TERN, Australia’s land observatory: addressing the global challenge of forecasting ecosystem responses to climate variability and change

The global challenge of understanding and forecasting ecosystem responses to climate extremes and climate change is addressed in this review of research enabled through environmental research infrastructure (RI) provided by Australia’s Terrestrial Ecosystem Research Network (TERN). Two primary climatic drivers of ecosystem structure and function in Australia are fire and aridity, to which Australian flora and fauna have shown marked adaptability. Australian vegetation shows resilience to climate extremes of flooding rains, droughts and heatwaves such that variability in primary productivity of Australian vegetation has a tangible effect on the global carbon cycle. Nonetheless, Australian flora and ecosystems could be vulnerable to projected climate change (e.g. to increasing vapour pressure deficit). Refugia are also vulnerable to climate change, with conditions in these areas already near the tipping point for a change in community composition. Ensuring genetic diversity during directional change in climate (e.g. increasing aridity) requires proactive approaches to conservation and restoration projects. To address these challenges, TERN provides environmental RI at three scales of observation: (i) environmental monitoring using remote sensing techniques at a landscape and continental scale; (ii) a spatially extensive network of ecosystem monitoring plots; and (iii) intensely measured sites collecting detailed data on ecosystem processes. Through partnerships with international environmental RIs, TERN enables research that addresses global challenges, on the first steps toward the forecasting of ecosystem–climate interactions.


Introduction
As global climate change becomes more difficult to ignore, there is an urgent need to understand how terrestrial ecosystems can be expected to respond to the changes they experience. Ecosystem responses to climate change and extremes of variability include However, our understanding of ecosystem responses to climate change and variability has lagged far behind our ability to predict those responses using models. Environmental research infrastructure (RI) is required at national, regional, continental and global scales to address important environmental challenges such as the impacts of climate change, coral bleaching, biodiversity threats, geohazards and extreme events. Establishing a coherent RI across a diverse range of scientific disciplines and contributing networks is a vital challenge to solve, thus creating a goal for developing cooperation amongst environmental RI organisations, government and industry for the shared purpose of addressing global challenges.
The Terrestrial Ecosystem Research Network (TERN) is Australia's terrestrial ecosystem observatory, providing environmental RI at three scales of observation (table 1 and figure 1): (i) ecosystem surveillance monitoring plots from which spatial Table 1. List of physical and data research infrastructure (RI) operated by TERN.  . Development of fire has been associated with increasing aridity, for which the earliest evidence of arid-adapted vegetation appeared more than 30 Mya (Martin 2006). Ecosystem responses to fire and aridity will thus be reviewed first, followed by a review of ecosystem responses to Australia's highly variable modern climate regime, which was more recently established (during the early Pleistocene, 2 Mya; Martin 2006). Ecosystem responses to climate extremes are further explored for each extreme, presented in sections focused upon (i) an extremely wet period which occurred since TERN's establishment in 2009 and (ii) the dry extreme, which is associated with drought and heatwave. We will then finish with a section on ecosystem responses to climate change, the most recent of forces to affect ecosystems in Australia and globally. Examples from the literature were obtained upon review of the TERN publications catalogue (https://tern.org.au/Brochures-Publications-pg27411.html#Publications). Refer to the supplementary information, available online at stacks.iop.org/ERL/14/095004/mmedia for references from the TERN catalogue which were cited in this letter.

