Landscape control of nitrous oxide emissions during the transition from conservation reserve program to perennial grasses for bioenergy

Future liquid fuel demand from renewable sources may, in part, be met by converting the seasonally wet portions of the landscape currently managed for soil and water conservation to perennial energy crops. However, this shift may increase nitrous oxide (N2O) emissions, thus limiting the carbon (C) benefits of energy crops. Particularly high emissions may occur during the transition period when the soil is disturbed, plants are establishing, and nitrate and water accumulation may favor emissions. We measured N2O emissions and associated environmental drivers during the transition of perennial grassland in a Conservation Reserve Program (CRP) to switchgrass (Panicum virgatum L.) and Miscanthus x giganteus in the bottom 3‐ha of a watershed in the Ridge and Valley ecoregion of the northeastern United States. Replicated treatments of CRP (unconverted), unfertilized switchgrass (switchgrass), nitrogen (N) fertilized switchgrass (switchgrass‐N), and Miscanthus were randomized in four blocks. Each plot was divided into shoulder, backslope, and footslope positions based on the slope and moisture gradient. Soil N2O flux, soil moisture, and soil mineral nitrogen availability were monitored during the growing season of 2013, the year after the land conversion. Growing season N2O flux showed a significant vegetation‐by‐landscape position interaction (P < 0.009). Switchgrass‐N and Miscanthus treatments had 3 and 6‐times higher cumulative flux respectively than the CRP in the footslope, but at other landscape positions fluxes were similar among land uses. A peak N2O emission event, contributing 26% of the cumulative flux, occurred after a 10.8‐cm of rain during early June. Prolonged subsoil saturation coinciding with high mineral N concentration fueled N2O emission hot spots in the footslopes under energy crops. Our results suggest that mitigating N2O emissions during the transition of CRP to energy crops would mostly require a site‐specific management of the footslopes.


Introduction
Renewable fuels are an increasing portion of the liquid fuels portfolio of the United States and some analyses suggest that to fulfill federal mandates, the nation should produce %80 billion liters of ethanol from cellulosic sources by 2022 (Kim et al., 2009). This will require planting up to 21 million ha with cellulosic crops (McLaughlin et al., 2002) like switchgrass (Panicum virgatum) and Miscanthus (M. 9 giganteus). Land seemingly suitable for energy crops is that currently enrolled in the Conservation Reserve Program (CRP), which amounts to %12 million hectares of agriculturally marginal and environmentally sensitive land (Farm Service Agency, 2012).
Energy crops like switchgrass and Miscanthus can provide crucial ecosystem services when placed in environmentally sensitive lands across the landscape. When planted in floodplains or in lower landscape positions within watersheds, these grasses can serve as riparian buffers to reduce nutrient and sediment load to surface water and groundwater (Perez-Suarez et al., 2014;Smith et al., 2014). Furthermore, due to the low soil disturbance and perennial rooting systems, they can store C in the soil due to their large below-ground C allocation (Lemus & Lal, 2005;Follett et al., 2012).
A critical component of the societal benefit of energy crops is that they produce fuel with a low C footprint (Tilman et al., 2006;Adler et al., 2007;Gelfand et al., 2013). The United States Environmental Protection Agency's revision to the National Renewable Fuel Standard requires that to qualify as renewable, advanced biofuels lifecycle greenhouse gas (GHG) emissions must be 50% of those from fossil fuel (EPA, 2010). To achieve this goal, emissions of nitrous oxide (N 2 O) must be low during the feedstock production phase, because N 2 O is the largest source of GHG for feedstock production (Adler et al., 2012). There are arguments to posit that converting historically managed CRP lands to agricultural crops may increase N 2 O emissions (Ruan & Robertson, 2013).
In the Ridge and Valley physiographic province of the Allegheny Plateau in the northeastern United States, many lands are considered marginal due to being seasonally wet. These landscapes have characteristically shallow, coarse, and rocky ridge top soils as well as fractured subsoils that drain water to the back and footslope positions (Fig. S1). The footslope soils are derived from mixed colluvial sandstone and shale, often with fragipans at shallow depth that limit drainage (Ciolkosz et al., 1995). Restricted drainage in the footslope causes a shallow, temporary water table, and extended periods of soil water content above field capacity or near saturation during spring snowmelt and rainstorm events (Buda et al., 2009). The steep soils with a semi-impermeable subsoil favor nitrate (NO À 3 ) transport to stream water (Kleinman et al., 2006;Zhu et al., 2011). The simultaneous occurrence of well drained upper lands and poorly drained footslope positions may create biogeochemical hot spots for denitrification and N 2 O emissions (Vilain et al., 2010).
The N 2 O emissions from soil predominantly originate from nitrification and denitrification (Firestone & Davidson, 1989). Soil oxygen, NO À 3 , ammonium (NH þ 4 ), and labile organic C concentration determine the contribution of nitrification and denitrification to the total N 2 O flux (Weier et al., 1993;Gillam et al., 2008). We hypothesize that in seasonally wet lands of the northeastern United States, N 2 O emissions could be severe during the transition from CRP to energy crops due to accelerated C and nitrogen (N) cycling of above and below-ground residues from the former CRP vegetation, soil disturbance during the land conversion Nikiema et al., 2012;Ruan & Robertson, 2013), and relatively low crop growth rate in the first two years of stand establishment. Furthermore, if the energy crops are fertilized, co-locating N fertilizer with decomposable organic residues along water flow paths will create conditions for increased N 2 O emissions. Thus, managing N and N 2 O emissions from energy crops in this region requires understanding the interactive controls of landscape, crop growth, and hydrology on soil N cycling.
In this research, we measured soil N 2 O emission during land transition from CRP to energy crops Miscanthus and switchgrass in a watershed in the Ridge and Valley region. The lower part of the watershed was under CRP since 1999. In 2012, we divided the landscape into replicated plots of: (i) unconverted CRP, (ii) Miscanthus, (iii) switchgrass, and (iv) fertilized switchgrass. The N 2 O fluxes from the CRP and establishing switchgrass and Miscanthus were monitored during the 2013 growing season, the second year after land conversion. In June of 2013, 10.8 cm of rain from hurricane 'Andrea' caused sudden soil saturation in parts of the watershed. This event gave us an opportunity to measure the eventbased response of N 2 O emissions during the transition from CRP to energy crops. To the best of our knowledge, this is the first time that N 2 O emissions from CRP, switchgrass, and Miscanthus have been simultaneously measured to assess the landscape by land management interaction during land use transition. Our research questions were: (i)

