Positive Feedbacks in the Fire Dynamic of Closed Canopy Tropical Forests

See allHide authors and affiliations

Science  11 Jun 1999:
Vol. 284, Issue 5421, pp. 1832-1835
DOI: 10.1126/science.284.5421.1832


The incidence and importance of fire in the Amazon have increased substantially during the past decade, but the effects of this disturbance force are still poorly understood. The forest fire dynamics in two regions of the eastern Amazon were studied. Accidental fires have affected nearly 50 percent of the remaining forests and have caused more deforestation than has intentional clearing in recent years. Forest fires create positive feedbacks in future fire susceptibility, fuel loading, and fire intensity. Unless current land use and fire use practices are changed, fire has the potential to transform large areas of tropical forest into scrub or savanna.

Fire is recognized as a historic but infrequent element of the Amazonian disturbance regime (1, 2). Currently, however, fires in Amazonian forests are frequent because of the accidental spread from nearby pastures and the increased susceptibility of partially logged or damaged forests (3–6). Here, positive feedbacks associated with accidental forest fires are reported; these constitute a threat to the integrity of a large part of the Amazonian forest.

Field studies were concentrated in the Tailândia region (Fig. 1). Ten 0.5-ha plots (eight fire-affected and two control), spread over 100 km2, were established in 1996 to study fire impacts on forest structure, biomass, and species composition (3). These plots were recensused after the dry season of 1997, during which eight of the plots burned to varying degrees. Fire recurrence, tree mortality, and biomass combustion levels within forests of different burn histories were quantified. In addition, combustible fuel mass was assessed with the planar intersect method (7) as adapted by Uhl and Kauffman (8, 9).

Figure 1

Study regions within the Brazilian Amazon (dark shaded area; see inset at lower left). Fire incidence in Tailândia and Paragominas (rectangles) was investigated with multitemporal Landsat TM imagery. Landowner interviews in Pará(PA) and Mato Grosso (MT) were used to assess fire in additional study regions [Alta Floresta, Santana do Araguaia, and Paragominas (square)]. All study regions are located along the development frontier in a region known as the Arc of Deforestation (shown with lighter shading).

We also examined characteristics of fires while they were occurring in four forest types (previously unburned, once-burned, twice-burned, and more than two previous burns) in December 1997. Direct observations of fires were made at widely scattered locations within a 150-km2 area south of Tailândia. For each observed fire, flame heights and depths (the width of the flaming front) were measured or estimated (10). The time the fireline took to move across a known distance was used to calculate the rate of spread and was combined with flame depth data to calculate the average range of flame residence times at a point. Flame height was used as a conservative estimate of total flame length for the calculation of fireline intensity (11) because wind and slope were minimal (12).

The first fire to enter a forest usually moves slowly along the ground (Table 1) and is similar to a prescribed burn (<50 kW m–1) in intensity (13). These fires consume little besides the dry leaf litter, but because of the characteristically thin tree bark [7.3 ± 3.7 mm for >20 cm diameter at breast height (dbh) (8)] protecting the cambium tissues, they still kill roughly 95% of the contacted stems >1 cm dbh. Large, thicker barked trees survive. After the fire, a rain of combustible fuels of all sizes falls from the standing dead trees (Table 1) (14). Fire damage and windthrow in these thinned forests continue to cause mortality for at least 2 years after the fire (4, 15). Fuel levels rise substantially and the open canopy (50 to 70% cover) allows greater solar heating and air movement to dry out the forest fuels. Previously burned forests thus become susceptible to fire during common dry season weather conditions (3).

Table 1

Forest, fuel, and fire characteristics of four different forest types within the Tailândia study region.

View this table:

Previously burned forests were much more likely to burn than were unburned forests in 1997 (Table 1). Burned forests are often adjacent to fire-maintained pasture and agricultural plots and are therefore frequently exposed to sources of ignition. Second fires are faster moving and much more intense. We estimate heat release (12) of <7500 kW m–2 in first burns but of 75,000 kW m–2 or more in subsequent burns. Because of the increased flame depth, the residence time increases despite faster rates of spread, resulting in greater tree mortality. Large trees have little survival advantage during these more intense fires. Fire-induced tree mortality can be modeled as a function of bark thickness and fire residence time (16). For the observed fire characteristics and bark thickness distribution (8), no more than 45% of trees over 20 cm dbh are susceptible to fire-induced mortality in the initial fires. However, in recurrent fires, up to 98% of the trees become susceptible to fire-induced mortality.

The impacts of recurrent fires are much worse than those of initial fires. Higher mortality results in a very open canopy (10 to 40% cover), large inputs of combustible fuels, and faster drying. During the 1997 fires, substantial amounts of carbon were released to the atmosphere, with combustion reducing onsite biomass by approximately 15, 90, and 140 Mg ha–1 in first, second, and recurrent burns, respectively. Invading grasses and weedy vines add highly combustible live fuels to the already fuel-laden forest (17). Fires in highly degraded areas are significantly more severe in all respects (flame height, intensity, depth, residence time, and rate of spread). Recurrent fires have the potential to eradicate trees from the landscape (18).

