Mulga, a major tropical dry open forest of Australia: recent insights to carbon and water fluxes

Mulga, comprised of a complex of closely related Acacia spp., grades from a low open forest to tall shrublands in tropical and sub-tropical arid and semi-arid regions of Australia and experiences warm-to-hot annual temperatures and a pronounced dry season. This short synthesis of current knowledge briefly outlines the causes of the extreme variability in rainfall characteristic of much of central Australia, and then discusses the patterns and drivers of variability in carbon and water fluxes of a central Australian low open Mulga forest. Variation in phenology and the impact of differences in the amount and timing of precipitation on vegetation function are then discussed. We use field observations, with particular emphasis on eddy covariance data, coupled with modelling and remote sensing products to interpret inter-seasonal and inter-annual patterns in the behaviour of this ecosystem. We show that Mulga can vary between periods of near carbon neutrality to periods of being a significant sink or source for carbon, depending on both the amount and timing of rainfall. Further, we demonstrate that Mulga contributed significantly to the 2011 global land sink anomaly, a result ascribed to the exceptional rainfall of 2010/2011. Finally, we compare and contrast the hydraulic traits of three tree species growing close to the Mulga and show how each species uses different combinations of trait strategies (for example, sapwood density, xylem vessel implosion resistance, phenological guild, access to groundwater and Huber value) to co-exist in this semi-arid environment. Understanding the inter-annual variability in functional behaviour of this important arid-zone biome and mechanisms underlying species co-existence will increase our ability to predict trajectories of carbon and water balances for future changing climates.


Introduction
Tropical dry forests (TDFs) are globally extensive and widely threatened (Pulla et al 2015), representing the first frontier of anthropogenic land-use change (Portillo-Quintero and Sanchez-Azofeifa 2010). Globally, research and conservation efforts for rainforests and tropical moist forests greatly exceed that for TDFs (Pulla et al 2015), despite the fact that the loss of TDFs arising from human activities exceeds that for rainforests and moist tropical forests (Miles et al 2006), probably because human population densities in TDFs tend to be larger than those in rainforests and moist tropical forests.
Tropical forests represent significant sites of C uptake and storage. In 2011 a global land sink anomaly occurred (Le Quéré et al 2014, Poulter et al 2014 and this was predominantly the result of increased productivity of southern hemisphere (especially Australian) semi-arid vegetation (Poulter et al 2014, Cleverly et al 2016a. Given the importance of Australian TDFs (including Mulga and riparian forests) to global atmospheric C dynamics and regional water and C budgets it is timely to provide a synthesis of recent developments in our understanding of the structure, function and behaviour of this biome and to discuss the specifics of the behaviour of Mulga during the land sink anomaly.
TDFs have two key climatic features: they experience warm or hot temperatures all year (mean annual biotemperature >18°C), and they experience a predictable dry season every year. The former attribute is the same as that for lowland rainforests but distinguishes them from tropical montane forests, which experience lower daytime temperatures because of the effect of altitude on temperature . Unlike many TDFs globally, Australian TDFs, as with most Australian woody ecosystems, are dominated by evergreen species.
In this short synthesis we summarise our current understanding of the ecophysiology and ecohydrology of a large (5500 km 2 ) endorheic semi-arid tropical low open forest in central Australia. In particular we address the following key questions: (1) What determines the climate of this region?
(2) How does climate influence phenology of this region?
(3) What are the principle drivers of daily, seasonal and inter-annual patterns of carbon (C) and water fluxes?
(4) How did the Mulga forests contribute to the 2010/ 2011 global land sink anomaly?
(5) How do hydraulic traits of Mulga compare with those of two co-occurring tree species and which traits may be associated with extreme drought resilience?

Mulga: a major dry open forest of tropical Australia
Mulga are comprised of a complex of closely related Acacia spp., especially A. aneura and A. aptaneuera (Maslin and Reid 2012). Acacia dominated landscapes grade from low open forests, through woodlands to low shrub lands (Specht and Specht 1999) and cover 20%-25% of the Australian continent (figure 1(a); Nicholas et al 2011). They are especially adapted to semi-arid and arid tropical and sub-tropical regions. As with all Acacias, Mulga forms symbioses with nitrogen-fixing cyanobacteria. Across the continent forest density (trees per hectare), height and growth form of Mulga trees are highly variable, ranging from relatively short (<3 m) to relatively tall (>10 m) trees, and ranging from a very open tree canopy to a relatively closed canopy. Mulga occurs on flat terrain and requires significant water storage capacities in the substrate (e.g., clay-rich soils or the presence of a hardpan within several meters of the surface; Pressland 1976), with low rates of drainage past the relatively shallow roots. Mulga usually forms a nonoverlapping mosaic with hummock (spinifex) grasslands (figure 1(b); Nicholas et al 2011). The boundaries between Mulga and adjacent spinifex grasslands have not moved in the past 1000 years (Bowman et al 2007), suggesting that they have persisted in roughly the same locations over a possibly large diversity of climatic conditions. In the Ti Tree basin in central Australia (approximately 22°14′ S, 133°17′ E), the site of the studies discussed here, the Mulga canopy is ca. 6.5 m tall with a basal area of 8 m 2 ha −1 and a stem density of approximately 3300 stems ha −1 (figure 1). Dominant understory forbs and herbs include Psydrax latifolia, Eremophila gilesii, Eremophila latrobei ssp glabra, Sida and Abutilon spp., and Solanum ellipticum. The dominant grasses form a nearly complete cover when conditions permit and include perennials Thyridolepis mitchelliana and Eragrostis eriopoda, and annual Eriachne pulchella ssp pulchella. The Ti Tree basin receives the majority of its annual rainfall (long term (1987-2016) median of 299 mm) in the Austral summer wet season (December to February; figure 2), although rainfall can occur in any month (figure 2). The low level of rainfall received in the months June to September inclusive (<20 mm month −1 ) is essentially ineffective in wetting the soil profile to any significant depth: almost all of this rain is intercepted by litter (and shallow roots) and evaporated rapidly back to the atmosphere. In the past 29 years annual rainfall has varied between 97.4 mm (1994) and 750.6 mm (2010).

