Previous estimates of the flux of carbon from land use change in sub-Saharan Africa have been based on highly aggregated data and have ignored important categories of land use. To improve these estimates, we divided the region into four subregions (east, west, central, and southern Africa), each with six types of natural vegetation and five types of land use (permanent crops, pastures, shifting cultivation, industrial wood harvest, and tree plantations). We reconstructed rates of land use change and rates of wood harvest from country-level statistics reported by the Food and Agriculture Organization (FAO) (1961–2000) and extrapolated the rates from 1961 to 1850 on the basis of qualitative histories of demography, economy, and land use. We used a bookkeeping model to calculate the annual flux of carbon associated with these changes in land use. Country-level estimates of average forest biomass from the FAO, together with changes in biomass calculated from the reconstructed rates of land use change, constrained the average biomass of forests in 1850. Comparison of potential (predisturbance) forest areas with the areas present in 1850 and 2000 suggests that 60% of Africa's forests were lost before 1850 and an additional 10% lost in the last 150 years. The annual net flux of carbon from changes in land use was probably small and variable before the early 1900s but increased to a source of 0.3 ± 0.2 PgC/yr by the end of the century. In the 1990s the source was equivalent to about 15% of the global net flux of carbon from land use change.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
 The annual sources and sinks of carbon from land use change play a major role in the rate at which carbon dioxide increases in the atmosphere, and are required for the construction of global and regional carbon budgets. There is probably no region where the sources and sinks of carbon are less well constrained than sub-Saharan Africa. Nowhere else are reported rates of deforestation as variable (Table 1), nowhere else has the estimated release of carbon from degradation of forests been larger than that from deforestation [Gaston et al., 1998], and nowhere else have historical changes in croplands yielded such divergent estimates [Ramankutty and Foley, 1999; Houghton and Hackler, 2001]. Furthermore, recent global estimates of the net flux of carbon from land use change have included a simplified accounting for sub-Saharan Africa [Houghton, 1999, 2003]. These analyses considered only two types of ecosystems (closed forests and open forests) [Houghton and Hackler, 2001] and only one type of land use change: the conversion of forests to croplands. Logging, fuel wood harvest, shifting cultivation, the establishment of tree plantations, and other forms of degradation and restoration, activities that were included in analyses of other regions, were not included in analyses of Africa.
Table 1. Average Rates of African Deforestation and Forestation (Including the Establishment of New Plantations) and Calculated Fluxes of Carbon for the 1990s
 The purpose of this analysis was to obtain an improved estimate of the annual net flux of carbon from land use change in Africa. Improvements included a greater number of ecosystems (five types of forest in each of four regions instead of two forest types in all of sub-Saharan Africa) and a greater number of land uses (five instead of one). We reconstructed a history of land use change that was consistent with a number of data sets from the Food and Agriculture Organization (FAO) and used independent data to construct a range of plausible estimates of flux. Finally, we compared the results with other recent estimates of deforestation and carbon emissions for the region. The analysis is more comprehensive than previous reconstructions of land use change in sub-Saharan Africa.
 This section describes, first, the geographical regions included in the analysis and the ecological zones considered, including the stocks of carbon in vegetation and soils and the per hectare changes in these stocks that result from growth and decay. Following that, we describe the data used to define rates of land use change from 1850 to 2000, dividing the types of land use into those that affect the area of forests and those that affect carbon stocks within forests without changing area. Finally, we briefly describe the bookkeeping model used to calculate the annual flux of carbon from these changes in land use and the procedure for initializing the model.
2.1. Natural Ecosystems
 We divided Africa into five geographical regions (Table 2) and included four of them in this analysis of sub-Saharan Africa (North Africa was not included). Despite the division, the regions are far from homogeneous. Most of them contain all five of the ecosystems described below, and land use histories have varied as well. Within each region we considered five types of woody formations [FAO, 2001].
 Tropical rain forest is centered about the equator, occupying much of central Africa and stretching along the coast of West Africa as far as Sierra Leone. A strip of rain forest also exists along the eastern coast of Madagascar. Precipitation is greater than 1000 mm per year, and any dry season is less than four months. The FAO  reports that little of the forest remains undisturbed. Rather, the zone contains large areas of secondary forests and grasslands. In central Africa, for example, which holds more than 80% of this forest (Table 3), only 65% of the tropical rain forest zone is still forested.
Table 3. Total Area (106 ha) of African Forests in Different Ecological Zones and Regions in the Year 2000 [from FAO, 2001]
Subtropical areas in southern Africa are included in these estimates.
 Tropical moist deciduous forest extends along the northern and southern borders of the rain forest, with patches in Mozambique and Madagascar. Precipitation is generally between 800 and 1500 mm/yr, with a pronounced dry season of up to six months. The forest generally appears as a woodland, Sudanese woodland to the north and miombo woodland in the south. Only about 31% of the deciduous forest zone in Africa remains forested [FAO, 2001]. Most of these forests are in southern Africa.
