Soil carbon stocks in ecoregions of Africa

The African continent plays a growing role in the carbon (C) cycle. However, Africa is one of the weakest links in our understanding of the global carbon cycle particularly when considering the soil compartment. Most of the soil organic carbon (SOC) estimates concern the global size of the soil C reservoir without indication on its distri5 bution, or if given, limited to the contribution of a large ecosystems or region. The aim of the study is i) to assess the original soil C stocks (SOC) of Africa for the different countries and Ecoregions, and ii) to compare SOC estimates from different soil properties databases and digital maps. Four recent global digital soil maps and five soil properties databases were used to estimate the SOC for different soil layers in Africa. 10 Those databases hold between 1799 and 4043 soil profiles which are considered to represent the soil units showed on a 0.5 by 0.5 degree soil maps of Africa. SOC of Africa ranged 133 420–184 116 Tg for 0–100 cm soil layer. The most recent databases estimated 166 397 Tg C which corresponded to 9% of the global SOC stock and 68% of the terrestrial C pool of Africa. Average SOC ranges 0.4–8.2 kg m−2 at 15 national scale and 1.53–6.61 kg m−2 for ecoregions. Using different soil database or spatial data leads to up to 30% of difference. These quantities indicate a great potential to develop sink activities through soil C sequestration activities.