What determines the climate of central Australia?
The semi-arid climate of central Australia is old. The development of a distinct dry season occurred as much as 35-50 Mya, preceding the separation of Australia and South America from Antarctica which opened the Southern Ocean and changed the world's climate to a cooler and drier regime (Crisp and Cook 2013). The modern climate regime was established by the end of the Pliocene (2.6 Mya), since which time the climate has dried by stages (Martin 2006) and was likely to have become semi-arid ca. 0.5 Mya. By using the term dried by stages, Martin (2006) meant that, whilst glacial periods have been drier than during inter-glacial periods, successive glacial and inter-glacial periods have been drier than those that occurred before. Thus, the longest and driest droughts in Australia spanned the most recent glacial period.
The second important feature of the Australian climate is its inter-annual variability in rainfall. Amongst the arid and semi-arid regions of the world, three of the four with the largest variability in rainfall occur along the rim of the Indian Ocean (the Thar Desert in India, Somalia and Australia north of 27°S; Van Etten 2009). In the Ti Tree basin over just the last 100 years, for example, annual rainfall has been as low as 25 mm yr −1 in 1928 and as high as 955 mm yr −1 in 1974 (Cleverly et al 2016a(Cleverly et al , 2016b. Together, the ancient and more recent climate histories underscore two critical features that impact the development of  Australian TDFs: an exceptional amount of interannual variability in rainfall and a long-term drying trend. Three climate drivers have the potential to have a direct influence on continental weather patterns in Australia: El Niño-Southern Oscillation (ENSO) in the equatorial Pacific Ocean, the Indian Ocean dipole (IOD), and the southern annular mode (in the Southern Ocean, SAM). These climate modes have been individually related to the occurrence of rainfall, but not to amount (Pui et al 2012). However, ENSO, IOD and SAM periodically synchronise their effects on Australia's weather patterns, resulting in extremes of rainfall and drought (Cleverly et al 2016a). One such period began in 1999, signalling an increase in extreme weather events like the Millennium Drought at the beginning of the 21st century and widespread flooding in Queensland in 2011.
Our recent analysis (Cleverly et al 2016a) has demonstrated that when climate modes synchronise, the general amount of atmospheric moisture formed by ENSO (i.e., dry in Australia during El Niño, wet during La Niña) combines with the strength of the monsoon depression driven by IOD and placement of weather systems by SAM to enhance or prevent rainfall across the continent (Cleverly et al 2016a). Facilitated by Australia's flat terrain, a strong monsoon depression can interact with the continental low (which is indirectly influenced by SAM) to generate widespread storms with the intensity of summer rainfall but the large spatial and temporal extent of winter storms (Kong andZhao 2010, Cleverly et al 2013). Alternatively, dry conditions induced by each climate mode individually can combine to further reduce rainfall beyond the effects of any single climate mode, leading to more intense Australian heatwaves (Perkins et al 2015). However, these interactions were not apparent in the early part of the measurement record (1982)(1983)(1984)(1985)(1986)(1987)(1988)(1989)(1990)(1991)(1992)(1993)(1994)(1995)(1996)(1997)(1998), suggesting that this interaction among climate modes is a periodic feature of regional climate. Thus, these three climate drivers interact in the Indian Ocean region to generate the large degree of rainfall variability observed across Australia.

How does climate determine phenological patterns of Mulga low open forest?