 Tropical dry forest borders the tropical moist deciduous forest to the north and south, with more than half of it in southern Africa as miombo, mopane, and acacia woodlands. Precipitation is 500–1000 mm/yr with a dry season of six to seven months. The vegetation type is woodland, and if agriculture occurs it is generally bush fallow.
 Tropical shrubland may exist where precipitation is 200–1000 mm/yr. Shrublands include scrub woodlands, thickets, and wooded grasslands. The tropical shrubland zone includes the Sahelian zone in the north and the Kalahari in the south, but most of these tropical shrublands/woodlands are in East Africa (Table 3).
 Tropical mountain forests include submontane forests (above 800–1200 m) and montane forests (above 1500–2000 m). The climate is cooler and usually wetter than the surrounding lowlands. East Africa has most of the montane forests, but small areas exist in each region except West Africa. Although some of the forests in southern Africa are subtropical, rather than tropical, their areas are small, and we included them in the corresponding tropical forest types.
2.2. Carbon Stocks
 For each region we assigned initial biomass values to each vegetation type so that the average forest biomass for a region at the end of a simulation (year 2000) was equal to the average biomass (weighted by forest area) reported by FAO [2001, Table 7]. When possible, we tried to find ecosystem-specific estimates that were consistent from region to region. We converted aboveground biomass [FAO, 2001] to units of total carbon by adding an additional 20% for roots and by multiplying by 0.5 for carbon content. The values of initial biomass, together with rates of biomass accumulation during forest growth and rates of wood decay, are given in Table 4. Rates of growth varied with vegetation type (i.e., precipitation and temperature) [Bellefontaine et al., 2000] as did rates of decomposition of coarse woody debris [Delaney et al., 1998; Chambers et al., 2000; Mackensen et al., 2003].
Table 4. Initial Carbon Stocks and Changes in the Stocks for Different Ecosystems in Response to Changes in Land Usea
Units are in MgC ha−1, MgC ha−1 yr−1, or years. Ranges include differences among regions. Each region has a single value. East Africa has lower values of biomass than the other regions for the same types of forest.
 Forest biomass is poorly known in many tropical regions, and we constructed an alternative estimate of flux (alternative 1) on the basis of values of biomass reported by Brown et al. . The area-weighted mean was 43% higher for all of sub-Saharan Africa than reported by FAO , and we increased the reference estimates of biomass by 43% in each forest type.
 The FAO 1990 Forest Resource Assessment [FAO, 1995] summarized the processes affecting African forests as follows: “The process of change in Africa appears to be formed by phases of progressive degradation, characterized by a variety of relatively small changes, both in area and in biomass. The main actor behind this process is rural population pressure for land (subsistence farming, pastures) and wood (fuel wood, construction material).” These small incremental changes in African land use are one reason why measuring the changes has been so difficult, and why estimates of carbon flux are so variable [Achard et al., 2002; DeFries et al., 2002; Houghton, 2003] (Table 1).
 Determining the rate of conversion of forests to permanent croplands is usually one of the best ways for estimating changes in forest area because the area of croplands is generally better documented than the area of forests. In Africa, however, the conversion of forests to permanent agriculture accounted for only 16% of the change in forest area between 1980 and 1990 [FAO, 1993]. Forests were degraded, fragmented, and replaced with long- and short-term fallows (shifting cultivation, or a rotational type of land use in which croplands alternate with forests). Degradation of forests results from grazing and related fire practices, selective logging, fuel wood harvest, and other human activities. Fragmentation results from the clearing of small patches of forest, leaving a mosaic of forest and nonforest. Over time, most fragmented forests in Africa are subsequently converted to permanent agriculture, suggesting that fragmentation is an intermediate step toward deforestation [FAO, 1993]. Shifting cultivation involves the annual clearing of small plots of forest, and creates a mosaic of cropped fields (often with trees) and fallows, intermixed with secondary and mature forests.
 Clearly, some ambiguity exists in distinguishing different categories along this continuum of change. For simplification, we distinguished two categories of land use change: changes in the area of forests (deforestation and the establishment of plantations) and changes in the per hectare stocks of carbon within forests (degradation and enhancement). Ecosystems other than forests may also lose or accumulate carbon as a result of human activity, but the largest per hectare changes involve forests, and, according to the FAO definitions of forest (see section 2.1), nearly all lands are forests.
2.3.1. Changes in Land Use That Affect the Area of Forests
18.104.22.168. Rates of Deforestation Between 1980 and 2000
 Because of revisions in successive FAO Forest Resource Assessments (FRAs), we reconstructed the “FAO” rates from the most recent FRAs. First, we obtained total forest areas for 1990 and 2000 from FAO . The assessment emphasizes net changes in forest area (natural forests plus plantations), but it also gives plantation areas for the year 2000, allowing plantations and natural forests to be distinguished in 2000. From the 1990 FRA [FAO, 1995] we obtained the areas of plantations in 1990. We assumed that the revisions in total forest area between the 1990 and 2000 FRAs were largely for natural forests, and that the 1990 FRA estimate of plantation area was accurate. Matthews  used a similar assumption.
