Dynamics of component carbon fluxes in a semi-arid Acacia woodland, central Australia


  • James Cleverly,

    Corresponding author
    • School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Nicolas Boulain,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Randol Villalobos-Vega,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Nicole Grant,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Ralph Faux,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Cameron Wood,

    1. National Centre for Groundwater Research and Training, School of Environment, Flinders University, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Peter G. Cook,

    1. National Centre for Groundwater Research and Training, School of Environment, Flinders University, Adelaide, South Australia, Australia
    2. Water for a Healthy Country National Research Flagship, Division of Land and Water, Commonwealth Scientific and Industrial Research Organization, Adelaide, South Australia, Australia
    Search for more papers by this author
  • Qiang Yu,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Andrea Leigh,

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author
  • Derek Eamus

    1. School of the Environment, University of Technology Sydney, Broadway, New South Wales, Australia
    2. National Centre for Groundwater Research and Training, University of Technology Sydney, Broadway, New South Wales, Australia
    Search for more papers by this author

Corresponding author: J. Cleverly, School of the Environment, University of Technology Sydney, PO Box 123, Broadway, NSW 2007, Australia. (James.Cleverly@UTS.edu.au)


[1] Vast areas in the interior of Australia are exposed to regular but infrequent periods of heavy rainfall, interspersed with long periods at high temperatures, but little is known of the carbon budget of these remote areas or how they respond to extreme precipitation. In this study, we applied three methods to partition net ecosystem photosynthesis into gross primary production (GPP) and ecosystem respiration (Re) during two years of contrasting rainfall. The first year was wet (>250 mm above average rainfall), while little precipitation fell during the second year (>100 mm below average). During the first year of study, rates of GPP were large (793 g C m−2 yr−1) in this semi-arid Mulga (Acacia aneura) and grass savanna due to complementary photosynthetic responses by the canopy and C4 understorey to cycles of heavy rainfall. Patterns in GPP during the summer and autumn matched those in leaf area index (LAI), photosynthetic activity, and autotrophic respiration. During the dry year, small but positive photosynthetic uptake by Mulga contributed to the neutral carbon budget (GPP / Re = 1.06 ± 0.03). Small rates of photosynthesis by evergreen Mulga when dry were supported by storage of soil moisture above a relatively shallow hardpan. Little soil organic matter (1.1%) was available to support heterotrophic respiration (Rh) without input of fresh substrate. The two largest sources of Re in this study were autotrophic respiration by the seasonal understorey and Rh through decomposition of fresh organic matter supplied by the senescent understorey.

1 Introduction

[2] Arid and semi-arid regions cover an estimated 20% of the globe [Prentice et al., 2001; Liu et al., 2012], exerting influence on global climate and atmospheric carbon dynamics. Semi-arid ecosystems experience extremes in temperature, moisture availability, and sunlight, which can act together to generate large photosynthetic and respiratory carbon fluxes [Huxman et al., 2004a; Baldocchi, 2008; Wohlfahrt et al., 2008; Yan et al., 2011]. In Australia, the driest permanently inhabited continent on Earth, 70% of the land area is arid or semi-arid where potential evapotranspiration rates are high relative to precipitation [Warner, 2004; Eamus et al., 2006]. Owing to the remoteness of much of Australia's interior, little is known about carbon dynamics in its semi-arid regions. The overall objective of this study was to describe ecosystem photosynthetic and respiratory carbon fluxes in central Australia by identifying responses of ecosystem component carbon fluxes to variations in meteorology, soil conditions, leaf physiology, and community composition of the active vegetation.

[3] Global flux networks measure the net ecosystem exchange (NEE) of CO2 between ecosystems and the atmosphere, wherein partitioning of NEE elucidates changes in ecosystem respiration (Re) and gross primary production (GPP) as ecosystems respond to a variety of stressors and disturbances [Baldocchi, 2008]. Net ecosystem photosynthesis (NEP) is analogous to net photosynthetic assimilation (A) in leaves, is positive during net carbon uptake (i.e., a carbon sink), and is the inverse of NEE [Wohlfahrt et al., 2005]:

display math(1)

where Re is the sum of autotrophic and heterotrophic respiration (Ra and Rh, respectively) and includes carbon losses due to geophysical processes (e.g., photodegradation). In practice, Re is estimated when GPP is known to be 0, either at night [Reichstein et al., 2005] or at the intercept of a photosynthetic light response function (LRF) [Lasslop et al., 2010; Lasslop et al., 2012]. When extrapolating from nocturnal to daytime temperatures, overestimation of Re can be avoided by estimating ecosystem “dark” respiration (Rd) from the intercept of LRFs [Wohlfahrt et al., 2005; Baldocchi, 2008; Lasslop et al., 2010]. In semi-arid regions, NEP increases with light as expected up to about half of full sunlight, but then it is not unusual for NEP to become negative such that Re in full sunlight may exceed the value of ecosystem Rd [Scott et al., 2010]. Thus, Re was estimated in our study during each of three time periods: night, daytime when solar irradiance (Es) was smaller than half of full sunlight, and daytime in high light.

[4] In arid and semi-arid ecosystems, Re and GPP are dependent upon delivery of rainfall and associated increases in soil wetness [Eamus, 2003; Baldocchi, 2008; Yan et al., 2011], although the seasonality of weather systems that deliver precipitation influences the strength and occurrence of photosynthetic and respiratory responses [Reynolds et al., 1999; Hastings et al., 2005]. Precipitation falling in winter storms is typically delivered by large frontal systems that can persist over hours or even days and generate low intensity rainfall of wide spatial extent, thereby recharging soil moisture. Both Re and GPP can be enhanced by storage of winter moisture, consequentially having a much greater effect on global carbon cycling than otherwise expected [Wohlfahrt et al., 2008]. In contrast to winter storms, summer convective storms develop in response to local surface heating, creating intense, short-lived, and spatially varied rainfall when sufficient atmospheric moisture is present [Berry et al., 2011]. Unpredictability in timing and intensity of summer precipitation restricts ecosystem GPP responses to large or, more importantly, cyclically recurring precipitation pulses that have an effect on soil moisture [Ma et al., 2012].

[5] Because of the short-lived and unpredictable nature of summer storms, responses of GPP or Re to summer rainfall pulses depend upon receipt of adequate rainfall to evoke a response (i.e., pulse-threshold-response) [Schwinning and Sala, 2004]. The response of Re to a pulse of rainfall typically occurs within a day when storm size is larger than a threshold (typically between 5 and 13 mm) [Huxman et al., 2004b; Baldocchi, 2008; Liu et al., 2012]. During long periods between precipitation pulses, soil microbial activity is limited by low soil moisture (θ), thus litter tends to accumulate beneath C3 woody vegetation into fertile islands [Cable et al., 2012]. Fertile islands contain labile sources of soil organic matter that are ready to support large pulses of Re upon priming by moisture [Cable et al., 2012; Ma et al., 2012]. When a large rainfall pulse then occurs, the response of Re precedes the response of GPP, which can result in a switch in sign of NEE from negative before rainfall, to positive immediately afterward, and sometimes back to negative [Huxman et al., 2004a; Luo et al., 2007; Wohlfahrt et al., 2008]. An understanding of the conditions under which component carbon fluxes are enhanced following precipitation pulses can better inform land surface models and improve simulations of carbon dynamics, which universally underestimate Re and overestimate Ra in semi-arid regions [Mitchell et al., 2011]. Consequently, the first aim of this study was to evaluate the functional responses of component carbon fluxes to potential meteorological and physical drivers.

[6] Rainfall is a primary driver for carbon and water fluxes in all semi-arid areas, but the patterns are especially apparent in central and northern Australia because of the large interannual variability in rainfall that is skewed toward infrequent, very wet years [Van Etten, 2009; Bowman et al., 2010; Morton et al., 2011; Papalexiou and Koutsoyiannis, 2013]. Storms can be formed in these regions by two mechanisms: the monsoon depression can bring moisture during the summer, while synoptic frontal systems can bring large cold fronts across the continent year-round [Zhang, 2010]. However, in central Australia, either storm system is unlikely to be strong in any given year [Berry et al., 2011]. It is during infrequent wet years that central Australia is distinguished by extreme precipitation. In contrast to other global semi-arid regions where frontal rainfall is restricted to winter, interactions between continental cold fronts and the monsoon depression can augment summer rainfall in central Australia, thereby generating storms with the persistence and spatial extent of winter storms but delivering higher intensity rainfall than is generated by convection storms alone [Kong and Zhao, 2010].