Site description
The experimental area, hereafter called Mattern (40°42 0 N, 76°36 0 W), is located near the town of Leck Kill in east-central Pennsylvania (PA). It is part of a long-term monitoring site of the USDA-Agricultural Research Service (Sharpley et al., 2008). The site has a temperate humid climate with annual mean temperature and precipitation of %9.2°C and %106 cm, respectively. Mattern is an 11-ha sub-watershed of the 726-ha watershed (WE-38) that drains to Mahantango Creek, a tributary of the Susquehanna River (Sharpley et al., 2008). The site is representative of the Ridge and Valley physiographic province.
Mattern has mixed land use with 57% cropland, 30% forest, 4% pasture, 9% meadow, and <1% buildings. The upper valley lands have rotations of soybean (Glycine max L.), wheat (Triticum aestivum L.), and corn (Zea mays L.). Slope ranges from 1% to 20%. The elevation above sea level varies from 267 m in the valley floor to 285 m near the summit. The topsoil typically has a silt loam texture with 20-40% rock volume (Table 1) that grades to a loam and/or silty clay loam texture at depth (%100 cm). Albrights soils are distributed along stream and valley floor and have a fragipan and argillic horizon beginning at a depth of 50-70 cm (Needelman et al., 2004). These soils experience prolonged soil saturation during spring snowmelt and after rain events. In contrast, Berks soils in the shoulder and backslope positions are relatively shallow and well drained.
The plots were established in the lower portion of the watershed, in an area of approximately 3-ha that includes the watershed outlet (Fig. 1). Mattern has an ephemeral stream that is active most of the year except during dry summers. Both the lowland and upland were under conventionally tilled corn, soybean, wheat, and alfalfa (Medicago sp.) rotation with poultry manure applied at the rate of 5 Mg ha À1 yr À1 , corresponding to 205 kg N ha À1 yr À1 until 1999 (Kleinman et al., 2006). The lowland was considered marginal due to seasonal wetness and high slope that impairs annual cropping, and was brought under CRP in 1999. The CRP vegetation comprises the perennial cool-season grasses orchard grass (Dactylis glomerata L.), tall fescue (Schedonorus arundinaceus (Schreb.) Dumort.), timothy (Phleum pratense L.), and legumes including alfalfa (Medicago sativa L.) and clovers (Trifolium spp.).