TERN and global cooperation
TERN was established in 2009 by the Australian government through the National Collaborative Research Infrastructure Strategy (NCRIS) to meet several objectives: to foster scientific interactions in the environmental sciences, to establish a national terrestrial site and observing network, to facilitate access to high-quality environmental data, and to provide a bridge between environmental science and policy (Thurgate et al 2017). TERN was originally developed as a network of networks, some of which were established wholly within TERN to fill gaps amongst existing networks (Thurgate et al 2017). This combined approach of joining pre-existing and new networks across the environmental space was particularly effective at avoiding duplication (Thurgate et al 2017) and thus reducing establishment costs. After expending the capital costs of establishing TERN's continental RI, the first challenge involved reducing the scope of the RI to fit within the given operations budget whilst retaining a consistent, continental scope. Integration of TERN RI began by extensive consultation which resulted in the grouping of observational infrastructure by scale of measurement, ultimately leading to TERN's current three scales of observation.
Integration across TERN's three scales of observation began in the research community With endorsement from the TERN Advisory Board, TERN continues to foster increasingly close integration across three scales of measurement through regular executive group meetings which include programme leaders of RI at each scale of observation and members of TERN's scientific advisory committee.
As the global research enterprise becomes increasingly interconnected, there is a growing need for internationalising national, regional and continental RI to serve a wider range of researchers as they join forces to tackle global challenges such as climate change, biodiversity loss, food security and infectious diseases. Growing internationalisation is facilitated by interactions with TERN's international counterparts, which share common objectives, structure and functioning (e.g. NEON, CERN, TERENO, CZO USA). For example, TERN's flux data are integrated into FLUXNET, a globally distributed environmental RI (Baldocchi et al 1996, Gu and Baldocchi 2002, Baldocchi 2008, Novick et al 2018 which is improving our understanding of ecosystem responses to fluctuations in environmental conditions (von Buttlar et al 2018). In addition, other global initiatives and policy frameworks have emerged in recent years to provide global access to data products collected at local or regional scales (e.g. GEOSS, IPBES). In another example of growing internationalisation, the Global Environmental Research Infrastructure (GERI) was formed to foster cooperation amongst RIs by founding members TERN (Australia), SAEON (South Africa), CERN (China), NEON (USA) and eLTER (Europe). TERN is part of an international consortium of environmental RI organisations, with the goal of providing spatially comprehensive and integrated data streams which are model-ready and publicly available for global syntheses. See the supplemental information for further details on TERN's international outreach activities and TERN's twenty-year vision.

Ecosystems and climate: fire and aridity
An understanding of ecosystem-climate interactions is the foundation of forecasting ecosystem responses to climate change (figure 2; Beringer et al 2015), assuming that such predictability is accurate enough to be feasible (Haughton et al 2018a). Environmental RI in the form of ecological observatories provides a platform for improved understanding of how ecosystems respond to climate across a diversity of vegetation types. In Australia, 32 major vegetation types have been identified in a national vegetation information system (NVIS; https://environment.gov.au/land/ native-vegetation/national-vegetation-informationsystem), although they can be simplified by combining similar types (e.g. arid and semi-arid Acacia forests, woodlands, shrublands and savannas are three major vegetation types which are all defined by a dominant canopy of Mulga Acacia). By example in a review of key findings from research using TERN RI at the Alice Mulga SuperSite of the Ti Tree basin, Eamus et al (2016) grouped NVIS vegetation types into forest, savanna, Mulga, shrubland, grassland and agriculture (figure 3). Across these landscapes, fire and aridity are key forces shaping ecological relationships with climate.
Bushfire is a primary attribute of the tropical wetseasonal savannas and semi-arid grasslands of western and northern Australia, where annual fire frequencies are common across the northern tropical savanna (see figures 2 and 4). The concentration of TERN infrastructure in northern Australia at intermediate longitudes is organised around the North Australian Tropical Transect (NATT; figure 1) to support savanna research across a very large precipitation gradient (320 to >1200 mm annual precipitation; Hutley et al 2011, Cleverly et al 2013, Ma et al 2013. Heavy rainfall during the wet season at northernmost locations along the NATT is associated with a large accumulation of biomass (Hutley et al 2011), and this grassy biomass cures over the subsequent dry season that lacks rainfall, leading to the very high frequency of fire re-occurrence at the northern end of the NATT (ca. 1-2 years; figure 4). Fire is responsible for the majority of productivity losses in the northern savanna (63%), whereas large weather events such as cyclones contribute very little to the long-term net biome carbon budget (Hutley et al 2013). Burning is furthermore largely responsible for greenhouse gas emissions from savannas and consequential greenhouse gas forcing of climate (Bristow et al 2016), amongst a cascade of indirect feedbacks between climate and ecosystems which are mediated through local atmospheric dynamics (figure 2; Beringer et al 2015). The outcome of this work is that it has contributed to a better understanding of tropical savanna   Outside of the tropics, bushfire tends to follow two patterns. In drylands, fires occur as a result of fuel accumulation directly following the conclusion of very wet periods, whereas in sclerophyllous eucalypt forests, wildfires occur following drought, once fuel has cured sufficiently (Griffin et al 1983, Bradstock 2010. At TERN's Calperum Mallee SuperSite (FLUXNET code AU-Cpr) in a Mediterranean climate, bushfire can have little or no effect on soil respiration, but net ecosystem productivity (NEP) and thus gross primary production (GPP) can be reduced following fire (Sun et al 2015(Sun et al , 2016(Sun et al , 2017b. Bushfire plays an important role in shaping Australian landscapes, but its observation by environmental RI is limited in time and space, and each event provides a few more hints toward a better understanding of ecosystem responses to fire. Seventy per cent of Australia is arid or semi-arid (Eamus et al 2006), where aridity is likely to dominate over warming and low levels of soil phosphorus in determining adaptation to future climate (Steane et al 2017). For example, fauna such as ants, termites and lizards in the Australian tropical savanna are arid-adapted and are thus likely to be resistant to future increases in aridity (Andersen et al 2015). By contrast, species in the arid-Mediterranean ecotone are fully adapted to neither climate and are thus sensitive to variations in climate, both spatial and those projected for the future (Guerin et al 2016). With aridity increasing globally, international integration of environmental RI organisations create further opportunities for discovering diversity responses to aridity.
Climate factors associated with aridity include temperature, vapour pressure deficit, solar radiation, precipitation and water availability. Aridity is moreover associated with patterns of water-use efficiency, lightuse efficiency, species richness, productivity and adaptability of leaf traits to native growth conditions (

