Journal of Geophysical Research: Atmospheres

Greenhouse gas emissions from shifting cultivation in the tropics, including uncertainty and sensitivity analysis

Authors


Abstract

[1] Annual emissions of CO2, CH4, CO, N2O, and NOx from biomass burning in shifting cultivation systems in tropical Asia, Africa, and America were estimated at national and continental levels as the product of area burned, aboveground biomass, combustion completeness, and emission factor. The total area of shifting cultivation in each country was derived from the Global Land Cover 2000 map, while the area cleared and burned annually was obtained by multiplying the total area by the rotation cycle of shifting cultivation, calculated using cropping and fallow lengths reported in the literature. Aboveground biomass accumulation was estimated as a function of the duration and mean temperature of the growing season, soil texture type, and length of the fallow period. The uncertainty associated with each model variable was estimated, and an uncertainty and sensitivity analysis of greenhouse gas estimates was performed with Monte Carlo and variance decomposition techniques. Our results reveal large uncertainty in emission estimates for all five gases. In the case of CO2, mean (standard deviation) emissions from shifting cultivation in Asia, Africa, and America were estimated at 241 (132), 205 (139), and 295 (197) Tg yr−1, respectively. Combustion completeness and emission factors were the model inputs that contributed the most to the uncertainty of estimates. Our mean estimates are lower than the literature values for atmospheric emission from biomass burning in shifting cultivation systems. Only mean values could be compared since other studies do not provide any measure of uncertainty.

1. Introduction

[2] The burning of secondary vegetation in shifting cultivation releases CO2 and other trace gases to the atmosphere. If the shifting cultivation system is in equilibrium (as for other natural system in a steady state), carbon dioxide released by fire will be reincorporated into the secondary vegetation biomass regrowing in the fallow areas that make up the system and will not therefore contribute to variation in the atmospheric CO2 concentration on longer time scales, while the other trace gas emissions will represent a net addition to the atmosphere [Fearnside, 2000; Lehsten et al., 2009]. However, shifting cultivation systems are rarely in equilibrium thus resulting in net emission of CO2 and other trace gases [Fearnside, 2000]. The goal of the present study is to use available information about the spatial distribution and characterization of shifting cultivation in the tropics to derive improved estimates of greenhouse gas emissions from biomass burning in shifting cultivation systems, at national and continental level (tropical Asia, Africa and America). Estimates are given for carbon dioxide (CO2), methane (CH4), carbon monoxide (CO), nitrous oxide (N2O) and nitrogen oxides (NOx).

[3] Shifting cultivation has been practiced for thousands of years in forests around the world, especially in the tropics, where it provides livelihood for rural populations [Pedroso et al., 2008]. It was estimated that, in 1980, 500 million people may have been dependent upon shifting cultivation, out of a total agricultural population of 1200 million [Food and Agriculture Organization of the United Nations and United Nations Environment Programme (FAO/UNEP), 1982]. Ruthenberg [1976, p. 28] defined shifting cultivation as an “agricultural system that involves an alternation between cropping for a few years on selected and cleared plots and a lengthy period when the soil is rested.” The main rationale for shifting cultivation is the effective use of the existing soil fertility. In the most common system, vegetation is cut, allowed to dry for some time and then is burned, raising the level of nutrients just before the preparation of soil for planting. In long fallow shifting cultivation systems, after 2–4 years of cultivation the land is abandoned and there is a regrowth of the natural vegetation for 1–3 decades. Fallow vegetation, especially when consisting of forest (secondary forest) or bush, reduces leaching and stores nutrients that will be made available for crops by fire in the next cropping cycle. Shifting cultivation is usually practiced without manure or inorganic fertilizers and allows relatively high yields in poor soils [Ruthenberg, 1976]. Variation on human population density is one of the main causes of changes on shifting cultivation systems. If the human population were constant, very little primary forest would be cleared because secondary forest accumulate sufficient nutrients during the fallow period and are less difficult to clear [Gleave, 1996]. Increasing population, with its need for a larger area under cultivation, leads to the clearing of new areas of primary forest or to a decrease in the fallow period [Hall et al., 1985]. Shifting cultivation practices are adapted to local, social, and environmental characteristics and there is a large number of shifting cultivation systems described in the literature. A characterization of the diversity of shifting cultivation systems in the tropics can be found in the reviews of Warner [1991] and Thrupp et al. [1997].

[4] Emissions from land use and land cover change are the most uncertain component of the global carbon cycle and estimates vary greatly and are difficult to compare due to differences in data sources, assumptions, and methods [Ramankutty et al., 2007]. During the last decade several estimates of carbon emissions from land cover change have emerged [Fearnside, 2000; DeFries et al., 2002; Houghton, 2003; Achard et al., 2004]. These studies provided some measure of error or variation of estimates, but they did not follow a proper uncertainty analysis. Fearnside [2000] used the lower and higher values of emission factors appearing in the literature to establish a low trace gas scenario and a high trace gas scenario, respectively. The error associated to DeFries et al. [2002] estimates were derived from sensitivity to initial biomass values at ±25% of the value given by the bookkeeping model used, and the ranges derived from the correction factors between percent tree cover estimates obtained with low- and high-resolution satellite data. In the study of Houghton [2003], errors were considered to be approximately ±50% for tropical regions and were estimated based on the author's expert opinion. Achard et al. [2004] used deforestation rate estimates based on a statistical sampling (standard errors of pantropical change estimates ranging from 13% to 15%) and tested the sensitivity of their model to the biomass parameter by adding or subtracting 20% to or from the forest biomass figures to obtain regional estimates of minimum and maximum biomass, respectively. The studies also differed in the consideration of land use practices. Only Fearnside [2000] and Houghton [2003] considered shifting cultivation but they did not disaggregate emission estimates at a continental level.

[5] In global estimates of biomass burning emissions from shifting cultivation, the area of concern is obtained by multiplying the population engaged in shifting cultivation by the average land requirement per capita for this type of agriculture [Seiler and Crutzen, 1980], or from the Food and Agriculture Organization (FAO) statistics [Fearnside, 2000; Houghton, 2003; Lauk and Erb, 2009]. Population engaged in shifting cultivation is roughly estimated as a fraction of rural population or obtained also from data published by FAO. Estimates are derived at global or continental level since the FAO Forest Resources Assessment data on shifting cultivation are not disaggregated by country. Moreover, the most recent data provided by FAO on the global area under shifting cultivation are from 1980 [FAO/UNEP, 1982]. Another important factor is the length of the fallow period, because it affects the amount of biomass that will be present at the time the fallow secondary vegetation is cut. Fearnside [2000] assumed a single fallow period of 12.5 and 4 years in long and short fallow shifting cultivation, respectively, to estimate global emissions of greenhouse gases. Houghton [2003] and Houghton and Hackler [2006] assumed different fallow periods for Asia, Africa and America, but did not disaggregate at country level. In the present study, the area of shifting cultivation is obtained from land cover products derived from satellite data and the cropping and fallow periods are defined for each country using available literature sources, which allow the estimation of emissions at the national level. The present work relies on a more thorough and sophisticated uncertainty and sensitivity analysis approach than those previously used for global pyrogenic emissions from shifting cultivation.

2. Data

2.1. Global Land Cover 2000

[6] The Global Land Cover 2000 (GLC2000) data set consists of a global land cover map explicitly linked to more thematically detailed regional maps [Bartholomé and Belward, 2005]. Teams of regional experts mapped each continent independently. The regional classes were defined using the Land Cover Classification System (LCCS), developed by FAO, which allowed the aggregation of the regionally detailed classes into a simpler, global legend. The GLC2000 map was derived with 14 months of global daily images acquired by the 1 km spatial resolution VEGETATION instrument on board SPOT-4 satellite between 1 November 1999 and 31 December 2000.

[7] The GLC2000 product was selected for this study because is the only available global land cover with regionally optimized maps whose legends include classes explicitly concerning shifting cultivation. Four regional land cover maps were used in order to cover the tropical regions of Asia, Africa and America: “South and South East Asia,” “China,” “Africa,” and “South America.” Table 1 provides a brief description of the GLC2000 regional land cover classes selected to characterize the spatial distribution of shifting cultivation in the tropics.

Table 1. GLC2000 Regional Maps Used in This Study, With a Brief Description of the Selected Land Cover Classes
Regional MapClass NameDescriptiona
  • a

    Extracted from the GLC2000 report [Fritz et al., 2003].

  • b

    In China, this class was mapped only for Yunnan Province.

South and Southeast AsiaEvergreen shrubland and regrowth/abandoned shifting cultivation/extensive shifting cultivation (class 9)“Shrub cover and regrowth of predominant evergreen appearance during the dry season including (1) shrub and regrowth patterns resulting from abandoned or less intense shifting cultivation, (2) tree cover of low height or sparse canopy, e.g. resulting from over-logging, (3) bamboo invaded areas, (4) forest remnants, not mapped at coarse spatial resolution.”
 Mosaic of cropping, regrowth, or other natural vegetation/intensive shifting cultivation (class 16)“Mainly intensive shifting cultivation, comprising recent burns, recent and last year's cropping, abandoned fields with regrowth, and forest remnants usually on steep and less accessible slopes.”
ChinaMosaic of croppingb (class 23)“In the tropical part of the Yunnan province the broadleaved/needleleaved forest is heavily fragmented due to shifting cultivation under 1500 meters.”
AfricaMosaic forest/croplands (class 7)“A major feature of the Central African forest biome is the presence of ribbons of secondary forest formations along the road network, either old or recent. These formations correspond to a pattern of land management—the former ‘paysannats,’ which since colonial times follow the road network. The vegetation found here is formed by a complex of secondary regrowth, fallow, home gardens, food crops and village plantations.”
South AmericaMosaic of agriculture and degraded forest formations (class 52)“This is a common class across South America and corresponds to shifting cultivations, agroforestry, fragmented forests and secondary forest and rural complex.”