 Finally, to obtain an estimate of natural forest area in 1980, we used the revised 1990 natural forest area [from FAO, 2001], and increased it by the amount of deforestation reported for the 1980s [FAO/UNEP, 1981]. We assumed that rates of change were better known than total areas of forest, as the latter have been substantially revised in successive FRAs. We used the 1980 area in plantations as reported in the 1980 FRA [FAO/UNEP, 1981].
 It is noteworthy that the areas of plantations are different in the two publications reporting the 1990 FRA [FAO, 1993, 1995]. The estimates given by FAO  are ∼70% of those given by FAO  because, on average, only 70% of established plantations survived [FAO, 1993]. The areas used in this analysis are from FAO ; they include the 70% reduction. The difference is small in sub-Saharan Africa because the areas in plantations are relatively small, but the difference affects our reconstructed rates of deforestation. The estimated rate of deforestation (of natural forests) is 0.2 × 106 ha higher than it would be without the 70% correction to 1990 plantation areas.
 With these assumptions we reconstructed the areas of natural forests and plantations in 1980, 1990, and 2000. We also calculated average decadal rates of deforestation (natural forests), as well as average rates of plantation establishment for the periods between these dates. The procedure was applied to each of the four regions of sub-Saharan Africa.
22.214.171.124. What is the Purpose of Deforestation?
 The changes in carbon that accompany deforestation depend on the use of the deforested lands: permanent cultivation, shifting cultivation, pasture, or other use. For the years 1961–2000 we obtained annual changes in cropland and pasture areas from the FAO Statistical Database (FAOSTAT) (available at http://faostat.fao.org). The same source also reported areas of forests and woodlands and areas of “other lands” but only from 1961 to 1994.
 The areas of forest reported by FAOSTAT are very different from the areas reconstructed from the FRAs, and the changes in area are even more different (Figure 1). We believe that the different rates of change are explained by the ways the two estimates account for shifting cultivation. Apparently, the FRAs define the transformation of forests to shifting cultivation as deforestation, while FAOSTAT considers shifting cultivation (crops and fallow) as part of the forest system (unless the period of fallow become less than 5 years, in which case the lands are classified as croplands or arable lands). We used the difference between the two estimates of forest loss to define the rate at which forests were converted to shifting cultivation. We used estimates of the areas in fallow forests from the 1980 FRA [FAO/UNEP, 1981] to define the absolute areas in shifting cultivation at that time. For calculating changes in terrestrial carbon storage, the area in shifting cultivation, including fallow, is less important than changes in that area.
 In sum, we used rates of deforestation from the FRAs to define an updated “FAO” estimate of deforestation from 1980 to 2000; data from FAOSTAT to define changes in croplands and pastures between 1961 and 2000; and differences between the two estimates of forest change to define changes in the area of shifting cultivation (including fallows).
 The FAO estimate of deforestation is higher than other estimates (Table 1), and to construct alternative 2 we used rates of deforestation reported by Achard et al. . We not only reduced the rates of deforestation for croplands and shifting cultivation but also changed its distribution between humid and nonhumid forests to be consistent with the findings of Achard et al. .
126.96.36.199. Historic Changes in Land Use
 For the years preceding 1961 (that is, prior to the data of FAOSTAT), we used a general history of sub-Saharan Africa to estimate changes in land use. Before the 19th century, land in sub-Saharan Africa was used largely for hunting, gathering, herding, and shifting cultivation. Settled agriculture existed, but before the colonial period began late in the 19th century, demographic and economic needs were such that land cleared for cultivation could be left fallow for long periods as cultivators moved on and cleared new land. Estimates of cropland areas before ∼1900 are variable, in part because “croplands” were largely in shifting cultivation, where the distinctions between cropped areas, fallows, and forests are ambiguous. Nevertheless, net changes in cropland areas were probably small before ∼1900. In fact, the population of Africa is thought to have declined following European colonization late in the 19th century [Kimble, 1962]. Before colonization, agricultural lands are believed to have been in a “quasi-equilibrium” for thousands of years, increasing and decreasing as a result of wars, epidemics, famines, and slave trade (both intra-African and trans-Atlantic). Despite the changes that colonization initiated, it was generally not until ∼1930 that the changes began to accelerate. By the 1930s, the railroads and most of the other major transport routes were in place, and it became feasible to begin development of areas that had hitherto been inaccessible.
 The expansion of croplands after ∼1930 was driven by two factors: population growth and European demand for export crops. Populations increased as a result of improved public health, as well as the absence of the wars, epidemics, and famines that had characterized the late 19th century [Kimble, 1962]. Export crops were expanded because colonial governments needed cash to recover from worldwide economic depression. This combination of demographic pressure and economic incentive has continued to the present despite independence from colonial rule by the 1960s.
