Spatial and temporal variability of fires in relation to ecosystems, land tenure and rainfall in savannas of northern South America: SPATIAL AND TEMPORAL VARIABILITY OF FIRES

Spatial and temporal variability of fires in relation toecosystems, land tenure and rainfall in savannas ofnorthern South America

M . R O M E R O-R U I Z *, A . E T T E R w , A . S A R M I E N T O w and K . T A N S E Y *

*Department of Geography, University of Leicester, University Road, Leicester LE1 7RH, UK, wDepartamento de Ecologıa

y Territorio, Facultad Estudios Ambientales y Rurales, Universidad Javeriana, Tr. 4 Nr 42-00 piso 8, Bogota DC, Colombia

Abstract

Fire is a predominant factor forcing global terrestrial biomass dynamics, with more than 30%of the land surface showing frequent burning, particularly in the tropics, where it mostlyaffects savannas ecosystems annually. Savannas, which cover approximately 269 million ha inSouth America, play a major role in the global carbon cycle. They are affected by increasinghuman pressures and global climate change. Using satellite data, this study quantifiesvegetation burning in the Colombian Llanos savannas for the period 2000–2008, and analyzeshow fire spatial pattern, frequency and extent vary with ecosystem type, land tenure andrainfall. On average 2.75� 0.5 million ha (24� 4.2%) of the savannas burn each year. Burnedarea is highly variable, with 3.4 million ha burned in 2002–2003 and o1.9 million ha in 2005–2006. However, during the 2000–2008 period near of 3.7 million ha (33.5%) of the savannasnever burned. Compared with the average 8–10 years of fire return time for the tropics andsubtropics, these savannas burn twice as often. In addition, the average burn size figure fortropical and subtropical grassland savannas (with o5% trees) of 7000 ha (median 5000 ha), isabout seven times the average burned patch size we found in our study. Fires predominate inthe well-drained high plain savannas, lowest figures occurring along the Andean foothills, inforested areas and in pasture and croplands. Annual proportion burned varies with landtenure, being highest in National Parks. This study is the first complete regional map of firedisturbance in a South American savanna. This detailed regional data provides a uniqueopportunity for increasing the accuracy of global carbon emission calculations.

Keywords: burned area, Colombia, fire, Llanos Orientales, Modis, satellite

Received 23 June 2009 and accepted 17 August 2009

Introduction

Human induced land cover change is considered to be

one of the most important processes affecting ecosystem

structure, carbon and nutrient cycling, greenhouse gas

emissions and land–atmosphere energy exchange, all

contributing to global ecological change (Foley et al.,

2005). Factors driving land cover dynamics such as fire

are key actors in these processes. More than 30% of the

global land surface shows significant fire frequencies

particularly in the tropical belt 201N–301S (Chuvieco

et al., 2008a; Tansey et al., 2008). Between 2000 and 2007

more than 3900 million ha of burned area was detected

globally, 80% occurring in woodlands and scrublands,

and 17% in grasslands and croplands (Cochrane, 2002;

Tansey et al., 2004, 2008). Most of these fires are asso-

ciated with savanna vegetation taking place in Africa,

Australia, and South America (Andreae, 1992; Hao &

Liu, 1994).

Savannas play a role in the global carbon cycle given

their belowground storage capacity, seasonal burning,

regrowth and tree-grass dynamics. Savanna fires are a

major source of CO2 release to the atmosphere (Mouillot

et al., 2006), and fires associated with tropical grazing

systems are thought to have been contributing to 440%

of globally burned phytomass (Hall & Scurlock, 1991).

Van der Werf et al. (2003) in turn, estimates carbon

emissions from fires in the tropics to represent 9% of

the NPP (4 Pg C yr�1), and the portion of NPP con-

sumed by in for tropical savannas to be around 25%.

Emissions from fires have increased over the last 100

years, from 30% to 50% mainly in the tropical areas

(Mouillot et al., 2006).

Fire is a seasonal phenomenon with peaks associated

with dry seasons. During the wet season normally thereCorrespondence: M. Romero, tel. 1 44 571 2576639, fax 1 44 571

2018326, e-mail: [email protected]

Global Change Biology (2010) 16, 2013–2023, doi: 10.1111/j.1365-2486.2009.02081.x

r 2009 Blackwell Publishing Ltd 2013

is an establishment of an herbaceous carpet that dries

quickly in the dry leading to increased fire risks. In

northern South America, specifically in Colombia and

Venezuela, fires occur in a dry season between Decem-

ber and early April (Dwyer et al., 2000). Besides climate,

land management practices are a major factor determin-

ing soil and vegetation characteristics, and driving the

ecological processes of savannas that control the

amounts of C within the terrestrial pool, such as fire

frequencies and biomass accumulation (Hoffmann et al.,

2002; Grace et al., 2006). Humans have used fire to

develop roads, establish crop areas, eliminate grass

and other types of nondesirable vegetation, eliminate

culture wastes, hunt, and to develop methods to pre-

vent natural disasters (Laris, 2002; Wardell et al., 2004).

The global characterization of fire activity by Chuvieco

et al. (2008a) concludes on the importance of socio-

economic drivers such as population density, accessi-

bility and economic availability in determining the

density, intensity and variability of fire patterns.

In Latin America there is an increasing human pres-

sure on savannas due to the role such ecosystems are

playing in the world food supply (Brannstrom et al.,

2008). In the Brazilian savannas burning frequency

increased due to more intensive management of the

native grasslands for cattle grazing (Coutinho, 1990),

while in areas of intensive agriculture and improved

pastures burning has been disappearing (Brannstrom

et al., 2008). In Colombia there has been an increasing

expansion of the agricultural front into the savannas of

the Llanos during the last 30 years. Such human pres-

sures are driving big changes in the ecological processes

by affecting fire regimes, hydrology and nutrient cycles

including the capture and emission of green house

gasses (Asner et al., 2004).

