Introduction

Carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) are important greenhouse gases (GHGs) exchanged between soils and the atmosphere. Combined, all three GHGs constitute almost 90 % of the total radiative forcing from long-lived GHGs in the atmosphere (Houghton et al. 1996; Shine and Sturges 2007; Montzka et al. 2011), and N2O contributes to stratospheric ozone depletion (Crutzen 1974). The atmospheric concentrations of CO2, CH4, and N2O have risen sharply since the pre-industrial era, likely as a result of land use change, fossil fuel consumption, and large-scale animal husbandry (Ciais et al. 2013).

Well-drained (aerobic) soils are important sources and sinks of CO2, CH4, and N2O. The production of CO2 in soil by plant root and microbial respiration is one of the largest carbon (C) fluxes globally (Hashimoto et al. 2015). The oxidation of atmospheric CH4 by methanotrophic bacteria in aerobic soil is the only known terrestrial sink of atmospheric CH4 (Le Mer and Roger 2001). However, anaerobic soil conditions cause net soil CH4 production (Segers 1998). Given sufficient N availability, soil may either produce N2O as a byproduct of nitrification (an aerobic process) and denitrification (an anaerobic process; Webster and Hopkins 1996; Bremner 1997), or consume N2O when the reduction of N2O to dinitrogen gas (N2) exceeds N2O production (Chapuis-Lardy et al. 2007).

Despite the significance of these gases to the global energy balance, rates of soil GHG fluxes and the proximate mechanisms that control soil GHG fluxes are unclear in many terrestrial ecosystems. This is especially true in arid and semiarid ecosystems that are generally understudied relative to their land area (Martin et al. 2012). For example, in a global inventory of studies that measured soil CH4 uptake, only five out of a total 318 studies occurred in hot or cold deserts, and three more in ecosystems classified as chaparral (Dutaur and Verchot 2007). In total, these eight studies accounted for ecosystems that cover ~40 % of the Earth’s land surface, provide food and shelter to 41 % of the human population, and store 27 % of the global C stock (Reynolds 2001; Lal 2004; Safriel and Adeel 2008).

Soil temperature, nitrogen (N), phosphorus (P), and water are often broadly described as first-order controls on plant- and microbial-mediated processes like soil GHG fluxes (Vitousek and Howarth 1991; Running et al. 2004; Elser et al. 2007; Mahecha et al. 2010). In unfrozen soil, soil CO2 efflux is a function of the availability of C sources (driven by gross primary productivity; Raich and Schlesinger 1992), sufficient available N and P (Neff et al. 2002; Cleveland and Townsend 2006), warm temperatures (e.g., Q10 functions; Mahecha et al. 2010), and increasing water availability (provided soils stay aerobic; Orchard and Cook 1983; Raich and Schlesinger 1992). Soil CH4 uptake has been modeled strictly as a function of soil texture and soil water content (Potter et al. 1996; Striegl 1993) based on evidence that methanotrophic bacteria are chiefly limited by the supply of CH4 and oxygen (O2) into soil (which requires abundant soil air-filled pore space) and water (which limits gaseous diffusion and cellular activity). Soil N2O production is often attributed to denitrification, which requires sufficient soil water content for anaerobic conditions, sufficient N availability for nitrate (NO3 −) production, and an energy source like dissolved organic carbon (DOC; Nömmik 1956). On the other hand, soil N2O production can also result from nitrification, an aerobic process that is typically highest at intermediate soil water contents (Stark and Firestone 1995), and soil N2O consumption can occur when N availability is low (Chapuis-Lardy et al. 2007).

Most of the relationships between temperature, soil water availability, soil nutrient availability, and soil GHG fluxes were identified in humid climates. However, these relationships may not hold for more xeric climates because these ecosystems often have different biogeochemistry and limitations than more mesic ecosystems (Austin et al. 2004; Austin 2011), with unexpected and counter-intuitive consequences. For example, in six of seven terrestrial biomes, litter decomposition (an important source of soil CO2 efflux) was predicted by litter quality and climatic conditions. In arid grasslands, however, these variables failed to predict litter decomposition rates due to the effects of photo-oxidation (Parton et al. 2007). High rates of soil CH4 uptake have been repeatedly measured in dry environments with soil water contents less than one percent, by mass (Striegl et al. 1992; Galbally et al. 2010; Sullivan et al. 2013)—a result attributed to the adaptation of specialized methanotrophic bacteria to arid soils. Furthermore, available N pools and transformations, such as nitrification (a means of soil N2O production), may be significantly greater during dry seasons than wet seasons (Parker and Schimel 2011; Sullivan et al. 2012). Such results, and a perhaps misplaced emphasis on water as a limiting factor in arid soils (Austin 2011), cast doubt on the applicability of the aforementioned mechanisms controlling GHG fluxes in arid soils.