2.2. Map of the Ecosystems of Central America

[8] Contrary to the other regions, the GLC2000 map for North and Central America does not include an agriculture and forest mosaic class or a shifting cultivation class. However, shifting cultivation is a common agriculture system in Central America, spreading from southern Mexico to Panama [e.g., Pulido and Caballero, 2006; Tschakert et al., 2007]. In order to fill this gap in the GLC2000 information, the Map of the Ecosystems of Central America [Vreugdenhil et al., 2002] was used to extract the spatial distribution of shifting cultivation in Central America. The primary objective of that project was to map and describe the ecosystems of Mesoamerica (Belize, Guatemala, El Salvador, Honduras, Nicaragua, Costa Rica, and Panama) using a comprehensive, regionally endorsed, classification system. The map and associated data represent a 1997–99 baseline for the region and were derived adapting the UNESCO physiognomic–ecological classification system to Central America.

[9] Shifting cultivation is the main cause for fires in lowland broadleaf forests of Central America and fires take place in areas under agriculture and in areas bordering remaining natural vegetation [Vreugdenhil et al., 2002]. Shifting cultivation is also the main type of agriculture practice in the highlands of Honduras, where pine forests predominate [Williams, 1964]. In this region, most vegetation disturbance is the direct or indirect effect of anthropogenic action, mainly shifting cultivation. Therefore, we selected the classes broadleaf and needle forest labeled as “intervened” in the Map of the Ecosystems of Central America as the best available information about the spatial distribution of shifting cultivation in the region.

2.3. Soil Texture

[10] A global data set of soil types is available at 1° spatial resolution [Zobler, 1986]. Soil types are described at two levels of detail, including 106 types based on Zobler's assessment of FAO Soil Units [Zobler, 1986] and an aggregated list of 27 soil types. A more extensive version of these data, including soil surface texture, surface slope, and other phase information, is also available in the work of Staub and Rosenzweig [1987]. We used the later version to obtain a data set of soil texture (1° spatial resolution), required by the method used in this study to estimate biomass accumulation in fallow areas. According to Staub and Rosenzweig [1987], each element of the data set represents the near-surface texture (upper 30 cm) of the dominant soil unit in a 1° square cell of the Earth's surface. The texture classes of this data set and our aggregation into sandy versus nonsandy texture soils are shown in Table 2. The criterion used, following closely that of Zarin et al. [2001], was that coarse and medium texture soils were assigned to the sandy soil category, given that fine texture components were not present. The presence of fine texture on a soil texture class assigned it to the nonsandy texture class.

Table 2. Soil Texture Classes as Indicated by Staub and Rosenzweig [1987] and Aggregation Used in This Study
Soil Texture ClassAggregation to Sandy Versus Nonsandy Soil
CoarseSandy
MediumSandy
FineNonsandy
Coarse-mediumSandy
Coarse-fineNonsandy
Medium-fineNonsandy
Coarse-medium-fineNonsandy
OrganicNonsandy
Land-iceland-ice
WaterWater

2.4. Climate Data

[11] The method used in this study to estimate biomass accumulation in fallow areas required an estimate of the duration of the growing season and of the growing season temperature. Therefore, to estimate these variables we used basic information on land surface state and flux parameters generated by the Global Land Data Assimilation System (GLDAS) (Goddard Earth Sciences (GES) Data and Information Services Center (DISC), National Aeronautics and Space Administration (NASA)) [Rodell et al., 2004]. GLDAS uses multiple land surface models (not coupled to the atmosphere), integrates a high number of observation based data, and executes globally at spatial resolutions of 0.25° and 1° [Kumar et al., 2006]. Currently, GLDAS uses four land surface models (LSMs): Mosaic, Noah, Community Land Model (CLM), and Variable Infiltration Capacity (VIC). We used the 1° spatial resolution data set, which has a higher temporal coverage (1979–2008) than the 0.25° data set (2000 to the present). The temporal resolution of this data set is either 3-hourly or monthly. We used the monthly resolution, which is just a temporal averaging of the 3-hourly products, as we were interested in estimating the growth season on a monthly basis. The parameters of interest were: mean surface temperature (K), total evapotranspiration (kg m−2 s−1), rainfall rate (kg m−2 s−1), and snowfall rate (kg m−2 s−1). VIC outputs do not include mean surface temperature because the model was run in “water balance mode”; in this mode, the model does not solve the full energy balance, thus not all variables are recorded. Total precipitation is the sum of rainfall and snowfall, but over our regions of interest total precipitation is mostly rainfall. The conversion of rates (rainfall, snowfall, evapotranspiration) to total monthly values was done using the equation

equation image

where Xi is the value of variable X in the ith month, and Xri is the variable X rate (i.e., kg m−2 s−1) in the ith month. The three constants are for the number of seconds in a 3 h period (10800), the number of 3 h periods in 1 day (8), and the number of days in 1 month (30). Obviously, this transformation was not carried out for the mean surface temperature.

3. Methods

[12] Estimates of trace gas and aerosol emissions from biomass burning require data on burned area, amount of biomass per unit area, combustion completeness (the proportion of biomass consumed during the fire) and emission factors (the amount of atmospheric species released per unit mass of biomass burned). The classical equation of Seiler and Crutzen [1980] for pyrogenic emissions estimation is

equation image

where Ei is the emission of species i (g), A is the burned area (ha), AGB is the aboveground biomass (Mg ha−1), CC is the combustion completeness (fraction), and EFi is the specific emission factor (g kg−1). In our study, equation (2) is adapted to the special case of pyrogenic gas emission from shifting cultivation, given that there are no direct estimates of the area of forest cleared and burned annually in this type of agricultural system, and that the biomass present at the time of clearing depends on how many years natural vegetation is allowed to recover, i.e., the length of the fallow period. Estimation of the area cleared annually by shifting cultivation is described in section 3.2. Section 3.3 describes the equations used to estimate aboveground biomass. Area and biomass estimates depend on data about cropping and fallow periods, which are given in section 3.1.

3.1. Cropping and Fallow Periods

[13] The relation between cropping period and fallow period is a measure of the intensity of shifting cultivation. The shorter the fallow period, the more stationary does farming become [Ruthenberg, 1976]. Within the continuum from long fallow rotation to permanent cultivation, length of fallow is highly variable. It can differ from village to village and among households of a particular village, but can also vary among the different plots of the same household [Gleave, 1996]. It is problematic to find reliable and consistent data for large regions, like the three regions covered in this study, for an agricultural system that is considered the most complex and multifaceted form of agriculture in the world [Thrupp et al., 1997]. We conducted an extensive literature review about shifting cultivation systems in the tropics. Tables 35 show the cropping and fallow periods reported in the literature for the countries of tropical Asia, Africa and America, respectively. These data on cropping and fallow lengths were used in two different ways to estimate emissions from shifting cultivation. At national level (section 4.2), the minimum and maximum fallow length reported for each country was used to set two scenarios for the emissions estimation: short fallow and long fallow; as most studies report cropping periods of 1, 2, or 3 years, we used 2 years as an average length for the cropping period. Lauk and Erb [2009] also assumed an average length of 2 years for the cropping period. At continental level (section 4.3), the full range of data on cropping and fallow periods available (Tables 35) was used to produce total emissions estimates of each gas for each continent and to perform the uncertainty and sensitivity analysis. For some countries, shifting cultivation was mapped by the land cover products but no information was available about the cropping and fallow periods. In those cases we assumed the values reported for a neighboring country.

Table 3. Cropping and Fallow Periods in Tropical Asian Countries as Reported in the Literature
CountryCropping Period (years)Fallow Period (years)References
  • a

    Not available.

Bhutan13, 5, 7, 8,Upadhyay [1995]
Cambodia2, 3, 53, 10, 25Baird and Dearden [2003], Maxwell [2004]
China1, 2, 53, 5, 15, 30Huijun et al. [2002], Liang et al. [2009], Sturgeon [2000]
India13, 4, 5, 12Gupta [2000], Prasad et al. [2001a, 2003]
Indonesia1, 24, 5, 8, 13, 20, 27Christanty et al. [1997], Lawrence et al. [1998]
Laos1, 25, 6, 7, 18Ducourtieux et al. [2006], Seidenberg et al. [2003]
Malaysia15, 10, 15Nielsen et al. [2006], Ohtsuka [1999]
Myanmarnaa13, 15, 18Shinya et al. [2005]
Papua New Guinea1, 33, 4, 5, 10, 14Hartemink [2001], Manner [1981]
Philippines1, 2naaLawrence [1997]
Thailand15, 8, 9, 10, 15Delang [2006], Rerkasem and Rerkasem [1994]
Vietnam12, 4, 10, 15Folving and Christensen [2007], Sam [1994]
Table 4. Cropping and Fallow Periods in Tropical African Countries as Reported in the Literature
CountryCropping Period (years)Fallow Period (years)References
  • a

    Not available.