Experimental design
The lower portion of the watershed was divided in four blocks with relatively uniform slope and aspect. Each block contained four plots measuring %60 9 30 m each ( Fig. 1) one each of the following treatments: (i) unconverted CRP, (ii) N-fertilized switchgrass (switchgrass-N), (iii) unfertilized switchgrass (switchgrass), and (iv) unfertilized Miscanthus. In the 12 plots designated for energy crops, CRP vegetation was killed with glyphosate [N-(phosphonomethyl) Glycine] in the summer of 2011, and the aboveground biomass baled and removed from the field. The glyphosate-treated area was no-till planted with winter rye (as a cover crop) in the fall of 2011, which was in turn killed with glyphosate prior to planting the energy crops in the spring of 2012. The switchgrass plots were no-till drilled in rows 20-cm apart. Rhizomes of Miscanthus plants were hand planted 76 cm apart in chisel-made furrows. Thus, the Miscanthus plots had a higher level of soil disturbance. Throughout 2012, the plots of both species were hand replanted, using plants, in areas with poor establishment, especially in wet footslope positions. By 2013, plot establishment was satisfactory except in the footslopes, especially in Miscanthus. In 2012, broadleaf weeds and annual grasses were controlled with herbicide; weed pressure was low in 2013. The switchgrass-N plots received 50 kg ha À1 of N as broadcasted urea on May 29th, 2013, a year after planting.
The upland portion of each plot borders cropland; the bottom portion merges into the ephemeral stream that drains the watershed. Based on the increasing soil wetness from top to bottom, each plot was divided into three segments: shoulder, backslope, and footslope ( Fig. 1). This is customary when studying topographic effects on soil N 2 O emissions (e.g. Pennock et al., 1992;Vilain et al., 2010). The combination of four treatments, three landscape positions, and four replications (blocks) yielded a total of 48 monitoring points.

Measurement of soil water content and air filled soil volume
We continuously monitored volumetric soil water content (h V , m 3 m À3 ) with CS-616 soil moisture sensors (Campbell Scientific Inc., Logan, UT, USA). The sensors were installed at three depths (0-20, 20-40, and 40-60 cm) in each of the 48 monitoring points, for a total of 144-sensors. Each sensor was connected to one of four dataloggers through a network of buried cables.
The h V was used to calculate the volumetric air content in the soil layer (h A, m 3 m À3 ): where h S is the total porosity of a layer after correcting for rock volume (m 3 m À3 ).

Measurement of N 2 O emissions
We measured N 2 O emissions from May to September of 2013. The sampling frequency varied from weekly to biweekly, and Each value represents mean AE standard deviation (n = 16). Different letters within a column indicate a statistically significant difference among the landscape positions at P < 0.05.

Landscape position Vegetation
Shoulder Backslope  increased after fertilization and precipitation events. The soilatmosphere N 2 O flux was measured by the static, non-steady state, vented, aluminum-foil insulated chamber method (Hutchinson & Mosier, 1981). In 2012, PVC collars of 30-cm diameter were inserted 5 cm in the soil in each of the 48 monitoring points. Each collar supports a gas-tight chamber of 10-cm height. The chamber interior was free of any vegetation.
At measurement time, chambers were placed on top of the collars and left there for 45 min. Gas samples of 20 ml were drawn from each chamber through a rubber septum connected to a manifold inside the chamber. Samples were taken 15, 30, and 45 min after chamber closure. Soil temperature near the chamber was measured during gas sampling.
The 20-ml gas samples were transferred to 12-ml pre-evacuated Labco exetainer vials (Labco Limited, Lampeter, UK). The N 2 O concentration in the gas samples was measured with a Varian CP3800 (Varian, Walnut Creek, CA, USA) gas chromatograph (with Compi-Pal autosampler) equipped with a 63 Ni electron capture detector that operates at 300°C to detect N 2 O. The ideal gas law was used to calculate the lg N 2 O-N and the flux was calculated from the rate of increase in N 2 O concentration in the chamber headspace.
Cumulative N 2 O flux (May 9th to September 13th) was calculated by interpolation and linear integration. We refer to this as the growing season flux as it roughly corresponds to the period when the grasses were green and soils were not frozen. The N 2 O flux on the days in between two sampling days was estimated as: where, F L is the estimated flux of N 2 O, F o and F f are the measured fluxes bracketing the days interpolated, d o and d f are the days corresponding to F o and F f , and d i is the ith day in between d o and d f .

Soil sampling
Soil samples (0-20 cm) were taken on every other gas flux measurement day to determine NH þ 4 and NO À 3 concentration (five sampling days during the growing season). We were not able to sample soil for mineral N on the day immediately after the cessation of 'Andrea' on June 11th because the soils were flooded. Fresh soil samples were extracted with 100 ml 2M KCl and the inorganic N species were determined by colorimetric analysis (Lachat Quick Chem 8000). All values were corrected for gravimetric water content measured on a 10 g subsample oven dried at 105°C for 48 h.
Soil cores of 50-120 cm in length, depending on the location, were collected from each sampling point in spring 2012. The cores were taken with a SIMCO Earthprobe 200 (SIMCO Drilling Equipment, Inc. Osceola, IL, USA) using plastic tube liners of 4.5 cm in diameter and a tip diameter of 3.8-cm. The soil core was used for geomorphic description, determination of bulk density, particle size distribution (USDA-pipette method; Day, 1965), total soil C and N concentration (Elemental Combustion System auto-analyzer, Costech Analytical Technologies, Inc., Valencia, CA, USA), and rock volume (water displacement method).