[7] Differences in photosynthetic capacity and leaf area index (LAI) between C3 and C4 vegetation dictate the timing and size of ecosystem photosynthetic and respiratory responses to precipitation pulses [Hattersley, 1983; Whitley et al., 2011; Barron-Gafford et al., 2012]. For example, woody vegetation is known to maintain photosynthesis over a wide range of temperatures and soil moistures, while the C4 grass layer is likely to contribute most of the annual GPP but only during the wet season [Hutley et al., 2005; O'Grady et al., 2009; Whitley et al., 2011; Barron-Gafford et al., 2012]. The vegetation over much of semi-arid, central Australia contains both C3 and C4 vegetation in a mosaic of two types of savanna, a biome containing an almost continuous grassy understorey and a discontinuous tree layer [Eamus and Prior, 2001]: (1) Mulga (Acacia spp.) woodland and (2) low, open eucalypt savanna with extensive Spinifex grass (Triodia spp.) and sparse Corymbia and Eucalyptus trees [Hutley et al., 2005; Eamus et al., 2006; O'Grady et al., 2009; Eamus et al., 2013]. The second aim of this work was to investigate the relationships between leaf function (A, Ra) and ecosystem carbon fluxes (NEP, GPP, and Re) as they respond to seasonality, light, humidity, and soil moisture content.

[8] This study will focus on a Mulga woodland savanna during a 2 year period that was characterized by a juxtaposition of extremely wet and dry years. We tested three research hypotheses in this study. First, the large, cyclically recurring storms of central Australia were hypothesized to strongly enhance GPP by providing substantial soil moisture and reduced vapor pressure deficit interspersed with periods of clear skies and full sunlight. Consequently, GPP was expected to be larger than Re in the wet year (i.e., carbon sink) but smaller than Re in the dry year (i.e., carbon source), which is generally observed in other semi-arid regions [Huxman et al., 2004a; Baldocchi, 2008; Wohlfahrt et al., 2008; Yan et al., 2011]. Third, we hypothesized that ecosystem photosynthesis would reflect the seasonal activity of C4 vegetation in the summer and C3 vegetation during the remainder of the year.

2 Material and Methods

2.1 Site Description

[9] This study site is part of the Terrestrial Ecosystem Research Network's (TERN) OzFlux and Australian Supersite Network (ASN) programs and is located on Pine Hill cattle station near Ti Tree NT, central Australia (22.28°S, 133.25°E, 549 m asl). At the nearby Bureau of Meteorology station on the Territory Grape Farm (BOM TGF), mean and median annual rainfall has been 318.3 and 297.8 mm, respectively, since recording began in 1987 (http://www.bom.gov.au/). Of the annual median rainfall, 72% falls during the summer months (DJF) and 86% falls during monsoon season (Nov–Apr), placing the site just inside the Australian Monsoon Tropics [Bowman et al., 2010; Hutley et al., 2011]. Air temperature at the site can range from a low of −4°C (11 August 1994) to a high of 46°C (5 January 1994) (http://www.bom.gov.au/), although mean minimal and maximal temperatures are in the range of 5°C and 22.6°C in July to 22°C and 37.5°C in January.

[10] The soil is a heavily weathered red kandosol, which is typical of large portions of semi-arid Australia and has a high potential for drainage [Schmidt et al., 2010; Morton et al., 2011]. The soil characteristics of this site have been described in Eamus et al. [2013] and are summarized here. The surface bulk density is 1.69 ± 0.02 g cm−3. Soil texture is a sandy loam (74/11/15% sand/silt/clay), and depth to groundwater is 49 m. Soil organic matter content (g) in 100 g of soil is equal to 1.1% at the surface and declines to 0.7 ± 0.1% at 1 m deep. The minimal soil organic matter content (0.5%) was observed where hardpan was encountered. Bare soil patches are likely surface expressions of the hardpan. Hardpan formation in the top meter, and possibly deeper, is commonly observed with this type of soil [Morton et al., 2011]. Variability in the depth of the hardpan implies that root development, near-surface storage of soil moisture, and infiltration are spatially heterogeneous in this flat (slope = 0.2%) landscape.

[11] This study was located in a Mulga savanna woodland at the southern end of the North Australian tropical transect [Hutley et al., 2011]. Mulga woodlands are widespread across Australia, covering 20–25% of the land surface area [Nicholas et al., 2011]. This site contains three varieties of Mulga and their hybrids, all of which form symbioses with nitrogen fixing cyanobacteria: Acacia aneura var. aneura, A. aneura var. intermedia, and A. aneura var. tenuis. The woodland canopy is 6.5 m tall and forms an incomplete cover with a basal area of 8 m2 ha−1. A patch of Spinifex grass (Triodia sp.), which crosses sharp ecotones to form a mosaic with Mulga [Nicholas et al., 2011], is located in a small gap (ca. 150 m2) within the measurement footprint of the eddy covariance system. The dominant understorey forbs and shrubs are distributed sparsely throughout the measurement fetch and include Psydrax latifolia, Eremophila gilesii, E. latrobei ssp glabra (Crimson turkey bush), Sida and Abutilon spp., and Solanum ellipticum (Potato bush). The dominant grasses form a nearly complete cover when conditions permit and include perennials Thyridolepis mitchelliana (Window mulga-grass), Eragrostis eriopoda (Naked woolybutt grass), and annual Eriachne pulchella ssp pulchella (Pretty wanderrie). Like the forbs and shrubs, mulga-grass uses the C3 photosynthetic pathway [Christie, 1975]. The remaining grasses use the C4 photosynthetic pathway [Hattersley, 1983; Simon and Alfonso, 2011]. Growth and photosynthesis in the understorey are moisture dependent, and the C4 component is more likely to be active in the summer than in other seasons [Cable et al., 2012]. Below the understorey, cryptobiotic crusts form a widespread ground cover that may contribute to rapid carbon turnover [Jameson, 2012] in response to small precipitation events [Schwinning and Sala, 2004].

2.2 Eddy Covariance, Meteorology, and Soil Measurements

[12] An eddy covariance system was employed to measure fluxes of carbon, water vapor, and energy. Beginning 2 September 2010, measurements of three-dimensional wind speed (CSAT3 sonic anemometer, Campbell Scientific Inc., Logan, UT, USA), virtual temperature (CSAT3), moisture density in air (Licor LI7500, LiCor Biosciences, Lincoln, NE, USA), and CO2 density in air (LI7500) were sampled at a frequency of 10 Hz. Covariances were computed every 30 min to generate fluxes following standard OzFlux QA/QC and corrections procedures, which include but are not limited to [Cleverly, 2011; Eamus et al., 2013]: outlier and spike removal, co-ordinate rotation, frequency attenuation correction, conversion of virtual heat flux to sensible heat flux, and the WPL correction [Wesely, 1970; Webb et al., 1980; Schotanus et al., 1983; Campbell Scientific Inc., 2004; Massman and Clement, 2004]. The eddy covariance system was mounted to a tower (Nally Instrument Towers, Adelaide, SA, Australia) at a height of 11.7 m and oriented toward the southeast, centered on the historically predominant wind direction ranging from easterlies to southerlies [Chen et al., 1991]. Energy balance closure, as determined by the ratio of daily total outgoing radiation and energy fluxes to total daily incoming radiation and energy fluxes, was 0.97 [Eamus et al., 2013].

[13] A threshold for friction coefficient (u*) filtering was identified to be 0.25 m s−1 from the relationship between NEE and u*. To do this, values of NEE were grouped into u* classes of 0.05 m s−1 and plotted against average NEE in that class. When applying the u* filter, 56% of NEE measurements were discarded when u* was smaller than the threshold. Failure to apply the u* filter results in underestimation of nocturnal fluxes in the presence of drainage flow or when turbulence within and above the canopy are decoupled [Aubinet, 2008; van Gorsel et al., 2008; van Gorsel et al., 2011; Oliveira et al., 2013]. However, wind measured beneath the top of the canopy was negligible whenever u* was below the threshold (data not shown), thus implying that (1) within and above canopy conditions were closely coupled, (2) cold air drainage was unlikely, and (3) large ejections during turbulent periods that follow nonturbulent conditions reflect storage fluxes. The effect of the u* filter on estimation of nocturnal respiration was evaluated by comparing filtered and nonfiltered fluxes (section 2.5.2).