Beninnaa12, 15Versteeg et al. [1998]
Cameroon33, 10Brown [2006]
Côte d'Ivoire6, 1010, 20Bassett [1988]
D. R. Congo1, 3>20Makana and Thomas [2006]
Ghana2, 32, 4Amanor and Pabi [2007], Braimoh [2009]
Guinea18, 17Sirois et al. [1998]
Guinea-Bissau25, 6Temudo [1998]
Madagascar13, 4Styger et al. [2007]
Mozambique1, 4, 82, 3, 7, 20, 25Silva et al. [2009], Williams et al. [2008]
Nigeria1, 51, 2, 3Aweto and Iyanda [2003], Tian et al. [2005]
Sierra Leone1, 23, 6, 7, 9, 10, 14Gleave [1996], Nyerges [1989]
Ugandanaa1Carswell [2002]
Zambia2, 5, 6, 7, 10, 159, 26Murao [2005], Oyama and Kondo [2007], Stromgaard [1989]
Table 5. Cropping and Fallow Periods in Tropical American Countries as Reported in the Literature
CountryCropping Period (years)Fallow Period (years)References
  • a

    Not available.

Belize1, 31, 2, 3, 10Kupfer et al. [2004]
Bolivia3, 515, 20Kennard [2002]
Brazil1, 2, 3, 6, 82, 3, 4, 5, 9, 15, 20Aboim et al. [2008], Lewis et al. [2002], Metzger [2002], Porro [2005], Silva-Forsberg and Fearnside [1997], Siminski and Fantini [2007]
Colombia3, 5naaBarrios et al. [2005]
Costa Rica1, 2, 3naaGuggenberger and Zech [1999]
Guatemalanaa2, 7Renner et al. [2006]
Honduras14, 25House [1997]
Mexico1, 2, 3, 56, 7, 10Lambert [1996], Pulido and Caballero [2006]
Nicaragua2, 3naaSmith [2001]
Panama2, 32, 31Tschakert et al. [2007]
Paraguay310, 15Kammesheidt [1998]
Peru1, 3, 82, 5, 6, 7, 20Coomes et al. [2000], Staver [1989], White et al. [2005]
Venezuela2, 4<15Hernández et al. [2003]

3.2. Area of Shifting Cultivation

[14] The area of shifting cultivation was derived from the Global Land Cover 2000 map for Asia, Africa and South America and from the Map of the Ecosystems of Central America. Both land cover maps were produced from satellite imagery. Satellite sensors provide detailed information regarding land cover and the spatial distribution and extent of land use changes. However, there are temporal and spatial limitations related with land cover mapping in regions where shifting cultivation is the dominant system of agricultural production. Shifting cultivation systems are highly dynamic, changing on an annual basis. During each dry season, new areas of primary or secondary forest are selected and cleared by the households that usually have several plots in different stages of cultivation [Coomes et al., 2000]. Spatially, it is an agricultural system that leads to a complex mosaic of primary or mature forest, cultivation fields and fallow vegetation [Fox et al., 2000; Metzger, 2003; Brown, 2006], which may consist of bush or secondary forest.

[15] The data from the GLC2000 maps and from the Map of the Ecosystems of Central America were reprojected into the Albers Equal Area Conic projection and the area of shifting cultivation was extracted for each country of the three continents. We assumed that this area is the total area affected by shifting cultivation in each country, i.e., includes the area of fields under cultivation and the area of fallow vegetation. We also assumed that, sooner or later, depending on the cultivation cycle, all areas of fallow vegetation or secondary forest will be cleared for cultivation. These assumptions may lead to some overestimation of the area of shifting cultivation in some countries, since not all secondary forests result from shifting cultivation. In Brazil, for example, much secondary vegetation is not part of a shifting cultivation system, but rather results from abandonment of cattle pastures [Fearnside, 2000].

[16] The area of fallow vegetation cleared and burned annually per country (A in equation (2)) was calculated as

equation image

where A is the area cleared and burned annually (ha), TA is the total area of shifting cultivation (ha), CP is the cropping period (years), and FP is the fallow period (years). CP + FP is the length of the rotation cycle and 1/(CP + FP) represents the fraction of the total area of shifting cultivation in a given country that is cleared and burned for cultivation each year. TA was computed from the selected land cover maps (GLC2000 and Map of the Ecosystems of Central America) and the values of CP and FP by country are presented in Tables 35. This estimation of the area burned annually assumes that the area burned is equated to the area felled, which according to Fearnside et al. [1999] is a reasonable assumption in tropical forests, namely in the case of Brazil.

3.3. Aboveground Biomass

[17] Biomass accumulation following disturbance caused by shifting agriculture practices was estimated based on the equations developed by Zarin et al. [2001], following the work of Johnson et al. [2000]. Zarin et al. [2001] used the data from Johnson et al. [2000], which compiled global data on the aboveground biomass accumulation in secondary forests. Zarin et al. [2001] developed global equations and tested their regional applicability with validation data from the Amazon, not included in the global data set from which equations were derived. To develop those equations (equations (4) and (5)), data that relate broadleaved aboveground biomass accumulation following disturbance with the growing season degree-years (GSDY) variable were used. See the annexes of Johnson et al. [2000] for the data used. The analysis was separated by soil texture (i.e., sandy versus nonsandy), and the data used relate to 40 plots located in sandy soils and 147 plots located in nonsandy soils, all with broadleaved vegetation. We fitted again the equations to the available data, because the standard errors of the parameter estimates (SEE) were not reported by Zarin et al. [2001], and they were necessary for our uncertainty and sensitivity analysis (section 3.6). The resulting equations are

equation image
equation image

where AGBs and AGBns are the aboveground biomass accumulation in sandy and nonsandy soils (Mg ha−1), respectively, FP is the fallow period (years) (values from Tables 35), T is the growing season temperature (°C), and L is the duration of the growing season (days) (L/365 is the duration of the growing season as a fraction of the year, which is estimated from the climate data, as described in section 2.4). The product FP × T × L/365 is the GSDY variable, an index that represents the increase in metabolic processes commonly associated with increased temperature in the presence of adequate moisture [O'Neill and DeAngelis, 1981; Johnson et al., 2000]. Equation (4) has an R2 = 0.63 (n = 40) and the SEE are 15.4520 and 2.9625, for the intercept and slope, respectively. Equation (5) has an R2 = 0.92 (n = 147) and the SEE is 0.0102. Note that the equation for nonsandy soils shown by Zarin et al. [2001] included an intercept, but our analysis indicated that it was not significantly different from zero, so we fitted an equation without intercept. The aboveground biomass data used (again, see the Johnson et al. [2000] annexes) refer to dry matter.

[18] To estimate the duration of the growing season and the mean surface temperature of the growing season, the average for the period 1979–2008, for each variable and for each LSM (i.e., Mosaic, Noah, CLM), was calculated. On a monthly basis, it was assumed that a given month supported vegetation growth only if the precipitation was greater than the evapotranspiration, given that in tropical regions temperature is not a limiting factor for vegetation growth (Zarin et al. [2001] used the same assumption). Therefore, the mean surface temperature of the growing season was computed as the mean surface temperature of those months when precipitation was greater than evapotranspiration. This was done for each LSM and finally the resulting duration of the growing season (as a fraction of the year) and mean surface temperature were calculated as the mean of the three LSMs (i.e., Mosaic, Noah, CLM).

[19] For the purpose of this study, biomass accumulation was estimated on a country basis. The following data sets were used: spatial distribution of shifting agriculture, country limits, soil texture data, climate data (i.e., length of growing season as a fraction of the year, and growing season average surface temperature). As most of the countries have both sandy and nonsandy texture soils, the total aboveground biomass accumulation was estimated by areal weighting of those two classes, i.e., the AGB of equation (2) was computed as

equation image

where AGB is the aboveground biomass present at the time the fallow vegetation is cut in shifting cultivation, AGBs and AGBns are the aboveground biomass accumulation in sandy and nonsandy soils (equations (4) and (5)), respectively, and PAreas and PAreans are the proportion of the country area with sandy and nonsandy soils, respectively.

3.4. Combustion Completeness

[20] Combustion completeness or combustion factor is the fraction of biomass exposed to fire that is actually consumed (or volatilized) [Araújo et al., 1999]. Combustion completeness depends on vegetation type, moisture content and meteorological conditions during the fire. In the case of shifting cultivation, vegetation is cut at the end of the wet season or at the beginning of the dry season, allowed to dry for some time and then burned, raising the level of nutrients just before the preparation of soil for planting. The proper time to burn is very important. If the land is burned too soon after clearing, vegetation will be too moist and burning will be rather incomplete, soil fertility may not be enough to guarantee agricultural productivity and weeds may start establishing in the burned field before planting [Warner, 1991]. Ideally, a field will be burned at the end of the dry season [Gupta, 2000; Sirois et al., 1998].

[21] Biomass loadings and fractions of biomass burned (combustion completeness) are difficult and expensive to measure in tropical forests. The few studies available indicate high variability among years and among sites, on a micro scale, at any given site [Fearnside et al., 1999]. Most studies describing tropical forest clearing experiments were conducted in Brazil. Table 6 presents estimates of combustion completeness from seven experiments performed in Brazil and two in India, concerning 22 burns in forests with different land use histories: primary forests and second- and third-growth forests. Prasad et al. [2001b] and Gupta et al. [2001] specifically describe biomass burning experiments in forests affected by shifting cultivation. The estimates refer to overall combustion completeness, i.e., they include several biomass compartments (different trunk and branch sizes, leaves, litter, etc.). We assumed a combustion completeness value of 40.6%, which is the mean of the values in Table 6.