Introduction
Carbon dioxide (CO 2 ) is, by far, the largest contributor to the anthropogenically en- 20 hanced greenhouse effect (IPCC, 2007a). The importance of CO 2 to the climate has provided the impetus for research on the global C cycle with particular attention on C stocks in main terrestrial compartments, mainly soils and phytomass. The increasing interest of the contribution of terrestrial ecosystems to mitigate climate change has given rise to the possibility of emission credits for soil organic carbon (SOC) seques-to mitigate the greenhouse effect and, simultaneously, combat land degradation (Lal, 2004).
Soil organic matter is a key component of any terrestrial ecosystem, and any variation in its abundance and composition has important effects on many of the processes that occur within the system. The magnitude of organic matter and soil carbon stock 5 result from an equilibrium between the inputs (mostly from biomass detritus) and outputs to the system (mostly decomposition and transport), which are driven by various parameters of natural or human origins (Schlesinger and Palmer Winkler, 2000;Amundson, 2001). The decrease of organic matter in topsoils can have dramatic negative effects on water holding capacity of the soil, on structure stability and compact-10 ness, nutrient storage and supply and on soil biological life such as mycorrhizas and nitrogen-fixing bacteria (Sombroek et al., 1993).
The carbon balance of terrestrial ecosystems can be changed markedly by the direct impact of human activities. Land use change is responsible for 20% of the global anthropogenic CO 2 emissions during the 1990s (IPCC, 2007b) and is the main primary 15 net C release in Africa, much of it through burning of forests (Williams et al., 2007). The impact of land use change varies according to the land use types. The clearing of forests or woodlands and their conversion into farmland in the tropics reduces the soil-carbon content, mainly through reduced production of detritus, increased erosion rates and decomposition of sol organic matter by oxidation. Various reviews agree that 20 the loss amounts to 20 to 50% of the original carbon in the topsoil, but deeper layers would be little affected, if at all (Sombroek et al., 1993;Murty et al., 2002;Guo and Gifford, 2002). Conversion of forests to pasture did not change soil carbon (Guo and Gifford, 2002) or may actually increase the soil organic matter content (Sombroek et al., 1993). Changes in soil carbon under shifting cultivation were half as large (Detwiler, 25 1986). Commercial logging and tree harvesting did not result in long-term decreases in soil organic matter (SOM) (Knoepp and Swank, 1997;Houghton et al., 2001;Yanai et al., 2003). Changes in the amount of soil organic matter following conversion of natural forests to other land uses depend on several factors such as the type of forest 799 ecosystem undergoing change (Rhoades et al., 2000), the post conversion land management, the climate (Pastor and Post, 1986) and the soil type and texture (Schjonning et al., 1999).
In Sub-Saharan Africa, the increasing demand for food can encourage farmers to reduce the length of fallow periods, cultivate continuously, overgraze fields, or remove 5 much of the above-ground biomass through fuel collection or for building materials. Such practices can result in the reduction of SOC, water holding capacity, nutrients, as well as enhance soil erosion (Lal, 2004). Nevertheless, appropriate land management could revert this trend and contribute to soil carbon sequestration. Increasing the SOM could help reverse these problems and may be crucial for future African agriculture and 10 food production (Bationo et al., 2007;Sanchez, 2000). Some operations that increase organic matter inputs such as reforestation of agricultural lands, improved fallow, reduced tillage or fertilization can increase SOM levels. Several studies have shown that a synergetic effect exists between mineral fertilizers and organic amendments in Sub-Saharan Africa, that leads both to higher yields and SOC content (Palm et al., 2001;15 Vagen et al., 2004;Bationo et al., 2007). However, soil protection or conservation decisions cannot be made without maps of land properties. If projects have to be implemented with the objective of enhancing the storage of carbon in the terrestrial biosphere, potentially for carbon trading related to the Kyoto protocol, it is important to have the capabilities of verifying and monitoring the 20 changes of soil organic carbon (SOC) over time and space (Post et al., 2001). While, Annex I country of the Kyoto protocol are committed to report national estimates of changes in the amount of soil organic carbon (SOC) in cropland, we can expect that in the future, all the countries will have to report their emissions.
Several attempts were developed to report the soil organic mass of the world. 25 The global soil C storage estimates over the past 70 years range 400-9120 Pg C (Amundson, 2001). However, most recent studies give estimates ranging from 1115 to 2200 Pg C in the first meter and generally converge on a value of about 1500 Pg (Post et al., 1982;Batjes, 1996;Schlesinger and Palmer Winkler, 2000) while plant biomass is estimated to range between 560 and 835 Pg C (Whittaker and Likens, 1975;Bouwman, 1990). Furthermore, the world's mineral soils represent a large reservoir of C of about two third of the global terrestrial C stocks. Soils contain nearly as much carbon as the vegetation under rainforest, but considerably exceed the biomass in other ecosystems, by a factor 2 to 10 (Sombroek et al., 1993). According to Sombroek 5 (1993), this large variation of SOC estimates between authors is explained by the difference in base maps selected (FAO/Unesco Soil Map of the World, Holdridge Life Zones, Vegetation Maps) and the various assumptions made about soil attributes directly related to the total content of organic-matter per soil type such as organic matter content and distribution in the soil profile, bulk density of soil layer, and average soil depths. Nonetheless, the size and dynamics of the soil carbon pool of the world, it is still poorly known (IPCC, 2007b). Uncertainties concerning (1) estimates of C in ecosystems, (2) per hectare changes in C stocks in response to different types of landuse change and (3) legacy effects, are important particularly in Africa (Houghton and Goodale, 2004). There are great needs to improve the understanding comprehension 15 of the C cycle in Africa. Most of the time, the reported values consist of estimates of the global size of the soil C reservoir without indication on its distribution, or if given, limited to the contribution of a large ecosystems or region. Models of the global C cycle require accurate estimations of the masses in the different reservoirs. Regarding the soil compartment, 20 global C pools are difficult to estimate because of still limited knowledge about specific properties of soil types (Sombroek et al., 1993;Batjes, 1996), the high spatial variability of soil C even within one soil map unit (Cerri et al., 2000), and the different effects of the factors controlling the soil organic C cycle (Pastor and Post, 1986;Parton et al., 1987). Thus, regional studies are being proved necessary in order to refine global esti-25 mations obtained by aggregation of regional estimates, mainly at country scale (Milne et al., 2007). Aggregate of more precise studies, at a country or regional scale, certainly would improve the global estimates. But only few and not homogeneous (different layers used to report results) data exists.

801
In order to provide reliable predictions, successful global carbon cycle models must accurately represent fundamental controlling processes and conditions in robust model structures that can be validated by comprehensive data sets collected over a wide range of controlling environmental conditions. However, global observational data sets of SOC are currently incomplete and large-scale validating data sets are needed in 5 global carbon cycle models (Bombelli et al, this issue). Models of global climate change need accurate and complete SOC inventories because the SOC pool represents the largest component of the global C pool and acts as a regulator of atmospheric CO 2 levels (Amundson, 2001).
The objectives of this study are 1) to assess the soil C stocks of Africa for the different 10 countries and ecoregions, and 2) to compare SOC estimates from different digital map and soil properties database.
The Digital Soil Map of the World (DSMW -FAO, 1995), the ISRIC-WISE and ETOPOS-DSMW spatial data are 0.5 by 0.5 degree grid version of the edited and corrected 1:5M Soil Map of the World. The spatial data used for Harmonized World Soil Database is a mosaic of the Digital Soil Map of the World and the SOTER regional studies (Northeastern Africa, and Southern Africa). The numbers of maps units within 5 those four spatial data for Africa are 1598, 7964, 1588 and 6994 respectively. The FAO-UNESCO (1974) legend was used to aggregate available soil profile data and to link derived interpretations of soil properties with the soil units on the grid maps.