Mulga has a highly variable phenology that is largely rainfall-pulse driven. Key features of this phenological variability include large fluctuations in the dates of start-of-growing season, growing season peak and end-of-growing season, along with season length and the presence of a growing season in a given year The phenology of Australia's semi-arid ecosystems is very sensitive to extremes of climate such that variance in total annual amount of rainfall alone explains 80% of variation in the length of the growing season (Ma et al 2013(Ma et al , 2015. However, uneven distribution of rainfall across the year can inhibit phenological responses in Mulga of central Australia due to the juxtaposition of wet and dry months (Nano and Clarke 2010). Thus, greenness in the vegetation of central Australia as determined from Moderate Resolution Imaging Spectroradiometer (MODIS) EVI is more highly correlated to the synchronised climate signal amongst ENSO, IOD and SAM than to rainfall (Cleverly et al 2016a), implying that the suite of meteorological conditions associated with the state of the climate system (e.g., cool and wet or hot and dry) exerts a larger influence on the phenology of Mulga than any individual environmental driver.
Seasonal and inter-annual phenological variation across the landscape is large. Seasonal variations in leaf area index (LAI) and landscape-scale photosynthetic capacity are primarily driven by seasonal variations in understory grasses and forbs (Eamus et al 2013a, Ma et al 2013. Seasonal changes in LAI of the Acacia overstory, in contrast, are relatively minor (Eamus et al 2013a). In the understorey of Acacia Mulga, annual grasses (such as Aristida spp.) green up in response to summer rain (∼February) and set seed in autumn (Mott and McComb 1975). In contrast annual forbs, such as species of Helichrysum spp. and Helipterum spp. germinate in response to rain in early winter (May-June) (Mott and McComb 1975), and perennial grasses green up into the summer monsoon periods (Flora of Australia 2002). Collectively, these different phenological patterns of understorey in Acacia low open forests enable a long greening period (figure 3) at the community scale during the wet periods and are observable by satellite sensors (Ma et al 2013).
Despite the principal control of rainfall on phenology, the response of Mulga to rainfall can be inhibited during the summer wet season by canopy temperatures that exceed Acacia's 38°C thermal limit for growth (Nix andAustin 1973, Cleverly et al 2016b). Thus, growing-season phenology and canopy structure of this low open forest are strongly related to its physiological phenology (e.g., the seasonal and annual cycle of ecosystem productivity and light use efficiency; Migliavacca et al 2015), which provides a basis for the strong relationships amongst gross primary production and EVI during the green-up phase and hysteresis during autumnal brown-down (figure 4; Ma et al 2013Ma et al , 2014. Hysteresis means that the relationships between GPP and EVI during greenup differs from that during browndown. In Mulga the lower GPP for a given EVI during browndown presumably reflects a loss of photosynthetic potential in the canopy during browndown. Although direct empirical relationships between photosynthetic activity, ecosystem productivity and MODIS vegetation indices have been observed in various temperate and tropical ecosystems, MODIS vegetation products have been less successful in capturing seasonality of productivity in many Australian evergreen forests and sclerophyll woodlands (Restrepo-Coupe et al 2016). This prompted a re-evaluation of the connection between satellite observations and various flux productivity measures with the expectation that satellite greenness products constituted a combined measurement of ecosystem structure (e.g. LAI) and function (e.g. leaf level photosynthetic assimilation capacity), rather than productivity per se. Restrepo-Coupe et al (2016) showed the lowest correlations between MODIS vegetation products and productivity at locations where meteorological variables (drivers) and vegetation phenology were asynchronous, as in Mediterranean ecosystems (schelophyll forests). In contrast, the largest correlations were observed at locations where synchrony across meteorological variables and vegetation phenology was apparent, as observed in the Mulga TDF. They concluded that remote sensing products would not follow productivity when: either phenology is asynchronous with key meteorological drivers, and productivity is  Ma et al (2013). The date of the SGS is defined as the date halfway between the date of the minimum EVI and date of the fastest rate of greenup (maximum slope in the EVI response in the greenup). The date of the PGS is defined as the date when EVI is maximum in each year. The EGS is defined as the date half-way between the fastest decline in EVI and minimum EVI. therefore driven by one or more meteorologic drivers over different times of the year; or when phenology is relatively aseasonal, and productivity is solely driven by meteorology (especially solar radiation, air temperature, soil water availability, VPD, or different combinations).

Drivers of seasonal and annual patterns of C and water flux
The environmental drivers associated with patterns of C and water flux are closely related to those driving phenology in Australia (Ma et al 2014). In this section, we will address how patterns of rainfall, soil moisture content and meteorological drivers affect dynamics in C and water fluxes during a drier-than-average year (hydrologic year 2012-2013) and two years with average amounts of rainfall (2013-2014 and 2014-2015). Evaluation of drivers in this section will begin with a comparison of rainfall, evapotranspiration (ET) and net ecosystem productivity (NEP) between a drier-than-average year (2012-2013) and the subsequent year with average rainfall (2013-2014; Cleverly et al 2016c).