Table 6. Combustion Completeness Values Reported in Studies Describing Clearing Experiments in Tropical Forests
Forest TypeLocationCombustion Completeness (%)Reference
Primary rain forestPará, Brazil20.1Araújo et al. [1999]
Rain forestMato Grosso, Brazil19.5; 22.7; 41.8; 47.5; 61.5Carvalho et al. [2001]
Primary rain forestAmazonas, Brazil20.5Carvalho et al. [1998]
Rain forestAmazonas, Brazil30.0Fearnside et al. [2001]
Rain forestPará, Brazil43Fearnside et al. [1999]
Secondary rain forest (second and third growth)Pará and Rondônia, Brazil42.5; 47.3; 52.7; 62.5; 63.2; 87.5Hughes et al. [2000]
Primary rain forestRondônia, Brazil47.0; 54.0Guild et al. [1998]
Moist secondary mixed deciduous forestAndhra Pradesh, India16.1; 20.2; 30.0Prasad et al. [2001b]
Dry deciduous forestAndhra Pradesh, India27.0; 36.0Gupta et al. [2001]

[22] We also assumed that all aboveground biomass present at the time of clearing is cut and exposed to fire. This may lead to some overestimation, because in some shifting cultivation systems very large trees are not cut [e.g., Nyerges, 1989; Gupta, 2000; Kupfer et al., 2004] and part of the biomass may be removed for house construction, charcoal making or as firewood [e.g., Lawrence, 1997; Liang et al., 2009]. Emissions from biomass removed for firewood and charcoal are implicitly included, by using estimates of the biomass exposed to fire that have not been reduced to reflect removal of these products, although the combustion completeness and emission factors of these types of biomass burning would be different from the ones of the slash burned in cleared forests. In the special case of chitemene system in Zambia, in which trees are chopped in a larger area (the outfield), and the slash piled in a smaller area (the infield), and burned [Stromgaard, 1989], the amount of biomass exposed to fire is larger than the biomass present in the cultivation site at the time of clearing. Unfortunately, the present study does not cover shifting cultivation in Southern Africa due to the lack of spatial distribution data.

3.5. Emission Factors

[23] After estimating biomass loading, on a dry weight basis, and biomass consumption (biomass loading times combustion completeness), the emission factor allows the estimation of the total emission of each compound. An emission factor is defined as the amount of a compound released per amount of dry biomass consumed, expressed in units of grams per kilogram. Andreae and Merlet [2001] evaluated information on emissions from various types of biomass burning (savanna and grassland, tropical forest, extratropical forest, biofuel burning, charcoal making, charcoal burning, agricultural residues) reported in a large number of publications and presented the emission factors for a large variety of species. We used an updated version of these data provided by M. O. Andreae (personal communication, 2009). The emission factors (mean and standard deviation) for CO2, CO, CH4, NOx, and N2O, for biomass burning in tropical forests, are: 1626 ± 39 g kg−1, 101 ± 16 g kg−1, 6.6 ± 1.8 g kg−1, 2.26 ± 1.26 g kg−1, and 0.2 ± 0.1 g kg−1, respectively.

3.6. Uncertainty and Sensitivity Analysis

[24] In order to assess model output variability, an uncertainty analysis followed by a sensitivity analysis was performed [Saltelli et al., 2004]. Uncertainty analysis aims to quantify the overall uncertainty associated with the model response as a result of uncertainties in the model inputs, while sensitivity analysis shows how the variation in the model output can be apportioned to different sources of variations [Crosetto et al., 2000]. Assuming a deterministic model with output variable Y and k scalar input factors, Y = f(X1, X2, … Xk), for the purpose of uncertainty and sensitivity analysis, input factors are treated as random variables with a known probability density function (pdf). The output Y has its own pdf, whose estimation is the purpose of uncertainty analysis [Crosetto et al., 2000].

[25] The Monte Carlo method, which allows exploring the full range of variation for the input factors and does not require assumptions upon the model structure, is used to perform multiple evaluations of the model with randomly selected model inputs. The Monte Carlo based uncertainty analysis involves four steps [Crosetto et al., 2000]: assign a distribution (pdf) to each input factor Xi; generate a sample of size N (Xj, j = 1, …, N) from the factors' distributions according to an appropriate design; evaluate the model at each sample point Xj; analyze the resulting output values Yj. In the definition of the pdfs one must take advantage of all available information about the stochastic properties of each input factor (e.g., data measurement techniques, published data, expert opinion). The generation of samples of the input variable space is performed using quasi-random sequences [Sobol', 1967], which allow a better input space exploration than simple random sampling techniques. In the model evaluation step, the output Yj is computed for each sample point Xj, obtaining a sequence of Yj. The last step is the estimation of the expected value and variance for the output variable Y.

[26] Concerning the sensitivity analysis, a variance-based decomposition technique is considered. This technique performs a decomposition of the function Y = f(X1, X2, … Xk) into main effects and interactions, and allows the quantification of the contribution of each variable to the model output total variance [Saltelli et al., 2010]. One measure of sensitivity of Y to an individual input variable Xi (the main effect) is V[E(YXi)], i.e., the expected amount of variance that would be removed from the total output variance if we were able to learn the true value of Xi (within its uncertainty range). After dividing by the total variance the first-order sensitivity index is obtained:

equation image

This measure indicates the relative importance of an individual input variable in driving the uncertainty and can be seen as indicating where to direct effort in the future to reduce uncertainty.

[27] The total effect for the input variable Xi is linked to E[V(YXi)], which is the expected amount of output variance that would remain unexplained (residual variance) if Xi, and only Xi, were left free to vary over its uncertainty range, all the other variables having been fixed at their true values. After dividing by the total variance, the total sensitivity index is obtained:

equation image

This index represents the overall effect of Xi, including the interactions with all other variables. The total sensitivity index is used to identify unessential variables, those that are not important neither singularly nor in combination with others. The estimation of the pair (Si, STi) is important to appreciate the difference between the impact of factor Xi alone on Y (Si) and the overall impact of factor Xi through interactions with the other factors on Y (STi). The difference STiSi is a measure of how much Xi is involved in interactions with the other factors. The measure 1 − ∑ Si is the fraction of the output variance that is not explained by the single variables, i.e., it is due to interactions between variables [Saltelli et al., 2004].

[28] In the present study, the uncertainty and sensitivity analysis was done for the estimation of the total emission of each gas (Ei) from each continent (Asia, Africa and America) using the following model:

equation image

In this equation, b1AGBns, b0AGBs, and b1AGBs are the regression coefficients of equations (4) and (5). All other parameters were defined previously. We defined TA as a normally distributed variable and used the commission error reported for the class “Mosaic: Cropland/Tree Cover/Other Natural Vegetation” (18%) in the validation of the Global Land Cover 2000 Map [Mayaux et al., 2006], as the coefficient of variation of this variable. A discrete uniform distribution was assumed for the variables CP and FP, since they represent the length of cultivation and fallow, in years (Tables 35). Regarding biomass estimation, uncertainty is represented in the regression coefficients (b1AGBns, b0AGBs, and b1AGBs), using a normally distributed variable with the standard error of the parameter estimates as a dispersion measure (see SEE values in section 3.3). Concerning combustion completeness (CC), the Rayleigh distribution was fitted to the values shown on Table 6 (σ = 0.313881, using the MATLAB Distribution Fitting Tool). Rayleigh distribution allows a good fit to the CC data, which are skewed to the right. Last, uncertainty in the EF is represented by a normal distribution using the mean and standard values provided by Andreae and Merlet [2001]. Climatic data (T and L) and data about the area of each country with sandy and nonsandy soils were excluded from the uncertainty and sensitivity analysis. Using this method, model is evaluated N(k + 2) times and because accuracy of estimates increase with sample size, N was set to 25600. The number of input factors (k) was 35 for Asia, 47 for Africa, and 47 for America. Rosa et al. [2011] applied the uncertainty and sensitivity assessment techniques followed in this study to the estimation of atmospheric emissions from vegetation fires in Portugal.

4. Results

4.1. Spatial Distribution of Shifting Cultivation

[32] Figures 13 show the spatial distribution of shifting cultivation in Asia, Africa and America, respectively, according to the GLC2000 regional maps (see Table 1) and the Map of the Ecosystems of Central America. The location of the studies describing fallow and cropping periods practiced in shifting cultivation systems in the tropics (references included in Tables 35) is overlaid. For most of the countries in the three regions there is an agreement between the spatial data and the literature review. Two types of inconsistency may be observed: countries where shifting cultivation is mapped by the land cover products but for which we could not find any reference reporting cropping or fallow periods (e.g., Ecuador, Liberia), and countries for which there are studies describing shifting cultivation and the associated cropping and fallow periods, but where this land cover class is not mapped (e.g., Mexico, Zambia, Philippines). Three countries from Eastern Africa were excluded from the analysis: Uganda, Kenya and Ethiopia. According to Nandwa [2001], in Eastern and Southern Africa there have been considerable changes in the traditional shifting cultivation type of agriculture, with the fallow practice almost absent in Malawi, Zimbabwe and Kenya and fallow lengths substantially shorter in Zambia, Mozambique and Tanzania. Concerning Uganda, fallow period reported for this country is 1 year (see Table 4), i.e., agriculture could be considered very similar to permanent cultivation.

Figure 1.

Spatial distribution of shifting cultivation in Asia (shading) according to the Global Land Cover 2000 map. The locations of case studies (circles) describing shifting cultivation systems are overlaid.

Figure 2.

Spatial distribution of shifting cultivation in Africa (shading) according to the Global Land Cover 2000 map. The location of case studies (circles) describing shifting cultivation systems is overlaid.

Figure 3.

Spatial distribution of shifting cultivation in America (shading) according to the Global Land Cover 2000 map and the Map of the Ecosystems of Central America. The location of case studies (circles) describing shifting cultivation systems is overlaid.