[14] Auxiliary observations of solar irradiance (Es), soil moisture content (θ), soil temperature (Ts), vapor pressure deficit (D), and rainfall were collected to identify relationships between meteorology and carbon fluxes. Incident Es was observed from a four-component radiometer that was positioned at a height of 12.6 m (CNR1, Kipp and Zonen, Delft, The Netherlends). D was determined as the difference between atmospheric vapor pressure (kPa) and saturation vapor pressure at air temperature (HMP45C, Vaisala, Helsinki, Finland) at a height of 11.7 m. Onsite rainfall (CS7000, Hydrologic services, Warwick, NSW, Australia) was measured at a height of 2.5 m from the center of a 10 × 15 m clearing within the footprint of the tower. Surface θ was measured using soil water content reflectometers (CS616, Campbell Scientific Inc., Logan, UT) that were inserted horizontally with the upper and lower rods at 0.025 and 0.057 m deep, respectively. Surface Ts was observed using averaging soil thermocouples at depths of 2 and 6 cm. Temperature-corrected CS616 output was compared to gravimetric measurements from intact soil cores to empirically correct observed soil moisture to actual soil moisture (as described in Eamus et al. [2013]). Profile measurements of θ to a depth to 100 cm were collected using time domain reflectometry (CS605 and TDR100, Campbell Scientific Australia, QLD, Australia). To represent on-site variability in soil measurements, sensors were placed in microhabitats identified as bare soil, beneath Mulga canopies, and beneath understorey vegetation.

[15] To obtain a full, 2 year record of NEP, gaps that were introduced during QA/QC procedures were filled using a self-organizing linear output (SOLO) model. The SOLO model is a type of artificial neural network (ANN) that was trained on a self-organizing feature map (SOFM) of interrelationships between regressors: seven in this case [Eamus et al., 2013]. These regressors were determined through stepwise procedures and were: soil heat flux including heat storage above the sensors, net radiation, Ts, θ, atmospheric specific humidity, air temperature, and D [Hsu et al., 2002; Abramowitz et al., 2006; Eamus et al., 2013]. The combination of SOLO and SOFM procedures is the ANN analogue of multivariate analysis of covariance (MANCOVA), which results in low model error estimates (as small as 8 mmol m−2 day−1 RMSE [Eamus et al., 2013]).

2.3 Leaf Area Index, Leaf Physiology, and Autotrophic Respiration

[16] Leaf area index of the canopy and understorey was measured optically using the digital photography method [Macfarlane et al., 2007] as described in Eamus et al. [2013]. In short, 100 m transects were established within the tower fetch in cardinal quarter-directions: i.e., east, southeast, south, etc. Photographs were taken using a Canon DSLR camera fitted with a lens of short focal length (18 mm). The camera was oriented to 0° nadir (viewing upward) and 57.5° to the vertical (viewing downward) for canopy and understorey LAI, respectively.

[17] Satellite remote sensing was used to estimate LAI during the periods between site visits. Total site LAI was estimated using the MODIS product MOD15A2, which is provided as an 8 day composite average and matches total LAI measured photographically at this site [Eamus et al., 2013]. A spline function was fit to MODIS LAI to interpolate daily total LAI. Spline functions were also fit to the temporal sequence of canopy and understorey LAI observations to construct interpolated time series of proportional contributions to total LAI. A lower limit of 0.1 was set for the canopy to avoid unrealistic estimates of LAI in evergreen Mulga.

[18] Leaf gas exchange measurements were collected from Mulga and representative understorey leaves. Leaf photosynthetic LRFs to incident photosynthetically active radiation (PAR) were collected with a Licor 6400 (Licor LI6400XT, Li-Cor Biosciences, Lincoln, NE) on clusters of leaves of terminal branches during two measurement campaigns. To obtain LRFs, an external light source was used to increase PAR systematically from dark to a maximum of 3000 µmol m−2 s−1 after dark adapting in the cuvette for a minimum of 5 min to 10 min before the first measurement. LRFs on Mulga leaves were measured in the morning and at midday in December 2010 to identify A in full sunlight (Asat) early in the growing season before NEP became large.

[19] The y-intercept where the LRF goes to PAR of 0 was used as an estimate of leaf dark respiration (Rd [µmol m−2 s−1]). Measurements on Mulga leaves were collected 30 March to 2 April 2011 between 7.00 and 14.00. LRFs were measured for understorey leaves 4–7 April 2011, also between 7.00 and 14.00. Measured values of Rd are sensitive to enhancement of carbon fluxes due to predarkening illumination (i.e., light enhanced dark respiration, LEDR) [Atkin et al., 1998; Barbour et al., 2007; Barbour et al., 2011]. While LEDR can cause overestimation of Rd for 50–100 min after darkening [Atkin et al., 1998; Barbour et al., 2007], the effect is minimal in low light (e.g., during the 7.00 measurements) [Atkin et al., 1998; Barbour et al., 2011] and during a short period ca. 5–10 min after darkening, which is when LEDR is compensated by inhibition of Rd in the light [Atkin et al., 1998]. Thus, overestimation of Rd due to LEDR was expected to be present but minimal in our study.

[20] Measurements of leaf-level Rd were scaled by LAI to estimate daily Ra. The diurnal course of Rd was integrated across the daily series of LRFs (7.00 to 14.00) to estimate total daily Rd. Daily fraction of Ra from each vegetation layer (i.e., understorey or canopy) was determined as the product of LAI and total Rd. Proportions of canopy and understorey LAI were extrapolated to the beginning of the eddy covariance time series by maintaining a constant proportionality with the first measurement date. The daily sum of canopy and understorey values of total Rd × LAI was taken to be an estimate of total ecosystem Ra.

2.4 Light Response Functions and Ecosystem Metabolism

[21] The rectangular hyperbolic light response function was chosen for its simplicity, common use in photosynthesis modeling, and previous demonstration of its applicability in eddy covariance carbon research [Wohlfahrt et al., 2005; Lasslop et al., 2010]:

display math(2)

where α is the slope at low light and represents ecosystem quantum yield, β is maximum NEP (NEPmax) and is analogous to the maximum photosynthetic rate (Amax), and γ is the y-intercept where Es is zero.

[22] Short-term analysis of ecosystem metabolism included 30 min observations of Es and NEP, averaged over four 6 day periods: (1) springtime when the ecosystem was a carbon source (23–29 Nov 2010), (2) following an early summer rainfall event that resulted in the carbon balance switching from a source to a sink (10–16 Dec 2011), (3) at the end of a very wet summer (20–26 Feb 2011), and (4) shortly after a large springtime rainfall event when the ecosystem was again a carbon source (25–31 Oct 2011). LRFs using EC flux data were created individually for morning (before 12:00 local standard time, LST) and afternoon measurements during these four periods.

2.5 Carbon Flux Partitioning

[23] Carbon fluxes were partitioned from NEE in two steps (Figure 1): first Re was estimated, which was then used to estimate GPP in equation (1). Re was estimated from the responses of NEE to temperature when GPP was assumed to be negligible. The manner in which GPP was assumed to be negligible depended on time of day and was grouped into three classes or “methods” (Figure 1): (1) night, (2) sunlit at dawn and dusk, and (3) under full sunlight periods (sections 2.5.1–2.5.4). Ts rather than Ta was chosen for functional response curves because (1) more than 60% of Re originates from the soil [Janssens et al., 2001], (2) Ts was more representative of canopy and subcanopy Ta than measurements at 11.7 m (data not shown), and (3) use of a single temperature facilitates comparison of Re rates among methods (section 2.5.5). Only flux data with gaps that were unfilled were used to estimate Re to avoid the confounding effects of gap filling procedures, which are modeled in part on Ts, θ, and net radiation. All of the following subsections (2.5.1–2.5.6) refer to Figure 1.

Figure 1.

Schematic flow chart showing the methods for partitioning of carbon fluxes.

2.5.1 Nocturnal Respiration: Nightly Sums

[24] Observations of NEE were accumulated over a full nocturnal period to avoid double accounting of carbon that is stored at one time of night and released at another. It was assumed that nocturnal GPP was 0; thus, Renight was assumed to be equal to total nocturnal NEE on nights without gaps in the flux data. These estimates of Renight include storage fluxes during intermittent turbulence and the morning transition from calm to turbulent conditions [Aubinet, 2008]. Sources of error in this approach include flux loss due to unaccounted horizontal losses of stored CO2, which was minimal because of the flat topography (0.2% slope) and extensive fetch (> 10 km) at this site, and due to secondary uptake of respired CO2 by early morning photosynthesis. Underestimation of Re due to secondary uptake is more likely during the morning transition than during nocturnal episodes of intermittent turbulence. Thus, error of Renight that was estimated from nightly sums is expected to be proportional to the regression error term in thermal sensitivity functions (Figure 2).