Statistical analysis
We used parametric and non-parametric statistical methods for data analysis. All statistical analyses were performed with the R statistical software (R Development Core Team, 2012). First, we used an analysis of variance (ANOVA) with block, vegetation type, and landscape position effects. The residuals of the daily N 2 O fluxes were not always normally distributed and were log-transformed for ANOVA. The residuals of the cumulative N 2 O fluxes were not normally distributed and the original fluxes were transformed using the reciprocal square root transformation based on the ladder of powers method (Tukey, 1977) to achieve normality (Shapiro-Wilk, W = 0.96, P = 0.17) and variance homogeneity. Data were back transformed for tabular presentation. Marginal means were compared using Tukeyadjusted P-values at a significance level of P < 0.05. The main effect and the interactions on daily and cumulative N 2 O flux were assessed with the following linear model: where, Y ijk represents the response variable N 2 O flux; ⍺, b, and c represent the main effect of vegetation type, landscape position, and block, respectively; e is the error term, and i, j, k denote the respective levels of the main factors. Second, we used Random Forest method to analyze which variables best explain the observed variation in log 10 transformed N 2 O flux (Breiman, 2001). Random Forest was applied to the pooled data including only the days when soil mineral N was measured along with soil N 2 O flux, h A , and weather variables. We used the function randomForest from the package randomForest in R (Liaw & Wiener, 2002). The control parameters for random forest were seed = 500 (set random number) and ntree = 1000 (number of trees). The control parameter mtry indicates the number of variables available for splitting at each node and was calculated as square root of total number of variables (Strobl et al., 2009). The variable importance was plotted by using the function varImpPlot. The variables used for this analysis are: vegetation type, landscape position, soil NO À 3 and soil NH þ 4 concentration in the top 20 cm layer (mg N kg À1 dry soil), volumetric soil air content at the 0-20, 20-40, and 40-60 cm soil depth (h A20, h A40 , and h A60 ; m 3 m À3 ), and cumulative precipitation in the last two days (PPT 2 , cm).
Finally, using the tree package in R (seed = 500), we constructed a conditional inference tree of N 2 O emissions using observations from the five dates in which mineral N was measured. Each tree terminal node has an average N 2 O flux and a number of observations (n).

Weather, soil properties, and biomass yield
During the study period, the mean daily air temperature was 19.1°C and the cumulative precipitation was 40.2 cm, with a large, 10.8 cm rain event during hurricane 'Andrea' on June 10th. The landscape positions had similar C and N distribution and bulk density in the top 20 cm soil layer (P > 0.05); all the treatments started with the same organic C and N in the top 20 cm of soil (Table 1). Soils had mean C and N concentration of 19 and 2.1 g kg À1 soil. The soil bulk density (free of rocks) varied from 1.0 to 1.5 g cm À3 , with a mean of 1.2 g cm À3 . Rocks occupy a substantial soil volume (mean 0.32 m 3 m À3 , range 0.19 to 0.43 m 3 m À3 ), with a significantly higher rock fraction in the shoulder position that decreases down the slope (P < 0.05). The clay concentration varied from 130 to 290 g kg À1 , with the highest figures in the footslopes (Table 1).
The aboveground biomass yield of the energy crops harvested in fall 2013, after the second year of growth, exceeded 10 Mg ha À1 , and was 2-4 times higher than that of the CRP (Table 2). While the figure for the CRP reflects the harvestable biomass in fall, it does not represent total aboveground biomass produced because a large part is produced in spring, senesces in early summer and is not harvestable in fall. There was no statistically significant effect of the landscape position and N-fertilization on biomass yield.