Here, we assessed how soil CO2, CH4, and N2O fluxes changed as a function of soil fertility and the soil physical environment along the three million year Substrate Age Gradient of Arizona (SAGA). The SAGA is well suited to elucidate proximate controls on GHG fluxes in semiarid ecosystems because it has strong, naturally occurring gradients of soil texture, water holding capacity, and soil C, N, and P (Table 1; Selmants and Hart 2008, 2010), in addition to substantial seasonal climatic variability (Sheppard et al. 2002). Further, the three oldest SAGA sites experienced seven years of fertilization with N, P, and N and P in combination (N + P) prior to our study, allowing for experimental evaluation of the relationships between soil GHG fluxes and soil nutrient availability. Importantly for this project, the SAGA has repeatedly shown soil C and N patterns consistent with ecosystem retrogression (Selmants and Hart 2008; Sullivan et al. 2012), a unimodal pattern of C and N pools and fluxes that has been attributed to reduced P availability during soil genesis (Selmants and Hart 2008; Peltzer et al. 2010). Similarly, Selmants and Hart (2008) found increasing rates of fractionating soil nitrogen transfomations, such as nitrification and denitrification, and a corresponding increase in soil and plant 15N isotope signatures, all of which indicate greater N availability with substrate age. Finally, after 1 year of N and P fertilization among these sites, N fertilization increased grass primary production on younger land surfaces and P fertilization increased grass primary production on older land surfaces (Newman and Hart 2015).

Table 1 Substrate age, location, elevation, mean annual precipitation (MAP), mean annual temperature (MAT),  % clay content, field capacity, soil total carbon (C) to nitrogen (N) ratio, total C, total N, total (Kjeldahl) phosphorus (P), and tree basal area along the Substrate Age Gradient of Arizona (SAGA)
Full size table

Methods

Study sites

This study was conducted on the SAGA within the San Francisco Volcanic Field (SFVF) in a woodland ecosystem. The entire SFVF is ~5000 km2 in size (Priest et al. 2001) and is located along the southern margin of the Colorado Plateau in central-northern Arizona, USA. Since its formation during the Pliocene Epoch, volcanic activity has migrated in an east-northeast direction at a rate of ~2 cm per year due to the North American Plate moving over a stationary magmatic hot spot (Tanaka et al. 1986). The volcanism resulted in over 600 monogenetic basaltic cinder cone volcanoes ranging in age from ~6,000,000 to <1000 years (Tanaka et al. 1986).

The SAGA is comprised of four sites within the extent of the SFVF. The substrate ages of these sites are approximately 0.93, 55, 750, and 3000 ky, yet other soil forming factors such as current climate, topography, vegetation, and parent material are constant (sensu lato, Jenny 1941; Selmants and Hart 2008). The sites were aged using dendrochronology and archaeology at the 0.93 ky site and potassium argon dating at the three older sites (Selmants and Hart 2008). The underlying substrate is a pyroclastic sheet of volcanic cinders, consisting primarily of microporphyritic basalt (Moore and Wolfe 1987). Relatively flat topography (<1 % slope) and stable landscape positions allow for in situ weathering of the basalt parent material (Selmants and Hart 2008) and an increase in the soil fine-textured fraction with increasing substrate age. Mean annual air temperature of the SAGA sites is ~11 °C and mean annual precipitation is ~340 mm (Selmants and Hart 2008); air temperatures in the region can range between 37 and −33 °C (1981–2010 data from nearby Sunset Crater National Monument (SCNM) weather station; ncdc.noaa.gov). Annual precipitation ranged between 236 and 660 mm across thirty years (1981–2010; SCNM weather station). Seasonal precipitation dynamics cause roughly half the annual precipitation to fall as snow between December and March, and the other half to fall as rain during monsoonal thunderstorms between July and September (Sheppard et al. 2002). The four sites are open piñon pine (Pinus edulis Engelm.) and one-seed juniper (Juniperus monosperma Engelm.) woodlands. Blue gramma grass (Bouteloua gracilis (Wild. Ex Dunth) Lag. Ex Griffiths), a C4 perennial grass, dominates the inter-tree canopy vegetation of the three oldest sites, while woody shrubs (Rhus trilobata Nutt., Fallugia paradoxical (D. Don) Endl., Ephedra viridi Coville) are present in inter-canopy spaces at the youngest site.

Nutrient availability varies among the SAGA sites in a manner consistent with biogeochemical theory (Walker and Syers 1976; Vitousek and Farrington 1997). Pools and fluxes of atmospherically derived nutrients (e.g., C and N) increased with substrate age to a maximum and then declined (Selmants and Hart 2008) during ecosystem retrogression (Peltzer et al. 2010). Retrogression has been attributed to the steady decline in P availability as a function of increasing substrate age (Selmants and Hart 2010).

