Savanna fires and their impact on net ecosystem productivity in North Australia JASON BERINGER *, LINDSAY B. HUTLEY w , NIGEL J. TAPPER * and LUCAS A. CERNUSAK w 1 *School of Geography and Environmental Science, Monash University, PO Box 11A, Clayton, Vic. 3800, Australia, wSchool of Science and Primary Industries, Charles Darwin University, Darwin, NT 0909, Australia Abstract Savannas comprise a large area of the global land surface and are subject to frequent disturbance through fire. The role of fire as one of the primary natural carbon cycling mechanisms is a key issue in considering global change feedbacks. The savannas of Northern Australia burn regularly and we aimed to determine their annual net ecosystem productivity (NEP) and the impact of fire on productivity. We established a long-term eddy covariance flux tower at Howard Springs, Australia and present here 5 years of data from 2001 to 2005. Fire has direct impacts through emissions but also has indirect effects through the loss of productivity due to reduced functional leaf area index and the carbon costs of rebuilding the canopy. The impact of fire on the canopy latent energy exchange was evident for 40 days while the canopy was rebuilt; however, the carbon balance took approximately 70 days to recover. The annual fire free NEP at Howard Springs was estimated at 4.3 t C ha 1 yr 1 with a range of 3.5 to 5.1 t C ha 1 yr 1 across years. We calculated the average annual indirect fire effect as 10.7 t C ha 1 yr 1 using a neural network model approach and estimated average emissions of fine and coarse fuels as 11.6 t C ha 1 yr -1 . This allowed us to calculate a net biome production of 2.0 t C ha 1 yr -1 . We then partitioned this remaining sink and suggest that most of this can be accounted for by woody increment (1.2 t C ha 1 yr -1 ) and shrub encroachment (0.5 t C ha 1 yr -1 ). Given the consistent sink at this site, even under an almost annual fire regime, there may be management options to increase carbon sequestration by reducing fire frequency. Keywords: CO 2 fluxes, eddy covariance, eucalyptus, fire, Howard Springs, net biome productivity, Savanna Received 7 July 2005; revised version received 3 November 2006 and accepted 29 November 2006 Introduction Savanna ecosystems are characterised by the coexis- tence of ‘carbon-rich’ woody and ‘carbon-poor’ herbac- eous plants dominated by grasses. Savannas occur in over 20 countries, mostly in the seasonal tropics, with a limited distribution in temperate regions (Hutley & Setterfield, 2007). Approximately 20% of the world’s land surface is covered with savanna vegetation and this biome is responsible for almost 30% of global net primary production (NPP) (Grace et al., 2006). For tropical savannas, carbon uptake from the woody com- ponent largely occurs via the C3 photosynthetic path- way and via the C4 pathway for the grass layer. The two coexisting lifeforms (trees and grasses), thus have dif- fering responses to available moisture and nutrient, light, changes to atmospheric CO 2 concentration and temperature and thus mechanistic modelling of savan- na responses to climate change is complex. Stochastic events such as fire and climatic variation maintain tree– grass coexistence in savanna, although a unifying eco- logical explanation for the wide range of observed savanna physiognomy is yet to emerge (Sankaran et al., 2004, 2005). Fire is possibly the most significant driver of savanna function in Australia and modification of savanna fire regimes influences the tree and grass balance. Reduc- tions in fire occurrence or intensity tend to enhance tree recruitment over herbaceous species, dramatically alter- Correspondence: Jason Beringer, tel. 161 3 9905 9352, fax 161 3 9905 2948, e-mail: jason.beringer@arts.monash.edu.au 1 Present address: Smithsonian Tropical Research Institute, PO Box, 0843-03092, Balboa, Ancon, Republic of Panama. Global Change Biology (2007) 13, 990–1004, doi: 10.1111/j.1365-2486.2007.01334.x r 2007 The Authors 990 Journal compilation r 2007 Blackwell Publishing Ltd