4.2. Country-Level Emissions

[33] Annual estimates of greenhouse gas emissions from biomass burning in shifting cultivation at country level for tropical Asia, Africa and America, and for two scenarios (short and long fallow period) are summarized in Tables 712. In Asia (Tables 7 and 8), the emissions under the long fallow scenario are larger for almost all countries. Only in Bangladesh and Sri Lanka gas emissions are smaller under the long fallow scenario. Myanmar, Laos and Vietnam have the largest emissions of greenhouse gases, while small countries such as Brunei, Bhutan and Sri Lanka have the smallest emissions. In America (Tables 11 and 12), most of the countries also have larger emissions under the long fallow scenario. Brazil, Colombia and Venezuela have the largest emissions, while Belize, El Salvador and the Guianas (French Guiana, Suriname and Guyana) have the smallest. Contrary to what happens in Asia and America, in Africa (Tables 9 and 10) most countries have larger emissions under the short fallow scenario. Democratic Republic of Congo and Côte d'Ivoire have the largest emissions, while Angola, Guinea-Bissau and Central African Republic have the smallest. The small emissions of greenhouse gases estimated for Angola are, most certainly, due to the small area under shifting cultivation mapped by the GLC2000 map, which mapped this class only in the most northern province of Angola, the exclave of Cabinda.

Table 7. Greenhouse Gas Emissions From Shifting Cultivation in Tropical Asian Countries: Short Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara (ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow period assumed to be equal to India.

  • d

    Fallow period assumed to be equal to Malaysia.

Bangladeshc1,508,1763301,63526.78,038,7003,263,7125,306,796329,63521,5407,376653
Bruneid15,80952,25854.2122,42449,70480,8195,02032811210
Bhutan472,348394,47012.51,180,874479,435779,56148,4233,1641,08496
China6,765,94031,353,18817.423,605,6469,583,89215,583,409967,97363,25421,6601,917
Indonesia4,703,0504783,84241.232,277,33213,104,59721,308,0741,323,56486,49029,6162,621
India7,582,04231,516,40821.632,768,16613,303,87521,632,1011,343,69187,80630,0672,661
Cambodia3,234,7673646,95322.914,815,4966,015,0919,780,539607,52439,70013,5941,203
Laos10,901,07751,557,29728.744,770,85018,176,96529,555,7451,835,873119,96841,0803,635
Sri Lankac785,8513157,17024.13,787,7071,537,8092,500,477155,31910,1503,475308
Myanmar18,621,787131,241,45270.086,865,00435,267,19257,344,4533,561,986232,76379,7047,053
Malaysia671,869595,98149.14,715,1381,914,3463,112,726193,34912,6354,326383
Thailand666,11635951,59531.229,714,81512,064,21519,616,4131,218,48679,62427,2652,413
Vietnam9,928,43722,482,10914.335,376,33014,362,79023,353,8961,450,64294,79432,4602,873
Table 8. Greenhouse Gas Emissions From Shifting Cultivation in Tropical Asian Countries: Long Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara (ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow period assumed to be equal to India.

  • d

    Fallow period assumed to be equal to Malaysia.

Bangladeshc1,508,17612107,72759.76,433,3742,611,9504,247,030263,80717,2395,903522
Bruneid15,80915930162.6151,23061,39999,8366,20140513912
Bhutan472,348847,23535.91,695,584688,4071,119,35069,5294,5431,556138
China6,765,94030211,436170.536,048,21314,635,57423,797,4441,478,19396,59533,0762,927
Indonesia4,703,05027162,174247.240,089,73816,276,43426,465,4811,643,920107,42436,7853,255
India7,582,04212541,57461.433,237,61213,494,47121,942,0091,362,94289,06430,4982,699
Cambodia3,234,76725119,806127.715,295,9486,210,15510,097,712627,22640,98714,0351,242
Laos10,901,07718545,05494.351,372,28120,857,14633,913,7192,106,572137,65747,1374,171
Sri Lankac785,8511256,13257.23,208,9831,302,8472,118,429131,5888,5992,944261
Myanmar18,621,78718931,08994.888,246,72135,828,16958,256,6023,618,645236,46680,9727,166
Malaysia671,8691539,522142.25,621,2382,282,2233,710,894230,50415,0635,158456
Thailand6,661,16315391,83385.033,296,59213,518,41621,980,9451,365,36089,22230,5522,704
Vietnam9,928,43715584,02691.253,266,65521,626,26235,164,3022,184,252142,73348,8754,325
Table 9. Greenhouse Gas Emissions From Shifting Cultivation in Tropical African Countries: Short Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara (ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow period assumed to be equal to Cameroon.

  • d

    Fallow period assumed to be equal to Sierra Leone.

  • e

    Fallow period assumed to be equal to Ghana.

Angolac121,270324,25426.7647,990263,084427,77426,5711,73659553
Benin450,6401232,18957.31,845,606749,3161,218,38875,6814,9451,693150
Cameroon2,142,5283428,50626.611,407,6984,631,5257,530,860467,78430,56810,467926
Central African Republicc279,215355,84322.61,259,673511,427831,58151,6543,3751,156102
Congoc1,927,7833385,55727.110,454,1224,244,3746,901,351428,68228,0139,592849
Côte d'Ivoire14,566,638101,213,88654.966,653,79827,061,44244,001,9052,733,206178,60661,1595,412
D. R. Congo16,486,29520749,377106.179,498,32532,276,32052,481,2963,259,908213,02472,9446,455
Equatorial Guineac432,844386,56930.92,677,4331,087,0381,767,524109,7917,1742,457217
Gabonc1,398,2233279,64528.37,913,4003,212,8405,224,079324,49721,2057,261643
Ghana6,620,09821,655,02416.627,435,74011,138,91018,111,8681,125,03073,51725,1742,228
Guinea6,339,4728633,94745.128,573,87911,600,99518,863,2181,171,70076,56726,2182,320
Guinea-Bissau255,228536,46130.91,126,966457,548743,97346,2123,0201,03492
Liberiad5,669,41331,133,88324.427,635,10011,219,85118,243,4771,133,20574,05125,3572,244
Madagascar6,630,22931,326,04620.226,840,72710,897,33517,719,0671,100,63171,92224,6282,179
Nigeria7,038,63012,346,2102.96,758,0932,743,7864,461,396277,12218,1096,201549
Sierra Leone5,505,54031,101,10823.225,498,54010,352,40716,833,0141,045,59368,32623,3962,070
Togoe213,325253,33160.03,198,771129,87012,111,688131,1698,5712,935260
Table 10. Greenhouse Gas Emissions From Shifting Cultivation in Tropical African Countries: Long Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara(ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow period assumed to be equal to Cameroon.

  • d

    Only one value of fallow period was found for D. R. Congo (>20 years); the value 40 years was set for the long fallow period scenario.

  • e

    Fallow period assumed to be equal to Sierra Leone.

  • f

    Fallow period assumed to be equal to Ghana.

Angolac121,2701010,10656.0565,828229,726373,53523,2021,51651946
Benin450,6401526,50862.71,661,011674,3711,096,52768,1114,4511,524135
Cameroon2,142,52810178,54457.010,170,5154,129,2296,714,12741,705227,2539,332826
Central African Republicc279,2151023,26857.21,331,029540,398878,68754,5803,5671,221108
Congoc1,927,78310160,64959.19,487,1693,851,7916,263,012389,03125,4228,705770
Côte d'Ivoire14,566,63820662,12071.447,304,41019,205,59131,228,2901,939,765126,75743,4053,841
D. R. Congod16,486,29540392,531182.371,553,91929,050,89147,236,7492,934,140191,73665,6555,810
Equatorial Guineac432,8441036,07059.62,151,529873,5211,420,34588,2265,7651,974175
Gabonc1,398,22310116,51957.26,660,8562,704,3084,397,204273,13517,8486,112541
Ghana6,620,09841,103,35033.136,533,78114,832,71524,117,995149,810497,89633,5222,967
Guinea6,339,47217333,65670.723,586,9599,576,30515,571,073967,20763,20421,6421,915
Guinea-Bissau255,228631,90335.31,124,847456,688742,57546,1253,0141,03291
Liberiae5,669,41314354,33868.624,324,3209,875,67416,057,846997,44365,17922,3191,975
Madagascar6,630,22941,105,03827.029,823,03012,108,15019,687,8521,222,92379,91427,3642,422
Nigeria7,038,63031,407,72626.336,963,58315,007,21524,401,7311,515,72999,04833,9163,001
Sierra Leone5,505,54014344,09663.221,750,1118,830,54514,358,466891,88558,28219,9571,766
Togof213,325435,55465.32,321,766942,6371,532,72895,2066,2212,130189
Table 11. Greenhouse Gas Emissions From Shifting Cultivation in Tropical American Countries: Short Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara (ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow periods assumed to be equal to Brazil.

  • d

    Fallow periods assumed to be equal to Peru.

  • e

    Fallow periods assumed to be equal to Honduras.

  • f

    Only one value of fallow period was found for Venezuela (<15 years); the value 7 years was set for the short fallow period scenario.