Nitrogen, P, and N + P were applied annually in B. gracilis-dominated intercanopy spaces of the three older sites (55, 750, and 3000 ky) between 2004 and 2010. In this experiment, the youngest site did not receive nutrient additions because of the relative scarcity of B. gracilis. The fertilization methodology is described in detail by Newman and Hart (2015). Nitrogen was applied as ammonium nitrate at a rate of 7.5 g N m−2 y−1 to the N and N + P plots; P was applied as a triple super phosphate at a rate of 5 g P m−2 y−1 to the P and N + P plots. Ten g N m−2 y−1 induced piñon tree mortality in a semiarid ecosystem in New Mexico, USA, so the N addition here was designed to minimize potential mortality while maximizing N inputs (Newman and Hart 2015). Phosphorus addition exceeded the sorption capacity of the soil and biological N:P requirements (Newman and Hart 2015). Nutrients were added by hand as pre-weighed granulated solids to the soil surface shortly before the onset of summer monsoonal rains (early July) each year. The timing of this addition was intended to increase the likelihood that the nutrients would be incorporated into the soil by the first monsoonal rains. A 1 m boundary separated each 1.5 m2 plot from other plots, and each plot was trenched to 0.5 m mineral soil depth with a tile spade every year before fertilizer was applied. Nutrient additions were made to the 1.5 m2 plots in a randomized complete block design with 8 blocks per site, from which we randomly selected 5 for GHG flux measurement. Blocks were fenced with barbed wire to minimize grazing by ungulates and free-range cattle. One plot in each of the five blocks was unamended and served as a control. We used these long-term fertilization plots to evaluate experimentally if nutrient availability limited the soil fluxes of CO2, CH4, and N2O along the SAGA (see below).

Static chamber measures of GHGs

We measured CO2, CH4, and N2O fluxes between the soil and atmosphere during days representative of the distinct conditions of the growing season of northern Arizona: the cool and wet post-snowmelt spring season, the warm and dry summer season, the warm and wet summer monsoon season, and the cool and dry fall season. We measured GHG fluxes in the middle of each distinct season: March 30th, June 1st, August 1st, and October 1st, 2010. The August 1st sampling point occurred approximately 25 days after fertilization at each of the three oldest sites. We used this temporal sampling design for the purpose of capturing seasonal dynamics in soil gas fluxes during each season, and correlating these fluxes with soil physical and chemical properties measured at that time. It was not our intention to scale these values to estimate annual fluxes, given the potential for errors in extrapolating the temporally limited dataset.

At the unfertilized 0.93 ky site, we randomly selected five plots from the eight intercanopy plots utilized by Selmants and Hart (2008). At the three older substrate ages, we randomly selected five of the eight fertilization treatment blocks described above. To measure soil GHG fluxes, we inserted one 30 cm diameter polyvinylchloride (PVC) collar 2 cm into the mineral soil in the center of the 1.5 m2 plots. In March, we placed the collars in the soil several days before we sampled to minimize soil disturbance. The collars remained in the soil for the duration of the growing season. During each measurement period, we placed a 30 cm diameter PVC vented static chamber over the collar and sealed the chamber to the collar using a butyl rubber gasket. We measured changes in gas concentration over time by sampling chamber headspace gas 0, 15, 30, and 45 min after chamber installation with evacuated Silonite Minicans (Entech Instruments, Simi Valley, CA, USA). Each 100 mL Minican was over-pressurized to 160 mL. To reduce the effect of diel variability, we collected gas samples between the hours of 10:00 and 15:00 h local time during each measurement period. We quantified GHG concentrations from the headspace samples with an Agilent 6890 gas chromatograph equipped with a methanizer (Agilent Technologies, Palo Alto, CA, USA) that used a Haysep Q 60/80 column and a Porapak Q 60/80 column; both CO2 and CH4 concentrations were measured using a flame-ionization detector, while N2O concentrations were measured using an electron capture detector. We calculated fluxes using a linear regression of CO2, CH4, and N2O concentrations against the sampling interval and verified the linearity of the fluxes using the coefficient of determination. When an individual gas concentration measurement reduced the coefficient of determination below 0.50, we eliminated that concentration and calculated the flux based on three data points.

Soil physical environment

We used a soil thermometer (VWR Scientific, Inc., West Chester, PA, USA) to measure temperature at a 10 cm depth within a 0.5 m radius of the chamber collars inside each plot. Within the same radius, we also measured soil gravimetric water content (GWC) in the top 10 cm of mineral soil by collecting samples with a soil sampling tube (Oakfield Apparatus, Fon du Lac, WI, USA). Immediately after transport to the laboratory, a subsample was weighed, dried in an oven at 105 °C until it had reached a stable mass, and reweighed. We calculated the soil water potential (WP) of each sample using site-specific soil water release curves and our measurements of GWC. The water release curves were developed using a WP4 Dewpoint Potentiameter (Decagon Devices, Pullman, WA, USA), which measured soil WP at known GWCs.

Soil nutrient availability

At each sampling period, we measured available pools of N and P in the same top 10 cm of mineral soil used to measure WP and GWC from each plot where we measured GHG fluxes. We measured soil ammonium (NH4 +) and NO3 − concentrations by extracting 10 g of field moist soil with 50 mL of 2 M KCl, shaking for 1 h on a mechanical reciprocating shaker, and filtering through a Whatman no. 1 filter paper pre-leached with deionized water. Available N was assumed equivalent to total inorganic N (NH4 + + NO3 −) concentrations. We estimated the soil available P pool as labile orthophosphate (PO4 3−) using anion exchange membrane (AEM) strips. Approximately 1 g of field moist soil was placed in a 50 mL centrifuge tube with two AEM strips (10 × 50 mm; Ionic, Inc., Watertown, MA, USA) and 30 mL of deionized water, and the solution was rotated on an overhead shaker for 16 h at 30 rpm. We then removed and rinsed the AEM strips with deionized water, placed them in a clean 50 mL centrifuge tube, added 20 mL of 0.5 M HCl, and rotated the solution on an overhead shaker for 16 h. We froze the extracts for later analysis of NH4 +, NO3 −, and PO4 3− concentrations using a QuickChem 8000 Flow Injection Autoanalyzer (Lachat Instruments, Loveland, CO, USA).