Comparing a drier-and wetter-than average year
Mulga shows a disproportionate response to extremely high inputs of rainfall (e.g. during the global land sink anomaly; Cleverly et al 2013, Eamus et al 2013a). This is driven by a tropical-extra-tropical connection in the Great Australian Bight which creates a regular pattern of storm-interstorm intervals (Berry et al 2011) and thereby provides periods of high light, soil moisture content and humidity during wetter-than-average years that are highly favourable for photosynthesis (Cleverely et al 2013). By contrast, inter-storm periods during dry-to-average years are hot and VPD is large, exceeding Mulga's thermal tolerance of 38°C (Cleverly et al 2016b). In this section, we explore these issues more fully.
Rainfall in hydrologic years 2012-2013 (193 mm) and 2013-2014 (295 mm) was approximately 65% and 99% of the long-term average (median 299 mm), respectively. As is typical of this site, rainfall predominantly occurred in the summer wet season. Rates of ET are extremely responsive to pulses of rain (Cleverly et al 2016c). After rains have concluded, however, ET declines rapidly (and exponentially) to zero (Eamus et al 2013a, Cleverly et al 2016c. Peak rates of daily ET in the wet season range between three and four millimetres, whilst cumulative ET in the two years was 150 and 250 mm, or 50% and 84% of average annual rainfall (figure 5).
Over these two years, the Mulga low open forest oscillated across the seasons between being a C sink and C source (figure 6). During the winter and early spring, the Mulga was a small sink (positive values of NEP ranged from 0.1 to 0.3 gC m −2 d −1 ). The largest sink activity during this two-year period (2012-2014) occurred in late summer/early autumn of 2014 when temperatures, vapour pressure deficit and soil moisture content were optimal for NEP. Source activity in the Mulga occurred in pulses during the summer/ autumn of 2013 and the summer of 2014 (figure 6). On a hydrological annual time-scale, the Mulga was a small source in the drier-than-average hydrologic year (−26 gC m −2 d −1 in 2012-2013) and a small sink in the average year (12 gC m −2 d −1 in 2013-2014). It was not until the wet summer of the second hydrologic year that Mulga became a moderate-to-strong sink. Importantly, rainfall in the first hydrologic year (193 mm) was significantly less than in the second hydrologic year (295 mm), thereby revealing the importance of the amount of rainfall received as a determinant of productivity of this semi-arid tropical ecosystem.
Different combinations of rainfall, temperature, solar radiation and vapour pressure deficit are the principle determinants of NEP and GPP . However, this does not mean that these other drivers are not important for explaining inter-annual variation in NEP; instead, the environmental drivers are themselves related to rainfall in a series of complexities which we will now discuss.
3.2. The importance of wet season rainfall distribution in driving patterns of NEP Cleverly et al (2016b) compared NEP during two years with near-average annual rainfall (2013-2014 and 2014-2015) but with different temporal patterns to evaluate the impact of differences in the distribution of precipitation within the wet season. Each wet season received about 280 mm, but it was distributed more evenly across the wet season of 2013-2014 than that of 2014-2015. In 2014-2015 most of the rainfall had accumulated by mid-January, and the site received very little rainfall for months afterward (figure 7). These differences had significant effects on summer and autumn patterns of soil moisture content and hence the response of NEP across these two years. In the summer wet season 2013-2014, there was a  This resulted in the Mulga canopy experiencing conditions that exceeded the thermal tolerance of the phyllodes. It was not until cooler temperatures, a smaller VPD and the positive impact of transpirational cooling on canopy temperature (once soil moisture levels were sufficient) were evident, that NEP became positive in late February and March 2014. In contrast, rain in December 2014-January 2015 produced a positive response in NEP almost immediately, and a large C sink was maintained throughout January and February of 2015. Cooler air temperatures and sufficient soil moisture above the hard pan to allow transpirational canopy cooling resulted in positive NEP in mid-January to mid-March 2015, much earlier than during the previous year.
We will now demonstrate how differences in soil moisture content were formed due to differences in the weekly-to-monthly pattern of rainfall in the Mulga TDF.
Soil moisture content measured at a point approximately 1 m below the ground surface is highly dependent upon the (i) amount, timing and intensity of rainfall and (ii) the presence of a subsurface hardpan at this site (Cleverly et al 2013(Cleverly et al , 2016b(Cleverly et al , 2016c. The siliceous hardpan is buried below the surface and prevents drainage of water to the regional water table, which is 49 m deep at this site (Cleverly et al 2016c). Soil moisture content in the hardpan is 0.05 m 3 m −3 and has shown only minor fluctuations since measurements began in 2010. Conversely, soil moisture content in the unconsolidated loamy sand above the hardpan has fluctuated between ca. 0.08 m 3 m −3 and 0.35 m 3 m −3 (saturated) since 2010 (Cleverly et al 2013(Cleverly et al , 2016b. The fluctuations in soil moisture content in the unconsolidated soil above the hardpan relate strongly to the amount and pattern of rainfall, hence providing a link for responses of NEP to rainfall (Cleverly et al 2016b).
