any of these environmental variables, with the exception of soil 651 temperature, which showed only a weak positive relationship to N 2 O flux when the data was 652 analysed using the GLM (P < 0.05). This relationship was not detected using ANCOVA. varied significantly between seasons in the dataset as a whole, the absolute difference in 712 WFPS between dry season and wet season were relatively small (i.e. 7.4 %). Indeed, some 713 habitats showed much smaller variations in soil moisture, such as premontane forest and 714 montane grassland that showed no significant seasonal variation in WFPS whatsoever (Table  715 3). 716 One critical factor contributing to these weak seasonal trends in N 2 O flux is the atypical 718 response of N 2 O flux to changes in soil moisture. Nitrous oxide flux showed a weak but 719 negative correlation with WFPS in the field dataset (r 2 = 0.01, P <0.06 for the pooled dataset), 720 rather than following a curvilinear pattern predicted by denitrification theory (Firestone and 721 Davidson, 1989;Firestone et al., 1980;Weier et al., 1993;Davidson, 1991)  Collectively, these findings suggest that the role of soil moisture in regulating N 2 O flux is 731 more complex than predicted by existing theory, falsifying our first two hypotheses. 732 What could explain these unexpected trends? We believe that these patterns occurred due 734 to the complex interplay between environmental conditions and the microbial processes 735 that produce N 2 O in soil (i.e. ammonia oxidation by archaea, ammonia oxidation by bacteria, 736 denitrification, dissimilatory nitrate reduction to ammonium). We suspect that the action of 737 lesser-known microbial processes, such as oxidation of ammonia by archaea and 738 dissimilatory nitrate reduction to ammonium (DNRA), may explain the divergence from 739 theoretical norms. Our expectations of how N 2 O production should respond to variations in 740 soil moisture are predicated on the assumption that N 2 O is produced almost exclusively by 741 AOB and denitrifying bacteria, with the former operating at lower soil moisture content (i.e. 742 30-60 % WFPS) and the latter at higher soil moisture content (i.e. >60 % WFPS) (Firestone 743 and Davidson, 1989;Firestone et al., 1980;Weier et al., 1993;Davidson, 1991). More recent 744 advances in soil N research, however, have highlighted the importance of other microbial 745 taxa or processes, not previously considered in conceptual or process-based models. exercise to more fully account for these two sources of variation (Table 4) (Table 4). Next, we multiplied the unweighted seasonal 862 mean flux by the areal fraction for each habitat to derive area-weighted seasonal flux 863 estimates (Table 4). We subsequently multiplied the area-weighted seasonal flux by the 864 fraction of the year accounted for by either season, in order to produce an area-weighted 865 and seasonally-weighted annual flux estimate for each habitat (Table 4). The final step of this 866 process was to sum the area-weighted and seasonally-weighted flux estimates for each 867 habitat, to drive an overall weighted flux estimate for the Kosñipata Valley as a whole (Table  868 4). Weighted annual estimates of N 2 flux were calculated using the N 2 O yield values for each 869 habitat as determined in our soil moisture manipulation experiment (   Habitat -a proxy for NO 3 availability under field conditions -was the best predictor for N 2 O 900 flux, with more N-rich habitats (i.e. premontane forest) showing significantly higher N 2 O flux 901 than habitats with lower N availability (i.e. upper montane forest, montane grassland). 902 Nitrous oxide flux originated primarily from nitrate reduction rather than from nitrification, 903 probably due to low pH soil conditions which may have inhibited the activity of AOB. 904 Contrary to our prior research, we found only weak evidence for seasonal trends in field N 2 O 905 flux, with the exception of lower montane forest, which showed significantly higher N 2 O flux 906 during the dry season compared to the wet season. Weak seasonal trends in field N 2 O flux 907 among the other montane habitats probably stems from relatively modest seasonal 908 variation in key environmental drivers (e.g. temperature, WFPS, NO 3 -), combined with a soil moisture response that was complex and non-linear. Nitrous oxide flux was significantly 910 influenced by soil moisture content, but the trends in N 2 O production and flux diverged from 911 theoretical norms. For example, we saw little evidence of N 2 O production from ammonia-912 oxidation, even though the field measurement (i.e. resin bags) indicate that nitrification 913 occurs. This may be due to the predominance of AOA, which produce significantly N 2 O than 914     shown in dark-grey, while data from the +100 % and +150 % treatments are shown in mid-