Data analysis and statistical methods

To analyze the effect of season and substrate age across all four SAGA sites on unfertilized soil nutrient availability, the soil physical environment, and soil GHG fluxes, we used two-way repeated measures analysis of variance (RMANOVA). To analyze the effect of fertilization, substrate age, and season on soil nutrient availability, the soil physical environment, and soil GHG fluxes, we used three-way RMANOVA using data from the three oldest sites only. In both cases, we used RMANOVA to overcome violations of independence associated with autocorrelation between seasonal measurements. We expressed the effect of fertilization on soil nutrients, the soil physical environment, and soil GHG fluxes, as a relative effect size:

$$\frac{(Fertilized - Unfertilized)}{Unfertilized} \times 100\%.$$

We used two-tailed t-test to ascertain if soil nutrients, the soil physical environment, or soil GHG fluxes significantly differed between fertilized and unfertilized soil (if the effect size significantly differed from zero). Before performing RMANOVA and t test analyses, we took the natural log of soil NH4 +, NO3 −, total available N, available P, and WP (but not soil GHG fluxes, GWC, or temperature, which met statistical assumptions) to transform the data and meet statistical model assumptions of normality and homoscedasticity. However, only untransformed data are presented for clarity. For all parametric statistics, we set our alpha, a priori, at 0.10 because we were concerned of the possibility of Type II errors given our small sample size (n = 5) and anticipated large spatial and temporal variability in this study. To explore relationships between soil GHG fluxes, substrate age, season, soil nutrients, soil temperature, and soil water availability, we used regression tree analysis. Regression tree analysis is a machine learning recursive partitioning statistical approach that is robust for violations of assumptions of the true model, and allows for visual, simplified interpretation of the often complex multiple interactions between predictive and independent variables (De’ath and Fabricius 2000). We chose to use regression tree analysis rather than, for example, multiple or stepwise regression, because it represents the most important mechanistic variables related to soil GHG fluxes independently of a priori hypotheses that could bias our choice of initial model conditions (e.g., parabolic and linear relationships between temperature, water content, and soil CO2 efflux; Sullivan et al. 2011). All statistical analyses were performed using open source R software (version 3.1.1; R Core Team 2014); regression tree analysis was performed using the “rpart” package in R (version 4.1–8; Therneau and Atkinson 2014).

Though substrate ages across the SAGA are unreplicated, the use of unreplicated substrate age gradients nonetheless provides important opportunities for the study of soil and ecosystem development and associated biogeochemical processes across long time scales (Vitousek 2002; Wardle et al. 2004). While some have questioned the utility of chronosequences (e.g., Johnson and Miyanishi 2008), such concerns often revolve around short- and medium-length chronosequences focused on vegetation succession. Rather, we focus on the effects of long-term soil development on clear, demonstrated soil physical and chemical characteristics—a “valid” use of chronosequences (Walker et al. 2010). Therefore, we proceeded to use parametric statistics to infer differences among sites, but the results and inferences herein should not be extrapolated to other sites without caution. All raw data associated with fluxes and soil properties, from each season, will be archived in the free online data repository Dryad (http://www.datadryad.org).

Results

Substrate age and seasonal effects on soil temperature, water content, nutrients, and GHG fluxes in unfertilized soil

Soil temperature at the 10 cm mineral soil depth in unfertilized plots varied significantly by season (F(3,48) = 569, p < 0.001) but the seasonal trends depended on substrate age (season by substrate age interaction F(9,48) = 44.0, p < 0.001; Fig. 1). At the three oldest sites, March was the coldest month, but at the youngest site, October was coldest. At all sites, June had the warmest soil temperature. Soil temperature did not vary significantly by substrate age alone (F(3,16) = 0.94, p = 0.446).

Fig. 1
figure 1

Soil temperature (top left panel), gravimetric water content (top right panel), available nitrogen (N; sum of ammonium (NH4 +) and nitrate (NO3 −); center left panel), available phosphorus (P; resin-extractable orthophosphate; center right panel), soil NH4 + concentration (bottom left panel), and soil NO3 − concentration (bottom right panel) measured during four sampling periods (March, June, August, and October) at the four sites that comprise the Substrate Age Gradient of Arizona. Data are means ± one standard error (n = 5)

Full size image

Soil GWC in the top 10 cm mineral soil also varied significantly by season (F(3,48) = 51.0, p < 0.001) and there was a significant interaction between season and substrate age (F(9,48) = 3.72, p = 0.001). June was consistently the driest season and August was the wettest season at all sites except the 750 ky site (Fig. 1). Soil GWC varied significantly by substrate age (F(3,16) = 22.8, p < 0.001); mean soil GWC across all seasons increased consistently from 0.07 kg kg−1 at the 0.93 ky site to 0.17 kg kg−1 at the 3000 ky site. Patterns of soil WP were generally similar to GWC. Soil WP varied significantly by substrate age (F(3,16) = 29.18, p < 0.001) and season (F(3,48) = 562, p < 0.001), and there was a significant substrate age by season interaction (F(9,48) = 6.268, p < 0.001; Table 2).

