Aim  (1) To estimate the local and global magnitude of carbon fluxes between savanna and the atmosphere, and to suggest the significance of savannas in the global carbon cycle. (2) To suggest the extent to which protection of savannas could contribute to a global carbon sequestration initiative. Location  Tropical savanna ecosystems in Africa, Australia, India and South America. Methods  A literature search was carried out using the ISI Web of Knowledge, and a compilation of extra data was obtained from other literature, including national reports accessed through the personal collections of the authors. Savanna is here defined as any tropical ecosystem containing grasses, including woodland and grassland types. From these data it was possible to estimate the fluxes of carbon dioxide between the entire savanna biome on a global scale. Results  Tropical savannas can be remarkably productive, with a net primary productivity that ranges from 1 to 12 t C ha−1 year−1. The lower values are found in the arid and semi‐arid savannas occurring in extensive regions of Africa, Australia and South America. The global average of the cases reviewed here was 7.2 t C ha−1 year−1. The carbon sequestration rate (net ecosystem productivity) may average 0.14 t C ha−1 year−1 or 0.39 Gt C year−1. If savannas were to be protected from fire and grazing, most of them would accumulate substantial carbon and the sink would be larger. Savannas are under anthropogenic pressure, but this has been much less publicized than deforestation in the rain forest biome. The rate of loss is not well established, but may exceed 1% per year, approximately twice as fast as that of rain forests. Globally, this is likely to constitute a flux to the atmosphere that is at least as large as that arising from deforestation of the rain forest. Main conclusions  The current rate of loss impacts appreciably on the global carbon balance. There is considerable scope for using many of the savannas as sites for carbon sequestration, by simply protecting them from burning and grazing, and permitting them to increase in stature and carbon content over periods of several decades.

Measurements of carbon dioxide flux over undisturbed tropical rain forest in Brazil for 55 days in the wet and dry seasons of 1992 to 1993 show that this ecosystem is a net absorber of carbon dioxide. Photosynthetic gains of carbon dioxide exceeded respiratory losses irrespective of the season. These gains cannot be attributed to measurement error, nor to loss of carbon dioxide by drainage of cold air at night. A process-based model, fitted to the data, enabled estimation of the carbon absorbed by the ecosystem over the year as 8.5 ± 2.0 moles per square meter per year.

A simple 'big leaf ecosystem gas exchange model was developed, using eddy covariance data collected at an undisturbed tropical rainforest in south-western Amazonia (Brazil). The model used mechanistic equations of canopy biochemistry combined with an empirical stomatal model describing responses to light, temperature and humidity. After calibration, the model was driven using hourly data from a weather station at the top of the tower at the measurement site, yielding an estimate of gross primary productivity (annual photosynthesis) in 1992/1993 of about 200 mol C m~^ year"\ Although incoming photon flux density emerged as the major control on photosynthesis in this forest, at a given PAR CO2 assimilation rates were higher in the mornings than in the afternoons. This was attributable to stomatal closure in the afternoon in response to increasing canopy-to-air vapour pressure differences. Although most morning gas exchange was clearly limited by the rate of electron transport, afternoon gas exchange was generaWy observed to be very nearly co-limited by both Rubisco activity (V^^^) and electron transport rate. The sensitivity of the model to changes in nitrogen allocation showed that the modelled ratio of V^^^^ to electron transport (•/max) served nearly to maximize the annual carbon gain, and indeed, would have resulted in almost maximum annual carbon gain at the pre-industrial revolution atmospheric CO2 concentration of 27 Pa. Modelled gross primary productivity (GPP) was somewhat lower at 27 Pa, being about 160mol C m~^ year"'. The model suggests that, in the absence of any negative feedbacks on GPP, future higher concentrations of atmospheric CO2 will continue to increase the GPP of this rainforest, up to about 230 mol C m"^ year' at 70 Pa.