Because savannas are naturally subject to frequent

fires, they also are dynamic aggrading systems and will

tend to sequester carbon in periods of fire exclusion. It

has been shown for Africa, that fires play an important

role as a disturbance factor affecting vegetation struc-

ture, woody cover and biomass accumulation, espe-

cially in mesic savannas with mean annual rainfall

4350 mm (Sankaran et al., 2005). Grace et al. (2006)

argue that in the framework of the global climate

change savannas could potentially help with carbon

sequestration if emissions from fires and grazing were

controlled by management options. Scurlock & Hall

(1998) contend that savannas account globally for an

estimated annual carbon sink of up to 0.5 Pg C.

However, carbon accounting still relies too much on

global average figures with few regional studies. Emis-

sions from fires can vary significantly due to local and

regional differences in ecosystem type, biomass and

combustion efficiencies (Van Der Werf et al., 2003).

Regional differences are highly relevant because savan-

nas vary much depending on the proportions of grass

and shrub/tree components as a response to soil, cli-

mate and management (Eiten, 1971). Delmas et al. (1995)

and Barbosa & Fearnside (2005a) have shown the com-

plexity of the burning processes in savanna fires, with

fire density and severity influenced by physical controls

(combustion), biological characteristics (biomass quan-

tity and quality), environmental factors (topography

and climate) and management practices. With global

warming some of these factors will change due to the

expected temperature increases and longer droughts in

this biome. Hoffmann et al. (2002) conclude that there is

an apparent positive feedback between rising carbon

emissions, global warming and fires, which could

increase fire frequency by 40% in the near future.

There are, however, large uncertainties in the stocks

and fluxes of carbon, and especially in the fluxes

attributable to burning and the impacts of land use

change (Grace et al., 2006). Burned area is a most

significant uncertainty factor of up to 50% of the mean

annual estimates (Van Der Werf et al., 2003). A funda-

mental problem regarding the estimates of carbon emis-

sions is the lack of spatially explicit regional analyses

that can be included in calculations, under the wide

range of biophysical and management conditions in

savannas. Wassmann & Vlek (2004) argue that in order

to get better estimates of fluxes, it is essential to inte-

grate GIS databases that can capture the heterogeneity

of natural factors and management, rather than look to

providing more detailed field data. In their study of the

global characterization of fire regimes Chuvieco et al.

(2008a) point to the need to address two aspects: (i)

exploring the relations between human and biophysical

variables at the regional scale, and (ii) the study of

the spatial variations of the fire patterns using spatial

metrics.

South American savannas cover approximately

269 million ha or in other words 8–10% of the global

savanna biome, and are subject to intense human pres-

sure (Grace et al., 2006; Chuvieco et al., 2008b). The

Llanos of Colombia and Venezuela is an important

savanna region in South America covering some 45 mil-

lion ha. The Llanos exhibit high fire frequencies (Tansey

et al., 2008), but lack a comprehensive regional knowl-

edge on the spatial and temporal burning patterns. This

paper addresses a first spatially explicit regional analy-

sis study of fire patterns for the Colombian Llanos

savannas. We address the variability of the burning

patterns using the best available satellite data for the

period of 2000–2008. A specific aim was to establish the

extent, recurrence and area patterns in relation to bio-

physical and management factors, such as rainfall,

ecosystem type and land tenure.

2014 M . R O M E R O-R U I Z et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

Materials and methods

Study area

In Colombia, the Llanos region covers almost 17

million ha located between 2.51 to 71N and �741 to

�671W with an altitude ranging between 200 and

400 m. Annual rainfall varies between 1500 and

2500 mm in an east–west gradient. The dry season runs

from December through the end of March. The mean

annual temperature fluctuates between 27 1C and 30 1C

in the dry months and 23 1C and 26 1C in the rainy

season (Etter, 1998a). Relative humidity varies between

60% and 90%. In the year 2000, nearly 11 million ha

(65%) of the Llanos region correspond to grassland

savanna ecosystems and the remaining 6 million ha

are distributed between gallery forest (a narrow strip

of forest found next to rivers), crops, introduced pas-

tures and water bodies. The region is composed of two

main physiographic units divided by the River Meta:

the Andean piedmont to the north and the high-plains

to the south. The savanna landscapes are composed of flat

to rolling plains and comprise three major geomorpholo-

gic units: (i) the quaternary foothills of the Andes; (ii) the

tertiary sedimentary plains with various incision levels

south of the Meta and Orinoco rivers; and (iii) the sandy

foothills and plains west of the Orinoco river. The Andean

piedmont (i) includes large areas of wetlands and season-

ally flooded savannas in the lower parts close to the River

Meta, which run northeast into Venezuelan Apure region.

The Llanos savannas are generally chemically poor, soil

limited ecosystems(pedobiomes), where Ultisols and Oxi-

sols predominate (Instituto Geografico Agustin Codazzi –

IGAC, 1983). However, savannas in the region are quite

heterogeneous and include several ecosystem types ac-

cording to their geomorphology, soils, flooding, and ve-

getation type (Fig. 1, Table 1).

Until the arrival of the Spanish in the 1500s the region

was occupied for several thousand years by aboriginal

hunter gatherer and few farmer communities (Rausch,

1994). Spanish missions introduced cattle and the

Llanos stayed until a few decades ago, a frontier area

Fig. 1 Location of the study area (black - Colombian Llanos), and the continuation of the Llanos into Venezuela (brown). Land tenure,

ecosystems and cleared areas are for the year 2000 [Derived from (Etter, 1998b) and Landsat images].