Belize16,88615,6295.128,75311,67418,9811,17977262
Bolivia2,077,35215122,19767.98,295,5543,367,9955,476,360340,16722,2297,612674
Brazil70,713,991217,678,49816.5292,078,170118,583,737192,817,15711,976,957782,65326,799923,717
Colombiac17,257,85924,314,4659.540,823,29616,574,25826,949,7441,674,000109,39037,4583,315
Ecuadord5,248,75321,312,1889.312,251,1044,973,9488,087,640502,36932,82811,241995
El Salvadore110,411418,40228.2518,605210,554342,36021,2661,39047642
French Guianac63,759215,94010.3164,55766,810108,6336,74844115113
Guyanac318,455279,6148.8704,070285,852464,79628,8711,88764657
Honduras858,5834143,09732.74,676,3981,898,6183,087,152191,76012,5314,291380
Nicaraguae834,6654139,11126.83,727,9451,513,5462,461,025152,8689,9893,421303
Panama1,253,5212313,38016.05,028,8582,041,7163,319,831206,21313,4754,614408
Paraguay3,210,38510267,53246.412,416,4285,041,0708,196,779509,14833,27111,3931,008
Peru2,718,2342679,5588.35,669,7252,301,9083,742,903232,49315,1935,202460
Surinamec216,729254,18210.3560,095227,398369,75022,9671,50151445
Venezuelaf5,806,2977645,14436.123,288,4439,455,10815,374,005954,96662,40421,3691,891
Table 12. Greenhouse Gas Emissions From Shifting Cultivation in Tropical American Countries: Long Fallow Scenario
CountryTotal Area of Shifting Cultivation (ha)Fallow Period (years)Area Cleared per Yeara (ha yr−1)Aboveground Biomass at the Time of Clearing (Mg ha−1)Biomass Cleared and Exposed to Fire per Year (Mg yr−1)Biomass Burned per Yearb (Mg yr−1)Emission of Gases (Mg yr−1)
CO2COCH4NOxN2O
  • a

    Calculated with equation (3), assuming a cropping period of 2 years.

  • b

    Assuming a combustion completeness of 40.6%.

  • c

    Fallow periods assumed to be equal to Brazil.

  • d

    Fallow periods assumed to be equal to Peru.

  • e

    Fallow periods assumed to be equal to Honduras.

Belize16,886101,40763.789,57336,36659,1323,673240827
Bolivia2,077,3522094,42586.18,131,3923,301,3455,367,988333,43621,7897,461660
Brazil70,713,991203,214,27271.5229,696,21793,256,664151,635,3369,418,923615,494210,76018,651
Colombiac17,257,85920784,44897.076,114,62530,902,53850,247,5273,121,156203,95769,8406,181
Ecuadord5,248,75320238,58096.122,920,5379,305,73815,131,130939,88061,41821,0311,861
El Salvadore110,411254,08971.9294,009119,368194,09212,05678827024
French Guianac63,759202,898103.6300,272121,910198,22612,31380527624
Guyanac318,4552014,47592.31,336,180542,489882,08754,7913,5801,226108
Honduras858,5832531,799107.33,412,8361,385,6122,253,004139,9479,1453,131277
Nicaraguae834,6652530,914116.73,607,1251,464,4932,381,265147,9149,6663,310293
Panama1,253,5213137,985190.37,230,1202,935,4294,773,007296,47819,3746,634587
Paraguay3,210,38515188,84662.211,737,0324,765,2357,748,272481,28931,45110,769953
Peru2,718,23420123,55688.510,933,2294,438,8917,217,637448,32829,29710,032888
Surinamec216,729209,851103.71,021,703414,812674,48441,8962,73893783
Venezuela5,806,29715341,54771.924,544,8309,965,20116,203,4171,006,48565,77022,5211,993

4.3. Continental Emissions

[34] Total annual emissions of CO2, CO, CH4, NOx, and N2O over tropical Asia, Africa and America are reported in Table 13. Mean and standard deviation values were calculated using the Monte Carlo based uncertainty analysis. Figures 46 show the frequency distribution of CO2 emissions as model output (N = 25600) in Asia, Africa and America, respectively. From the factors included in our model to estimate greenhouse gas emissions, the area of shifting cultivation, the cropping and fallow periods and the aboveground biomass accumulation estimates have different values for each country of each continent. The total area of shifting cultivation mapped by the land cover maps in America, Africa and Asia is 110 705 881 ha, 76 077 372 ha, and 71 852 315 ha, respectively. This explains the order of the continental estimates observed in Table 13, with the largest emissions from shifting cultivation occurring in America, followed by emissions from Asia and Africa. The other factors (cropping and fallow periods and biomass accumulation) contribute to differences in the magnitude of the emissions. The emission of CO2, for example, is about 15% lower in Africa comparing to Asia, but the total area of shifting cultivation is about 6% higher.

Figure 4.

Uncertainty analysis of total annual CO2 emissions from shifting cultivation in Asia.

Figure 5.

Uncertainty analysis of total annual CO2 emissions from shifting cultivation in Africa.

Figure 6.

Uncertainty analysis of total annual CO2 emissions from shifting cultivation in America.

Table 13. Total Annual Greenhouse Gas Emissions From Shifting Cultivation in Tropical Asia, Africa and Americaa
 AsiaAfricaAmerica
MeanSDMeanSDMeanSD
  • a

    Emissions given in Tg (1012 g) for CO2 and CO and in Gg (109 g) for the other species.

CO2 (Tg yr−1)241132205139295197
CO (Tg yr−1)1591391813
CH4 (Gg yr−1)9796188326261197892
NOx (Gg yr−1)335282285272410389
N2O (Gg yr−1)302425233633

[35] As observed in Table 13, the standard deviation values are very large comparing to the mean values, indicating large uncertainty in emission estimates from biomass burning in shifting cultivation systems. The sensitivity analysis (Tables 1416) allows the quantification of the contribution of each variable (total area of shifting cultivation, cropping and fallow periods, biomass accumulation estimates, combustion completeness and emission factors) to the model output total variance. For the three continents, the variables with the highest values of the first-order sensitivity index (Si), and thus contributing most to the uncertainty of the estimates, are the coefficients of the equations used to estimate the aboveground biomass on sandy soils (b0AGBs and b1AGBs), combustion completeness, and emission factors. For the gases with large emissions (CO2 and CO) and for CH4 the Si values of CC are very high, especially in Asia, comparing with the Si values of all other variables. For the gases with the smallest emissions (NOx and N2O), EF and CC are the factors that contribute most to the uncertainty of the estimates, with higher Si values of EF than those of CC in the case of NOx. In America, all the variables concerning Brazil (TA Brazil, FP Brazil, CP Brazil) have a larger contribution to the output variance than the variables associated to the other countries.

Table 14. Sensitivity Analysis of Greenhouse Gas Emissions From Shifting Cultivation in Tropical Asia: First-Order Sensitivity Index (Si) and Total Sensitivity Index (STi)a
VariableCO2COCH4NOxN2O
SiSTiSiSTiSiSTiSiSTiSiSTi
  • a

    Abbreviations are as follows: CC, combustion completeness; b1AGBns, b0AGBs, and b1AGBs, regression coefficients of the biomass accumulation equations; CP, cropping period; EF, emission factor; FP, fallow period; TA, total area of shifting cultivation. Boldface denotes highest values, which are mentioned in section 4.

TA Bangladesh0.00010.00010.00010.00010.00010.00010.00000.00010.00000.0001
TA Brunei0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Bhutan0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA China0.00060.00090.00060.00080.00050.00070.00030.00050.00030.0005
TA Indonesia0.00120.00150.00110.00140.00100.00120.00050.00080.00060.0009
TA India0.00120.00190.00100.00170.00080.00150.00040.00110.00050.0011
TA Cambodia0.00010.00020.00010.00020.00010.00020.00000.00010.00000.0001
TA Laos0.00190.00250.00170.00230.00140.00200.00080.00140.00090.0015
TA Sri Lanka0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Myanmar0.00660.00840.00620.00780.00560.00680.00440.00470.00460.0050
TA Malaysia0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Thailand0.00090.00120.00080.00110.00070.00100.00050.00070.00050.0007
TA Vietnam0.00210.00290.00190.00270.00160.00230.00080.00160.00100.0017
FP Bangladesh0.00000.00010.00000.00010.00000.00010.00000.00010.00000.0001
FP Brunei0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Bhutan0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP China0.00090.00090.00180.00180.00220.00220.00220.00220.00230.0023
FP Indonesia0.00000.00020.00000.00020.00000.00010.00000.00010.00000.0001
FP India0.00020.00120.00020.00110.00020.00100.00020.00070.00020.0007
FP Cambodia0.00010.00010.00010.00010.00010.00010.00010.00010.00010.0001
FP Laos0.00030.00030.00030.00030.00020.00020.00010.00010.00010.0001
FP Sri Lanka0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Myanmar0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Malaysia0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Thailand0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Vietnam0.00030.00090.00030.00080.00020.00070.00010.00050.00010.0005
CP China0.00030.00060.00030.00060.00020.00050.00010.00030.00010.0004
CP Indonesia0.00000.00010.00000.00010.00000.00010.00000.00010.00000.0001
CP Cambodia0.00010.00010.00010.00010.00010.00010.00000.00000.00000.0000
CP Laos0.00020.00030.00020.00030.00020.00020.00010.00020.00020.0002
b1AGBns0.00120.00140.00120.00130.00110.00110.00070.00080.00080.0008
b0AGBs0.03080.04060.02840.03760.02420.03310.01420.02280.01590.0244
b1AGBs0.01910.02770.01730.02570.01460.02260.00830.01550.00930.0167
CC0.90670.92510.81750.85620.68330.75150.38140.51500.43050.5535
EF0.00170.00250.07460.09890.18650.24570.44040.57740.39890.5233
ΣSi0.9766 0.9558 0.9249 0.8556 0.8669 
Table 15. Sensitivity Analysis of Greenhouse Gas Emissions From Shifting Cultivation in Tropical Africa: First-Order Sensitivity Index (Si) and Total Sensitivity Index (STi)a
VariableCO2COCH4NOxN2O
SiSTiSiSTiSiSTiSiSTiSiSTi
  • a

    Abbreviations are as follows: CC, combustion completeness; b1AGBns, b0AGBs, and b1AGBs, regression coefficients of the biomass accumulation equations; CP, cropping period; EF, emission factor; FP, fallow period; TA, total area of shifting cultivation. Boldface denotes highest values, which are mentioned in section 4.