Ecosystems and climate: climate variability and extreme events
Australia's climate is highly variable, with the cultural and economic significance of this highly variable climate illustrated in the well-known common parlance as 'a land K of droughts and flooding rains' (Dorothea Mackellar, https://dorotheamackellar. com.au/archive/mycountry.htm). A full range of vegetation and climate conditions are currently under-sampled by environmental RI globally (Jones et al 2017), thus the high temperature anomalies experienced by Australian ecosystems provides the world with an important end-member for developing an understanding of ecosystem responses to climate extremes (e.g. heatwaves; De Kauwe et al 2019).
Australia has recently experienced an increase in the frequency and severity of climate extremes (e.g. Ultimately, climate and weather are influenced by ocean-atmosphere interactions within ocean-basin modes of variability. For example, El Niño-Southern Oscillation (ENSO) is driven by a seasonal gradient of sea-surface temperature along the equatorial Pacific Ocean (Trenberth 1997), and fluctuations between the warm phase (El Niño) and the cold phase (La Niña) generally bring contrasting conditions to Northern and Southern Hemispheres. El Niño is associated with maritime wet conditions to the Northern Hemisphere and dry conditions across the Southern Hemisphere, whereas La Niña is associated with reversed impacts on precipitation and consequently ecosystems (Holmgren et al 2001). Data from TERN RI have been used to demonstrate that Australia shows continental phenological responses to ENSO-driven climate variability (Broich et al 2015) and that litterfall in the tropical rainforest of northeastern Australia is mainly driven by fluctuations in maximum temperature, which are related to ENSO (Edwards et al 2018).
Although ENSO provides the dominant climate signal for global weather patterns, it is becoming apparent that ENSO alone cannot fully explain differences in regional climate variability. Strong coupling amongst the tropical Pacific, Atlantic and Indian Oceans can impact the state of the climate, although limitations still exist in our ability to project future climate without including these teleconnections in climate models (Cai et al 2019). In Australia, extreme climate variability (floods, droughts, heatwaves) and resultant effects on water resources have been explained by interactions of the three nearest climate modes: ENSO, the Indian Ocean dipole (IOD) and the Southern annular mode (SAM) in the Southern Ocean  Multiple aspects of the climate contributed equally to increased CO 2 uptake in Australia , showing the 2011 global land C sink anomaly to be an integrated climatological, meteorological and ecosystem event (Cleverly et al 2016a). However, there were two restrictions on Australia's contribution to the land C sink anomaly. First, the asymmetric response of photosynthetic productivity to precipitation is dependent upon antecedent conditions, either amplifying or dampening their relationship (Sun et al 2017a). The land C sink anomaly followed the driest and hottest year of the Millennium Drought, thus antecedent water resources were at a minimum (van Dijk et al 2013). Second, energy-limited ecosystems did not show a similar asymmetric response to extrinsic forcing by precipitation and thus did not respond to climate forcing during the land C sink anomaly in the same way that semi-arid ecosystems did ( figure 5, Haverd et al 2017). Thus, the contribution of enhanced productivity in coastal, energy-limited ecosystems was expected to be small, whereas Australia's vast drylands have a high capacity for enhanced productivity during wet extremes (figure 5). Even with these limitations, the land C sink provided an ecosystem service which might have helped to slow the rate of climate change (Keenan and Williams 2018).
An inevitable outcome of increased productivity is biomass accumulation, especially across grasslands globally, where biomass accumulation is related to climate variability (Morgan et al 2016). Reduced diversity can result with the presence of a single species of invasive grass (e.g. buffel grass, Cenchrus ciliaris), which burns hotter and more completely than native grasses (Schlesinger et al 2013). For hummock grasslands which cover one-quarter of the Australian land area (Bowman et al 2008), the legacy of biomass accumulated during the 2011 land C sink anomaly persisted in the absence of burning for years as a strong carbon source due to photodegradation of the standing leaf litter (Cleverly et al 2016c). C budgets like those of hummock grasslands are very difficult to predict using land surface models (Haughton et al 2018a) due to a lack of theoretical foundation for the modelling of abiotic decomposition. This is thus an active area of research for which TERN's environmental RI will play an important role in integrating measurements, monitoring, modelling and remote sensing of carbon and water balances (Eamus et al 2016).