Multitemporal analyses of satellite imagery [Landsat Thematic Mapper (TM)] were used to extend our study of fire in space and time in the Tailândia region and also in the Paragominas region (Fig. 1). Both regions have similar forests, pronounced dry seasons, and average annual rainfall of 1500 to 1800 mm (19). A linear mixture modeling methodology (20) was used to separate forest from nonforest and to classify burned forests in a series of images for 1280 km2 near Paragominas (in 1984, 1991, 1993, and 1995) and for 2640 km2 around Tailândia (in 1984, 1991, 1993, 1995, and 1997). The forest location and the area affected by fire were determined for the images of each region. Cross tabulation of the classified images provided a history of deforestation and forest burning throughout the study regions. The fire rotation, which is the amount of time required to burn an area equivalent to the entire forested area (21), was calculated for each region.

Areas that are minimally forested because of the recurrence of fire are likely to appear deforested in satellite imagery analyses [Landsat TM, Système Probatoire d'Observation de la Terre (SPOT), and Advanced Very High Resolution Radiometer (AVHRR)]. Visual inspection of paper Landsat TM images, as used to monitor deforestation in the Brazilian Amazon (22, 23), is also expected to misclassify many burned forests as deforested areas. In this study, cross tabulation showed that, in comparison to unburned forest, once-burned forests were twice as likely to be classified as having been deforested, whereas twice- and thrice-burned forests were 11 and 15 times as likely to be classified as deforested.

We conducted a detailed study of deforestation in burned forests. Imagery of Paragominas for the period from 1993 to 1995 was used to test whether the deforestation of forests that had burned in 1992 was intentional (for example, slashed and burned for cattle pasture and crops) or accidentally induced by fire (that is, extremely thinned). Areas of forest that burned in 1992 that became either new slash or pasture were classified as intentional deforestation. These areas were generally adjacent to existing forest edges and had regular shapes. Forests that became “degraded pasture” (that is, second growth), an unlikely transition in just 2 years, were classified as accidental fire-induced deforestation. Fire-induced deforestation was generally irregular in shape and often occurred far from forest edges (Fig. 2).

Figure 2

Two 64-km2 imagery subsections illustrating the differences in location and form of normal deforestation (for example, caused by slash and burn for pasture and crops) and fire-induced deforestation, caused by accidental forest fires.

In the Paragominas region, we estimate that accidental fire-induced deforestation increased deforestation estimates by 129% between 1993 and 1995. Correcting the deforestation estimate for this factor yields an intentional (that is, slash and burn) deforestation rate of 1.7% for the period from 1993 to 1995, which is in accord with the average (1.8%) deforestation rate before the El Niño–induced fires of 1992/1993 (Table 2). This surprising result implies that the basin-wide jump in estimated deforestation rates from 1993 to 1995 (23,24) may have occurred largely because of the widespread forest fires of 1992 and 1993.

Table 2

Deforestation and forest burning in several study regions determined with two different methodologies: Imagery analyses were based on multitemporal analyses of Landsat TM imagery; interview-based data were from a 1996 landowner interview study of fire and deforestation. Comparison of these data with satellite imagery has shown them to be very conservative with regard to fire and deforestation (6, 29). An additional 85 interviews were conducted in Rondonia and Acre. Rondonia experienced fire but was excluded because of the small amount of total forested area (72 km2). Acre had no recorded occurrence of accidental forest fire and is apparently wetter and less seasonal.

View this table:

There have been no known analyses of the natural fire rotation in lowland tropical rainforests, but the limited data from charcoal studies (1) imply a fire rotation of hundreds or thousands of years. Fire-return intervals of less than 90 years can eliminate rainforest tree species, whereas intervals of less than 20 years may eradicate trees entirely (25). On the basis of our time series analysis of imagery, we calculate that Paragominas' and Tailândia's forests are currently experiencing fire rotations of between 7 and 14 years. Previously burned forests are even more prone to burning, with calculated fire rotations of less than 5 years.

Results derived from satellite images were compared to data from three other regions (7290 km2) of the Amazon. A total of 117 randomly chosen landowners from Pará and Mato Grosso were asked to mark deforested and burned forest areas on Landsat TM image prints (Fig. 1 and Table 2). Calculated fire rotations for each region were similar to one another (Table 2), indicating that the entire region between Mato Grosso and northeastern Pará may be experiencing the same fire regime. Though rotation times were longer than those determined by analysis of TM images, the landowner estimates of forest area burned were for non–El Niño years (1994 and 1995). The resultant fire rotation calculations are therefore very conservative, because this study's multitemporal analyses show that 90% of forest burning has occurred during El Niño years.

The average rate and intensity of forest burning and deforestation can be expected to increase as previously burned forest area expands. A positive feedback exists between forest fires, future fire susceptibility, fuel loading, and fire severity. In the past several years, roughly 50% of the remaining forests in the study regions around Paragominas and Tailândia have burned, 20% having burned more than once. First-time burns can be controlled and put out manually with minimal equipment, but more than 30% of the observed fires in previously burned forest had fireline intensities that were beyond the limits of manual control (13). Left unchecked, the current fire regime will result in an inexorable transition of the entire area to either scrub or grassland (25). Effects on the regional climate, biodiversity, and economy are likely to be extreme. These fire-induced changes will take several years to occur but are likely to be irreversible (26,27) under current climatic conditions.

  • * To whom correspondence should be addressed. E-mail: cochrane{at}


Stay Connected to Science

Navigate This Article