Soil ammonium and nitrate concentration
Mean soil NH þ 4 concentration during the growing season was 10 AE 12 mg N kg À1 soil (Fig. 2), lowest in Miscanthus (6 AE 5 mg N kg À1 soil) and highest in switchgrass-N (14 AE 20 mg N kg À1 soil). The variation in NH þ 4 concentration among landscape positions was very narrow (8.8 to 10.4 mg N kg À1 soil). The N-fertilization in switchgrass-N increased the mean NH þ 4 concentration from 0.8 AE 0.5 mg N kg À1 soil before the fertilization to 48 AE 22 mg N kg À1 soil on June 7th, a week after fertilization (Fig. 2). However, this peak NH þ 4 concentration faded away after the hurricane. A substantial increase in NH þ 4 concentration was also observed in unfertilized grasses (and higher in CRP than energy crops) during summer and tended to decrease over the growing season. No plot showed N stress as vegetation growth was thick and vigorous, but the N-fertilized plots had a darker green color and the plants lodged in mid-July.
Mean growing season soil NO À 3 concentrations were 19 AE 14 mg N kg À1 soil (Fig. 2). In contrast to NH þ 4 , the NO À 3 concentrations were highest in Miscanthus plots (25 AE 17 mg N kg À1 soil), which were twice as high as that of CRP (13 AE 13 mg N kg À1 soil). Again, the variation among landscape positions was narrow (footslope 17 and shoulder 20 mg N kg À1 soil, respectively). In the switchgrass-N plots, the N-fertilization had a minor effect on NO À 3 dynamics in shoulder and backslope positions, while the footslope position showed an increase in soil NO À 3 concentration (30 mg N kg À1 soil) during early June. This NO À 3 concentration was fivetimes higher than that in CRP (Fig. 2). On June 7th, just before the hurricane, Miscanthus plots had the highest NO À 3 concentration in all landscape positions (mean 43 AE 20 mg N kg À1 soil), while CRP plots had the lowest NO À 3 concentration (8 AE 3 mg N kg À1 soil). These differences were not discernible after the hurricane and NO À 3 concentration became comparable among treatments.

Soil air content
Soil profile air content ranged from 0 m 3 m À3 (soil water saturation) to 0.32 m 3 m À3 (Fig. 3). The greatest variation in h A among the treatments was in the footslope. Mean h A in the profile was lowest in the 20-40 cm layer (0.14 AE 0.03 m 3 m À3 ) and highest in the top 20 cm (0.19 AE 0.04 m 3 m À3 ). The start of the 2013, h A flux measurements coincided with a rain event that reduced h A in the soil profile. There was a steady increase in h A from May 9th onwards until the onset of the hurricane in June 10th.
The hurricane driven rain saturated the soil and decreased h A in the top soil layer of the footslope (0.02 AE 0.02 m 3 m À3 ), which was significantly lower than that in the backslope (0.08 AE 0.07 m 3 m À3 ) and shoulder (0.12 AE 0.06 m 3 m À3 ) positions on June 11th (P < 0.05). A prolonged period of low h A was noticeable in the footslope positions, especially in the subsurface Each value represents mean AE standard error of mean (n = 4). Mean biomass yields followed by same letter within a row are not significantly different at P < 0.05. The above-ground biomass yields are based on post-senescence manual harvesting of 3 m 9 1 m quadrats at 10 cm height from the ground surface at each monitoring point. The CRP biomass was harvested in September, while switchgrass and Miscanthus were harvested in early November.
soil layers (Fig. 3c). Soil aeration started to increase after June 14th as drainage progressed. Subsequent rain events decreased soil aeration, but never to the extent or length of the hurricane driven storm.

N 2 O emissions
The N 2 O flux from soil to atmosphere widely varied among the vegetation and landscape positions, ranging from 4 to 305 g N ha À1 day À1 among vegetation types (Fig. 4). The mean flux was lowest in CRP and highest in Miscanthus (8 vs. 16 g N ha À1 day À1 ). The footslope positions were the hot spots, with greater variability of N 2 O emissions than other landscape positions, and a mean flux (18 g N ha À1 day À1 , range of 60 g N ha À1 day À1 ) that was more than two-times higher than the shoulder positions (8 g N ha À1 day À1 ). Footslope positions under Miscanthus had the highest mean growing season N 2 O emission (26 g N ha À1 day À1 ) followed by switchgrass-N, switchgrass, and CRP (23, 16, and 7 g N ha À1 day À1 , respectively). During early May, emissions from the footslope positions were significantly higher than that from the shoulder positions (55 vs. 19 g N ha À1 day À1 , P < 0.05); emissions from backslope positions were intermediate and not significantly different from the other landscape positions (30 g N ha À1 day À1 , Fig. 4). In addition, during the same period, the average N 2 O emission from energy crops of 41 g N ha À1 day À1 was 2.6-times higher than that from CRP. The N-fertilization did not have an immediate effect on N 2 O flux from switchgrass-N in the shoulder and backslope positions; however, a small peak of 35 g N ha À1 day À1 occurred in the footslope position (Fig. 4).
Peak N 2 O emissions were triggered by the hurricane rains on June 10th (Fig. 4). The rain saturated the soil and reduced h A (Fig. 3), coinciding with a period of high mineral N concentration in soil (Fig. 2). Averaged over all plots, the N 2 O flux was 84 g N ha À1 day À1 on June 11th, a day after the large precipitation event. This flux was roughly four-times higher than the pre-hurricane flux of 22 g N ha À1 day À1 on June 7th. Furthermore, emissions from the footslope (156 g N ha À1 day À1 ) and backslope (76 g N ha À1 day À1 ) positions were significantly higher than those from the shoulder positions (20 g N ha À1 day À1 , P < 0.05). Immediately after the hurricane rains, the N 2 O emissions from the grasses were in the order of: Miscanthus > switchgrass-N > switchgrass > CRP (130, 103, 69, and 34 g N ha À1 day À1 , respectively). A significant interaction between vegetation type-by-landscape position was only observed during this hot moment (P = 0.03). The emissions from lower landscape positions of Miscanthus (305 g N ha À1 day À1 ) and switchgrass-N (233 g N ha À1 day À1 ) was significantly higher than that in CRP (20 g N ha À1 day À1 , P < 0.05). The hurricane induced peak N 2 O emission waned to background level after June 18th as the water drained and oxic conditions prevailed in the landscape (Fig. 3).