We present the energy and mass balance of cerrado sensu stricto (a Brazilian form of savanna), in which a mixture of shrubs, trees and grasses forms a vegetation with a leaf area index of 1-0 in the wet season and 0-4 in the dry season. In the wet season the available energy was equally dissipated between sensible heat and evaporation, but in the dry season at high irradiance the sensible heat greatly exceeded evaporation. Ecosystem surface conductance gî n the wet season rose abruptly to 0-3 mol m"^ s"^ and fell gradually as the day progressed. Much of the total variation in gs was associated with variation in the leaf-to-air vapour pressure deficit of water and the solar irradiance. In the dry season the maximal g^ values were only 0-1 mol m"^ s~\ Maximal net ecosystem fluxes of CO2 in the wet and dry season were -10 and -15 /Jmol CO2 m'^ s~', respectively (sign convention: negative denotes fluxes from atmosphere to vegetation). The canopy was well coupled to the atmosphere, and there was rarely a significant build-up of respiratory CO2 during the night. For observations in the wet season, the vegetation was a carbon dioxide sink, of maximal strength 0*15 mol m"^ d~\ However, it was a source of carbon dioxide for a brief period at the height of the dry season. Leaf carbon isotopic composition showed all the grasses except for one species to be C4, and all the palms and woody plants to be C3. The CO2 coming from the soil had an isotopic composition that suggested 40% of it was of C4 origin.

Air and soil temperatures were measured during dry season heading fires in three different physiognomic forms of native vegetation common in Central Brazil: cerrado sensu stricto (dense scrub of shrubs and trees), campo cerrado (open scrub), and campo sujo (open grassland with scattered shrubs). The vegetation was protected from fire for 15 y in some areas, had been burned once every two years, and once each year in other areas. The temperatures were measured with type-k thermocouples and recorded at intervals of 22.5 sees. Air temperature measurements were taken at 1, 60 and 160 cm. Maximum air temperatures ranged from 85°C to 840°C, and the duration above 60°C varied from 20 to 270 seconds. In the soil, negligible temperature increases were recorded below 5 cm depth, whereas at 2 cm maximum soil temperatures varied from 29 to 38°C. Possible influences of fuel load and moisture on the recorded data and on the behaviour of fire in the cerrado ecosystems are discussed.

To estimate the relative contributions of woody and herbaceous vegetation to savanna productivity, we measured the 13C/12C isotopic ratios of leaves from trees, shrubs, grasses and the surface soil carbon pool for 22 savannas in Australia, Brazil and Ghana covering the full savanna spectrum ranging from almost pure grassland to closed woodlands on all three continents. All trees and shrubs sampled were of the C3 pathway and all grasses of the C4 pathway with the exception of Echinolaena inflexa (Poir.) Chase, a common C3 grass of the Brazilian cerrado. By comparing the carbon isotopic compositions of the plant and carbon pools, a simple model relating soil delta 13C to the relative abundances of trees + shrubs (woody plants) and grasses was developed. The model suggests that the relative proportions of a savanna ecosystem's total foliar projected cover attributable to grasses versus woody plants is a simple and reliable index of the relative contributions of grasses and woody plants to savanna net productivity. Model calibrations against woody tree canopy cover made it possible to estimate the proportion of savanna productivity in the major regions of the world attributable to trees + shrubs and grasses from ground-based observational maps of savanna woodiness. Overall, it was estimated that 59% of the net primary productivity (Np) of tropical savannas is attributable to C4 grasses, but that this proportion varies significantly within and between regions. The C4 grasses make their greatest relative contribution to savanna Np in the Neotropics, whereas in African regions, a greater proportion of savanna Np is attributable to woody plants. The relative contribution of C4 grasses in Australian savannas is intermediate between those in the Neotropics and Africa. These differences can be broadly ascribed to large scale differences in soil fertility and rainfall.

No abstract

1 Carbon dioxide and water vapour fluxes were measured for 55 days by eddy covariance over an undisturbed tropical rain forest in Rondonia, Brazil. Profiles of CO2 inside the canopy were also measured. 2 During the night, CO2 concentration frequently built up to 500 ppm throughout the canopy as a result of low rates of exchange with the atmosphere. In the early morning hours, ventilation of the canopy occurred. 3 Ecosystem gas exchange was calculated from a knowledge of fluxes above the canopy and changes of CO2 stored inside the canopy. Typically, uptake by the canopy was 15 |imol m"^ s"^ in bright sunlight and dark respiration was 6-7 pmol m^^ s"\ The quantum requirement at low irradiance was: 40 mol photons per mol of CO2. 4 Bulk stomatal conductance of the ecosystem was maximal in the early moming (0.4-1.0 mol m^^ s'^) and declined over the course of the day as leaf-to-air vapour pressure difference increased.