Calculation of carbon stock
The estimates correspond to the SOC present in the layer after correction using the 10 volume percentage of fragments coarser than 2 mm. SOC was estimated using the following formula: where C is carbon density (kg m −2 ), V is the volume of soil per square meter (m 3 ), Gr is the volume of gravel (m 3 ), Bd is the bulk density (kg dm −3 ) and Cc is the carbon 15 content (g C kg −1 ). Organic carbon content is universally determined by oxidation of CO 2 , and is directly measured as CO 2 or by weight loss of the sample or by back-titration of the excess of the added oxidant. Bulk density is critical for converting organic carbon percentage by weight to content by volume, but it varies with the structural condition of the soil, 20 in particular the mineralogy, water content and packing. In general, bulk density is determined by the core sampling method which is comparable with values obtained by the clod method. The mean bulk density was computed for each profile and when there was no bulk density for a particular soil profile the mean bulk density of the corresponding soil unit was used in the derived database. Gravel is not considered to 25 contain organic carbon (Anderson and Ingram, 1993) and its volume is subtracted from the total volume. 803

Computation of soil carbon stocks
The soil properties are given per soil classes. The spatial data are identified as map units. Within one map unit different soil classes could be found. In order to report the soil properties of one map unit, the SOC estimate was calculated according to the relative proportion of all soil classes per soil layer. The soil composition of map units is 5 homogenized per soil profiles types. Furthemore, the calculation of C density (kg m −2 ) for each soil profile results from the soil properties per soil classes and the composition of soil classes per layer. For an individual soil profile with k layers, the organic carbon by volume is: where C d is the total amount of organic carbon (kg m −2 ) over depth, Bd i is the bulk density (kg dm −3 ) of layer i , Cc i is the carbon content (g C kg −1 ) of the layer i and Gr i is the volume of gravel (m 3 ) of the layer i .
The C stock representative for each soil mapping unit was determined by: where M ud is the total mass of organic carbon (kg of C) held in the upper d cm of the soil, n is the total number of map unit, A is the area of the map unit (m 2 ), C i d is the carbon density of the soil profile (kg m −2 ). It is assumed that each soil profile is homogenous in each map unit.
Soil carbon mass for one biome, one country or the continent is determined by ag- 20 gregating the soil carbon stock of the corresponding map units: where M d is the total mass of carbon (ton of C) held in the upper d cm of the soil and j is the number of map units within one biome, country or the continent.

The soil carbon map of Africa
Five different soil carbon maps for Africa were obtained using the different data sources (Table 1). It represents the evolution of the available data for the years 1996, 2002, 5 2005, 2006 and 2008. While the first map was obtained using the soil properties reported by Batjes (1996) and the digital soil map of the world (FAO, 1995), the most recent and precise map is considered to be the Harmonized World Soil Database (FAO/IIASA/ISRIC/ISSCAS/JRC, 2008). The digital soil map units were linked to the soil properties in order to represent the spatial distribution of the C stocks.

10
Each map unit is linked into a geographic information system to it respective soil profile type and soil properties using ArcGis 9.2.

Calculation of the carbon stocks per biomes and per country
Various biome classification could be found for Africa (Holdridge, 1947;White, 1981;FAO, 1999;WWF, 2000;Olson et al., 2001). The Terrestrial Ecoregions of the World Average SOC was 2.37 kg m −2 and 4.60 kg m −2 for 0-30 and 0-100 cm soil layers, respectively. However, SOC varied between countries. Results for each country are reported in Table 2. The maximum SOC density was found in Congo (8.20 kg m −2 ) while the minimum SOC density was found in Western Sahara (0.40 kg m −2 ). Most of the C stock 10 in Africa (0-100 cm) was located in Middle Africa (33%) while only 6% was located in southern Africa. soils. On the other hand, only 26.6% was contained in 0-30 cm soil layer in Histosol soils. In average, 52% of the SOC contained in the whole profile was located in 0-30 cm soil layer. While the horizon 0-30 cm is considered to be the most important SOC pool, it is also the most susceptible to be affected by land management (Bernoux et al., 2006).