Dry extremes: drought and heatwave
Interannual variability in the global carbon cycle is strongly related to the large variability of the semi-arid land C sink (Ahlström et al 2015). In a single example, the 2011 land C sink in Australia was immediately followed by the return of drought and associated heat, which shut down the Australian land C sink even whilst total water storage on the continent had persisted Thus, photosynthetic productivity of Australia is sensitive to meteorological drought (i.e. months to years of below-normal precipitation) and agricultural drought (i.e. yield reduction due to soil drying), but Australia's hydroclimatic variability can moderate against a more severe hydrological drought (i.e. decline of water storage below the long-term mean; figure 6; Dai 2011). Even during long-term drought, moderately wetter-than-average years can have a strong positive effect on total water storage and productivity across Australia (figure 6). The Millennium Drought of 2001-2009 was the longest meteorological drought on record in Australia, but antecedent hydrological drought began in 1994 (van Dijk et al 2013). During the Millennium Drought, a worldwide reduction in photosynthetic productivity was attributed to drying in the Southern Hemisphere (Zhao and Running 2010). Despite a widespread and severe reduction in CO 2 uptake during the Millennium Drought, Australian vegetation has shown resilience to a drought as extreme as even it was (figure 6; Campos et al 2013, Ma et al 2013Ma et al , 2015. Australia has experienced much dryer, longer droughts during previous glacial maxima (Martin 2006), and this long history of drought has conferred a level of adaptation in Australia's vegetation. At the mesic end of the aridity gradient, the structure of tropical rainforest trees is adapted to the maximum historical water deficit (Pfeifer et al 2018). In the more arid-adapted Mallee, a multi-stemmed eucalypt which is also fire adapted, survival of common dry periods is achieved by maintaining conservative (i.e. very small) rates of transpiration . Two sequential years of much below-average precipitation in central Australia Ecosystem functional responses to both phases of the 'Angry Summer' heatwave (dry followed by wet) were evaluated using TERN's flux tower infrastructure at seven TERN ecosystem processes sites across southern Australia (van Gorsel et al 2016). NEP and GPP declined sharply during the heatwave in Mediterranean woodlands and dry sclerophyll forests, although reductions were smaller during the wet phase than during the preceding dry phase (figure 7). By contrast, the wet sclerophyll forest at the Tumbarumba Super-Site maintained NEP and GPP at constant levels as before the heatwave, with evaporative cooling ameliorating the heatwave and weakening land-atmosphere feedbacks (figure 7; van Gorsel et al 2016). However, soil moisture reserves were nearly depleted in the wet sclerophyll forest during the relatively short 'Angry Summer' heatwave, showing afternoon reductions of NEP and GPP during the dry portion of the heatwave, which is consistent with a photosynthetic and stomatal down-regulation due to stress (see figure 7, Cowan and Farquhar 1977). Thus, this heatwave provided a second example of the associated effects of drought and heatwave on ecosystem productivity of Australian ecosystems (see figures 6 and 7) and further demonstrates that increases in the intensity, frequency or duration of heatwaves in future might have seriously detrimental consequences for even Australia's wettest forests (van Gorsel et al 2016).  8). The presence of a tipping point carries a further risk from climate change as the locations of tipping points contract toward the centre of a species' range. Long-term ecological RI from various countries worldwide, including Australia, has been joined into the International Long Term Ecological Network (ILTER), which addresses the grand challenge of climate change and the resultant loss of biodiversity which is likely to occur (Mirtl et al 2018).