S PA T I A L A N D T E M P O R A L VA R I A B I L I T Y O F F I R E S 2015

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

dominated by extensive cattle grazing. With the exploita-

tion of oil fields in the 1980s, development of infrastructure

in the region lead to increasing the regional economy and

to more profitable agricultural production. Large areas

of natural savannas changed from the predominantly

extensive grazing to more intensive agricultural uses,

with nowadays 2.5 million ha of introduced pastures

(Brachiaria sp.), 120 000 ha of oil palm and timber planta-

tions, and 180 000 ha of rice and other crops. However,

extensive cattle grazing is currently still the dominant

land use in area, with a cattle herd of approximately

5.5 million head, which relies on burning to provide

regrowth forage for the cattle in the dry season.

The Llanos are still sparsely populated with current

population of the region approximately 1.4 million or

o5% of the country total, 31% living in rural areas.

Most of the population is concentrated within 50 km of

the Andean mountain range, where the major center

Villavicencio with 350 000 inhabitants, is located. Also

20 minor towns with up to 20 000 inhabitants exist. The

major economic center, Villavicencio is well connected

to the capital Bogota. The economy of the region de-

pends mainly from the exploitation of oil fields, cattle

ranching and intensive agriculture (irrigated rice, sor-

ghum), and increasingly also on oil palm and forestry

plantations, mainly along the western part of the region.

Land tenure in the region includes privately owned

ranches (480% of area), as well as indigenous reserves

formally established in the 1990s with a total area (TA)

of 1 340 000 ha, and the National Park ‘El Tuparro’ of

548 000 ha created in 1973 (Fig. 1).

Burned area mapping

To analyze fire patterns we used the MODerate-resolu-

tion Imaging Spectroradiometer (MODIS) composite

surface reflectance (MOD09A1) Level 2 product,

which has a 500 m spatial and 8-day temporal resolution.

Level 2, corresponds to the atmospherically corrected

surface reflectance product composed of seven bands:

band 1 – visible red (620–670� 10�9 m); band 2 – near

infrared (NIR) (841–876� 10�9 m); band 3 – blue (459–

479� 10�9 m); band 4 – green (545–565� 10�9 m); band

5 – short wave infrared SWIR (1.230–1.250� 10�9 m); band

6 – medium wave infrared MIR (1.628–1.652� 10�9 m) and

band 7 – long wave infrared LIR (2.105–2.155� 10�9 m).

It has been shown that 490% of fires occur in the

peak of the dry season (Dwyer et al., 2000; Barbosa &

Fearnside, 2005b). We used data from end December to

early April to ensure inclusion of the whole dry season,

for all available data for the years 2000 through 2008.

Data were downloaded from the WIST System (https://

wist.echo.nasa.gov/api/). Images were downloaded,

subset and reprojected to UTM zone 191N. To eliminate

nonfuel areas such as water bodies, wetlands, river

sand banks, rocky outcrops and urban areas we used

land cover obtained from Landsat ETM products to

mask these areas. In some cases it was necessary to

eliminate clouds and shadows, for which we adopted

an unsupervised classification. To extract the burned

and not-burned areas we used the burned area index

(BAIM) developed by Martin et al. (2005) based on the

near and short wave infrared bands relationship:

BAIM ¼ 1

ðrCnir � rnirÞ2 þ ðrCswir � rswirÞ

2ð1Þ

where, rnir 5 near-infrared pixel reflectance, rswir 5

short wave infrared pixel reflectance and rCnir 5 0.04

and rCswir 5 0.2.

The BAIM threshold value was based on the mean

and standard deviation of the index, with pixels defined

as burned whenever their index value fell within the

Table 1 Characteristics of savanna vegetation types in the Llanos (Etter 1985, 1998a)

Savanna type Physiognomy Dominant species

Aboveground

biomass

(ton DM ha�1)

Terrace savannas Dense tussock grassland

with scattered trees

Trachypogon ligularis, Andropogon bicornis,

Curatella americana

7

Flat highplain savannas Dense tussock grassland Trachypogon vestitus, Panicum spp.

Leptocoryphium lanatum

6

Rolling highplain savannas Open tussock grassland Paspalum carinatum, Trachypogon ligularis,

Bulbostylis spp.

3.5

Sandy Savannas Open grassland with

scattered trees

Bulbostylis spp., Paspalum carinatum,

Curatella Americana, Byrsonima crassifolia

3

Dune Savannas Open grassland with

scattered trees

Paspalum spp., Byrsonima crassifolia 4

Alluvial overflow plain savannas Dense tussock grassland Andropogon bicornis, Axonopus purpusii,

Leptocoryphyium lanatum

7

Aeolian overflow plain savannas Dense tussock grassland Trachypogon spp., Andropogon spp. 6.5

2016 M . R O M E R O-R U I Z et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

range of 3 SD. In a subsequent processing step, the pixel

clusters identified as burn scars and being smaller than

25 ha.

The accuracy of burned area product was done with

the use of high resolution satellite images for two

different periods: Landsat images for 2001 (456_January

9, 457_March 14, 556_January 8 and 557_February 2)

and CBERS images for 2006–2007 (183093 and 183094_

February 9, 2007, 185093 and 185094, January 1, 2006,

185095_February 03 and 187096_January 1, 2007)

(http://www.cbers.inpe.br/en/index_en.htm). Those

images are freeing available and present between 2%

and 3% of cover in clouds and shadows. This step was

based on NDVI classifications associated with super-

vised and visual interpretation, using 4, 5, 3 band

combination from which fire perimeters were derived.