 Assuming that land use change in Africa was variable (not unidirectional) before ∼1900, we defined changes between 1850 (the start of our analysis) and 1960 (the start of FAOSTAT) with the following assumptions, procedures, and rules: (1) The area in croplands was extrapolated back to 1900 exponentially, such that rates of change increased from 0 in 1900 to the rates reported by FAOSTAT in the 1960s. Cropland area was assumed not to have changed between 1850 and 1900. (2) The area in forests was extrapolated back to 1900 exponentially, such that rates of forest loss increased from 0 in 1900 to the rates reported for the 1980s by FAO/UNEP . Forest area was held constant between 1850 and 1900. (3) The area in pastures did not change between 1850 and 1961. (4) Changes in the areas under shifting cultivation, as discussed above for the period 1980–2000, were defined annually by the difference between forest loss and permanent cropland expansion. Forest loss always exceeded cropland expansion. The absolute area in shifting cultivation, including fallow, was defined by the 1980 areas in fallow reported by FAO/UNEP .
 Because past rates of deforestation are so poorly constrained, we constructed a pair of alternative assumptions for alterative 3. In one case we assumed a linear increase in forest loss between 1850 and 1925, after which the reference rates were used. In the other case we assumed an increase in forest area between 1880 and 1900 (to represent the abandonment of croplands and recovery of forests following population declines late in the 19th century) [Kimble, 1962]. The increase in forest area (1880–1900) was equivalent to the loss (1850–1925) in the first case of the alternative.
188.8.131.52. Which Forests Were Deforested?
 We distributed deforestation among ecological zones in each region on the basis of the zones deforested in the 1980s [FAO, 1993]. Within each region we calculated a “preference” index by dividing the percentage of the region's deforestation in each ecological zone by the percentage of the region's forest area in that zone. Thus, for example, if two zones had the same absolute rates of deforestation, the zone with the smaller forest area indicated a higher “preference” for deforestation. In general, we assumed that the preferences applied over the entire period (1850–2000), and, as a consequence, we had to modify the preferences for central Africa because there was not enough montane forest. Apparently the clearing of montane forests in central Africa was unusually intense during the 1980s. In all regions, moist tropical forests were the most preferred for land use (Table 5). Shrublands were deforested the least, followed by rain forests and dry forests. The creation of pastures on shrublands before 1850 was an exception to these preferences, but had no effect on carbon storage over the last 150 years. Pastures had to be taken from shrublands to meet the constraints imposed by (1) the area of potential ecological zones, (2) forest areas in 2000, and (3) the preferences observed during the 1980s (Table 5).
Table 5. Relative Preference for New Agricultural Land Within a Regiona
 The apparent “preference” of forest types in the 1980s reflected, in part, the distribution of forests at that time, and not necessarily the longer-term preference for drier conditions and more accessible forests. To some extent we captured this longer-term preference for drier forests in the reference analysis by taking most pastures from shrublands (before 1850) and by deforesting the humid forests of central Africa only in the most recent decades. Nevertheless, to test the effects of this assumption, in alternative 4 we based the “preference” for forest types on the long-term differences between natural (prehuman) forest area and forest area in 2000 (both estimates given by FAO ). Those forest types with the largest changes in area were assumed to have been preferred.
 In some regions, such as West Africa, plantations have been established to try to stop or reverse desertification. In other regions, such as southern Africa, plantations have been largely for timber production. In Zambia and Zimbabwe, for example, plantations met 50% of the countries' needs for industrial wood during the 1990s [FAO, 2001]. According to the FRAs, the area of plantations in sub-Saharan Africa grew from ∼1.8 × 106 ha in 1980 to ∼2.8 × 106 ha in 1990 to 8.036 × 106 ha in 2000. We extrapolated the average rate of change backward linearly, assuming that the first plantations were established in 1950.
2.3.2. Changes in Land Use That Affect Carbon Stocks Within Forests
184.108.40.206. Harvest of Wood
 We distinguished fuel wood from industrial wood because the types of forests exploited and the effects on carbon stocks are often different. In 2000, fuel wood accounted for about 89% of total wood use in Africa [FAO, 2001]. Nevertheless, we did not explicitly simulate harvest of fuel wood in the reference analysis. Rather, we assumed that fuel wood was supplied by (and already counted in) the process of deforestation for permanent and shifting cultivation. In dry forests with high population densities and around cities, shortages of fuel wood may occur, leading not only to reduced biomass, but to deforestation as well. We did not increase rates of deforestation to account for such losses but assumed they were captured in the reported rates of deforestation.
 As an alternative to this assumption, we constructed alternative 5, which included an explicit harvest of fuel wood in proportion to the area of cropland (assumed to be an index of population size). The total harvest of fuel wood from sub-Saharan Africa in 2000 was 235 TgC [Bailis et al., 2005]. We assumed a harvest rate of 3 MgC ha−1 yr−1 for each ecosystem, with regrowth rates similar to those of fallow forests (Table 4).
 For industrial wood, rates of harvest (m3/yr) were obtained from FAOSTAT . A wood density of 0.58 and a carbon content of 0.5 were used to convert m3 of wood to MgC. For the years before 1961 we extrapolated rates exponentially back to ∼1900 and assumed constant rates between 1850 and 1900. Although the FAO  reports areas harvested for industrial wood, the values are underestimates because they refer only to those forests with formal, nationally approved management plans, while industrial harvests also occur outside forests with management plans. Harvests are selective, removing only a fraction of the aboveground biomass per hectare [FAO, 1993].