Table 2 Soil water potentials (MPa) in unfertilized soil from each of the four sites at the Substrate Age Gradient of Arizona during each sampling date
Full size table

Soil nutrients always varied by season and generally varied by substrate age. Soil NH4 + (F(3,48) = 13.4, p < 0.001), NO3 − (F(3,48) = 2.84, p = 0.048), available N (F(3,48) = 3.14, p = 0.034), and available P (F(3,48) = 85.8, p < 0.001) varied by season. Soil NO3 − (F(3,16) = 10.9, p < 0.001), available N (F(3,16) = 4.26, p = 0.022), and available P (F(3,16) = 90.1, p < 0.001) varied by substrate age. Any seasonal influence varied by substrate age for every soil nutrient we measured: Soil NH4 + (F(9,48) = 2.40, p = 0.025), NO3 − (F(9,48) = 3.03, p = 0.006), available N (F(9,48) = 2.27, p = 0.033), and available P concentrations (F(9,48) = 3.66, p = 0.002) had significant substrate age by season interactions (Fig. 1). There were few consistent patterns in soil available N; for example, concentrations were greatest in June at the 0.93 ky site, August at the 55 ky site, October at the 750 ky site, and March at the 3000 ky site (Fig. 1). Soil available P increased to a maximum between 0.93 and 750 ky (from 2.66 mg kg−1 at the 0.93 ky site to 57.7 mg kg−1 at the 750 ky site, averaged across all seasons) before declining at the 3000 ky site (Fig. 1). August consistently had the lowest available P concentrations at all sites, whereas October had the greatest available P concentration at the three oldest sites (Fig. 1).

In unfertilized soil, soil CO2 efflux generally increased with increasing substrate age (F(3,16) = 12.1, p < 0.001) and tended to increase during the growing season (F(3,48) = 28.3, p < 0.001; Fig. 2). However, the seasonal effect varied by substrate age (F(9,48) = 7.17, p < 0.001). At the 0.93, 55, and 3000 ky sites, soil CO2 efflux declined at the end of the growing season (between August and October; Fig. 2).

Fig. 2
figure 2

Soil carbon dioxide (CO2; top panel), methane (CH4; middle panel), and nitrous oxide (N2O; bottom panel) measured during four sampling periods (March, June, August, and October) at the four sites that comprise the Substrate Age Gradient of Arizona. Data are means ± one standard error (n = 5). Positive values represent a net flux to the atmosphere and negative values represent a flux into the soil. Asterisks indicate the mean flux was significantly different from zero, measured using two-tailed t-test and α = 0.10

Full size image

As with soil CO2 efflux, unfertilized soil CH4 uptake varied significantly by substrate age (F(3,16) = 3.57, p = 0.038), season (F(3,48) = 6.66, p < 0.001), and exhibited a significant site by season interaction (F(9,48) = 2.77, p = 0.011). Mean soil CH4 uptake was greatest in March at the 0.93, 55, and 3000 ky sites (though not significantly different from zero at the 0.93 ky site, Fig. 2).

Unlike soil CO2 efflux and CH4 uptake, N2O flux did not vary significantly by substrate age (F(3,16) = 1.96, p = 0.161) or season (F(3,48) = 0.63, p = 0.599), and there was no significant interaction between substrate age and season (F(9,48) = 1.01, p = 0.444). Substantial within-site variability meant that most mean soil N2O fluxes were not significantly different from zero, even when flux rates were substantial (Fig. 2). We only measured significant soil N2O production in August at the 55 ky site and March and August at the 3000 ky site (Fig. 2).

Fertilization effects on nutrients, soil temperature, water content, and GHG fluxes

Fertilization significantly increased soil nutrient concentrations, but had no significant impacts on soil temperature, soil GWC, or WP. Nitrogen, but not P, fertilization significantly increased soil NH4 + concentrations (F(3,48) = 65.286, p < 0.001). There was a significant interaction between fertilization treatment and season on soil NH4 + concentrations (F(9,144) = 2.162, p = 0.028; Table 3). Soil NO3 − concentrations were significantly higher in N, but not P, fertilized plots (F(3,48) = 37.9, p < 0.001), and there were no significant interactions between fertilization or season or substrate age. The effect of N fertilization on soil NO3 − was stronger in March and June than August and October (Table 3).

Table 3 The effect size of nitrogen (N), phosphorus (P), and nitrogen and phosphorus (N + P) fertilization, calculated as a percent relative to unfertilized soils, on soil nutrient concentrations at each of the three older sites at the Substrate Age Gradient of Arizona that received fertilization, during each sampling date
Full size table

Soil available P concentrations were significantly different among fertilization treatments (F(3,48) = 31.2, p < 0.001), and there were no significant interactive effects between fertilization treatment and substrate age and season on soil available P concentrations. Soil available P concentrations were consistently greater in P and N + P fertilized plots than in unfertilized plots, and like in N-fertilized plots, the fertilization effect was stronger in March and June than August and October (Table 3). Nitrogen fertilization significantly increased soil available P content relative to unfertilized soils in March and June (but not August or October) at the 55 ky site (Table 3).