A 3� 3 filter was applied to remove the ‘Salt and

pepper’ effect of automatic classification. After that a

visual classification was done to separate old and new

burned areas, taking the darkest clusters as recent burn

scars. Fire scars were extracted and compared with the

8 days MODIS burned areas classification. All these

products were spatially compared with to a hexagonal

tessellation of 5 km� 5 km. The total burned scars area

inside of every grid cell was divided by the overall of

the grid cell area to obtain the percentage of burned

scars in every cell. The accuracy for both classifications

was validated using the Kappa coefficient (k), as well as

the cross-tabulation of results.

In a subsequent step, the spatial pattern of the burn-

ing scars for each mapping season (e.g. 2000–2001,

2001–2002) was analyzed using a set of landscape

metrics: TA, number of patches (NP), mean patch area

(MPA) and patch area coefficient of variation (CV) using

the FRAGSTATS software (McGarigal et al., 2002).

Burning patterns and their relation to rainfall, ecosystemtype and land tenure

Rainfall. We used daily rainfall data from the months

December through April for the 2000–2008 study period

from four rainfall stations spread out over the study

area (Fig. 1), obtained from the Instituto de Hidrologıa,

Meteorologıa y Estudios Ambientales (IDEAM) http://

www.ideam.gov.co/. To analyze the relationship of

rainfall data of climate stations with the fire patterns

in the study area, we extracted the burned area for each

month in a radius of 50 km around each rainfall station,

to represent the local conditions while capturing

enough spatial variability. The area being flat, the aver-

age rainfall variability is not be affected by relief. We

then calculated the regression between the burned area

and total rainfall for each month.

Ecosystem types and land tenure. The study area is

heterogeneous in ecological terms and land tenure. We

therefore analyzed the burning patterns in relation to

ecosystem type and the land tenure in a GIS. The Eco-

system map was built using the savanna ecosystem

types from the map of Etter (1998b). The area of

forests, and cleared savannas of this map was upda-

ted through a supervised classification of Landsat-ETM

and CBERS satellite images. Four general land cover

types were examined: (i) Forest, (ii) Savannas, (iii) Plan-

ted pastures and annual crops, and (iv) Plantations (oil

palm and forestry). Supervised classification was done

using the ERDAS-IMAGINE 9.1 software by selecting

control areas in the form of ‘training areas’ using an

algorithm that considers the spectral characteristics of a

group of pixels assigned to each one of the classes. The

area classified as savanna was subdivided into the

savanna types of the ecosystem map.

The land tenure map was obtained combining two

sources: (i) the National Parks map acquired from

National Park Office (Ministry of Environment-Colombia,

2000); and (ii) the map of Indigenous territories from

Geographic Institute Agustin Codazzi (IGAC).

The average annual burned area for the period

2000–2008 was calculated for each ecosystem and land

tenure type.

Results

Temporal and spatial variability of fires

Fire activity in the Colombian Llanos savannas is highly

concentrated in the dry season spanning the months

December through March, with a marked peak in the

month of February (see Fig. 5). Between 2000 and 2008

an average 2.75 � 0.5 million ha were burnt every year,

equivalent to 24 � 4.2% of the Llanos savanna vegeta-

tion (Fig. 2). The Kappa coefficient of validation for the

analyzed images had a value of 0.79. In areas larger

than 1000 ha this value reached a value of 0.88, decreas-

ing for areas of o1000 ha, which gives a good spatial

Fig. 2 Annual burned areas in the Colombian Llanos savannas

and their average for the 2000–2008 periods.

S PA T I A L A N D T E M P O R A L VA R I A B I L I T Y O F F I R E S 2017

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

assessment of burns, at least for medium to large size

fires.

However, annual burning patterns exhibit high varia-

bility, with 43.4 million ha burned in 2002–2003 and

o1.9 million ha in 2004–2005 period (Fig. 3a and b). The

map of burning recurrence shows areas of high recur-

rence such as the southwest and northeast of the study

area, while other areas closer to the mountain range and

savannas with high replacement by pastures and plan-

tations show much lower fire recurrence (Fig. 3c). Dur-

ing the 2000–2008 period some 3.7 million ha (33.5%) of

savannas were never burned, while 1.95 million ha

(18%) was burned more than four times in 8 years.

A further important characteristic of fires and their

impact concerns the spatial patterns of scars and the

variability of their metrics such as total NP, the mean

Fig. 3 Burned area maps: (a) Burn scars (black) for a strong (2002–2003) fire season; (b) burn scars for a weak (2004–2005) fire season;

(c) fire recurrence for the period 2000–2008 (hotspots in circles are areas with high concentration of >5 in 8 years fire recurrence).

2018 M . R O M E R O-R U I Z et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

burned patch size and patch density. Table 2 shows a

number of average spatial metrics for the burned areas

of the period 2000–2008. The average number of an-

nually burned patches in the region was 3486 patches,

with a mean size of around 1000 ha, and a CV of 990 ha.

This gives an annual average of about one burned patch

for every 5000 ha. In general, mean patch size is well

correlated with the proportion of area burned (Fig. 4).

Relations of the burning patterns with rainfall, ecosystemtype and land tenure

The monthly average burned area and rainfall for the dry

season showed a negative relationship for all four stations

analyzed as shown in Fig. 5. The average monthly rainfall

for the dry months of December through March in the cen-

ter of the Llanos is 28.5 mm, and 13.6 mm only for January–

February. Burned area strongly decreases with monthly

rainfall values 450 mm across the whole study area (Fig. 6).