Figure 2.

Average nocturnal ecosystem respiratory carbon flux (Renight) as a function of nightly average Ts. Data were categorized by date of measurement and within soil moisture (θ) classes of (a) 3–5%, (b) 5–10%, and (c) 10–22%. Curves 1–6: least squares regression curves representing exponential (Q10) functions in which those marked with the same letter do not have a significantly different slope. ns: slope is not significantly different from zero. Curve 7: Arrhenius function fit to Renight in the wettest θ class.

[25] Thermal sensitivity functions were used to estimate values of Renight during nights when gaps were present in the data. To account for moisture limitation on Renight, nightly average θ was used to group nightly observations into three classes: low (3–5%), medium (5–10%), and high (10–22%). Thermal sensitivity curves were developed from Renight and Ts in each θ class on nights lacking data gaps. Within a θ class, thermal sensitivity was found to vary by measurement date (lnRenight ANCOVA; F = 9.84; df = 5, 392; p < 0.0001; Figure 2). Exponential (i.e., Q10) functions were determined within each θ class and measurement period (Figure 2).

[26] Extrapolation of Renight to daytime temperatures for the purpose of comparing to estimates of Reday (sections 2.5.3–2.5.4) was done by fitting the Arrhenius function to Renight [Lloyd and Taylor, 1994]. Comparison between Q10 and Arrhenius functions is shown in the temperature sensitivity of wettest θ class (curves 6 and 7, respectively; Figure 2c). Because the Arrhenius function assumes that Re is not limited by soil moisture availability [Lloyd and Taylor, 1994], extrapolation of Renight was performed using data from the wettest θ class. Because thermal sensitivity in the wettest θ class was not significantly different from Q10 in drier classes (Figure 2), the Arrhenius function was assumed to represent a general extrapolation of Re to daytime temperature regimes.

2.5.2 Night: u* Filter

[27] Application of the u* filter to nocturnal fluxes resulted in a sample size of nights without gaps that was too small for nightly accumulation analysis. Consequently, thermal sensitivity was evaluated with 30 min fluxes rather than nightly total NEE (Figure 1). Extrapolation of u* filtered, nocturnal NEE fluxes to daytime temperature was performed using an Arrhenius function (section 2.5.1). In the presence of destorage fluxes during intermittent nocturnal turbulence, application of the u* filter results in overestimation of source fluxes [Aubinet, 2008]. Similarly, unrealistically large values of Re were obtained from application of the u* filter in this study (Figure 3); thus, u* filtered fluxes were not used in carbon partitioning.

Figure 3.

Daytime ecosystem respiration (Reday) as a function of soil temperature (Ts) class. Solid bars represent Reday estimated using the intercept of the light response curve (γ) and applied in low solar irradiance (Es < 500 W m−2). Striped bars represent the maximum observed daily net ecosystem exchange (NEEmax) in high light (Es > 500 W m−2). Crosses represent extrapolation of the nocturnal Arrhenius curve (curve 7, Figure 2) and triangles represent extrapolation of the nocturnal Arrhenius curve after application of the u* filter. Numbered lines: Q10 curves with the same letter do not have a significantly different slope.

2.5.3 Sunlit Fluxes: Dusk and Dawn

[28] Photosynthetic LRFs that were constructed from NEP and Es were used to estimate Reday. Daytime data (i.e., Es > 10 W m−2) were used to develop LRFs from the complete, 2 year data set. Observations of 30 min daytime NEP were grouped into 5°C Ts classes beginning at 6°C. Within a temperature class, a rectangular hyperbolic LRF was applied (nlinfit, Matlab) to obtain γ, an estimate of Reday at the point at which Es and GPP decline to 0 [Lasslop et al., 2010].

[29] During high-light conditions (Es > 500 W m−2) in semi-arid areas, midday carbon losses can be larger than γ [Scott et al., 2010]. These large midday respiratory fluxes occur at higher temperatures than γ and are the result of midday depression of photosynthesis. By assuming that depression of photosynthesis is least likely to confound LRF analysis in low light conditions, γ was taken to be an estimate of Reday when Es < 500 W m−2 (i.e., dusk and dawn). Double exponential functions were fit to Reday at the midpoint of each Ts class to determine Reday from Ts in each 30 min period (Figure 3).

2.5.4 Sunlit Fluxes: Full Sunlight

[30] A special application of the LRF method was used to estimate Reday when Es > 500 W m−2. By identifying periods when Asat can be assumed to equal zero, the LRF decomposes to the thermal sensitivity of NEE [Lasslop et al., 2010; Lasslop et al., 2012]. This is the first study to employ this application of LRFs to Reday. In our study, the 2 year maximal rate of NEE in high light (NEEmax) was identified in each of the 5°C Ts classes that were described in section 2.5.2. While not every incidence of positive NEE flux is associated with negligible GPP, the single value of NEEmax in each Ts class was assumed to occur when GPP = 0. This assumption is supported by previous observations of midday depression in Mulga leaves with consequential reduction of ET and stomatal conductance to zero [Eamus et al., 2013]. Errors in estimation of Reday due to GPP > 0 would lead to under-estimation of both Re and GPP.

2.5.5 Ecosystem Respiration: Method Comparison

[31] In summary, Re was determined from Q10 thermal sensitivity functions at night and light response functions during the day (Figure 1):

display math(3)

[32] All computational methods for estimating Re were evaluated as a function of temperature (e.g., γ(T)) with the exception of Renight on nights without gaps, which was totalled directly from measured NEE fluxes. Values of γ have been previously found to be in good agreement with combined measurements of nightly NEE and storage fluxes [van Gorsel et al., 2009]. Extrapolation of Renight via the Arrhenius function when the u* filter was applied resulted in values of Reday that were much higher (2 to 3.5 times) than those of any other method. Re estimated from each of these methods in the absence of the u* filter was in very close agreement at temperatures below their respective thermal optima (Figure 3), which implies that (1) the error implicit in estimated Re from each method is of approximately the same magnitude and direction and (2) the assumptions of each method are more likely correct than violated.

[33] Reday increased exponentially with soil temperature (Ts) to a critical (optimum) temperature, which occurred at a higher temperature in high light (Es > 500 W m−2) than in low light (lnReday ANCOVA; F = 17.75; df = 3, 11; p = 0.002; Figure 3). The difference in optimal temperature at dusk and dawn versus full sunlight is a result of the relatively cool morning and evening periods. Above the optimum temperature, Re declined exponentially with increasing Ts for both daytime methods of calculation (Figure 3).

2.5.6 Gross Primary Production and Daily Totals

[34] Following determination of Re, GPP was computed from the difference between Reday and NEE (Figure 1). By definition, Re and NEE are equal in the nocturnal measurements; thus, GPP was set to 0 at night. Determination of 30 min GPP from the residual of Reday and NEE [e.g., Reichstein et al., 2005] was chosen in favor of using LRFs to determine GPP [e.g., Lasslop et al., 2010] because of (1) the relative importance of Re in NEP at high light levels and (2) decoupling of LRF parameters and VPD (data not shown). Daily totals of GPP were determined from the sums of 30 min sunlit GPP and were computed from gap-filled NEE. Daily total Re was determined from the sum of 30 min Reday and total Renight.

2.6 Data Analysis

[35] The temperature sensitivity of Renight (i.e., the slope of the relationship between Renight and Ts) was compared among θ classes and seasons using analysis of covariance (ANCOVA) between ln(Renight) and Ts. In the event of rejecting the null hypothesis H0: global slope b = bi = 0 where the subscript i is the ith factor (observational period × θ class), post-hoc least significant differences (multiple comparisons) procedures were performed to test H0: bi = bj for all factorial pairs (i.e., H0: slopes for each grouping are parallel). All statistical comparisons were performed using Type III sums of squares and a presumed critical p = 0.05 (Matlab 7.9.0 R2009b, The Mathworks, Natick, MA, USA). Double-exponential curves were applied (nlinfit) to Reday and Ts under each light level, and slopes were compared using ANCOVA as with Renight. Summary statistics of thermal sensitivities in Reday and Renight are presented in supporting information Table 1.

