The amount of carbon stored in savannas represents a significant uncertainty in global carbon budgets, primarily because fire causes actual biomass to differ from potential biomass. We analyzed the structural response of woody plants to long-term experimental burning in savannas. The experiment uses a randomized block design to examine fire exclusion and the season and frequency of burn in 192 7-ha experimental plots located in four different savanna ecosystems. Although previous studies would lead us to expect tree density to respond to the fire regime, our results, obtained from four different savanna ecosystems, suggest that the density of woody individuals was unresponsive to fire. The relative dominance of small trees was, however, highly responsive to fire regime. The observed shift in the structure of tree populations has potentially large impacts on the carbon balance. However, the response of tree biomass to fire of the different savannas studied were different, making it difficult to generalize about the extent to which fire can be used to manipulate carbon sequestration in savannas. This study provides evidence that savannas are demographically resilient to fire, but structurally responsive.

Summary 1.Fire is important for the maintenance and conservation of African savanna ecosystems. Despite the importance of fire intensity as a key element of the fire regime, it is seldom measured or included in fire records. 2. We estimated fire intensity in the Kruger National Park, South Africa, by documenting fuel loads, fuel moisture contents, rates of fire spread and the heat yields of fuel in 956 experimental plot burns over 21 years. 3. Individual fires were conducted in five different months (February, April, August, October and December) and at five different return intervals (1, 2, 3, 4 and 6 years). Estimated fire intensities ranged from 28 to 17 905 kW m ); they were associated with a threefold difference between the mean moisture content of grass fuels in winter (28%) and summer (88%). 4. Mean fuel loads increased with post-fire age, from 2964 kg ha − 1 on annually burnt plots to 3972 kg ha − 1 on biennial, triennial and quadrennial burnt plots (which did not differ significantly), but decreased to 2881 kg ha − 1 on sexennial burnt plots. Fuel loads also increased with increasing rainfall over the previous 2 years. 5. Mean fire intensities showed no significant differences between annual burns and burns in the biennial, triennial and quadrennial categories, despite lower fuel loads in annual burns, suggesting that seasonal fuel moisture effects overrode those of fuel load. Mean fire intensity in sexennial burns was less than half that of other burns (638 vs. 1969 kW m − 1 ). 6. We used relationships between season of fire, fuel loads and fire intensity in conjunction with the park's fire records to reconstruct broad fire intensity regimes. Changes in management from regular prescribed burning to 'natural' fires over the past four decades have resulted in a decrease in moderate-intensity fires and an increase in high-intensity fires. 7. The highest fire intensities measured in our study (11 000 -> 17 500 kW m − 1 ) were significantly higher than those previously reported for African savannas, but were similar to those in South American cerrado vegetation. The mean fire intensity for late dry season (winter) fires in our study was less than half that reported for late dry season fires in savannas in northern Australia. 8. Synthesis and applications . Fire intensity has important effects on savanna vegetation, especially on the dynamics of the tree layer. Fire intensity varies with season (because of differences in fuel moisture) as well as with fuel load. Managers of African savannas can manipulate fire intensity by choosing the season of fire, and further by burning in years with higher or lower fuel loads. The basic relationships described here can also be used

Approaches to fire management in the savanna ecosystems of the 2-million ha Kruger NationalPark, South Africa, have changed several times over the past six decades. These approaches have included regular and flexible prescribed burning on fixed areas and a policy that sought to establish a lightningdominated fire regime. We sought to establish whether changes in management induced the desired variability in fire regimes over a large area. We used a spatial database of information on all fires in the park between 1957 and 2002 to determine elements of the fire regime associated with each management policy. The area that burned in any given year was independent of the management approach and was strongly related to rainfall (and therefore grass fuels) in the preceding 2 years. On the other hand, management did affect the spatial heterogeneity of fires and their seasonal distribution. Heterogeneity was higher at all scales during the era of prescribed burning, compared with the lightning-fire interval. The lightning-fire interval also resulted in a greater proportion (72% vs. 38%) of the area burning in the dry season. Mean fire-return intervals varied between 5.6 and 7.3 years, and variability in fire-return intervals was strongly influenced by the sequencing of annual rainfall rather than by management. The attempt at creating a lightning-dominated fire regime failed because most fires were ignited by humans, and the policy has been replaced by a more pragmatic approach that combines flexible prescribed burning with lightning-ignited fires.Respuesta de los Regimenes de Fuegos de Sabana a Cambios en las Políticas de Gestión de Fuego en un Parque Nacional Africano Resumen: Durante las seisúltimas décadas, los métodos de gestión de fuego en los ecosistemas de sabana han cambiado varias veces en el Parque Nacional Kruger (2 millones de ha), Sudáfrica. Estos métodos incluyeron quemas prescritas regulares y flexibles enáreas determinadas y una política que buscaba establecer un régimen de fuego dominado por relámpagos. Buscamos establecer si los cambios de gestión indujeron la variabilidad deseada en los regimenes de fuego en unárea extensa. Utilizamos una base de datos espacial de la informaciónde todos los incendios en el parque entre 1957 y 2002 para determinar elementos del régimen de fuego asociados con cada política de gestión. Elárea quemada en cualquier año fue independiente del método de gestión y se relacionaba cercanamente con la precipitación pluvial (y por lo tanto con pastos combustibles) en los dos años precedentes. Por otro lado, la gestión afectó a la heterogeneidad espacial de los incendios y su distribución estacional. En comparación con el intervalo de fuego por relámpagos, la heterogeneidad fue mayor en todas las escalas durante el período de quemas prescritas. El período de fuego por relámpagos también presentó una mayor proporción (72% vs. 38%) deárea quemada durante laépoca de secas. Los intervalos promedio de retorno de fuego variaron entre 5.6 y 7.3 años, y la secuencia de la precipitación anua...