In relation to the variability in land tenure/manage-

ment and biophysical conditions, fire patterns of the

Colombian Llanos savannas show marked differences.

Replacement of savannas by pastures and plantations

drastically reduces fires, which practically disappear in

improved pastures and plantations. In savannas the

proportion of area that is burned each year varies greatly

with land tenure, being highest within National Parks

with over 28%, while in ranches this was 19% (Fig. 7a).

Variability is also highest in National Parks. During 2000–

2008, 30–32% of the area of ranches and indigenous

reserves were never burned, while for National Parks

this figure was only 16%. Overall, savanna fires are more

frequent in Indigenous reserves than in ranches. The

patterns of fire recurrence are also variable with ranches

showing the lowest recurrence, and National Park the

highest recurrence of 3 in 8 years or more (Fig. 7b).

Table 2 Spatial metrics of burned areas in the Colombian

Llanos for the period 2000–2008

Year

Percent

land area

Numbers

of scars

Average

scar

size (ha)

Area

coefficient

of variation

2000–2001 0.26 3953 952 1050.6

2001–2002 0.24 3513 925 1144.2

2002–2003 0.30 3034 1408 982.0

2003–2004 0.26 3061 1189 1122.4

2004–2005 0.16 4005 565 898.8

2005–2006 0.24 3770 899 938.7

2006–2007 0.23 3185 1019 898.2

2007–2008 0.25 3369 1048 889.0

Average 0.24 3486 1005 990.5

Fig. 4 Relationship between the total burned area and the

average burn scar size for the Colombian Llanos savannas.

Fig. 5 Monthly rainfall in the dry season and area burned within 50 km radius for climate station Las Gaviotas (see Fig. 1).

Fig. 6 Regression between the dry season monthly rainfall and

area burned within 50 km radius for four rainfall stations (see

Fig. 1) across the Colombian Llanos in the period 2000–2008.

S PA T I A L A N D T E M P O R A L VA R I A B I L I T Y O F F I R E S 2019

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

Differences in savanna ecosystem type also had a

strong impact on fire patterns, with the average annual

proportion of burned area by ecosystem varying be-

tween 3% and 38% per year. The highest figures occur in

the well-drained high-plain savannas south of the River

Meta especially in the sandy savannas (39%) (Fig. 8a).

The lowest figures occur along the foothills of the

Andes. Also, between 22% and 30% of the area of

well-drained high-plain savannas were not burned dur-

ing the 2000–2008 period. In contrast more than 80% of

the alluvial overflow plain savannas were free from fire

in the same period (Fig. 8b). The ecosystem type with

the highest fire recurrence was again the sandy high-

plain savanna type, where 230 000 ha (36%) burned in

four or more years during the study period.

Discussion

Recent global studies have confirmed that the savanna

biome is the most continuously affected by burning

with strong environmental impacts (Chuvieco et al.,

2008b; Tansey et al., 2008). According to the figures of

Van der Werf et al. (2003) and Tansey et al. (2004) our

results indicate that burned areas of the Colombian

Llanos savannas correspond to between 3% and 6% of

the globally burned herbaceous vegetation area, and

about 4% of the annually burned area in South America.

Multi-satellite data has become a major means to

monitor changes in land cover, including fires at the

global, regional and local scales (Roy et al., 2005).

Although this has made possible to map areas affected

by fires and analyze burning dynamics, the differences

in the spatial, temporal and spectral resolutions of

satellite data, as well as in the processing algorithms

often leads to large differences in results (Roy et al.,

2005). A recent study by Chuvieco et al (2008b) using 32

day composition MODIS data, calculated that 1.2% of

the Colombian territory is burned annually, which is

only about half of the 2.75 million ha we found for the

Colombian Orinoco region alone. This discrepancy can

be attributed to the algorithms used for detection, the

definition of the dry season, the geographical location of

the affected areas and the type of sensor used (Loboda &

Csiszar, 2007). The algorithms used by these studies do

not take into account factors such as the meteorological

conditions where the fire occurs, the type and structure

of the landscape, land use and cover, biomass, distribu-

tion and heterogeneity and combustion rates (Loboda &

Csiszar, 2007). Our results agree with Roy et al. (2005)

who point to the need to develop algorithms that are

sensitive to ecosystem variability and to the character-

istics of the fire regime of specific environments.

The commission and omission errors in our results of

validation may be attributed to the timing of Landsat

and CBERS image acquisition dates and the periods of

MODIS image composition. Some burned areas were

not identified by our algorithm because the fire occur-

rence was at the beginning of the period of the 8-day

Fig. 7 Burning patterns for different land tenure types for the

period 2000–2008: (a) average annual proportion burned and

standard deviation; (b) fire recurrence in 8 years.

Fig. 8 Burning recurrence for different ecosystem types:

(a) average proportion annually burned and standard deviation;

(b) fire recurrence in 8 years.

2020 M . R O M E R O-R U I Z et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

MODIS composition, while the high-resolution images

(Landsat and Cbers) were taken at the end of the 8-day

composition. The precision in large burned areas shows

a high accuracy level leading to an acceptable overall

Kappa (k) value (0.79), which compensates for the

omission errors that arise from the fact that MODIS

images do not detect small scars (o10 ha).

Temporal and spatial variability of fires

Our results show that the fire activity in the Colombian

Llanos savannas produces a changing seasonal mosaic

of burned and unburned vegetation patches, such as

found in other savannas of the world (Laris, 2002).