 We did not have data defining which forests were harvested, and we distributed industrial harvests between rain forests and moist deciduous forests in proportion to the areas of those ecosystems in each region. We also set rules to specify the “age” of forests cleared and harvested. The four “ages” were undisturbed forest, secondary forest (75% of undisturbed biomass), mature fallow, and just-logged forest. For each type of land use (not only wood harvest) we defined which ages were available for use and the order in which they were available (Table 6).
Table 6. Forests Available for Different Types of Land Use Ranked in Order of Preference
Permanent croplands and pastures
mature fallow forests (defined by length of fallow); forests recently logged; secondary forests; primary forests
secondary forests; primary forests
Fuel wood harvests
not explicitly included in the reference analysis (see fractions burned in Table 4)
220.127.116.11. Other Activities That Change Carbon Stocks Within Forests
 Grazing and browsing, fires, and desertification all contribute to the degradation of forests, but data on the extent of such degradation are generally not available. Thus we did not explicitly include these forms of disturbance. Rather, we assumed that they were in a long-term equilibrium, in which sources and sinks of carbon were balanced. Clearly, this is not the case if these activities have either increased or decreased over time. In the absence of data, however, we assumed that the net flux of carbon from changes in these activities was small relative to the processes for which documentation exists.
 Wars, and especially mass movements of refugees, may also reduce the stocks of carbon on land, but the losses of carbon in one area may be offset by accumulations of carbon in other areas. In Angola and Mozambique (southern Africa), for example, wars led to local emigration, allowing abandoned croplands to revert to forests. After peace returned to Mozambique in 1992, deforestation accelerated [FAO, 2001]. We did not try to account for the effects of wars or refugees. The FRA estimates of deforestation presumably include such changes.
2.3.3. Other Activities: Trees Outside of Forests
 Not all trees are included in the FRA 2000 definition of forest or other wooded land. Trees outside the forests include trees in cities, around croplands, along roads and in many other locations, but, by definition, they are not in forests. Tree planting may also occur outside of plantations, and these trees may be extremely important in supplying fuel wood. As much as 30% of the wood resources of Burkina Faso, for example, are estimated to be in fallows and the sparse trees of agricultural lands [Jensen, 1995]. The use of these trees to supply fuel wood reduces the pressure on forests. We did not include in our analysis either the harvest or planting of trees outside of forests.
2.4. A Model for Calculation of Sources and Sinks of Carbon
 The model used in this analysis is the same bookkeeping model used in previous analyses [Houghton, 1999, 2003]. Briefly, the model is based on a series of response curves that define the per hectare changes in carbon that follow a change in land use (see section 2.2 and Table 4). The carbon initially held in biomass is released to the atmosphere either at the time of clearing or harvest, if wood is burned, or over a longer period, if wood is left to decay on site or is removed for wood products (with rate constants of 0.1 yr−1 for short-lasting products or 0.01 yr−1 for long-lasting products). The model also tracks the loss (and recovery) of soil carbon with cultivation (and in fallows).
 The model tracks only those hectares that have been cleared, abandoned, harvested, and planted. Intensification of land use and most forms of management other than these broad categories were not included. Ecosystems undisturbed by human activity were also ignored. Rates of growth and decay vary with land use and ecosystem (Table 4), but they did not change in this analysis as a result of changes in climate, CO2, or nutrient availability. Thus the calculated fluxes of carbon (gains and losses in biomass, soils, wood products, and woody debris) result only from changes in land use, including forestry.
2.5. Initializing the Model
 Logically, the sequence of this analysis would be as follows: initialize the areas and per hectare stocks of carbon in natural and managed ecosystems in 1850, reconstruct changes in land use between 1850 and 2000 (including both those that affect forest area and those that affect carbon stocks within forests), and use the model to calculate the transitions in area, the final areas in 2000, and the annual changes in regional carbon stocks (or net flux of carbon). However, estimates of forest area and carbon stocks in the year 1850 do not exist. They do exist for the year 2000, and thus the sequence of the analysis was as follows: document the areas of forest and the per hectare carbon stocks of natural and managed ecosystems in 2000, reconstruct changes in land use between 1850 and 2000, and use the 2000 areas, stocks, and the per hectare changes following a change in land use to calculate the initial (1850) areas and stocks. The model was run forward from 1850, but the data used to initialize the model in 1850 were derived from forest areas in 2000 and the changes in land use over the previous 150 years.
 In fact, the model was run from 1700 to 2000 to allow 150 years (1700–1850) for the areas in secondary forest and fallows, as well as the pools of carbon in woody debris and wood products, to equilibrate with background rates of disturbance. These background rates were the 1850 rates of land use change, assumed to apply between 1700 and 1850.
3.1. Changes in Land Use
 In the year 2000, pastures covered the largest area of sub-Saharan Africa, followed by forests (including dense forests (rain, moist, and montane forests), dry forests, and shrublands) (Figure 2). If the fallows of shifting cultivation are considered forests, forests covered the largest area. “Other lands” (largely deserts) also covered a large area of Africa but were assumed not to have lost or gained carbon as a result of land use change. Croplands and shifting cultivation, while extensive, covered only about 20% of the region in 2000.