Table 1. Summary of Leaf Gas Exchange Responses to Photosynthetically Active Radiation (PAR)a
  1. aCondition: before (7 Dec 2010) and following (8 Dec 2010) overnight rainfall at the midpoint of a six-day rainfall event (53 mm); α: initial slope of the light response function; Asat: net photosynthetic rate in full sunlight (2000 µmol m−2 s−1 PAR); Amax: maximum net photosynthetic rate; Rd: dark respiration (0 µmol m−2 s−1 PAR).
DryBefore 10.000.021391.893.16−1.05
DryAfter 10.000.021390.692.11−1.32
WetBefore 10.000.021393.927.37−2.37
WetAfter 10.000.021395.247.37−1.05

[36] The temporal dependence of Re and GPP upon Es, D, Ts, θ, and atmospheric pressure (P) was evaluated using power spectral and coherence analyses [Cleverly et al., 2002]. These analyses were performed on 26 daily observations of total Es, daily maximal and minimal D (Dmax and Dmin, respectively), maximal and minimal Ts (Tsmax and Tsmin, respectively), average θ, and average P. Three 64 day analysis windows covered the wet summer (30 Dec 2010 to 3 Mar 2011), winter 2011 (20 Jun to 22 Aug 2011), and the dry summer (30 Dec 2011 to 2 Mar 2012). Power spectral density (PSD) was computed for each time series using a fast Fourier transform (sptool, Matlab). Coherence between PSDs was computed using a Hanning window of length of 25 and window overlap of 24 (mscohere, Matlab). Coherence between PSDs is indicative of temporal squared correlation (0–1) between physical driving variables (e.g., θ) and Re or GPP. Storm cycle period was identified from the peak frequency or frequencies in the coherence between Es and θ to account for (1) periods of low pressure and cloudiness that do not produce rainfall and (2) phase shifting of θ to postrainfall exponential decay [Eamus et al., 2013].

[37] Envelope analysis was used to evaluate the net and gross carbon flux limit within the observed range of Es, D, Ts, and θ. In this example of an extreme value problem [Coles, 2001], observations of NEP, GPP, or Re were blocked into meteorological classes, within which the maximal value was identified. A function was then applied that circumscribes the carbon flux maxima as a function of meteorological classes (e.g., see Figure 2 in Whitley et al. [2009]). Block maxima were determined in Ts classes of 5°C in width (e.g., 5–10, 50–55°C), D in classes of 0.65 kPa (e.g., 0–0.65, 5.2–5.85 kPa), and Es in classes of 100 W m−2 (30 min observations) or 5 MJ m−2 day−1 (daily totals). All blocks were further divided into four θ classes: < 4%, 4–8%, 8–12%, and > 12% for daily average θ or < 5%, 5–11%, 11–17%, and > 17% in 30 min observations. The lowest Es class (0–100 W m−2) includes all nocturnal observations. For daily response functions, gap-filled data were used to avoid sample-size restrictions. In the 30 min observations, unfilled data sets were used to avoid autocorrelation with meteorological relationships that were incorporated during gap-filling procedures. Summary statistics of curves in the envelope analysis are presented in supporting information Table 2.

Table 2. Summary of Annual Meteorology and Component Carbon Fluxesa
  1. aSingle values represent annual totals; Binomial values represent daily average ± standard error, se; Ts: soil temperature; D: vapor pressure deficit; Es: solar irradiance; θ: % volumetric soil moisture content; Re: ecosystem respiration; γ: estimate from intercept of the light response function; NEEmax: estimate from NEEmax method; NEP: net ecosystem photosynthesis; GPP: gross primary production.
n [days]360347
Rainfall [mm]565.2183.8
Rainfall (BOM) [mm]542.6258.4
Ts [°C]24.3 ± 0.125.6 ± 0.1
D [kPa]1.5 ± 0.012.0 ± 0.01
Es [MJ m−2 day−1]22.0 ± 0.321.9 ± 0.3
θ [cm3 (100 cm3)−1]6.8 ± 0.044.7 ± 0.02
Reday (γ) [mol m−2 yr−1]6.35.8
Re (Nocturnal) [mol m−2 yr−1]16.86.8
Reday (NEEmax) [mol m−2 yr−1]21.521.2
Re [mol m−2 yr−1]44.633.9
NEP [mol m−2 yr−1]21.1−0.3
GPP [mol m−2 yr−1]66.233.2
GPP / Re1.49 ± 0.031.06 ± 0.03

3 Results

3.1 Ecosystem and Leaf-Scale Photosynthesis

[38] Figure 4 shows the daily pattern of NEP in four contrasting periods across the study. Average Ts was 33.8 ± 0.4, 30.7 ± 0.4, 30.3 ± 0.2, and 27.8 ± 0.3°C during each period in Figures 4a–4d, respectively. The largest rate of NEP was observed in February 2011 (NEP > 10 µmol m−2 s−1) and the largest negative extreme occurred during the early morning in December 2010 (NEP < −5 µmol m−2 s−1; Figure 4). NEP was smallest during springtime (< 2 µmol m−2 s−1), but positive photosynthetic activity was supported in the morning (Figures 4a and 4d). During midday, depression of NEP was common except during the late summer when NEP was largest (Figure 4). In February after two months of heavy rainfall, NEP peaked in the early afternoon rather than early morning (cf. Figures 4b and 4c). Periods of large nocturnal respiratory fluxes occurred during the morning transition (Figures 4b and 4c) and intermittently during the latter half of the night (Figure 4d).

Figure 4.

Average diel pattern of 30 min net ecosystem photosynthesis (NEP) ± se during four periods representing (a) the November carbon source period, (b) early heavy rainfall, (c) late heavy rainfall, and (d) at the end of extended drought.

[39] The ecosystem LRFs are shown in Figure 5. During the spring, NEP was positive in the morning under low light (Figures 5a and 5d). In the springtime, hysteresis in LRFs resulted in NEP that was smaller (i.e., more negative) when Es > 500 W m−2 than when Es was 0 (i.e., Figure 5a), which supports the assumption that GPP was negligible when NEEmax was measured in each of the nine Ts classes. Model parameters α and β were larger in the morning than in the afternoon, while γ was smaller in the morning (Figure 5). Ecosystem photosynthesis did not reach light saturation (i.e., the LRF did not reach a plateau) in February 2011. Values of γ ranged from −0.3 to −5 µmol m−2 s−1 across the four selected periods (Figure 5).

Figure 5.

Ecosystem light response functions. Values represent 30 min average NEP ± se and Es ± se. Curves in Figures 5a and 5d are third-order polynomials for illustrative purposes, while curves in Figures 5b and 5d are rectangular hyperbolic functions that were fit in the morning (AM; boxes and solid lines) and afternoon (PM; circles and dashed lines).

[40] Phyllode-scale A and Rd in Mulga showed large, short-term responses to rainfall (Table 1). Before rainfall resumed, small rates of Asat during the morning declined to near zero by midday (Table 1), which further confirms the assumption that A and GPP are negligible at midday under some conditions. Just before rainfall, Asat and Rd were small and then increased on the day following overnight rainfall. Before rainfall, Rd was slightly smaller in morning than at midday (Table 1). Conversely after rainfall, Rd was smaller at midday than in the morning (Table 1).

3.2 Meteorology and Soil Moisture

[41] More than three times the rainfall accumulated during the first year of study than during the second year (Table 2). Heavy precipitation fell in December 2010 through April 2011 (Figure 6). More than half of the dry year (2011–2012) rainfall (Table 2) fell in January and February 2012 (119.6 mm) (Figure 6). In 2011, April through September was a very dry period, receiving only 1.2 mm in total rainfall (Figure 6). Likewise, no precipitation was recorded during May through August 2012 (Figure 6). Total monthly rainfall was below average during 15 of 24 months, was 0 mm during 6 months, and was less than or equal to 1 mm in another 3 months. Meteorological drought, which is defined as three or more consecutive months of below average precipitation [AMS, 1997; Dai, 2011], was experienced three times: July–September 2011, November 2011 to February 2012, and May–August 2012 (Figure 6). The research site and nearby BOM TGF station did not receive substantially different rainfall amounts in the first year, but the difference between them was 74.6 mm in the second year (Table 2).

Figure 6.

Seasonal time series of rainfall, Es, vapor pressure deficit (D), Ts, and θ.

[42] Daily meteorology and soil water contents are shown in Figure 6. In the first year, mean daily surface θ was large (> 0.10 m3 m−3) and mean D small (Table 2), Ts exceeded 40°C in 4% of the total observations, and D exceeded 4 kPa during 4% of observations. During the second year, mean surface θ was small and mean D large (Table 2), Ts exceeded 40°C in 7% of observations, and D exceeded 4 kPa in 10% of observations. Neither average Ts nor average daily Es was substantially different between years 1 and 2 (Table 2). Daily total Es was largest in the summer during interstorm periods (e.g., 34.1 MJ m−2 day−1 on 21 Dec 2010) and was smallest during winter under deep cloud cover (e.g., 1.1 MJ m−2 day−1 on 11 Jul 2012), although Es could be nearly as small during summer storms (e.g., 3.0 MJ m−2 day−1 on 27 Feb 2012; Figure 6).