Growing season cumulative N 2 O flux
The ANOVA of cumulative N 2 O flux shows a significant vegetation-by-landscape position interaction (P = 0.009, Table 3). The differences in cumulative N 2 O emissions between the CRP and energy crops only expressed in the footslope positions (Table 3). While in CRP emissions were similar across landscape positions (%1 kg N ha À1 ), these emissions were higher in the footslope for the energy crops, in particular for Miscanthus (%5.5 kg N ha À1 ).

Analysis of conditions leading to N 2 O emissions
The Random Forest analysis identified landscape position, cumulative precipitation in the two days before N 2 O flux measurement (PPT 2 ), and soil NO À 3 concentration as the most important factors influencing N 2 O emissions (Fig. 5). It is the soil aeration (h A ) in subsurface layers (20-40 and 40-60 cm) that seem to be related to N 2 O emission, rather than the aeration in the top layer. Nonetheless, the Random Forest model explained only 28% of the variation in N 2 O flux. Clearly, there are other controls of N 2 O emissions that we did not measure and can include microsite properties that are difficult to characterize.
Using a regression tree, the predictor variables as identified by the Random Forest were used to classify the N 2 O fluxes in groups that can be identified by specific properties. The regression tree contains seven terminal nodes. The primary node shows a split based on h A40 = 0.03 m 3 m À3 , with the highest emission when h A40 < 0.03 m 3 m À3 (Fig. 6). The observations that belong to the primary node are associated with high NO 3 -N concentration (mean 19 mg N kg À1 soil) and poor aeration status (h A40 < 0.03 m 3 m À3 ) in the soil profile, and are mostly found in the footslope positions. The shoulder positions never sustained h A40 < 0.03 m 3 m À3 . Accordingly, node 4, which is the node with lowest average emission, is composed only of measurements in the shoulder positions. When h A40 > 0.03 m 3 m À3 , moderately high emissions (mean 28 g N ha À1 day À1 ) occur only in the footslope positions when soil NO 3 -N concentration is >10 mg N kg À1 soil (node 7). Since the construction of the regression tree excluded the highest N 2 O emission period after the hurricane event, the potential emissions can be much higher than the values predicted by the tree. However, the tree clearly points to the landscape position and the subsoil air content as the drivers of N 2 O emissions.