 The long-term changes in areas of croplands, pastures, shifting cultivation (including fallows), and forests, as a result of the data and assumptions described in section 2.3, are shown in Figure 3. Pasture and “other lands” are omitted from Figure 3 because the changes in area were small. Over the period 1850–2000, croplands doubled in area from 86 to 173 × 106 ha, while the area in pastures decreased slightly from 800 to 795 × 106 ha. Forest area (including dense forests, dry forests, and shrublands) decreased by 211 × 106 ha (25%), from 845 to 635 × 106 ha. In 2000 most forests were still primary (undisturbed) forests in the reference analysis. The cumulative area logged (since 1850) for industrial (nonfuel) wood was 162 × 106 ha, almost twice the increase in croplands. However, the area in secondary forests (60 × 106 ha in 2000) was considerably less than the area logged because logged forests were reharvested and cleared for shifting cultivation (Table 6). The loss of forest area to other land covers was greater than the expansion of permanent croplands by 134 × 106 ha, representing, according to our interpretation, an increase in the area in shifting cultivation and fallows by 135%. According to the data and interpretation used in this analysis, the area in shifting cultivation increased more rapidly than the area in permanent croplands in the last two decades.
3.2. Changes in Forest Biomass
 Despite the reduction in total forest biomass, the average biomass of African forests increased slightly from 62 to 67 MgC/ha over the last 150 years, according to the reference analysis (Figure 4). All regions except central Africa showed a reduction in average biomass (Figure 4b). The reason for this modest increase in average biomass (in central Africa and the entire region) is that large areas of primary forests persisted. Logging, which should have reduced biomass, was significant, but secondary forests, according to the data and assumptions of the reference analysis, were recycled. They were reharvested and converted to fallows for shifting cultivation (Table 6). As a consequence, the areas of secondary forests did not increase, and average forest biomass did not decline. Although the cumulative area logged was equivalent to 25% of African forest area in 2000, the area of secondary forests in 2000 was considerably less than this cumulative area. The slight increase in average biomass resulted from the relative loss of forests with lower than average biomass.
 When fuel wood was explicitly harvested (alternative 5), the area in secondary forests increased dramatically, especially in the drier forest types (dry forests, shrublands, and montane forests). Nevertheless, because the removals and regrowth were largely offsetting, average forest biomass was reduced in the region by only 0.6 MgC/ha as a result of fuel wood harvest over the last 150 years.
3.3. Net Flux of Carbon
 Over the period 1850–2000, changes in land use caused a release of 12.8 PgC to the atmosphere according to the reference analysis (about 6% of the total carbon held in the region in 1850 (Table 7). The largest loss was from forest biomass: 9.8 PgC, but 3.8 PgC were lost from soils as a result of cultivation. In contrast, the pool of dead organic matter (slash), generated as a result of disturbance, increased by 0.3 PgC, and the pool of wood products increased by 0.5 PgC.
Table 7. Net Fluxes of Carbon From Different Types of Land Use Change, and Net Changes in Carbon Stocks
Type of Land Use
1850–2000 Total, PgC
1990–1999 Average, PgC/yr
Positive values indicate emissions of carbon to the atmosphere.
Conversion to croplands
Conversion to pastures
Conversion to shifting cultivation
Establishment of plantations
Deforestation and afforestation (changes in forest area)
Degradation (changes within forests)
Change in biomass
Change in woody debris
Change in wood products
Change in soil carbon
 The conversion of forests to croplands released 6.5 PgC, and the expansion of shifting cultivation released 5.6 PgC (Table 7). Harvest of timber (including decay of wood products) released only 0.8 PgC, while the establishment of plantations sequestered an order of magnitude less (0.03 PgC). Thus deforestation accounted for 94% of the total loss of carbon; degradation (from logging) accounted for 6%.
 The annual net flux of carbon began increasing after ∼1900 in all regions, with steeply accelerated emissions after the 1970s (Figure 5). During the 1990s the net annual flux averaged 0.287 PgC/yr in the reference analysis. In this decade the proportions of the flux attributable to deforestation and degradation were 90% and 10%, respectively.
 The long-term (1850–2000) and recent (1990s) estimates of flux for the alternative analyses are shown in Table 8. The largest differences from the reference analysis resulted from lower rates of deforestation (alternative 2) and higher estimates for biomass (alternative 1). Estimates of flux for the 1990s varied from 0.08 PgC yr−1 to 0.47 PgC yr−1. Alternative rates of deforestation and reforestation before 1950 (alternative 3) had no effect on current estimates of flux, and the alternative preference for forest types (alternative 4) had relatively little effect. Although the gross emissions of carbon from combustion of fuel wood (0.173 PgC yr−1 during the 1990s) were more than half of the reference estimate of net flux, the net flux from fuel wood, alone, was only 0.007 PgC yr−1. The net contribution was small because the emissions from fuel wood burning were largely offset by the accumulation of carbon in forests recovering from harvest.
