Dryland vegetation response to wet episode, not inherent shift in sensitivity to rainfall, behind Australia's role in 2011 global carbon sink anomaly

There is compelling new evidence that semi-arid ecosystems are playing a pivotal role in the interannual variability and greening trend of the global carbon cycle (Ahlström et al., 2015). The situation is exemplified by the vast inland region of Australia, the driest inhabited continent. Using a global model, Poulter et al. (2014) inferred that Australian ecosystems contributed 57% of a record global carbon uptake anomaly in 2011 and have entered a regime of the enhanced sensitivity to rainfall since the mid-1990s. Here, we present new observation-based evidence confirming a significant role of Australian ecosystems in the 2011 carbon sink anomaly. Our results do not, however, support a shift in sensitivity of vegetation activity to rainfall. We upscaled carbon and water fluxes from 14 sites around Australia, using a fine spatial resolution (0.05°) implementation of the CABLE biogeochemical land surface model. We used model data fusion to constrain biospheric fluxes and stores using remotely sensed vegetation cover (Zhu et al., 2013) and formal parameter estimation using the Levenberg–Marquardt method (Doherty, 2004) to minimize residuals between predictions and observations of streamflow from 416 gauged catchments, (CO2 and H2O) ecosystem fluxes from 14 sites of the OzFlux network (Isaac, 2014), litterfall, soil, litter and biomass carbon pools. The analysis is updated from Haverd et al. (2013), the only published assessment of Australian biospheric carbon balance making the full use of relevant regional observation sources. Disregarding land-use effects, we estimate that terrestrial ecosystems in Australia absorbed 0.4 ± 0.2 [2σ] Pg C in 2011 (Fig. 1a), lower than the estimate by Poulter et al. (2014) (0.84 Pg C), but still a significant proportion (30%) of the record global residual land sink anomaly of 1.5 ± 0.9 [1σ] Pg C y−1 relative to the 2003–2012 decadal mean (Le Quéré et al., 2015). Poulter et al. (2014) suggested that Australian ecosystems have entered a regime of enhanced sensitivity of vegetation activity to precipitation since the mid-1990s, compared with the preceding 15-year period. Their finding of a fourfold increase in sensitivity was based on regressing modelled annual net ecosystem production (NEP) on precipitation totals for March–April–May (MAM), taken to represent the peak vegetation activity. However, vegetation activity in inland Australia (inset, Fig. 1b), shown to account for greater than 90% of variance in Australian NEP (Haverd et al., 2013), is governed by episodic rainfall (Broich et al., 2014), with the marked differences in seasonal distribution between years (Pickett-Heaps et al., 2014). Growth may be associated with rain falling at any time of the year; therefore, total annual rainfall – not any specific range of months – may be a more relevant predictor of ecosystem productivity. This is confirmed by our data, which show that annual NEP is more sensitive to whole-year rainfall: the slope of annual NEP for inland Australia (1982–2013) regressed against whole-year rainfall ((1.1 ± 0.2 [1 SE]) × 10−2 PgC × mm−1 month) is higher than for MAM ((0.4 ± 0.1) × 10−2 PgC × mm−1 month). No significant change (t-test of the equality of slopes; P = 0.45) in sensitivity of NEP to rainfall is apparent (1997–2010: (0.9 ± 0.1) × 10–2 PgC mm−1 month, 1982–1996: (0.8 ± 0.3) × 10−2 PgC mm−1 month) between the two periods considered by Poulter et al. (2014), when 2011 is omitted (Fig 1b). Inclusion of the exceptional year 2011 does increase the slope of the relationship between GPP and rainfall for the later period (1997–2011: (1.3 ± 0.2) × 10−2 PgC mm−1 month). This is an expected consequence of the response of drought-adapted vegetation to a strong rainfall pulse in the preceding year (Fig. 1a) (Huxman et al., 2004). This apparent increase in ecosystem sensitivity is thus based on a single wet episode and does not justify Poulter's interpretation of an inherent shift in the sensitivity of vegetation activity to moisture availability, mediated by increased grass cover and woody encroachment (Poulter et al., 2014). Carbon stocks in semi-arid systems are robust to native, year-to-year fluctuations of rainfall. Our evidence suggests that large episodes of carbon cycle variability, like the 2011 anomaly, are driven by pulse response behaviour of the drought-adapted biota in response to inter- or multi-annual variations in rainfall amount. VH and CMT thank the support of the Australian Climate Change Science Program. BS acknowledges funding as an OCE Distinguished Visiting Scientist to CSIRO Oceans & Atmosphere, Canberra.