Discussion
Converting CRP lands to energy crops only increased N 2 O emissions significantly in the footslopes during the second year of the land use transition (Table 3) no-till soybean converted from CRP grassland. In our case, the lower landscape positions occupy at most a third of the lower part of the watershed, and therefore at least more than two thirds of the area had N 2 O emissions comparable to those of CRP, hereby answering our first research question. The N 2 O emission differences among treatments built up over three distinctive periods: during early season growth prior to the hurricane, the week of the hurricane in mid-June, and the post hurricane period until the end of the growing season. These three periods have distinctive evolutions of soil mineral N and soil water in the different treatments. In the early season, the cool-season CRP vegetation starts growing before the warm season energy crops, which would reflect in comparatively more mineral N depletion in CRP. Accordingly, in early June, the soil NO À 3 concentration in CRP was lower than that of Miscanthus in all landscape positions; however, the switchgrass soil NO À 3 concentration was similar to Miscanthus in the footslope, but similar to CRP in the shoulder and backslope (Fig. 2). Furthermore, with the obvious exception of fertilized switchgrass, CRP had slightly higher NH þ 4 than unfertilized switchgrass and Miscanthus at all landscape positions. Thus, the behavior of mineral N in the early season is more nuanced than expected.
Several factors may have contributed to relatively high NO À 3 accumulation under energy crops, particularly Miscanthus. First, in the years prior to planting the CRP in 1999, the soils received > 200 kg N ha À1 yr À1 as manure. In addition, the CRP vegetation had a visible, but not dominant, proportion of legumes. Killing the pre-existing CRP vegetation during land conversion created fresh above and belowground pools of dead grass and legume root biomass that may have mineralized N during decomposition, adding to N mineralization from soil organic matter Ruan & Robertson, 2013). Killing the rye cover crop added additional decomposable residues, albeit in low quantities (<2 Mg ha À1 ). Second, chisel plowing prior to rhizome establishment caused a greater level of soil disturbance in Miscanthus compared to no-till switchgrass seeding, which may have accelerated soil organic matter decomposition (Grandy & Robertson, 2006;Mazzilli et al., 2015). Third, due to the different techniques used to establish the energy crops, the switchgrass stands were denser than that of Miscanthus early in the growing season, allowing switchgrass an earlier canopy closure and higher N uptake. This might have resulted in more N available for losses in Miscanthus plots. In short, killing the established CRP vegetation in C and N rich soils, can cause high concentration of soil mineral N, favoring N 2 O emissions even under no-till (switchgrass); the effect was more pronounced under Miscanthus possibly due to additional factors such as chisel plowing and slower establishment of the wide-spaced Miscanthus rhizomes (Table 3). A similar case was reported by Gauder et al. (2012), who found higher N 2 O emission from Miscanthus compared with shrub willow (Salix spp.) during early summer and attributed the difference to the partial soil cover by Miscanthus, in addition to high soil mineral N after fertilization. An alternative hypothesis for the lack of NO À 3 accumulation in CRP can be a limited nitrification potential due to the low disturbance, as suggested by the moderate accumulation of NH þ 4 (Fig. 2); it has been shown that the nitrification potential remarkably increased when long-term undisturbed pasture land was cleared and cultivated for short-rotation poplar (Populus spp.) and shrub willow (Nikiema et al., 2012;Palmer et al., 2014). According to our non-parametric statistical analysis, potentially higher N 2 O emissions from these energy crops are realized only when high mineral N and soil saturation converge in the landscape, which occurs predominantly in the footslopes (Figs 5 and 6). Furthermore, the analysis identified greater influence of subsoil aeration on N 2 O emissions in these landscapes, a fact that is often overlooked. At our site, limited drainage through the fragipans and argillic subsoil layers within the colluvial footslopes (Ciolkosz et al., 1995;Needelman et al., 2004) causes extended periods of water saturation, even when the top layer is relatively well drained (Fig. 3). Thus, we hypothesize that the high N 2 O emission are explained by the juxtaposition of an oxic top layer above an anoxic subsurface layer that creates a hot spot for N 2 O production from denitrification when NO À 3 is not limiting. The hot moment and the hot spots for N 2 O emissions were triggered by hurricane Andrea (June 10th to 13th), when soils under energy crops had high soil NO À 3 , and the landscape was conducive to accumulation of water in the footslopes. In this time period differences among treatments built up significantly, mostly in the footslopes (Fig. 4). Given the high level of soil NO À 3 before the event, denitrification was most likely the dominant contributor of peak N 2 O emissions under low soil aeration conditions following the hurricane (Weier et al., 1993;Gillam et al., 2008). This single peak emission event contributed on average 26% of the growing season cumulative N 2 O flux in 2013. This is similar to the findings of Parkin & Kaspar (2006), who reported that two peak events of 29 days-long accounted for 45% of annual N 2 O flux from corn in the Midwest. The remainder N 2 O emissions, which are grouped mostly in the right branch of the regression tree (Fig. 6), are either from nitrification or denitrification in anoxic microsites.
The comparison of unfertilized and N-fertilized switchgrass in this period also reflects the concurrence of hot spots and hot moments. The response of N 2 O flux to N fertilization was not very prominent or immediate because the soil was dry for more than a week after   (4) Backslope (1) Footslope (5) Backslope (1) Footslope (4) Shoulder (27) Backslope (29) Footslope (22) Shoulder (53) Backslope (48) Footslope (45) N-fertilization (May 29th to June 9th) (Fig. 3), limiting N transformations (Bateman & Baggs, 2005;Davidson et al., 2008). However, a small peak from the footslopes may be attributed to favorable soil water content for nitrification. A similar lag period between fertilization and peak emission due to a dry period has been reported by Baggs et al. (2003). Addressing our second research question regarding the effect of N-fertilization on N 2 O emission, we found that N 2 O emissions in this N-rich soils are similar for both switchgrass-N and switchgrass treatments (Table 3); nonetheless, switchgrass-N had significantly higher N 2 O emissions than CRP in the footslopes. During and after the hurricane, N 2 O emissions were more responsive in both Miscanthus and switchgrass-N in the footslopes (Fig. 4). The hurricane mobilized NO À 3 in all treatments (Fig. 2). In addition, shortly after the hurricane, N uptake and transpiration due to vigorous growth decreased the high NO À 3 and anoxia co-occurrences, which likely decreased N 2 O emissions later in the growing season.
Previous studies of landscape scale N 2 O emissions also identified the footslope positions as the hot spots for N 2 O emissions (van Kessel et al., 1993;Castellano et al., 2010;Vilain et al., 2010). The same processes seem to operate in this watershed, with localized zones of water accumulation causing spatially variable N 2 O emissions. Even after receiving a 10.8-cm of rainfall during the hurricane, fast drainage and redistribution of water from the convex shoulder positions precluded aeration limitations that favor denitrification. In fact, the critical h A of 0.03 m 3 m À3 threshold identified through the regression tree analysis is rarely, if ever, reached in the backslope and shoulder positions. However, the sources of NO À 3 for denitrification and N 2 O emission in the footslope may originate from upslope, with NO À 3 transferred by interflow to the footslopes. Regardless, and answering our third research question, while footslopes in the Ridge and Valley region are at risk of high N 2 O emissions, backslope and shoulder positions seem to posit a much lesser risk of emissions, even in the transitional years and in N-rich soils. Because the biomass yield did not vary significantly across the landscape, the N 2 O emission per unit of biomass harvested was higher in the footslopes (0.41 AE 0.15 kg N 2 O-N Mg biomass À1 ), in particular under Miscanthus (0.86 AE 0.09 kg N 2 O-N Mg biomass À1 ).
Interannual variability of N 2 O emissions as reported by many studies (Burchill et al., 2014;Oates et al., 2016), and the representativeness of the study period, limit the generalization of these results. The study period had a wet early summer (Fig. 4a) coinciding with high mineral N availability in the energy crop plots (Fig. 2), the perfect storm for a transitional system when N 2 O emissions are the concern. Thus, our results may represent an upper estimate of N 2 O emissions during a transition from CRP to energy crops in this landscape.