However, we found that the fire recurrence has a strong

spatial pattern, with some areas showing consistently

high burning rates while other areas exhibiting consis-

tently low fire recurrence. As observed by Chuvieco

et al. (2008a) for other high fire frequency areas in the

tropics, the Colombian Llanos savannas are character-

ized by low accessibility, mostly extensive low invest-

ment uses, low-density population and dominance of

grass cover, and fires are predominantly of high density,

low duration and high spatial variability.

Compared with the average 8–10 years of fire return

time for the tropics and subtropics that was estimated

by Van der Werf et al. (2003), the savannas in the

Colombian Llanos burn twice as often. Compared

with the savannas of northern Brazil the average

4 years fire return time in the Colombian Llanos

savannas is almost twice the 2.5 years reported by

Barbosa & Fearnside (2005b). However, if the compar-

ison is made with only the well-drained sandy savan-

nas, the figures of the Llanos savannas and Northern

Brazil turn to be very similar.

Van der Werf et al. (2003) give an average burn size

figure for tropical and subtropical grassland savannas

(with o5% trees) of 7000 ha (median 5000 ha), which is

about seven times the average burned patch size we

found in our study. We found that except for the 2004–

2005 fire season which was a low burn year, the

interannual average burned patch size was very con-

stant. Several studies have shown that fire intensity

changes throughout the dry season, with early fires

burning vegetation more superficially and leaving

a more fragmented pattern in the landscape, while

with the dry season progressing fires become more

intense, more continuous and larger (Louppe et al.,

1995; Russell-Smith et al., 1997). We found that fire

areas tend to be smaller at the beginning of the dry

season, becoming much larger towards the middle of

the dry season and again smaller at the end of the

season.

Relation of the burning patterns with rainfall, land tenureand ecosystem type

Fire frequency and pattern is one of the dominant

factors defining heterogeneity of savanna environ-

ments, and man has been identified as the main cause

of fire production in savanna ecosystems (Hudak et al.,

2004). This phenomenon has significant ecological im-

plications in terms of species composition and ecosys-

tem functionality. In a broad context it has been shown

that the extension, frequency and severity of fires,

depends factors such as climate, topography and soil,

is much influenced by the structure and composition of

the vegetation (Shvidenko & Nilsson, 2000). Our results

show a marked realtionship with ecosystem types (ve-

getation, soil, hydrology), confirming the above state-

ment. Structure and composition of vegetation tend to

determine fire activity in the long-term, while changes

in climatic factors have a strong relation with fire

activity in the short-term.

Temperature and precipitation are the two most

highly correlated climatic factors with the levels of

presence and hazard fires (Finney, 1998). Barbosa &

Fearnside (2005b) found a high correlation between

the area burned and rainfall in the Roraima savannas

of the Brazilian Amazon. These authors also discuss the

importance of precipitation and the different types of

ecosystems in the occurrence of a fire. The results of our

study show the same tends, where a sharp decrease in

fires occur in months where rainfall exceeds 50 mm.

Fire has been recognized as an indicator of anthro-

pogenic activity to help identify critical areas of defor-

estation and changes in land use (Malingreau &

Gregoire, 1996; Ehrlich et al., 1997). Barbosa & Fearnside

(2005b) found that the majority of the fires in the

savannas of northern Brazil are started by humans,

but that the spread of fires is independent of human

presence. We found that fires vary in terms of their

frequency and pattern for different savannas types in

the Colombian Llanos. The unexpectedly overall pro-

portion and large size of fires in uninhabited National

Park, can be explained by the periodic movement of

indigenous people living in Reservations around it that

move across the conservation area for communication

and hunting (Correa et al., 2006). According to our

results the fire frequency and burned proportion is

somewhat higher in indigenous reserves than in cattle

ranches. The constant burning in ranches is associated

with the need to deliver fresh regrowth for livestock

due to the low palatability and forage quality of grasses

in the native savannas and the control of woody plants.

In indigenous lands however, fire is more associated

to hunting activities and to the savannas ‘cleaning’ from

ticks and other. Studies concerning the effects of land

S PA T I A L A N D T E M P O R A L VA R I A B I L I T Y O F F I R E S 2021

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

tenure and land use differences on fire patterns in

savanna environments are very scarce. In other ecosys-

tem contexts such as the Amazon basin indigenous

reserves and national parks have contributed to the

reduction of deforestation rates associated to burning,

with fires four (indigenous reserves) to nine (parks)

times higher outside than inside them (Nepstad et al.,

2006).

In the Colombian savannas, transformation has been

relatively slow. However, during the last few years

increasingly large areas of savannas are being converted

into improved pastures, plantations (oil palm and tim-

ber) and annual crops, where fire activity is being

reduced or eliminated from these landscapes and their

immediate surroundings. The changes in the magni-

tude and frequency of disturbance variables such as

fires and biotic composition that normally drive the

state and cycles of these ecosystems is expected to have

large impacts on the local and regional ecological func-

tioning.

An unexplained finding of this study that needs

further study is the unexpectedly high proportion of

savannas not directly related with intensive land uses

that were not burned during the 2000–2008 period,

while some areas had a consistently high frequency.

This should be expected to have important conse-

quences for the local savanna vegetation dynamics

(Campos & Nepstad, 2006).