[43] Storm and interstorm intervals resulted in large, rapid and cyclic shifts of Es, Dmax, Tsmax, and θ (Figure 6). Daily Tsmax and Dmax increased through summer interstorm periods, reaching large values before dropping after the first major storm of the late summer or early autumn (Figure 6). Surface θ was regularly high (> 20%) in response to rainfall events in the first year, but only infrequently following rainfall in the second, drier year (Figure 6). Deep θ (100 cm) followed one of two patterns: (1) θ measured in one array was invariant and small (4.6 ± 0.005%) throughout the 2 year study period, or (2) θ measured in the other array was consistently larger (typically 10%) and increased slightly (2–4%) following series of storms, reaching a peak in the autumn (late Mar or early Apr) of both years of the study (Figure 6). This variability is the result of the spatial heterogeneity in the distribution of relatively shallow (typically 1 m) hardpans that prevent deep drainage of water.

3.3 Leaf Area Index, Respiration, and Gross Primary Production

[44] As with the meteorology, daily LAI and carbon fluxes were responsive to rainfall patterns (Figure 7). Canopy and understorey combined to result in a total LAI of 0.2–0.9 (Figure 7). Understorey LAI was low (but not 0 because of the presence of woody evergreen shrubs) in June and November 2011. Ra was smaller than Re during the early autumn 2011 and larger than Re at the end of that autumn (Figure 7). Fluctuations in Ra closely mirrored variation in LAI, from which Ra was scaled (Figure 7).

Figure 7.

Seasonal patterns of MODIS LAI, daily rainfall, total gross primary production (GPP), ecosystem respiration (Re), autotrophic respiration (Ra), and carbon balance (GPP / Re). MODIS LAI was interpolated to daily values using a spline function.

[45] Both Re and GPP were larger in the first year than in the following year (Table 2). Renight was larger in the first year than the second (201 g C m−2 yr−1 and 82 g C m−2 yr−1, respectively; Table 2). Total Re was 25% larger during the first year than during the second year (535 g C m−2 yr−1 and 407 g C m−2 yr−1, respectively; Table 2). During the first year, GPP was twice as large as in the following year (794 g C m−2 yr−1 and 398 g C m−2 yr−1, respectively; Table 2).

[46] Peak GPP was 465 mmol m−2 day−1 (5.6 g C m−2 day−1) in the first year (Sep 2010 to Aug 2011, n = 360 days) and 240 mmol m−2 day−1 (2.9 g C m−2 day−1) second year (Sep 2011 to Aug 2012, n = 347 days; Figure 7). Average daily GPP was 184.0 ± 5.7 mmol m−2 day−1 (2.2 ± 0.07 g C m−2 day−1) in the first year and 95.7 ± 2.4 mmol m−2 day−1 (1.1 ± 0.03 g C m−2 day−1) in the second year. On average, GPP exceeded Re by 49% in the first year, while GPP and Re were closely matched in the second year (Table 2 and Figure 7). Seasonally, GPP and Re were largest during the wet summer (DJF 2010–2011) and smallest in the following periods: during the dry summer (2011–2012), in November regardless of precipitation regime, and during both winters (JJA; Figure 7). Periods when GPP was consistently lower than Re (i.e., GPP / Re < 1) occurred in November 2010 and spring (SON) and summer 2011–2012 (Figure 7). Spikes of Re occurred 1–2 days after the start of a precipitation event throughout the 2 year period of the study, with one exception in June 2011 when Re did not increase in response to rainfall (Figure 7). GPP tended to be minimal during rainy periods and often but not always followed 2–5 days later by a large increase (Figure 7). Pulses in GPP were more frequent during the summer (DJF) and autumn (MAM) than in other seasons; for example, there was negligible increase in GPP in response to the first large rain in late October 2011 (Figure 7).

3.4 Ecosystem and Meteorological Drivers of GPP and Re

[47] The envelope functions describing meteorological limits on daily-integrated GPP and Re are shown in Figure 8. Daily GPP increased with increasing Ts across the low to moderate range (12–22°C) for all classes of soil moisture (Figure 8a), while there were few differences in GPP across moisture classes in this temperature range. However across the moderate-to-high range (22–38°C), GPP of the lowest θ was consistently smaller than that in the three larger classes of θ. A peak in GPP occurred in the range 22–28°C for the three largest classes of θ, but this was absent for the lowest class of θ (Figure 8a).

Figure 8.

Parameter envelopes for GPP or Re as functions of daily θ and Ts, D, or Es. Splines were used for all curve fits. Points represent the maximal flux in a given θ × meteorology block. N = 700 days.

[48] A similar qualitative pattern in GPP as a function of D across the four classes of θ was apparent (Figure 8c). Maximal GPP occurred at about 1.5 kPa for the three largest classes of θ and there was little difference in the response across these three classes of θ across the entire range of D. In contrast, GPP was essentially independent of D across the entire range for the smallest class of θ. As Es increased from very low to high values (2–25 MJ m−2 d−1), GPP increased with little difference in response between the three largest classes of θ (Figure 8e). Reductions in GPP at the largest values of Es (> 22 MJ m−2 d−1) were apparent in these three θ classes. For the lowest θ class, GPP was consistently smaller than in the other three classes when Es > 18 MJ m−2 d−1.

[49] Daily Re increased with increasing Ts in all classes of θ. However, small reductions in Re were observed at the largest values of Ts in the two intermediate classes of θ. In the smallest class of θ (< 4%), Re did not show a decline at high values of Ts and was consistently smaller than that of the other three classes when 36°C > Ts > 22°C (Figure 8b). Daily ecosystem respiration was maximal at intermediate values of D (1–2 kPa) but declined at smaller or larger values of D for the three largest classes of θ (Figure 8d). In contrast, Re increased approximately linearly with increasing D for the lowest class of θ (Figure 8d). Daily Re increased with increasing Es for low to moderate values of Es in the three largest classes of θ but declined at the highest values of Es (Figure 8f). In contrast in the smallest class of θ, Re increased approximately linearly with increasing Es (Figure 8f).

[50] As Ts increased, 30 min averaged values of GPP, NEP, and NEE increased to a peak value at moderate values of Ts but then declined with further increases in Ts across all four classes of θ (Figures 9a–9c). Temperature optima for GPP and NEP were in the range of approximately 30–38°C. The highest values of NEE fell into a narrow range of Ts (Figure 9c): 20–25°C (during the morning transition to turbulent conditions), 25–30°C (during nocturnal intermittent turbulence), 30–35°C (near-dusk turbulence events following calm conditions and little preceding photosynthesis). Maximal, 30 min averaged values of GPP and NEP were observed at moderate values of D (2–3 kPa), with smaller GPP and NEP observed at smaller or larger values of D (Figures 9d–f). GPP and NEP were larger for the intermediate class of θ (5–11%) for all except the largest D. In contrast, NEE tended to decrease with increasing D across the full range of D and in all classes of θ (Figure 9f). Thirty minute averaged values of GPP and NEP increased curvi-linearly with increasing Es (Figures 9g and 9h). Both GPP and NEP approached an asymptotic value at full sunlight, except for the response of NEP in the smallest θ class, which did not approach an asymptotic value (Figure 9h).

Figure 9.

Parameter envelopes for GPP, NEP, and NEE as functions of 30 min θ and meteorological variables (Ts, D, or Es). N = 26,922 30 min periods.

[51] Storm conditions (low Es, Ts, and D and high θ) typically persisted for 2–5 days, on average, while interstorm cycles were 5–11 days long (Table 3). During the wet summer (2010–2011), GPP was limited by reduced Es and Tsmax during the storm, but during the interstorm interval, GPP and Re responded positively only to increasing Es (Table 3). Storms were less frequent in the winter; thus, interstorm intervals were longer (10.7 days) (Table 3). Re responded to storm and interstorm fluctuations in Es and Tsmax during the winter (Table 3). GPP responded to cycles of cloudiness rather than rainfall in the dry summer (2011–2012), while Re responded only to fluctuations in Es (Table 3).