With the hardpan acting as a barrier to drainage, rainfall that delivers water in excess of that which is required by the vegetation can be stored and carriedover to other seasons or years. The importance of inter-seasonal and inter-annual carry-over of rainfall to sustaining NEP is well documented in arid and semi-arid ecosystems globally (Flanagan and Adkinson 2011). Even during a drier-than-average year (2011-2012, 184 mm of rainfall), drawdown of the storage reservoir in the unconsolidated soil above the hardpan (from 0.10 m 3 m −3 to 0.08 m 3 m −3 ) was associated with maintenance of gross primary production at low but non-negligible levels (Cleverly et al 2013). The reservoir was then rapidly re-filled and maintained soil moisture content of ca. 0.10 m 3 m −3 across the following autumn-winter (Cleverly et al 2013). We have found that Mulga trees have a small but significant proliferation of roots near the top of the hardpan where this soil moisture reservoir accumulates (Cleverly et al 2016b), and these roots are likely to remain active during all seasons because of the consistent elevation of soil moisture content in the reservoir relative to the hardpan (Cleverly et al 2013(Cleverly et al , 2016b. Whereas carry-over of soil moisture appeared to be an important feature of the site water budget in a drier-than-average year, it wasn't until the first year with average rainfall in the study (2014)(2015) when soil moisture content increased substantially (>0.2 m 3 m −3 ) such that delayed increases in NEP were associated with carry-over of that soil moisture ( figure 7).
In comparing the summer wet season during two years with average rainfall (2013-2014 versus 2014-2015), it became apparent that the pattern of rainfall distribution affected the degree to which soil moisture storage filled to capacity in the unconsolidated soil above the hardpan (Cleverly et al 2016b). During the summer-autumn of 2013-2014, soil moisture content increased to 0.26 m 3 m −3 following several weeks of rainfall (150 mm of rainfall by mid-January, figure 8), including one storm which delivered 105 mm of rainfall over six days (Cleverly et al 2016b). By contrast, in the summer-autumn of 2014-2015, rainfall was 250 mm by mid-January with 145 mm delivered over 12 days; soil moisture content increased to saturation (which is 0.35 m 3 m −3 in this soil) immediately thereafter; and NEP responded immediately and was positive by mid-January 2015. Saturation of the upper soil profile (top 1 m) in January 2015 resulted in the formation of a perched, ephemeral water table located in the unconsolidated soil above the hardpan (figure 8). These results are consistent with the threshold-delay model of Ogle and Reynolds (2004) whereby a pulse of rain that is insufficient to fill the 'bucket' (i.e., the soil moisture storage reservoir containing roots) causes a delay in the response of NEP, whereas sufficient rain to fill the bucket induces a rapid response in NEP (i.e., inhibition of the delay). When stomatal conductance is large, transpiration and transpirational cooling proceed at rates that exceed those observed when conductance is low. When the bucket was not filled during summer of 2013-2014 and NEP was delayed, transpiration would have been minimised; therefore, incoming solar radiation would have been partitioned into sensible heat flux instead of latent heat flux, thereby minimising transpirational cooling (Cleverly et al 2016b). In contrast, in the summer of 2014-2015 when NEP was positive and large rates of transpiration were occurring, less partitioning to sensible heat flux would be expected. In agreement with this is the observation that sensible heat flux in January-February 2014 was 307.1 W m −2 but in the same period in 2015, sensible heat flux was smaller (270.4 W m −2 ). Using imagery obtained from an unmanned drone with a thermal camera in March 2014 (i.e., during the autumn following the delay in NEP), it was revealed that a large fraction of the Mulga canopy still exceeded the 38°C thermal optimum for this species (Cleverly et al 2016b), further suggesting that a soil moisture reservoir that is not full is insufficient to provide the resources for transpirational cooling until the heat of the summer has passed.

Interactions of soil moisture content with temperature and VPD as drivers of variation in NEP
Having established the importance of the timing of rainfall as a determinant of the response of NEP, we extend this analysis to examine in detail the interactions among soil moisture content, VPD and solar radiation as determinants of NEP. Thus we evaluate meteorological drivers and NEP across a wide range of soil moisture contents during a very wet year (2010-2011) and the subsequent drier-than-average year (2011-2012). The effects of air and soil temperature, VPD and solar radiation on NEP were closely related to variations in soil moisture content as a consequence of correlations between these drivers and rainfall (Cleverly et al 2013). For example, the temperature at which peak NEP occurred and the reduction of NEP at high temperature were dependent upon soil moisture content ( figure 9). Similarly, the VPD at which NEP was maximised depended upon soil moisture, as was the response of NEP to light intensity. The thermal responses of NEP to soil moisture content in Mulga were not dissimilar to those of Eucalyptus leaves as a function of [CO 2 ], although NEP in Mulga appears to be more limited at low temperature than for leaf photosynthesis in Eucalyptus (Eamus et al 1995). Furthermore, the responses of NEP to VPD in Mulga imply a close link to ET via canopy conductance (as assimilation is linked to transpiration via stomatal conductance in leaves; Duursma et al 2014). Thus in Mulga, ecosystem inherent water-use efficiency (IWUE, the ratio of [NEP×VPD] to ET) is highly dependent upon soil moisture content during wet and dry seasons (Eamus et al 2013a), illustrating the coordinated control of NEP and ET by soil moisture, VPD and temperature.