TA Angola0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Benin0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Cameroon0.00010.00010.00010.00010.00010.00010.00010.00010.00010.0001
TA Central African Republic0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Congo0.00010.00010.00010.00010.00010.00010.00000.00010.00000.0001
TA Côte d'Ivoire0.00070.00170.00060.00160.00050.00150.00030.00110.00030.0012
TA D. R. Congo0.00390.00490.00360.00470.00320.00430.00200.00330.00220.0035
TA Equatorial Guinea0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Gabon0.00000.00000.00000.00000.00000.00000.00010.00010.00000.0000
TA Ghana0.00070.00130.00070.00120.00060.00110.00050.00080.00050.0009
TA Guinea0.00050.00080.00050.00080.00040.00070.00030.00050.00030.0006
TA Guinea-Bissau0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Liberia0.00070.00090.00060.00080.00050.00080.00030.00060.00040.0006
TA Madagascar0.00080.00110.00080.00100.00070.00090.00040.00070.00050.0007
TA Nigeria0.00050.00130.00050.00130.00050.00120.00030.00090.00040.0009
TA Sierra Leone0.00040.00080.00040.00070.00030.00070.00020.00050.00020.0005
TA Togo0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Angola0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Benin0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Cameroon0.00000.00010.00000.00010.00000.00010.00000.00000.00000.0000
FP Central African Republic0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Congo0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Côte d'Ivoire0.00060.00070.00060.00060.00060.00060.00060.00060.00060.0006
FP D. R. Congo0.00040.00060.00030.00060.00020.00050.00010.00040.00010.0004
FP Equatorial Guinea0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Gabon0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Ghana0.00040.00100.00030.00100.00020.00090.00000.00070.00010.0007
FP Guinea0.00020.00050.00020.00050.00010.00040.00010.00030.00010.0004
FP Guinea-Bissau0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Liberia0.00020.00060.00020.00060.00020.00050.00010.00040.00010.0004
FP Madagascar0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Nigeria0.00290.00570.00270.00540.00230.00490.00140.00370.00160.0039
FP Sierra Leone0.00030.00080.00030.00070.00030.00070.00030.00050.00030.0005
FP Togo0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Benin0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Côte d'Ivoire0.00020.00050.00020.00050.00020.00040.00010.00030.00020.0004
CP D. R. Congo0.00010.00020.00000.00020.00000.00020.00000.00010.00000.0002
CP Ghana0.00010.00040.00010.00030.00010.00030.00000.00020.00000.0003
CP Liberia0.00010.00010.00010.00010.00010.00010.00000.00010.00000.0001
CP Nigeria0.00290.00610.00260.00580.00230.00530.00150.00410.00160.0043
CP Sierra Leone0.00010.00010.00000.00010.00000.00010.00000.00010.00000.0001
CP Togo0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
b1AGBns0.00010.00020.00010.00020.00010.00020.00010.00010.00010.0001
b0AGBs0.17520.22990.16150.21770.14030.19870.08630.15040.09580.1589
b1AGBs0.11690.15110.10880.14350.09540.13130.06030.09980.06660.1053
CC0.60140.68390.55820.64910.48730.59370.30210.45030.33500.4757
EF0.00120.00180.05010.07390.12980.19120.33800.49650.30100.4424
ΣSi0.9117 0.8942 0.8664 0.7955 0.8081 
Table 16. Sensitivity Analysis of Greenhouse Gas Emissions From Shifting Cultivation in Tropical America: First-Order Sensitivity Index (Si) and Total Sensitivity Index (STi)a
VariableCO2COCH4NOxN2O
SiSTiSiSTiSiSTiSiSTiSiSTi
  • a

    Abbreviations are as follows: CC, combustion completeness; b1AGBns, b0AGBs, and b1AGBs, regression coefficients of the biomass accumulation equations; CP, cropping period; EF, emission factor; FP, fallow period; TA, total area of shifting cultivation.

  • b

    Boldface denotes highest values, which are mentioned in section 4.

TA Belize0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Bolivia0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Brazil0.02620.04160.02440.03940.02130.03590.01320.02710.01460.0286
TA Colombia0.00140.00260.00140.00240.00130.00220.00090.00170.00100.0018
TA Ecuador0.00010.00030.00010.00020.00010.00020.00010.00020.00010.0002
TA El Salvador0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA French Guiana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Guyana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Honduras0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Nicaragua0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Panama0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Paraguay0.00010.00010.00020.00020.00020.00020.00030.00030.00030.0003
TA Peru0.00000.00010.00000.00010.00000.00010.00000.00000.00000.0000
TA Suriname0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
TA Venezuela0.00020.00030.00020.00030.00020.00030.00010.00020.00020.0002
FP Belize0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Bolivia0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Brazil0.01640.04560.01540.04320.01350.03950.00830.02990.00920.0316
FP Colombia0.00000.00290.00000.00270.00000.00250.00000.00190.00000.0020
FP Ecuador0.00010.00050.00010.00040.00010.00040.00000.00030.00000.0003
FP El Salvador0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP French Guiana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Guyana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Honduras0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Nicaragua0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Panama0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Paraguay0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Peru0.00000.00010.00000.00010.00000.00010.00000.00010.00000.0001
FP Suriname0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
FP Venezuela0.00000.00010.00000.00010.00000.00010.00000.00010.00000.0001
CP Belize0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Bolivia0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Brazil0.05290.08670.04950.08250.04350.07550.02760.05750.03040.0607
CP Colombia0.00090.00110.00070.00100.00060.00090.00030.00070.00040.0008
CP Ecuador0.00030.00070.00020.00070.00020.00060.00000.00050.00000.0005
CP French Guiana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Guyana0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Nicaragua0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Panama0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Peru0.00010.00020.00010.00020.00010.00010.00010.00010.00010.0001
CP Suriname0.00000.00000.00000.00000.00000.00000.00000.00000.00000.0000
CP Venezuela0.00000.00010.00000.00010.00000.00010.00000.00000.00000.0000
b1AGBns0.00020.00040.00020.00040.00020.00040.00010.00030.00010.0003
b0AGBs0.10240.15420.09440.14600.08200.13320.05060.10070.05610.1064
b1AGBs0.07050.09730.06590.09340.05790.08620.03700.06650.04080.0701
CC0.60770.69100.56270.65470.49030.59760.30290.45140.33600.4772
EF0.00140.00180.05160.07360.13240.18990.34150.49160.30460.4382
ΣSi0.8809 0.8671 0.8439 0.7830 0.7939 

[36] The difference between the total sensitivity index (STi) and first-order sensitivity index (Si) values for the large majority of the variables of the three continents is small, indicating weak interaction of each variable with other variables in the model. The sum of the Si values of all the variables involved in the estimation of each gas emission (ΣSi) is close to 1 for the three continents, especially in Asia, meaning that the fraction of the output variance that is not explained by the main effects of the variables, i.e., it is due to interactions, is small.

5. Discussion

5.1. Area Estimation

[37] FAO/UNEP [1982] (summarized by Lanly [1985]) provides the only source available regarding the extent of shifting cultivation at global scale. The area of shifting cultivation, including cultivated fields and fallows, for the year 1980, was estimated at the country level through interpretation of remote sensing imagery, making a distinction between fallows of closed forest and fallows of open forest. The total area of shifting cultivation reported for tropical America, Africa and Asia-Oceania [Lanly, 1985] was 170 250 000 ha, 166 050 000 ha, and 73 250 000 ha, respectively. The estimates of total area used in our study are smaller for America and Africa and very similar for Asia. Shifting cultivation was not mapped in important areas, such as Southern Africa or Southern Mexico by the Global Land Cover 2000 and the Map of the Ecosystems of Central America, respectively, which could explain the difference between the two estimates of total area of shifting cultivation for America and Africa. However, this is only one possible cause for the difference between the two sources, since in Asia shifting cultivation was also not mapped in Philippines and Papua-New Guinea but the two estimates are similar.

[38] There is no available information about the area cleared in shifting cultivation each year, at global scale. Table 17 shows this area only for some countries of Asia, as we did not find similar studies for Africa and America. The area reported for India is within the range of our short and long fallow scenario estimates. For Malaysia (Sarawak only) and Vietnam the area reported is much larger than our estimates. For Bhutan it is similar to our short fallow scenario, while for Laos it is similar to our long fallow scenario.

Table 17. Area Cleared Annually in Shifting Cultivation Systems in Some Countries of Tropical Asiaa
CountryThis StudyOther StudiesReference
Short Fallow ScenarioLong Fallow Scenario
  • a

    Areas are in hectares.

India1,516,408541,574995,600Prasad et al. [2003]
Malaysia (Sarawak)95,98139,522282,000Uhlig et al. [1993]
Bhutan94,47047,235115,000Upadhyay [1995]
Vietnam2,482,109584,0263,500,000Sam [1994]
Laos1,557,297545,054450,000Hansen [1998]

5.2. Biomass Accumulation and Combustion Completeness

[39] Sensitivity analysis revealed that one of the factors that contribute most to uncertainty is combustion completeness. The combustion completeness values used in the present study to derive the probability density function included in the uncertainty analysis (Table 6), have a mean value of 40.6% and a large standard deviation of 18.4%. Combustion completeness in shifting cultivation systems is, in fact, highly variable. Upadhyay [1995], for example, refers that shifting cultivators in Bhutan try to achieve maximum burning efficiency but the unburned wood is left in the field for slow decay. Warner [1991] describes a shifting cultivation system in Philippines where a poor burn is followed by a secondary burn; in this case, vegetation that was been partially burned would be put in piles, sometimes mounted around unburned logs, and then burned again. In some wetter areas this will have to be done repeatedly until the field is judged to be adequately burned [Warner, 1991]. In a shifting cultivation system of Sierra Leone described by Nyerges [1989], the debris of felling are partially burned and farmers never entirely remove this debris that cover a part of the soil.