Implications for management of the transition to energy crops
Our results suggest that N 2 O emissions from energy crops during the transitional period, i.e. the first and second year after killing CRP, are contingent on land conversion method, landscape properties, and management strategies. The major insight of this research is that in the Ridge and Valley region, transitioning CRP land to energy crops presents a minor risk of increasing N 2 O emissions in well drained portions of the watersheds, except on the footslopes. As shown in Figs 2 and 4, once the mineral N concentrations drop due to crop uptake (and N losses), N 2 O emissions are low and comparable to that of CRP.
A potential outline of N management emerges from this research. First, to minimize N 2 O fluxes in N-rich environments, soil disturbance should be minimized to prevent an increase in organic matter mineralization. When switchgrass was no-till established and not fertilized, N 2 O emissions were comparable to CRP. Second, Miscanthus had higher N 2 O emissions during the establishment phase on footslopes when compared to switchgrass. This is likely due to a combination of tillage disturbance and slow establishment from wide-spaced plants. Higher biomass yield from Miscanthus in subsequent years (Arundale et al., 2014) may compensate for initial N 2 O emissions when considering lifecycle emissions of the crop. Thus, while our data point to the transition as a hot moment, longer term studies are also needed. Third, while it is not common to apply N fertilizer to Miscanthus or switchgrass in the first year of planting due to slow growth during establishment and to prevent favoring weed growth, later fertilization requires monitoring soil mineral N in spring to avoid over fertilization. In the second growing season of our study (2013), fertilizing switchgrass did not increase biomass yield. While our original concern was the need to increase N supply through fertilization, we did not foresee that soils at Mattern, which is rocky and marginal for agriculture, would continue to mineralize substantial amounts of N upon land conversion (Fig. 2).
Fourth, the risk of soil N 2 O emissions is particularly large only in footslope positions. Drainage from shoulders and backslopes is swift, which minimizes the risk of anoxic conditions. Detailed site specific delineation of seasonally saturated areas will be required to avoid management (e.g. fertilization) that could exacerbate N 2 O fluxes from footslope hotspots. However, while well drained landscape positions can be fertilized with lesser risks of enhancing emissions, NO À 3 can be transported to poorly drained areas prone to higher emissions, and it is unknown how much of the N 2 O emission in the footslope positions originated from the N transported from upslope areas.
In conclusion, managing the transition from CRP to energy crops maintaining a low N 2 O emission per unit of biomass produced in relation to the original vegetation might be optimized by designing a transition process that minimizes the co-occurrence of high mineral N and wet soils (e.g. no-till, minimum tillage, and minimum fertilization), particularly wet subsoils. Our research suggests that a relatively small portion of the watershed is prone to significant N 2 O emissions in the Ridge and Valley landscapes, and that perennial energy crops planted in well-drained portions of the landscape in current CRP land can be an important component of the portfolio of sustainable production options.