Table 8. Emissions of Carbon From Sub-Saharan Africa According to Alternative Data and Assumptions Used in the Analyses
Alternative rates of historic deforestation and reforestation
Alternative “preference” for forest types
Explicit harvest of fuel wood
 We can estimate the total amount of carbon lost from the region before 1850 by assuming that the initial biomass values (Table 4) are representative of their respective ecological zones before disturbance. The areas of ecological zones were obtained from FAO . The calculation suggests that the initial amount of carbon in the vegetation of sub-Saharan Africa was 105 Pg, that it had been reduced to 64 Pg by 1850 and to 56 Pg by 2000 (Table 9). Thus the total long-term loss in vegetation was almost 50%, with most of that loss (∼80% of it) occurring before 1850. The long-term loss of carbon from soils was absolutely less (40 Pg) and relatively less (19% of initial stocks).
 Forests had already been lost and degraded in Africa before 1850 (perhaps well before 1850). According to the areas of ecological zones [FAO, 2001] and the changes reconstructed in this analysis, almost 60% of the forests had already been converted to other lands by 1850, and another 10% were converted in the last 150 years. Although the estimates are uncertain, they are consistent with the long history of human occupation, indeed, evolution. Charcoal and pollen evidence suggests that deforestation and cultivation appeared at least as early as 4800 years BP [Hamilton et al., 1986], and the use of fire was certainly in existence long before that. The extent of clearing before 1850, as inferred from FAO ecological zones, is much greater than inferred with other maps of potential forest area [DeFries et al., 1999], reflecting the uncertainty of ecosystem distribution (and carbon stocks) before humans.
 Although the long-term loss of forest area suggests that sub-Saharan Africa must have been a long-term source of carbon to the atmosphere, the magnitude, and even the sign, of the flux is uncertain for any specific century. The net flux is unlikely to have been zero in the decades before 1900, but it might have been either a source or sink (Figure 6). The current rates of deforestation, however, despite their uncertainty, are almost certainly higher than at any time in the past; and the net annual flux has almost certainly been increasing over the last 100 years.
4.2. What Are the Major Uncertainties?
4.2.1. Estimates of Deforestation
 Recent rates of tropical deforestation as reported by the FAO are higher than other recent estimates (Table 1). The estimate by DeFries et al.  is highly uncertain for Africa because the technique is not reliable in savannas, although, according to the FAO , most recent deforestation was not in savannas. Nevertheless, other satellite-based estimates are also lower. An independent satellite-based survey of deforestation carried out by the FAO yielded an estimate of 2.2 × 106 ha yr−1, which is considerably lower than the country survey approach used in this analysis (5.2 × 106 ha yr−1) [FAO, 2001]. An intensive study of a 41.6 × 106 ha area in central Africa [Zhang et al., 2005] found an annual deforestation rate (0.42% yr−1), which is similar to the FAO's remote sensing-based estimate (0.43% yr−1) for all of Africa [FAO, 2001]. The studies by Zhang et al.  and by Achard et al.  also measured degradation, suggesting that some forms of degradation are observable from space.
 There is not much question that satellite-based approaches yield lower estimates of deforestation than the ground surveys reported by FAO . The real question is: Which estimates are correct? It is difficult to determine the accuracy of the ground-based estimates. The remote sensing-based estimates are sensitive to two processes. One is the spatial variability of deforestation. Samples generally consist of entire Landsat scenes, and the variability among scenes may be so high as to require >80% coverage of a region for an accurate estimate of deforestation [Tucker and Townshend, 2000]. In contrast, the sampling by Achard et al.  was only 6.5%, after stratification on the basis of regional expert opinion. It is also possible, especially in Africa, that the size of clearings is too small for a change in tree cover to be recognized. The fact that some forms of forest degradation are observed from space suggests the small plot size may not be a problem, but few studies of land use change have documented the distribution of patch sizes (i.e., the change in area as a function of patch size). In many parts of Africa the size of individual clearings or plantings may be the size of individual tree crowns, not readily observable with 30 m resolution Landsat TM, and certainly not observable when the minimal mapping unit is 3 × 3 pixels (i.e., 90 × 90 m).
4.2.2. Estimates of Biomass
 Estimates of net carbon flux from land use change are as sensitive to uncertainty in tropical forest biomass as they are to uncertainty in deforestation rates [Houghton, 2005]. The values of forest biomass simulated in this study were adjusted to match the average regional values reported in the 2000 FRA [FAO, 2001] (Figure 4a). They are lower than the values reported by others [Brown et al., 1989; Gaston et al., 1998] and lower than the values used previously to calculate carbon emissions [Houghton, 1999, 2003; Achard et al., 2002, 2004]. The higher estimates seem representative of dense forests (rain, moist, and montane forests), excluding dry forests and shrublands. Thus the differences among estimates may be explained more by differences in definition than differences in data.