The global land sink anomaly and Mulga
Having established general relationships among rainfall (timing and amount), temperature, VPD and NEP we now discuss the specific case of the 2010/2011 global land sink anomaly and Mulga.    2011 global sink anomaly. We now address the question: was there an observed response in C uptake for the Mulga during the 2010/2011 global land sink anomaly?
During the hydrologic year 2010-2011 rainfall in the Ti Tree basin was 565 mm, or 75% larger than the long-term average. Furthermore, a large fraction of this rainfall was received in the dry season of 2010, as is reflected in the repeated recharge of soil moisture in September-November 2010 (figure 11). Indeed the wet season for this hydrologic year was eight months long (figure 12) rather than the usual three or four months (Ma et al 2013).
The response of ET and NEE to this unusual pattern and amount of rainfall is shown in figure 13. Large rates of ET were maintained for the unseasonably wet dry season of 2010 through to the early autumn of 2011. By the winter of 2011, however, ET had declined to zero except for a brief increase in response to rain in June 2011. Across the 372-day study, total ET was 430.6 mm (76% of rainfall). The exponential decline in ET following rainfall exhibited a half-time of decay of 5.2 days, whilst a threshold amount of rainfall had to be exceeded (3.0 mm) for transpiration to increase. This half-time is similar to that observed in semi-arid grasslands and shrublands of central New Mexico (half-time approximately 2 days; Kurc and Small 2004) but shorter than the range in half-times observed across 15 sites of North America, Europe, Afric and New Zealand (15-35 days) by Teuling et al (2006). Interestingly, despite large differences across sites, there were no significant differences within sites across multiple years, and longer half-times were observed at sites experiencing seasonal droughts or at sites with woody vegetation. Our much shorter half-times  may reflect the sandy soils, the impact of the very shallow roots of Mulga and Mulga's rapid phenological response to precipitation when conditions are favourable (Cleverly et al 2016b).
The large rates of ET in the winter of 2010 and the late summer and autumn of 2011 were mirrored by large rates of C uptake (figure 13). The half-time of the decay in C uptake following rain was 7.2 days (Eamus et al 2013a). Unlike ET, which increased immediately following rain, C uptake showed a 1-3 day delay.
The large rates of C uptake sustained from September 2010 through to June 2011 are in marked contrast to the short duration of C uptake in figure 7 and in contrast to the annual totals for years 2011-2015 (table 1), and reflect the impact of large and sustained increase in soil moisture content during the 2010/2011 La Niña event. If we assume that all Australia's Mulga behave the same as the Mulga in the Ti Tree and all received similar increases in rainfall, it is possible to estimate the contribution of Australia's Mulga to the NEE anomaly quantified in Poulter et al (2014). Mulga covers approximately 33% of the 70% of Australia that is semi-arid. The anomalous increase in C uptake by the Ti Tree Mulga was 247 gC m −2 yr −1 (the difference between uptake in 2010-2011 and uptake in a 2013-2014 when rainfall was average). This represents 0.435 Pg C yr −1 for the entire Mulga estate, which accounts for about 50%-70% of the anomaly identified by Poulter et al (2014) as originating in Australia (0.6-0.9 Pg C yr −1 ). However, tree density of Mulga in the Ti Tree is larger than that of most Mulga in Australia and the contribution of all Mulga is therefore likely to range from 30% to 40% of the total anomaly.

Ecosystem C and water budgets: an ecosystem services perspective
Ecosystem services are the goods and services provided by ecosystems that are of benefit to humans (Eamus et al 2005). Examples of ecosystem services include erosion control, regulation of local and regional climate, C sequestration and the regulation of water flows (surface, sub-surface and groundwater recharge). For terrestrial ecosystems, the provision of ecosystem services is dependent on neutral or positive C and water budgets. A neutral or positive C budget means no net loss or an accumulation of C within the catchment. A neutral or positive water budget means that there is not a net export of water from the catchment; for example, by over-extraction of groundwater (by human extraction). This raises the question: over what timeframe should C and water budgets be determined? It is clear that seasonally, Mulga can oscillate between being a net sink or a net source of C. Similarly, at an annual time-scale, Mulga can be a net sink or source of C. Mulga in the Ti Tree basin has undoubtedly persisted for more than 50 yr, and probably for considerably longer. Table 1 shows how widely NEP and ET can vary across just four years of study. Interestingly, a regression of NEP versus rainfall using data in table 1 shows that when annual rainfall exceeded a threshold of 233 mm (R 2 =0.95), Mulga was a sink but when rainfall is <233 mm Mulga was a source. This is smaller than the thresholds of 275-350 mm identified for several semi-arid ecosystems in south-western USA (Scott et al 2015), but, is consistent with the approximately 100 mm difference between the threshold rainfall and the long-term average rainfall observed by Scott et al (2015). We estimate that this offset of 100 mm ensures that a positive C balance will be maintained in approximately 70% of years (because only about 30% of years have rainfall less than this threshold).