Thus far, we have discussed how adaptation, such as adaptation to aridity, can play an important role in protecting biodiversity from climate extremes, but considerations of adaptation and adaptability are also important for biodiversity conservation in a changing climate. Care must be taken, however, to avoid deprioritising refugia which are low in diversity and thereby less adapted to areas outside of their refugium (Costion et al 2015). Otherwise, several conservation approaches are available for promoting diversity and adaptability to climate change, including climateadjusted provenancing (figure 9), assisted migration, biodiversity corridors and ex situ strategies (Prober et al 2015, Christmas et al 2016b. In climate-adjusted provenancing, natural genetic variability is exploited to enhance climate resilience of restoration activities over time by predicting future changes in climate over incremental time steps (figure 9). This gradual approach over time allows for the detection of uncertainties (e.g. mismatch between predicted and actual climate change trajectories) before it is too late to correct for them (figure 9). TERN's plot-based monitoring infrastructure, especially those arrayed in transects along climate gradients, provide a powerful tool for evaluating community responses to climate change and for promoting resilience in biodiversity (Caddy-Retalic et al 2017).
There are several aspects of projected climate change in relation to ecosystem carbon cycles, such as changes in precipitation, vapour pressure deficit, temperature and elevated [CO 2 ]. Changes in precipitation seasonality are projected to reduce carbon stocks in the northern Australian savannas and rainforests, even with little change to annual total precipitation (Cook et al 2015). Increasing vapour pressure deficit during one key season as a result of decreasing seasonal precipitation and increasing temperature is expected to carry detrimental effects upon the carbon cycle of Australian alpine grasslands and tropical rainforests worldwide (Fu et al 2018, Marchin et al 2018). Photosynthetic production in tropical rainforests is currently restricted by high vapour pressure deficit, and they are unlikely to tolerate a much drier atmosphere (Fu et al 2018). In alpine grasslands of Australia, a tipping point has been identified wherein vapour pressure deficits which exceed this threshold can prevent the typical recovery of vegetational greenness at the end of the growing season (Marchin et al 2018). Increasingly elevated vapour pressure deficit is a serious risk for tree mortality and ecosystem function worldwide Warming is predicted to have strongly negative effects on Australia's temperate eucalypt forests, nearly 90% of which exist in temperature regimes above their thermal optimum for growth (11°C; Bowman et al 2014). In the absence of water or substrate limitations, respiration will increase along with rising temperatures, leading to global observations of increasing heterotrophic respiration and climate-driven loss of soil carbon (Bond-Lamberty et al 2018). Acting to counter the effects of increasing temperature, elevated [CO 2 ] can contribute to maintenance of photosynthesis at reduced stomatal conductance, thereby improving water-use efficiency whilst reducing pressure on limited water resources. CO 2 fertilisation is projected to be the main driver of savanna responses to climate, leading to increased carbon sequestration in vegetation, although the magnitude of the fertilisation effect will depend strongly upon changes in fire return interval and seasonality (Scheiter et al 2015). Increasing aridity and elevated [CO 2 ] are likely to affect vegetation dynamics of tropical savannas, which will alter fire regimes and provide further carbon feedbacks to climate (figure 2; Beringer et al 2015).

Final remarks
There is an urgent need for environmental information from RI in remote Australian landscapes, which would provide important outcomes and impact related to environmental reporting and fostering research in the framework of global challenges (van Dijk et al 2014). The risks of climate change as well as opportunities for conservation are emerging from this work, although there is still much to discover. Environmental RIs create the opportunity for identifying and evaluating the key drivers of ecosystem change by allowing researchers to observe state-changing events such as heatwaves, floods or droughts in locations which might be otherwise inaccessible. These are the first steps toward the development of an environmental forecasting system which can answer the global challenge of predicting ecosystem responses to climate change.

Acknowledgments
Australia's Terrestrial Ecosystem Research Network (TERN) is supported by the Australian government through the National Collaborative Research Infrastructure Strategy (NCRIS). We would like to thank two anonymous reviewers for providing substantive improvements to this letter.

Data availability statement
Data sharing is not applicable to this article as no new data were created or analysed in this study.