[40] The combustion completeness value used in our study is within the range of values used in global biomass burning calculations. Seiler and Crutzen [1980] assumed a combustion completeness of 25%. van der Werf et al. [2003] assumed a combustion completeness of 30% for closed forests and 53% for open moist woodlands, while Hoelzemann et al. [2004] assumed 50% for tropical forests and 60% for woody savannas. Lauk and Erb [2009] used a value of 53.8%, based on Fearnside [2000].

[41] The equation to estimate biomass accumulation on sandy soils is a factor with some contribution to the model uncertainty. The equation used for sandy soils was derived with fewer data (n = 40) than that for nonsandy soils (n = 147), quality of model fitting is lower, and according to Zarin et al. [2001], it overestimates aboveground biomass during early regrowth and underestimates it in later regrowth stages. These results about biomass accumulation and combustion completeness are in agreement with the findings of Fearnside et al. [1999]. According to these authors the amount of biomass exposed to burning and the proportion of that biomass consumed during the fire represents a major domain of uncertainty.

[42] With respect to the amount of biomass burned in shifting cultivation at country level we could only find the work of Ahuja [1991], reporting 41.3 Tg of biomass (dry matter) burned annually in India, which is larger than our short fallow and long fallow scenario estimates, 26.6 Tg and 27.0 Tg, respectively. This difference may be partially due to the fact that shifting cultivation was mapped by GLC2000 only in the northeast region of India (see Figure 1).

5.3. Effect of Fallow Length on Country-Level Emissions

[43] Fallow period is used in most studies related to shifting cultivation systems as a measure of intensity. Intensification of shifting cultivation, i.e., the shortening of the fallow period, due to population increase or market opportunities, and consequent deforestation, is a common narrative in many studies [e.g., Myers, 1992] but is also criticized by others [e.g., Ickowitz, 2006]. We derived estimates of greenhouse gas emissions at national level for two scenarios, the short and the long fallow period, in order to understand the effect of fallow length on the magnitude of pyrogenic emissions. Our estimates show the lack of a direct relationship between length of the fallow period and this magnitude. It will depend on the balance between the area cleared annually in each country (which in turn depends on the fallow period) and the biomass accumulation rate, which is a function of the length and temperature of the growing season and soil texture [Zarin et al., 2001]. In countries with a high rate of biomass accumulation (i.e., with a high growing season temperature and a long growing season, T and L in equations (4) and (5), respectively), aboveground biomass at the time of clearing can be much higher for long fallow periods than for short periods, overcompensating the smaller area cleared annually in the long fallow scenario, and resulting in larger amounts of biomass burned and larger emissions. For most countries of Asia and America the emissions under the long fallow scenario are larger than the emissions under the short fallow scenario, while in Africa it is the contrary. The average rate of biomass accumulation, defined as T × L/365 or the GSD index of Zarin et al. [2001] equations, for Asian, American and African countries is 16.2, 16 and 15.5, respectively. The results of Lauk and Erb [2009] also show a non linear relationship between the length of fallow and the emissions from biomass burning in shifting cultivation. These authors derived estimates based on three different assumptions on the length of the fallow period: 6, 12, and 18 years, and the highest estimate of biomass burned was obtained when the 12 year fallow period was considered.

[44] According to Gleave [1996] there are various problems in treating fallow length as a surrogate measure for cultivation intensity. It is normally assumed that the oldest fallow forests are used first, which is not always the case. For example, as the farm workforce becomes more elderly, young bush fallow is preferred for cultivation to denser, older bush, which is more difficult to clear. Also, accessibility is a factor influencing the frequency of cultivation and length of fallow period. It is common for land close to settlements and roads to be cultivated more frequently so that fallow periods are shorter than on land in less accessible areas [Gleave, 1996].

5.4. Comparison of Global and Continental Estimates

[45] According to our knowledge, the only three global estimates of biomass burning in shifting cultivation systems are those provided by Seiler and Crutzen [1980], Fearnside [2000], and Lauk and Erb [2009]. Houghton [2003] considered shifting cultivation when estimating annual net flux of carbon to the atmosphere from changes in land use, but did not disaggregate shifting cultivation emissions from other land uses. Table 18 compares global CO2 emission estimates from different inventories (including wildland fires, biofuel and agricultural waste burning and shifting cultivation) with our results. Our estimate (741 Tg CO2 yr−1) is lower than the literature values of global CO2 emissions from biomass burning in shifting cultivation, which range from 950 Tg CO2 yr−1 [Fearnside, 2000] to 2764 Tg CO2 yr−1 [Seiler and Crutzen, 1980]. Our study is the only one based on spatial information derived from satellite imagery. The pioneering study of Seiler and Crutzen [1980] derived the area of shifting cultivation indirectly, based on information on land requirements per capita. The work of Lauk and Erb [2009] is based on FAO data on the extent of shifting cultivation in 1980, while Fearnside [2000] uses FAO estimates of the area changes due to transition from forest to shifting cultivation in the 1981–1990 period. Comparing with other sources of greenhouse gases, our estimate of global CO2 emission represents from about 8% [Mieville et al., 2010] to about 16% [Ito and Penner, 2004] of global emissions from wildfires, and about 28% of the emissions of the Yevich and Logan [2003] assessment. The study of Lauk and Erb [2009] is the only one with a regional breakdown of the estimates. Our results for the continental emissions of CO2 (Table 13) are smaller for the three continents. Only our mean values can be compared to the estimates of other studies, since they do not provide any measure of uncertainty.

Table 18. Global CO2 Emission Estimates From Different Types of Biomass Burning
ReferenceEmissions (Tg CO2 yr−1)
  • a

    Values derived from estimates of burned biomass (dry matter) using an emission factor of 1.626.

  • b

    Minimum and maximum values obtained from different assumptions on the length of the fallow period.

Ito and Penner [2004], wildland fires, 20004548–6202
Hoelzemann et al. [2004], wildland fires, 20005716
van der Werf et al. [2006], wildland fires, 1997–20048903
Schultz et al. [2008], wildland fires, 1960–20005728–8650
Mieville et al. [2010], wildland fires, 1997–20059620
Yevich and Logan [2003], biofuel and agricultural field burning, 19852688
Seiler and Crutzen [1980], shifting cultivation, 1960sa2764
Fearnside [2000], shifting cultivation, 1981–1990950
Lauk and Erb [2009], shifting cultivation, 1980a,bTotal 1709–2345
   Asia436–597
   Africa524–722
   America750–1026
This study741

6. Conclusions

[46] All the variables used to derive the estimates presented in this study are subject to uncertainty. Moreover, the need to make assumptions in the modeling procedure is another source of uncertainty. The analysis we have presented here is the first to estimate greenhouse gas emissions from biomass burning in shifting cultivation systems at national level, and to provide continental estimates derived with an explicit technique of uncertainty and sensitivity analysis. Uncertainty analysis should be a prerequisite for model building in any field where models are used [Crosetto et al., 2000]. In the present study, uncertainty estimation is very important for modelers interested in atmospheric chemistry and gas emissions from biomass burning. The uncertainty analysis applied allowed the quantification of the overall uncertainty associated with the model response and constructing a confidence range. However, the actual bias of the estimates is unknown and several factors involved in the model may be a source of uncertainty and bias: the coarse spatial resolution of the land cover maps; the application of a global equation to estimate aboveground biomass accumulation; the large variability of combustion completeness; the different values of emission factors available in the literature.

[47] Unfortunately, there is a shortage of information on shifting cultivation at global scale to which we can compare our estimates and assumptions. The most relevant finding of our study is the large uncertainty in greenhouse gas emission estimates from biomass burning in shifting cultivation systems. Our results also show that future research should be focused on the combustion completeness and the emission factors in order to reduce uncertainty in the estimates, namely through burning experiments conducted in shifting cultivation systems. Over the last 2 decades, availability of global land cover and burned area data sets has contributed to reducing uncertainties associated with estimates of the fire affected area, which no longer appears to be the most uncertain factor of the model of Seiler and Crutzen [1980], like it was referred by Levine [1991]. Future research may also be focused on the estimation of net CO2 emissions from shifting cultivation, since the gross emissions estimated in the present work may be offset by the uptake of CO2 in recovering fallows.

Acknowledgments

[48] Part of the research was funded by a postdoctoral grant to J.M.N.S. (SFRH/BPD/26773/2006) from the Foundation for Science and Technology, Ministry for Science and Technology, Portugal. We would like to thank Pedro Viterbo, from the Portuguese Meteorological Institute (IM), for providing the knowledge that led us to choose the best climatic data for this study and for clarifying some issues regarding the models that make the GLAS data set. We also thank M. O. Andreae (Max Planck Institute for Chemistry, Germany), who provided the data on the emission factors, and Stefano Tarantola (Joint Research Centre of the European Commission, Italy), who provided the MATLAB code used in the uncertainty and sensitivity analysis. The soil texture data for this study are from the Research Data Archive (RDA), which is maintained by the Computational and Information Systems Laboratory (CISL) at the National Center for Atmospheric Research (NCAR). The original data are available from the RDA (http://dss.ucar.edu/) in data set ds770.0. Climate data used in this study were acquired as part of the mission of NASA's Earth Science Division and are archived and distributed by the Goddard Earth Sciences (GES) Data and Information Services Center (DISC).