In a recent analysis of the Ti Tree Mulga, Chen et al (2014) concluded that annual vegetation water use varied between 64 and 601 mm for the period 1981-2012. Most importantly (from an ecosystem services perspective), groundwater recharge (noting that the Ti Tree basin may become a significant source of potable groundwater for industrial, irrigation and domestic uses in central Australia) varied between 0 and 48 mm yr −1 but significant rates of recharge were very rare events, highlighting the need to extract groundwater at sustainable rates for the long-term preservation of this important resource.
It is clear that short-term (<5 years) studies of C balances for TDFs are unlikely to be adequate to provide sufficient information to determine whether particular locations are in a negative, neutral or positive C balance. This is because TDFs tend to have much larger variability in the timing, intensity and amount of rainfall than tropical rainforests and other mesic biomes. This large degree of variability is reflected in the large fluctuations between sink and source strength. Furthermore, large inter-annual variability in rainfall and hence ET impacts on the provision of water resources (surface and groundwater), and this should inform management decisions on the timing and volume of water resource extraction. The size and magnitude of changes in C storage pools (aboveground and below ground, biotic and abiotic) and the role of disturbance (for example fire and prolonged drought) in C and water balances remain poorly documented for much of the arid and semi-arid regions of Australia.

Mulga traits versus traits of co-occurring tree species
The Ti Tree basin contains three major ecosystems. The first is Mulga, dominated by Acacia spp., the second is riparian forest that follows the ephemeral rivers. The riparian forest is dominated by large Eucalyptus camaldulensis trees. Whilst the majority of tree species across the basin are evergreen (including Mulga and Eucalyptus camaldulensis), patches of the deciduous Erythrina vespertilio occur in paleochannels (a remnant of an inactive river or stream that has been subsequently infilled by younger sediment; Shanafield et al 2015) and along ephemeral rivers. Ecohydrological niche separation theory (Silvertown et al 2015) predicts that co-occurring species should partition water resources through variation in traits related to the uptake and transport of water. Table 2 summarises some hydraulic traits of these three contrasting species. Comparisons of traits amongst these species contribute to our understanding of niche separation and also to our understanding of how some species in semi-arid regions, such as Mulga, are so resilient. The third ecosystem in the Ti Tree is a sparse Corymbia savanna which we do not discuss further.
Mulga is shallow rooted (<2 m), evergreen and located above a shallow hard pan (Cleverly et al 2016b). Consequently it experiences very low phyllode water potentials, both pre-dawn and midday (the limit of our pressure chamber was 10 MPa, hence we can only state that midday water potentials were lower than −10 MPa). Several predictions can be made about the traits that should be associated with such low water potentials. A key prediction is that Mulga should exhibit a high resistance to xylem embolism to avoid catastrophic loss of the ability to transport water to the canopy (Sperry and Tyree 1988). A high resistance to embolism requires thick walled xylem conduits, which in turn results in a large wood density, coupled with a small sapwood conductive area and a concomitant small sapwood hydraulic conductivity. Similarly, we would predict a large resistance to vessel implosion (Hacke et al 2001). Finally, we would predict a large water-use-efficiency, as indicated by an enrichment (closer to zero) of δ 13 C of leaves. It is clear that Mulga does possess this suite of traits (table 2), and this contributes to the ability of this species to persist through multi-year periods of low annual rainfall and the dry season that occurs each year. These traits also confer a low rate of tree wateruse, as has been previously observed  (table 2). For every m 2 of sapwood, only about 880 m 2 of phyllode area is supported. This is much smaller than the 1800-2000 m 2 of leaf area supported by 1 m 2 of sapwood in Eucalyptus camaldulensis and Erythrina vespertilio. Changes in H v in response to changes in soil water availability are a common adaptive strategy to differences in water availability (Togashi et al 2015, Zolfaghar et al 2015).
A high resistance to xylem embolism and the very low leaf potentials experienced by Mulga are also associated with the lowest water potential at zero turgor of the three species discussed herein. Low water potential at zero turgor in Mulga is indicative of the accumulation of chemically inert compounds that confer protection from water stress, and these osmoprotective compounds are known to occur in Mulga as nitrogenrich compounds such as proline and glycine betaine (Erskine et al 1996). This is an important trait for shallow rooted evergreen species in semi-arid regions growing without access to groundwater. The maintenance of positive turgor is required to drive cell growth and maintain open stomata, thereby allowing photosynthesis to occur during periods of low soil moisture availability.