Seven years of fertilization had inconsistent effects on soil GHG fluxes. Soil CO2 efflux did not vary significantly due to fertilization (F(3,48) = 0.41, p = 0.745), nor were there any significant fertilizer interactions with substrate age (F(6,48) = 0.50, p = 0.806), season (F(9,144) = 0.44, p = 0.913), or substrate age and season (F(18, 144) = 0.816, p = 0.680). Fertilization only significantly increased soil CO2 efflux relative to unfertilized plots five times across all fertilization treatments, substrate ages, and seasons (36 possible combinations; Table 4).

Table 4 The effect size of nitrogen (N), phosphorus (P), and nitrogen and phosphorus (N + P) fertilization, calculated as a percent relative to unfertilized soils, on soil greenhouse gas fluxes at each of the three older sites at the Substrate Age Gradient of Arizona that received fertilization, during each sampling date
Full size table

Soil CH4 uptake was even less responsive to fertilization than soil CO2 efflux. However, while we found no overall fertilization effect (F(3,48) = 2.10, p = 0.117), we did measure a significant fertilization by season interaction effect on soil CH4 uptake (F(9,44) = 2.52, p = 0.010). Fertilization significantly changed soil CH4 uptake relative to unfertilized plots only twice; one effect was positive and the other was negative (Table 4). There were no significant fertilization by substrate age or substrate age by season by fertilization interactions on soil CH4 uptake.

Soil N2O fluxes were the more responsive to fertilization of the three GHGs measured, but, similar to the other GHGs, fertilization did not significantly affect N2O flux overall (F(3,48) = 1.44, p = 0.244). We did not find significant interactions between fertilization, substrate age, and season. However, fertilization increased mean soil N2O flux in 10 out of 36 possible fertilization, season, and substrate age combinations (Table 4). The most consistent fertilization impact on soil N2O flux was an increase in soil N2O production in June, August, and October at the 3000 ky site in N fertilized plots relative to control sites (Table 4). Although soil N2O production was greater on P fertilized plots than unfertilized plots in one-third of the cases, responses were inconsistent among all sites and seasons (Table 4).

Chamber-scale controls on GHG fluxes

As measured by regression tree analysis, the dominant controls of unfertilized soil CO2 efflux were soil GWC, substrate age, and season, in descending order of importance (Fig. 3). Soil CO2 efflux was greatest in moist soils. However, the dominant controls on fertilized soil CO2 efflux were, in order, season, soil temperature, substrate age, and total soil available N concentration (Fig. 4). Soil CO2 efflux in fertilized plots split into low rates in March and June and higher rates in August and October.

Fig. 3
figure 3

Regression tree diagram depicting the interactive relationships between carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) in unfertilized soils only and numerous possible predictor variables. GWC: gravimetric water content, Site age: substrate age, NO −3 : nitrate, WP: water potential, NH +4 : ammonium, M: March, J: July, A: August, O: October, n: the number of fluxes measured at this terminal node. Greater-than or less-than symbols (e.g., ≥, ≤ respectively) indicate a split at the node in which greater values are to the left or right indicated by the direction of the symbol

Full size image
Fig. 4
figure 4

Regression tree diagram depicting the interactive relationships between carbon dioxide (CO2), methane (CH4), and nitrous oxide (N2O) in fertilized soils and numerous possible predictor variables. Soil T: soil temperature, N: soil available nitrogen, NH +4 : ammonium, P: soil available P; GWC: gravimetric water content, Site age: substrate age, WP: water potential, M: March, J: July, A: August, O: October, n: the number of fluxes measured at this terminal node. Greater-than or less-than symbols (e.g., ≥, ≤ respectively) indicate a split at the node in which greater values are to the left or right indicated by the direction of the symbol

Full size image

Soil NO3 − concentration, season, WP, and site age were the dominant controls on unfertilized soil CH4 uptake, in descending order of importance (Fig. 3). Soil CH4 uptake was greatest when soil NO3 − concentrations were lower than 4.71 mg kg−1; in March, when soil NO3 − exceeded that threshold, CH4 uptake was greater than any of the other months. There were more factors that controlled soil CH4 uptake in fertilized plots than unfertilized plots. Soil NH4 + concentration, season, soil temperature, site age, total soil available N, soil available P concentration, WP, and GWC were all factors that demarcated branches of our regression tree (Fig. 4). Soil CH4 uptake was lowest when soil NH4 + concentrations were greater than 0.44 mg kg−1.

The dominant controls of unfertilized soil N2O flux were GWC, soil NH4 + concentration, site age, season, soil available N concentration, and soil NO3 − concentration, in descending order of importance (Fig. 3). Soil N2O uptake occurred with soil GWC below 0.21 kg kg−1, whereas soil N2O production predominated when soil GWC exceeded that threshold. By contrast, the regression tree analysis only identified two controls on fertilized soil N2O production: GWC and WP (Fig. 4).