Conclusions

Currently, change modelling is demanding increasingly

accurate and spatially explicit regional and local data

and knowledge on land cover dynamics and change

processes such as deforestation, burning, irrigation,

agricultural intensification and urbanization. This is

the more urgent in dynamically changing landscapes

of developing countries with fast population growth

and industrialization processes. Our study contributes

the first complete regionally accurate mapping for al-

most a decade of fire activity in the Colombian Llanos

savannas. Specifically, we show how fire occurrence

(pattern, frequency and extent) vary at the sub-regional

level due to ecosystem, land tenure and rainfall, essen-

tial to the understanding of fire patterns at the regional

scale. We have shown that the increasing trend of the

savanna clearing and intensification process is changing

the fire dynamics and burning patterns. However, these

land use and land cover changes need also be analyzed

in a broader context of the impacts on ecological com-

ponents and processes, including biodiversity, soils,

water and fauna, such as termites and ants. Currently,

Savannas are considered the most important biome in

terms of fire events in South America. Between 2000 and

2008 the savannas of northern South America contrib-

uted with 25% of the overall burned areas of the

continent (Chuvieco et al., 2008b). Of this, the Colom-

bian Llanos savannas contributed with 65% whereas

Venezuelan Llanos savannas with 35%. Burning in this

region constitutes a globally significant 3–6% of herb-

aceous vegetation, an annual average of 2.75 million ha.

Acknowledgements

The Alban Program of the European Union, the Colfuturo Fund(Colombia) and The Nature Conservancy – Northern AndesProgram provided funding to Milton Romero. The authors wouldlike to thank Juan Carlos Berrio for providing support in all thisprocess and Javier Medina for his comments to this document.

References

Andreae MO (1992) Biomass burning: its history, use and distribution and

its impact on environmental quality and global climate. In: Global

Biomass Burning, Atmospheric, Climatic, and Biospheric Implications (ed.

Levine J), pp. 3–21. MIT press, Cambridge, MA.

Asner GP, Elmore AJ, Olander LP, Martin RE, Harris AT (2004) Grazing

systems, ecosystem responses, and global change. Annual Review of

Environment and Resources, 29, 261–299.

Barbosa RI, Fearnside PM (2005a) Above-ground biomass and the fate of

carbon after burning in the savannas of Roraima, Brazilian Amazonia.

Forest Ecology and Management, 216, 295–316.

Barbosa RI, Fearnside PM (2005b) Fire frequency and area burned in the

Roraima savannas of Brazilian Amazonia. Forest Ecology and Manage-

ment, 204, 371–384.

Brannstrom C, Jepson W, Filippi AM, Redo D, Xu Z, Ganesh S (2008) Land

change in the Brazilian Savanna (Cerrado), 1986–2002: comparative ana-

lysis and implications for land-use policy. Land Use Policy, 25, 579–595.

Campos M, Nepstad D (2006) Smallholders, the Amazon’s new conserva-

tionists. Conservation Biology, 20, 1553–1556.

Cochrane M (2002) Wildfire Tropical Forest Fires in Latin America and the

Caribbean. Prevention, Assessment and Early Warning. United Nations

Environment Programme – UNEP, Washington.

Correa HD, Ruiz SL, Arevalo LM (2006) Plan de accion en Biodiversidad de la

cuenca del Orinoco – Colombia (2005–2015): Propuesta tecnica. Corporino-

quia, Cormacarena, Instituto Humbodt, Unitropico, Fundacion Oma-

cha, Fundacion Horizonte Verde, Universidad Javeriana, Unillanos,

WWF, Colombia GTZ, Bogota.

Coutinho LM (1990) Fire in the ecology of the Brazilian cerrado. In: Fire in

the Tropical Biota – Ecosystem Processes and Global Challenges (ed. Gold-

ammer JG), pp. 82–105. Springer-Verlag, Berlin.

Chuvieco E, Giglio L, Justice C (2008a) Global characterization of fire

activity: toward defining fire regimes from Earth observation data.

Global Change Biology, 14, 1488–1502.

Chuvieco ES, Opazo W, Sione H et al. (2008b) Global burned-land

estimation in Latin America using modis composite data. Ecological

Applications, 18, 64–79.

Delmas R, Lacaux JP, Brocard D (1995) Determination of biomass burning

emission factors: methods and results. Environmental Monitoring and

Assessment, 38, 181–204.

Dwyer E, Pereira JMC, Gregoire J-M, DaCamara CC (2000) Characteriza-

tion of the spatio-temporal patterns of global fire activity using satellite

imagery for the period April 1992 to March 1993. Journal of Biogeogra-

phy, 27, 57–69.

2022 M . R O M E R O-R U I Z et al.

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023

Ehrlich D, Lambin EF, Malingreau JP (1997) Biomass burning and broad-

scale land cover changes in western Africa. Remote Sensing of Environ-

ment, 61, 201–209.

Eiten G (1971) The Cerrado vegetation of Brazil. Botanical Review, 38,

201–341.

Etter A (1985) A Landscape Ecological Approach for Grazing Development: A

Case in the Colombian Llanos Orientales, pp. 187, Department of Land and

Rural Surveys. ITC, Enschede, The Netherlands.

Etter A (1998a) Ecosistemas de Sabanas. In: Informe Nacional sobre el Estado

de la Biodiversidad Colombia 1997 (eds Chaves ME, Arango N), pp. 76–95.

IAvH y PNUD, Bogota.

Etter A (1998b) Mapa general de ecosistemas de Colombia – 1:2 000 000.

(General ecosystem map of Colombia). In: Informe Nacional sobre el

Estado de la Biodiversidad en Colombia – 1997 (National Biodiversity Report

– 1997) (eds Chaves ME, Arango N), Instituto Alexander von Hum-

boldt, Bogota.

Finney M (1998) FARSITE: Fire Area Simulator – Model, Development and

Evaluation. Department of Agriculture, Forest Service, United States of

America.

Foley JA, DeFries R, Asner GP et al. (2005) Global consequences of land

use. Science, j309, 570–574.

Grace J, Jose JS, Meir P, Miranda HS, Montes RA (2006) Productivity and

carbon fluxes of tropical savannas. Journal of Biogeography, 33, 387–400.