Table 3. Summary of Coherence Statisticsa
 Period (Days)|r|
  1. aPeriod: 1/frequency [Hz]; |r|: magnitude of the correlation coefficient, calculated as the square root of the coherence in a given frequency range.
Summer (30 Dec 2010 to 3 Mar 2011)
Winter (20 Jun to 22 Aug 2011)
Summer (30 Dec 2011 to 2 Mar 2012)

4 Discussion

[52] Both gross primary production (GPP) and ecosystem respiration (Re) were enhanced by the magnitude and frequency of storms, particularly during the extremely wet year (section 4.1). As hypothesized, the enhanced soil moisture, reduced vapor pressure deficit (D), and large solar irradiance (Es) during cycles of storms and interstorms contributed to enhanced GPP in the wet year (section 4.2). As predicted in our third hypothesis, ecosystem responses to meteorological extremes in central Australia were related to large, seasonal responses of C4 vegetation during summertime, creating a large carbon sink in the wet year (section 4.3). However, the dry year did not produce the large carbon source that is observed in other semi-arid regions where more decomposable substrate tends to accumulate (section 4.1). Ecosystem fluxes were coupled as hypothesized to photosynthesis of the C3 Mulga canopy during dry periods, through which carbon neutrality (i.e., GPP:Re ≈ 1) was maintained by positive photosynthesis in low light and during the autumn after deep soil moisture had peaked (sections 4.2).

4.1 Large-Scale Patterns of Gross Primary Production, Ecosystem Respiration, and Carbon Budgets

[53] Average annual GPP at our site was large (50 mol m−2 yr−1, 601 g C m−2 yr−1) for a semi-arid ecosystem [cf. Prentice et al., 2001]. Relative to global biomes, carbon uptake in this study was larger than the annual average GPP in semi-arid or arid areas (19 mol m−2 yr−1, 228 g C m−2 yr−1) or tundra (24 mol m−2 yr−1, 264 g C m−2 yr−1); was about the same as the GPP of temperate grasslands and shrublands (40 mol m−2 yr−1, 480 g C m−2 yr−1) and boreal forests (50 mol m−2 yr−1, 601 g C m−2 yr−1); and was smaller than the GPP of temperate forests (79 mol m−2 yr−1, 949 g C m−2 yr−1), tropical savannas and grasslands (95 mol m−2 yr−1, 1.14 kg C m−2 yr−1), and tropical rainforests (193 mol m−2 yr−1, 2.32 kg C m−2 yr−1) [Prentice et al., 2001; Beer et al., 2010]. Intermediate in GPP between semi-arid and tropical savanna ecosystems, carbon uptake in this study was in general agreement with, although slightly larger than, GPP of a semi-arid C4 grass and Eucalyptus savanna in Queensland Australia (32 mol m−2 yr−1, 384 g C m−2 yr−1) [Hutley et al., 2005]but was substantially smaller than a tropical C4 grass and Eucalyptus savanna in northern Australia (113–173 mol m−2 yr−1, 1.36–2.08 kg C m−2 yr−1) [Chen et al., 2003; Hutley et al., 2005; Kanniah et al., 2011]. The large annual GPP in this study (relative to other semi-arid regions) was the result of significant photosynthetic responses by the largely C4 understorey to regular cycles of widespread, intense, and multiday storms (cf. Figures 6 and 7).

[54] Total annual GPP was larger in the wet year than in the dry year (Table 2). Reduction in productivity during dry periods has been extensively documented in annual- and continental-scale droughts in Europe, Australia, and the US [Campos et al., 2013], as well as in semi-arid shrublands in western North America [Baldocchi, 2008; Scott et al., 2009] where it is common for the carbon balance to switch from a sink to a source in response to drought [Luo et al., 2007; Bowling et al., 2010]. In the present study, a positive carbon balance was maintained during dry periods because reductions in GPP were compensated by proportional reductions in Re (Figure 7). In addition, availability of θ or groundwater at depth contributes to maintenance of a carbon sink during the dry season in tropical north Australian savannas [Hutley et al., 2005; Whitley et al., 2011], in the first year of drought in the boreal zone of Canada [Baldocchi, 2008], and in tropical and humid forests [Saleska et al., 2003].

[55] Annual Re was neither small nor large during the wet year but was small in the dry year (Table 2). In other semi-arid ecosystems having a similar mean annual temperature (25.0°C in this study), estimates of Re range between 29 and 63 mol m−2 yr−1 (348–757 g C m−2 yr−1) [Bahn et al., 2010], which is similar to, but generally larger than, Re in the present study. There are two explanations for the small rates of Re that were observed during the dry year: (1) θ was reduced, which inhibits heterotrophic respiration (Rh) [Baldocchi, 2008; Bowling et al., 2010; Yan et al., 2011; Cable et al., 2012], and (2) GPP was smaller in the dry year, with consequential reductions in substrate supply and hence autotrophic respiration (Ra) [Janssens et al., 2001; Migliavacca et al., 2011]. In the soils at our study site that are poor in organic matter: (a) Rh was small, (b) Ra was the dominant component of Re, (c) the “basal” respiration rate (i.e., the decomposition rate of soil organic matter [Kuzyakov and Gavrichkova, 2010]) was negligible, and (d) reduction of substrate supply during the dry year contributed to maintaining a positive or neutral carbon budget (i.e., GPP ≥ Re), in contrast to grass-dominated savannas in semi-arid regions, which are a source of carbon during drought [Scott et al., 2009].

4.2 Daily to Seasonal Patterns of Production and Respiration

[56] Seasonally, GPP was smallest in the winter (JJA), largest in summer (DJF) and autumn (MAM), and mostly constrained to pulses following precipitation during the dry summer and autumn (Figure 7). Generally in semi-arid regions, the presence of a GPP response to a rainfall pulse is infrequently achieved because of the high frequency of very small (< 5 mm) events and unpredictability in the timing of large events [Huxman et al., 2004b; Schwinning and Sala, 2004; Baldocchi, 2008]. Summer rainfall in central and northern Australia, in contrast, occurs as large, multiday storm cycles (Figure 6) that are created by cyclic weakening of the monsoon depression [Berry et al., 2011]. This combination of soil moisture recharge following large storms and intervening periods of abundant sunlight acts to enhance GPP in central Australia.

[57] In the summer, pulses of GPP were synchronized with increasing leaf area index (LAI) of the understorey (Figure 7). Growth and assimilation in C4 grasses are more likely to respond rapidly to rainfall than in C3 woody vegetation when Ts and D are high [Huxman et al., 2004b; Barron-Gafford et al., 2007]. In contrast, Mulga is tolerant of low xylem water potentials (< −7 MPa) [O'Grady et al., 2009], which allows for small but continued physiological function at very low θ. After the sharp decline in Ts and D that occurred in February of each year (Figure 6), pulses of GPP were sustained longer and extended through the autumn across a month without substantial rainfall (Figure 7), thus reflecting continued use of soil water that is stored above the hardpan. C3 woody vegetation can be active across a wide range of temperatures as a consequence of access to groundwater or θ reserves within the root zone [Hastings et al., 2005; Baldocchi, 2008; Barron-Gafford et al., 2012]. Likewise in this study, the magnitude of GPP pulses following summer rainfall depended upon (1) C4 understorey growth in response to antecedent rainfall events and (2) Mulga production when it was not limited by high summer temperature or reduced θ.

[58] Like GPP, Re was small in the winter, larger in the summer and autumn, and generally restricted to brief pulses in the dry year (Figure 7). Thermal sensitivity (i.e., the slope of the Ts-Renight curve, Q10) was highest in the autumn and winter following the wet summer (curve 1, Figure 2), indicative of a larger decomposable substrate supply [Lloyd and Taylor, 1994; Janssens et al., 2001]. Substrate supply was large following senescence of the summer-active understorey and remained high through most of the following winter when microbial Rh was limited by low Ts and low θ (cf. Figures 2, 6, and 7). In contrast to other semi-arid regions where decomposable substrate accumulates under C3 shrubs and trees that are typically deciduous [Cable et al., 2012], the majority of Australian woody vegetation is evergreen with sclerophyllous leaves that decompose slowly and contribute less to the availability of decomposable substrate than the episodic understorey with its large seasonal fluctuations in LAI and root turnover.

[59] Although Re tended to be larger in the summer, Re declined with very high temperature (Figures 3 and 8). Two explanations have been proposed for the apparent reduction of Re at high temperatures in semi-arid ecosystems [Anderson-Teixeira et al., 2011]: (1) reduced substrate is available during dry or drought conditions due to reduced GPP at high temperature and (2) inverse autocorrelation exists between high temperature and soil moisture such that microbial respiration is limited by low θ during periods of high temperature. Reduction of Re at high temperatures (i.e., thermal optimisation) in this study was the likely result of moisture restrictions on microbial decomposition, and that restriction occurred at a higher temperature in full sunlight than near dawn or dusk.