Discussion

Our results indicate that the mechanisms that control soil GHG fluxes in these semiarid soils are similar to those in more humid ecosystems, with some important distinctions. Nutrient fertilization appears to alter the mechanistic controls over soil GHG fluxes in semiarid soils.

Unfertilized soils

We expected soil CO2 efflux rates to be driven by a combination of seasonal climate dynamics and soil nutrient availability. Therefore, we correctly anticipated that soil CO2 efflux would be greatest during the August sampling dates, which had warm and wet conditions (brought on by the late-summer “monsoonal” precipitation pattern that dominates the southwestern U.S. climate; Sheppard et al. 2002). Yet, among the SAGA sites, we expected soil CO2 efflux to reflect the retrogressive pattern of C and N pools and fluxes that has been repeatedly demonstrated among these sites (Table 1; Selmants and Hart 2008; Sullivan et al. 2012; Looney et al. 2012; Newman and Hart 2015). The retrogressive decline in C and N availability between the 750 and 3000 ky sites has been ascribed to P limitation (Peltzer et al. 2010) brought on by a decline in plant and microbially available P pools (Selmants and Hart 2010). Therefore, we were surprised to measure higher rates of soil CO2 efflux at the 3000 ky than the 750 ky site in every season. Though it would seem that both C supply and nutrients were optimal for heterotrophic and autotrophic soil CO2 efflux at the 750 ky site, the 3000 ky site had the finest-textured soil of the SAGA sites. The higher silt and clay content of this soil allows for greater water holding capacity than the other sites (Selmants and Hart 2008, Sullivan et al. 2013). Our regression tree analysis suggested that the highest rates of soil CO2 efflux in unfertilized soil occurred in wetter soils. While each site, on average, receives similar precipitation, the finer-textured soil at the 3000 ky site likely led to the two highest water contents we measured (in August and October, Fig. 1) when soil CO2 efflux was also highest (Fig. 2). These results suggest that soil CO2 efflux is more limited by water availability than C availability, nutrient availability, or temperature among these semiarid sites. Further, the trend of increasing soil CO2 efflux with increasing substrate age (and water holding capacity) is consistent with more humid ecosystems where soil CO2 efflux increases linearly with soil water availability (Raich and Schlesinger 1992).

In both humid and arid ecosystems, soil water availability is an important driver of soil CH4 uptake. Low soil water content can limit CH4 uptake (Striegl et al. 1992). For example, models of CH4 uptake predict a sharp decline below 5–10 % GWC, depending on the soil texture (Del Grosso et al. 2000). By contrast, too much soil water reduces the air-filled pore space through which soil gas diffuses, and therefore can inhibit atmospheric CH4 diffusion into soil (Striegl 1993). The patterns and proximate controls of soil CH4 uptake we observed were at least partially contrary to our expectations. First, in three of four sites, the highest rate of soil CH4 uptake occurred in March, when temperatures were cold and water contents were intermediate. Second, the regression tree analysis provided little evidence that either soil water content or soil temperature were strong controls on soil CH4 uptake. We measured significant rates of soil CH4 uptake in both dry soil (e.g., June at the 55 ky site) and wet soil (e.g., October at the 3000 ky site). Furthermore, soil NO3 − concentration was the first-order control of soil CH4 uptake. While NO3 − fertilizer has been shown to reduce soil CH4 uptake and soil CH4 uptake has been negatively correlated with nitrification rates (Aronson and Helliker 2010; Neff et al. 1994), the mechanism behind the relationship between soil CH4 uptake and soil NO3 − concentration is not well understood, especially in arid soils.

We measured a wide range of soil N2O production and consumption that reflected the wide range of soil water content and N availability among sites and seasons. Though we only measured three instances of significant N2O production, in part because of the spatial and temporal variability of the process, two of those instances occurred at the 3000 ky site. Selmants and Hart (2008) measured greater soil 15N enrichment (~7 ‰) at this site than any of the other SAGA sites, and hypothesized that this was due to the fractionating effects of nitrification and denitrification at a site that was N-rich and P-poor. While similar hypotheses have been robust at other age gradients in humid ecosystems (e.g., Martinelli et al. 1999), our data provide limited evidence for greater rates of gaseous N losses late in primary succession in semiarid environments. It also appears that N2O production is almost entirely dependent on soil water content and N availability (Fig. 3). In most cases, the regression tree analysis showed that soil N2O production increased as GWC and soil N availability increased.

Fertilized soil

Seven years of annual N, P, and N + P fertilization at the three oldest SAGA sites had substantial effects on soil nutrient availability, but surprisingly few effects on soil GHG fluxes. Our results suggest that our experimental design was largely successful for two reasons. First, there was very little cross-plot movement of fertilizer (Table 3). Second, annual fertilizer additions raise the possibility that impacts on transient processes like soil GHG fluxes would only be observed immediately after fertilization (Sullivan et al. 2014). However, in this case, 7 years of fertilization appear to have caused persistent increases in soil nutrients, even in June, 11 months after the last fertilizer application. In fact, the sampling date with the least number of significant fertilization effects on soil nutrients was August, only 1 month after fertilization. This seemingly paradoxical result is likely due to plant utilization of fertilizer being greatest during the August growing season, when monsoonal rains and warm temperatures are ideal for plant growth; during our October sampling period, soil nutrients were again significantly greater in fertilized plots than unfertilized plots after grass aboveground senescence in late September. Nevertheless, our data raise the question: by what mechanism do the grasses release the nutrients, derived initially from fertilizer uptake, back into available soil pools after the growing season?