By contrast, Eucalyptus camaldulensis and Erythrina vespertilio exhibit markedly different strategies from Mulga for persistence in this semi-arid environment. Eucalyptus camaldulensis has access to groundwater and thus does not experience the very low predawn and midday water potentials that occur in Mulga. Eucalyptus camaldulensis is highly groundwater dependent and occurs predominantly as riparian forest above shallow unconfined groundwater (<6 m depth). Because it is less reliant on rainfall, with yearlong access to groundwater, this species does not experience low leaf water potentials, and thus does not require thick xylem cell walls to resists xylem embolism. Consequently the sapwood density and vessel implosion resistance (Hacke et al 2001) are smaller in Eucalyptus camaldulensis than in Mulga (table 2). This easy access to groundwater and large conductive sapwood area support much larger rates of transpiration than observed in Mulga, despite high canopy water potential (relative to Mulga; O'Grady et al 2009). To maintain a large rate of transpiration and high canopy water potential requires a large sapwood hydraulic conductivity, as is observed (table 2). Furthermore, Eucalyptus camaldulensis has a larger (closer to zero) water potential at zero turgor than Mulga because it experiences much larger canopy water potentials than Mulga. Finally, the δ 13 C of its leaves are much larger than those of Mulga, reflecting its relatively low WUE: the value recorded is close to that observed for leaves of Eucalypts growing in regions receiving >1500 mm rain per year. However, it does possess the capacity for a reduction in leaf area as a mechanism for adjusting water use during prolonged dry periods (O'Grady et al 2009), and mortality can occur in small trees during extended dry periods (Horner et al 2009).
Erythrina vespertilio has a similar strategy to Eucalyptus camaldulensis for surviving semi-arid conditions, but with one important difference. Like Eucalyptus camaldulensis, Erythrina vespertilio does not experience low canopy water potentials and therefore has a low wood density, narrow fibre walls and a very low vessel implosion resistance. It also possesses large cross-sectional areas of parenchyma, indicative of a large capacity for water storage. Similarly, it has a large theoretical hydraulic conductivity (table 2). However, it has ephemeral access to groundwater, occurring in paleochannels which are sites of small amounts of runon from precipitation and which contain perched shallow water tables for short periods of time. Consequently the δ 13 C of leaves of Erythrina vespertilio (and hence WUE) is intermediate between Mulga and Eucalyptus camaldulensis, and its H v is also intermediate between these two species. However, the key trait exhibited by this species is its deciduous habit. By being an obligate deciduous species its canopy entirely avoids periods of low soil and atmospheric water content. Consequently it never experiences low canopy water potential and therefore its water potential at zero turgor is very high (close to zero)-a trait that is associated with deciduous and mesic species.

Conclusions and future directions
TDFs experience warm/hot temperatures and a predictable dry season. However, as we have shown for our Mulga site in central Australia and in agreement with most other semi-arid and arid ecosystems, rainfall is extremely variable in terms of the amount and timing of precipitation. This has major impacts on phenological behaviour and concomitant productivity of such ecosystems. Mulga in this climate zone can oscillate repeatedly between being approximately C neutral to being large C sinks and large C sources. Contrary to widely held expectations the behaviour of semi-arid ecosystems can impact significantly on global C and water cycles. An important feature of arid and semi-arid regions is the importance of groundwater and associated terrestrial groundwater dependent ecosystems both to human well-being and the health and functioning of these landscapes. This synthesis has reiterated the importance of field ecophysiology, remote sensing and modelling to the generation of an in-depth understanding of the behaviour and functioning of these regions . Future work is required to examine the above-and below ground C storage pools and how these change at annual and decadal time-scales and to determine the resilience, sensitivities and and tipping-points of Mulga in the face of a changing climate. Since 1940, mean annual temperature in Alice Springs (200 km south) has increased by approximately 1.5°C, with associated increases in VPD during all except the wettest years. Since increased VPD and temperature increase the likelihood of forest mortality, (Eamus et al 2013b), the probability of mortality of Mulga is likely to also increase, thereby exerting a significant impact on regional C and water budgets and global productivity (Poulter et al 2014), especially during droughts in the Southern Hemisphere. Although mortality and drought are not unusual for Mulga, which was found to experience up to 40% mortality during the 1964-1966 drought (Cunningham and Walker 1973), thus far, Mulga has shown resilience to drought and high temperatures, thereby maintaining a neutral carbon balance through drier-than-average years. Determining the climate thresholds that induce mortality (and recruitment; Mulga is a masting species, producing large seed crops only after exceptionally high rainfall years) and determining how these events might change in the future remains a significant research challenge.
Recent studies have found semi-arid ecosystems (including Mulga) to be particularly sensitive to hydroclimatic variability and are thus vulnerable to future climatic extremes (Ma et al 2015). Such sensitivity suggests that maintaining ecosystem services and functioning of Mulga may be threatened in the future. Further studies are needed to examine the impact of climatic extremes on Mulga ecosystems to better understand climate thresholds of this important Australian TDF. Finally, further investigations of the mechanisms underpinning ecohydrological niche separation within these biomes is required to inform land surface models and global dynamic vegetation models, which are increasingly used in ecological and ecohydrological studies of TDFs.