 The trouble with any average, however, is the possibility that deforestation and biomass are not independent. One might have expected the calculated emissions of carbon to have been lower in this analysis than in the previous analysis where average forest biomass was higher [Houghton, 2003]. On the contrary, the annual net flux of carbon was not proportionally lower (Figure 7). The reason seems to be that most deforestation was of high-biomass forests (primarily moist and montane forests) (Table 5). This observation confirms the obvious, but often overlooked, fact that average forest biomass is not a good indicator of carbon emissions from deforestation [Houghton, 2005]. The forest biomass that determines the net flux of carbon is the biomass of the forests actually deforested.
4.2.3. Estimates of Degradation
 According to this analysis, deforestation accounted for 94% of the carbon lost between 1850 and 2000 (Table 7), and degradation accounted for only 6%. The latter is much lower than other estimates of degradation in Africa. For example, the FAO  reported that deforestation and degradation accounted for 75% and 25%, respectively, of changes in African forests between 1980 and 1990. Gaston et al.  calculated that degradation released 32% more carbon to the atmosphere than outright deforestation. The difference may be explained by differences in definition. If the conversion of forests to the fallows of shifting cultivation is called degradation rather than deforestation, the proportion of the net flux explained by each process becomes 50% in this analysis (Table 7).
 Alternatively, differences may be explained by processes omitted from this analysis. The fact that our simulations ended in 2000 with large areas of primary forest suggests that we did not account for all of the activities that affect forests in this region. We may have underestimated forest use by assuming that fuel wood was obtained through deforestation, and not through independent harvesting. The emissions of carbon from burning associated with deforestation was 0.22 PgC/yr during the 1990s, somewhat higher than the 0.13 PgC/yr emitted from fuel wood use (460 million m3/yr harvested (FAOSTAT), wood density 0.58 Mg/m3, carbon content 0.5), but the collection of fuel wood is not restricted to areas being deforested, as our approach assumed. When we included fuel wood harvest explicitly (alternative 5), the area of secondary forests in 2000 was much increased. Gross emissions from fuel wood were 0.17 PgC/yr during the 1990s. Other estimates vary between 0.166 PgC/yr [Yevich and Logan, 2003] and 0.235 PgC/yr [Bailis et al., 2005]. The net flux resulting from including fuel wood harvest, however, was much less because emissions were largely offset by forest regrowth.
 Other activities that reduce forest biomass, but that we did not consider, include fires and livestock grazing and browsing. We may also have underestimated rates of industrial harvest. Finally, we may have underestimated the formation of secondary forests by assuming that croplands are permanent. Instead, if croplands are in a long-term system of fallow, with 2–10 ha of recovering (secondary) forest for every hectare in crops, the area of secondary forests will be greater than we calculated. That is, the net change in area of croplands misses the fact that both croplands and forests are turning over. The simulated areas of primary forest in 2000 suggest that we underestimated disturbance. Whether this underestimation of disturbance means we underestimated carbon flux, however, depends on whether the disturbances were balanced by recovery. Most of the forests of Africa have probably been used for hundreds of years [van Gemerden et al., 2003], and the net flux of carbon at any time will depend on whether the rates of disturbance are increasing or decreasing. More than likely, they have been increasing in recent decades, releasing carbon to the atmosphere.
 In sum, we estimate that uncertainties in deforestation rates and forest biomass (especially the biomass of the forests actually deforested) contribute approximately equally to the uncertainty in estimates of carbon flux (Table 8). Management activities that affect carbon stocks without changing forest area rank third in contributing to the uncertainty of flux.
 Over the period 1850–2000, the reference analysis yielded a total flux for Africa remarkably similar to the earlier estimate based on aggregated data (12.8 and 13.0 PgC, respectively), although the historical pattern of the flux was different (Figure 7). Annual emissions in the new estimate are higher in the first half of the period and lower in the second half. For the decade of the 1990s the new estimate of flux is 0.29 PgC yr−1; the earlier estimate was 0.35 PgC yr−1. The similarity of the estimates for sub-Saharan Africa is largely fortuitous and does not imply a high level of confidence. The error for the 1990s is likely ±70% (Table 8). The new estimate of 0.29 PgC/yr is based on an estimate of deforestation that may be high [FAO, 2001] (Table 1). On the other hand, the large area of primary forest remaining at the end of these simulations and the small contribution of forest degradation to the estimated flux suggest that our estimate may be low. The range of 0.3 ± 0.2 PgC/yr includes the values reported previously by Achard et al. [2002, 2004], DeFries et al. , and Houghton  (Table 1).
4.3. How Do the Results Change Global Estimates of Terrestrial Carbon Flux?
 The lower estimate of flux for sub-Saharan Africa changes the global estimate for the 1990s to 2.1 PgC/yr (instead of 2.2 PgC/yr) [Houghton, 2003]. Sub-Saharan Africa accounted for about 15% of the global net emission of carbon from land use change in the 1990s.
 The analysis was improved as a result of suggestions from two anonymous reviewers. Research was supported by grants from NASA's programs in Terrestrial Ecology (grant NAG5-13351) and Land Cover/Land Use Change (grant NNG05GD146).