Fertilizer addition had surprisingly few significant impacts on soil CO2 efflux and CH4 uptake. There was no evidence that fertilizer alleviated nutrient constraints on soil CO2 efflux in a manner consistent with biogeochemical theory. Based on the retrogressive nutrient patterns demonstrated among the SAGA sites by Selmants and Hart (2008, 2010), and the response of grass primary production to 1 year of fertilization (Newman and Hart 2015), we were surprised that N fertilization did not increase soil CO2 efflux at the 55 ky site and P fertilization did not increase soil CO2 efflux at the 3000 ky site (Table 4). Like in this study, N and P fertilization across a retrogressive, tropical montane chronosequence had only muted effects on litter decomposition (an important source of CO2 efflux) across 4100 ky of soil development despite responses of aboveground productivity (Vitousek and Farrington 1997; Hobbie and Vitousek 2000). Soil temperature and water content exhibited stronger controls on soil CO2 efflux than nutrient availability in these semiarid soils.

In fertilized soils, NH4 + availability appeared to be the first-order control over CH4 uptake; soil CH4 uptake declined with increasing NH4 + availability. Relationships between inorganic N and soil CH4 uptake have been previously documented in more humid ecosystems. For instance, the suppressive effects of N addition on soil CH4 uptake increase with increasing duration of fertilization (Aronson and Helliker 2010), but arid ecosystems were underrepresented in that analysis. Here, we observed two different N influences on CH4 uptake. Across naturally occurring gradients of N availability in unfertilized soil, soil CH4 uptake appeared to increase with increasing NO3 − availability, whereas among fertilized plots, NH4 + suppressed CH4 uptake. In both cases, N availability was a more proximate control of soil CH4 uptake than either soil water content or soil texture. These patterns point to complex, poorly understood mechanisms that control soil CH4 uptake in semiarid ecosystems (Sullivan et al. 2013).

Based simply on the number of significant effect sizes, fertilization had a greater effect on soil N2O production than the other two GHG fluxes. Given that N availability must be sufficient for nitrification and denitrification, the two dominant sources of N2O production in soil (Davidson et al. 1986), we expected N and N + P fertilization to increase N2O production. However, we found an equal number of cases (four) with significant positive P effects on N2O production as N effects, and a surprising number (four) of non-significant negative N fertilization effects on N2O production (Table 4). While meta-analysis has demonstrated that N additions nearly always resulted in an increase in N2O production (Aronson and Allison 2012), semiarid and arid ecosystems were underrepresented, as they were for soil CH4 fluxes. At present, any link between P fertilization and N2O production is uncertain. For example, a synthesis highlighting the need for denitrification studies does not mention the role of P in denitrification (Davidson and Seitzinger 2006). The simplest explanation of these results may be that, in some conditions, P fertilization alleviated plant or microbial limitation and increased biological activity and N cycling. Regardless, fertilization removed N constraints on N2O production. While N2O fluxes in unfertilized soils indicated appeared to be controlled by N and water availability, fertilized soils were only controlled by water availability (Fig. 4). Aside from the unknown mechanism that caused P fertilization to increase N2O fluxes 25 % of the time, N2O production in both humid and arid ecosystems, whether from nitrification or denitrification, seems to depend on both soil N availability and water.

Conclusions

A paucity of data and a corresponding lack of a mechanistic understanding of soil GHG fluxes in arid environments have hindered consideration of these environments in many global biogeochemical models, despite their spatial extent. Substrate age gradients, combined with long-term fertilizations, can provide valuable information about limitation of ecosystem processes because they seek to constrain many state factors while allowing nutrients and soil physical properties to change over time with soil development (Vitousek 2004; Sullivan et al. 2014). By using a long-term in situ nutrient manipulation experiment at the SAGA to explore interactions between pedogenesis, soil physics, biogeochemistry, and soil GHG fluxes, our results highlight important controls on soil GHG fluxes in semiarid soils and raise several questions. Our results demonstrate that the mechanistic understanding of soil GHG fluxes derived in humid ecosystems does not apply consistently to the three GHG fluxes we measured here. We suggest that the applicability of mechanisms that drive GHG fluxes in humid ecosystems to semiarid ecosystems declines in the following order: N2O > CO2 > CH4. Clearly, N and water availability controlled unfertilized soil N2O fluxes (and fertilization reduced N limitation of N2O flux). However, we are unable to adequately explain relationships between soil N availability and soil CH4 flux, and we note the lack of apparent relationships between soil CH4 uptake and soil physical properties. By combining in situ observation with experimentation and regression tree analysis, our results have elucidated the separate and complex controls on soil GHG fluxes in semiarid soils that we hope will lead to development of better mechanistic models of soil GHG fluxes in arid climates.