[60] In semi-arid regions, photodegradation is the physical means by which plant litter is degraded in full light, at high Ts, and in the absence of θ [Gallo et al., 2009; Rutledge et al., 2010]. Wintertime Tsmax was high enough to promote photodegradation of a substantial fraction of the undecomposed litter, which was observed by the temporal correlation of wintertime Re to Tsmax (Table 3). Photodegradation does not lead to full decomposition, and the remaining “photodegraded” litter can be a labile source of carbon [Ma et al., 2012], contributing to observed Re responses to the early storms of spring and summer (Figure 7) in combination with flushing of CO2 from soil pores through infiltration immediately following the first storm of the wet season (cf. Figure 7 and Huxman et al. [2004b]). Thus, during periods when GPP and Ra were small, Rh was responsive to microbial decomposition of the senescent understorey, photodegradation on warm days in the winter, displacement of CO2 from soil pore spaces, and decomposition of photodegraded substrate following each of the first two large, springtime storms (Figure 7).

4.3 Subdaily Net Photosynthesis, Photosynthetic Light Responses, and Leaf Area Index

[61] Over subdaily time periods, GPP and NEP were limited by high and low values of Ts (> 40°C and < 20°C, respectively), large D (> 4 kPa), and low Es (< 100 W m−2; Figure 9). Low θ (< 5%) limited NEP at low to moderately high values of Ts (< 35°C; Figure 9) and wetter soil (5–10%) was associated with enhanced GPP and NEP (Figure 9). Apparent increases in GPP with increasing but low values of D reflected the autocorrelation of D and Ts in the early daylight hours. Thus, increasing D in the morning is caused by increased Ts, and this resulted in an apparent increase in GPP with increasing D (Figures 9a and 9d). Interestingly, the responses of GPP to D resembles the three-phase response of transpiration to increasing D [Eamus and Shanahan, 2002]: namely, increasing GPP at low values of D (Figures 9a and 9d), reaching a plateau or peak at low-to-moderate values of D, and finally declining with high D due to stomatal limitations, which suggests that it is likely that some features of the responses of GPP to D are attributable to stomatal function. Limitations on GPP and NEP due to low θ, high and low Ts, low Es, and high D were autocorrelated over the course of a day and illustrative of the effects of meteorology on plant and ecosystem photosynthesis in this Mulga savanna (cf. Figures 6 and 9).

[62] When understorey LAI was small (< 0.5), NEP declined after midmorning, which lead to positive respiratory fluxes (i.e., NEE > 0 µmol m−2 s−1) at midday in the spring (cf. Figures 4 and 7). Scott et al. [2010] observed moderate levels of positive net ecosystem respiration (i.e., NEE = 0–5 µmol m−2 s−1) in full sunlight while the soil dried, comparable to NEEmax observed in this study (Figure 3). Midday depression of NEP in water-limited ecosystems is the result of three factors, singly or in combination [Fu et al., 2006]: (1) photoinhibition in chloroplasts and consequentially elevated Ra, (2) reduced stomatal opening in response to high D and low θ, or (3) increased Rh in the midday and afternoon heat. Midday depression of NEP during periods of low understorey LAI and following drought (cf. Figures 4 and 7) was offset by positive assimilation rates in the morning by Mulga (cf. Figure 4 and Table 1) and possibly by photosynthesis in the cryptobiotic soil crust [Jameson, 2012].

[63] Photosynthetic responses to light were largest when LAI of the canopy and understorey was high (total LAI > 0.5; Figure 7). When understorey LAI was high, NEP did not saturate in full sunlight (Figure 5c), which is a characteristic of C4 vegetation [Hattersley, 1983]. Conversely, photosynthesis in Mulga, which are relicts of former central Australian rainforest understorey vegetation [Martin, 2006], saturated at moderate light levels (i.e., a large quantum yield, α; Table 1). The combination of these photosynthetic responses resulted in ecosystem LRFs that were characterized by large quantum yield (i.e., the initial slope of the LRF, α), large NEPmax, and failure to reach photosynthetic saturation in full sunlight (Figure 5c). Because of these photosynthetic characteristics of Mulga and C4 grasses, Mulga savannas of central Australia have the potential to generate large, positive carbon budgets (i.e., GPP > Re) across vast areas of the Australian continent, particularly following infrequent rainfall events.

5 Conclusions

[64] Among semi-arid and arid regions of the world, the tropical Mulga savannas of central and northern Australia are distinguished by large gross primary production (GPP) and low-to-moderate rates of ecosystem respiration (Re). Over the 2 years of this study, the amount of rainfall was more than 250 mm above the long-term average in the first year and more than 100 mm below the long-term average in the second year. Across these 2 years, the carbon budget shifted from a strong carbon sink to carbon neutral (i.e., GPP = Re). Three main conclusions were drawn from this study:

  1. [65] Ecosystem photosynthesis was enhanced by extreme and cyclic rainfall events that occur across northern and central Australia. Even during one of these wet (much higher than average rainfall) years, the monsoon depression experiences cyclic weakening that leads to interstorm periods of large solar irradiance (Es), high soil temperature (Ts), and high vapor pressure deficit (D). It is the abundant light and soil moisture that characterizes these interstorm periods and stimulates growth and photosynthesis in the C4 understorey (cf. Tables 1 and 3, Figure 7). In contrast, Mulga photosynthesis was stimulated when moisture stores below 1 m were recharged (cf. Figures 6 and 7) and when neither high Ts nor D limited photosynthesis (cf. Table 3, Figures 6 and 7).

  2. [66] Spatial heterogeneity of soil hardpan at depth and resultant soil moisture (θ) provides storage reservoirs at various depths, facilitating photosynthesis by both deep- and shallow-rooted species. Recharge events like the first year are uncommon, and decades may pass between the largest storm seasons [Ganter and Tobin, 2012]. In contrast to semi-arid areas with limited recharge below 50 cm [Kurc and Small, 2007], θ was recharged regularly in the dry year (Figure 6) because of the large infiltration rates in kandosol soil. Mulga has survived through Australia's drying climate over the past 0.5 Myr [Martin, 2006] in the face of an indeterminately long wait for the return of favorable conditions through (a) tolerance of low xylem water potential and (b) efficient and nearly exclusive use of stored soil water. The small but positive morning photosynthesis in Mulga, which is supported by storage of θ in shallow soil depths (in the top several meters), contributes to maintenance of a neutral carbon balance during moisture-limited periods.

  3. [67] The two largest sources of Re in this study were autotrophic respiration (Ra) by the seasonal understorey and heterotrophic respiration (Rh) by decomposition of fresh or “photodegraded” organic matter (i.e., priming) following understorey senescence. The pattern of Re followed closely that of understorey LAI and Ra, except in March 2011 when Re was larger than Ra and thereby indicative of large Rh due to the decomposition of senescent vegetation. In general, Re was limited by availability of decomposable biomass (i.e., substrate limitation), θ (i.e., moisture limitation), and temperature (i.e., thermal sensitivity), although these factors varied in synchrony with storm cycles so that no single factor explained variations in Re.

[68] The close association between GPP, Re, and precipitation pulses in central Australia suggests that large areas of Australia may be more of a carbon sink, and a smaller carbon source, than previously expected. Rainfall caused corresponding pulses of GPP and Re. These pulses were distinguished in wet and dry years by their size and the relative proportion of canopy and understorey LAI that contributed to the response. All primary producer types (trees, seasonal grasses and shrubs, and cryptobiotic crust) contributed to maintaining the carbon balance in drought conditions and across seasons.


[69] This work was supported by grants from the Australian Government's Terrestrial Ecosystems Research Network (TERN) (www.tern.gov.au) and the National Centre for Groundwater Research and Training (NCGRT, an Australian Government initiative supported by the Australian Research Council and the National Water Commission). This work was supported also by OzFlux and the Australian Supersite Network, both parts of TERN and the latter of which is a research infrastructure facility established under the National Collaborative Research Infrastructure Strategy and Education Infrastructure Fund, Super Science Initiative, through the Department of Industry, Innovation, Science, Research, and Tertiary Education. We would like to thank Peter Isaac for his assistance with data analysis. We would also like to thank A. O'Grady and an anonymous reviewer for their assistance in improving this manuscript.