Hall DO, Scurlock JMO (1991) Climate change and productivity of natural

grasslands. Annals of Botany, 67, 49–55.

Hao WM, Liu M-H (1994) Spatial and temporal distribution of tropical

biomass burning. Global Biogeochemical Cycles, 8, 495–504.

Hoffmann WA, Schroeder W, Jackson RB (2002) Positive feedbacks of fire,

climate, and vegetation and the conversion of tropical savanna. Geo-

physical Research Letters, 29, 2052.

Hudak A, Fairbanks D, Brockett B (2004) Trends in fire patterns in a

southern African savanna under alternative land use practices. Agri-

culture, Ecosystems and Environment, 101, 307–325.

Instituto Geogr a fico Agustin Codazzi – IGAC (Geographic Institute

Agustin Codazzi) (1983) Mapa general de suelos de Colombia – 1:1,500,000

(General soils map of Colombia). IGAC, Bogota.

Laris P (2002) Burning the seasonal mosaic: preventive burning strategies

in the wooded savanna of southern Mali. Human Ecology, 30, 155–186.

Loboda TV, Csiszar IA (2007) Reconstruction of fire spread within wild-

land fire events in Northern Eurasia from the MODIS active fire

product. Global and Planetary Change, 56, 258–273.

Louppe D, Oattara N, Coulibaly A (1995) The effects of brush fire on

vegetation: the Aubreville fire plots after 60 years. Commonwealth

Forestry Review, 74, 288–291.

Malingreau JP, Gregoire JM (1996) Developing a global vegetation fire

monitoring system for global change studies: a framework. In: Biomass

Burning and Global Change: Remote Sensing and Modeling of Biomass

Burning in the Boreal Forest (ed. Levine JS), pp. 14–24. MIT Press,

Cambridge, MA.

Martin MP, Gomez I, Chuvieco E (2005). Performance of a burned-area index

(BAIM) for mapping Mediterranean burned scars from MODIS data. Pro-

ceedings of the 5th International Workshop of Remote Sensing and

GIS applications to Forest Fire Management: Fire Effects Assess-

ment, Univesidad de Zaragoza, GOFC-GOLD, EARSel. pp. 193–

198.

McGarigal K, Cushman SA, Neel MC, Ene E (2002). FRAGSTATS: Spatial

pattern analysis program for categorical maps. Computer software pro-

gram produced by the authors at the University of Massachusetts,

Amherst. Available at : www.umass.edu/landeco/research/fragstats/

fragstats.html/ (accessed 23 May 2009).

Ministry of Environment-Colombia (2000). Mapas de las areas protegidas de

Colombia (1:1,500,000). Bogota.

Mouillot F, Narasimha A, Balkanski Y, Lamarque J, Field CB (2006) Global

carbon emissions from biomass burning in the 20th century. Geophysical

Research Letters, 33, L01801, doi: 10.1029/2005GL024707.

Nepstad D, Schwartzman S, Bamberger B et al. (2006) Inhibition of

Amazon Deforestation and Fire by Parks and Indigenous Lands.

Conservation Biology, 20, 65–73.

Rausch JM (1994) Una frontera de la sabana tropical los Llanos de Colombia

1531–1831. Banco de la Republica, Bogota.

Roy DP, Jin Y, Lewis PE, Justice CO (2005) Prototyping a global algorithm

for systematic fire-affected area mapping using MODIS time series

data. Remote Sensing of Environment, 97, 137–162.

Russell-Smith J, Ryan PG, Durieu R (1997) A Landsat MSS-derived fire

history of Kakadu National Park monsoonal northern Australia 1980–

94. Seasonal extend, frequency and patchiness. Journal of Applied

Ecology, 34, 748–766.

Sankaran M, Hanan NP, Scholes RJ et al. (2005) Determinants of woody

cover in African savannas. Nature, 438, 846–849.

Scurlock JMO, Hall DO (1998) The global carbon sink: a grassland

perspective. Global Change Biology, 4, 229–233.

Shvidenko AZ, Nilsson S (2000) Extent, distribution and ecological role of

fire in Russian forest. In: Fire Climate Change and Carbon Cycling in the

Boreal Forest (eds Kasiscke E, Sticks B), Springer, New York.

Tansey K, Gregoire J-M, Binaghi E et al. (2004) A global inventory of

burned areas at 1 km resolution for the year 2000 derived from spot

vegetation data. Climatic Change, 67, 345–377.

Tansey K, Gregoire J-M, Defourny P, Leigh R, Pekel J-F, van Bogaert E,

Bartholome E (2008) A new, global, multi-annual (2000–2007) burnt

area product at 1 km resolution. Geophysical Research Letters, 35, L01401,

doi: 10.1029/2007GL031567.

Van Der Werf GR, Randerson JT, Collatz GJ, Giglio L (2003) Carbon

emissions from fires in tropical and subtropical ecosystems. Global

Change Biology, 9, 547–562.

Wardell DA, Nielsen TT, Rasmussen K, Mbon C (2004) Fire history, fire

regimes and fire management in West Africa: on overview. In: Wildland

Fire Management Handbook for SubSaharan Africa (eds Goldammer JG, de

Ronde C), pp. 350–381. Global Fire Management Center, Frieburg/

Oneworld.

Wassmann R, Vlek PLG (2004) Mitigating greenhouse gas emissions from

tropical agriculture: scope and research priorities. Environment, Devel-

opment and Sustainability, 6, 1–9.

S PA T I A L A N D T E M P O R A L VA R I A B I L I T Y O F F I R E S 2023

r 2009 Blackwell Publishing Ltd, Global Change Biology, 16, 2013–2023