African SOC and the global cycle
Global carbon stocks were also calculated with the DSMW and the HWSD database (Fig. 2). Using the DSMW, these stocks were respectively 803 Pg, 1101 Pg, 1589 Pg and 2521 Pg of C for the layer 0-30 cm, 0-50 cm, 0-100 cm and 0-200 cm. Using the HWSD, the SOC of the world represented 814 Pg and 1850 Pg for 0-30 and 0-5 100 cm, respectively. Africa represented 8.6 and 9% of the SOC in the 0-100 cm soil layer according to the DSMW and the HWSD database respectively. Those results are lower than the 13% reported by Williams et al., 2007). On the other hand, Williams et al. (2007) reported that live plant carbon contained 610±47 and 80±28 at global scale and in Africa respectively. It means that 75 and 68% of the C stocks are contained in 10 the 0-100 cm soil layer in the world and in Africa, respectively.

Soil C stocks in ecoregions of Africa
The corresponding stocks and areas for each biome are reported in Table 3 (2001) reported that cultivation reduces the original soil C content by 30% but it was not mentioned under which ecological region. Impact of land use change depends on biomass productivity, mainly on ecosystem types, structure and composition, the soil substrate and the climate (Keeling and Phillips, 2007) 5 and management, mainly fertility management and tillage (Shepherd and Soule, 1998;Reicosky et al., 1999). Moreover, estimating the impact of land use change should consider the ecology and the soil types, the management and the farming practices for each ecological regions.

Variation of soil C stocks of Africa estimates
10 Table 4 reports the SOC estimates using four spatial data and five soil databases. It represents the evolution of the SOC estimates based on the most precise data for 1996,2002,2005,2006 and 2008. The SOC estimates for 0-100 cm soil layer of Africa, based on the soil databases and spatial data reported in  (Sombroek et al., 1993). It is important to highlight that SOC estimates can be very imprecise mostly for the small islands i.e. Cape Verde 20 or Canarias. However, the coastal boundaries that were used to correct the spatial data were the same. Furthermore, the difference of SOC estimates is attributed to the spatial scale of the spatial data and the soil properties of the soil database.
The data contained in the soil databases have changed over time. The main differences observed when estimating SOC stocks are (1) the number of soil profiles, (2) the 25 data of organic content, bulk density and gravel, (3) the soil layer that are considered. Increasing the number of soil profiles and the spatial resolution of the soil maps should increase the SOC estimates. Since 1995 the number of soil profiles increased from 808 4353 to 9607 (Table 1). The soil properties changed between databases and led to different SOC density estimates for an identical soil type. For example, Lithosols contained 2.7, 0, 4.1, 3.6 and 4.5 kg C m −2 in the databases 1996, 2002, 2005, 2006 and 2008, respectively. While the database of 2005 considered Lithosols as rocks and were included into the miscellaneous, Lithosols were considered only for the 0-10 cm 5 layer in the database of 2006 and 2008. Acrisols contained 6.50, 4.83, 6.83, 6.80 and 6.05 kg C m −2 in the databases 1996, 2002, 2005, 2006 and 2008, respectively. Small difference in the soil properties can induce important errors when considering the continental scale. The soil properties estimates have to be accurates and consider the spatial variability of SOC. When changing scale, few soil profiles are available for 10 Africa and are not randomly distributed. It is moreover difficult to increase the precision of the soil properties estimates without increasing the precision of the spatial data and vice versa. Both spatial and soil properties improvements have to be achieved to increase the precision of the SOC estimates at continental scale. The soil layers are also different between the databases. While the SOC estimates 15 of Batjes (1996) reported SOC for four soil layers, and Batjes (2006)

Error of SOC estimates
In order to assess the error due to the soil properties estimates, five maps were built using the same digital map (FAO, 1995) and different soil databases (Table 5). The SOC estimates ranged 211 956-117 356 Tg C. It appeared that the highest SOC estimate was obtained using the most recent HWSD database. When comparing the SOC esti-25 mates to the mean SOC estimates it appeared that the SOC estimates ranged +32% and −27%.
When estimating the error due to spatial data it was not possible to compare SOC 809 from the different maps using the same database because the spatial data uses different legends. The error due to spatial data was estimated making a comparison with the SOC estimates from Tables 4 and 5. The SOC estimates using the same soil properties and different soil maps varied a lot (Tables 4 and 5): i.e. the SOC estimates using the soil properties from HWSD and the spatial map from DSMW and the HWSD were 5 166 450 and 211 956 Tg. It appeared that the percentage of error due to spatial data ranged −30% to +27%. It means that the spatial data could over or underestimate the SOC estimates at continental scale of about 30%. When developing or monitoring land use change project, SOC data should be available for regional, national or local scale. According to our results we can assume that the estimate of SOC, based on the soil map of the world from FAO and IIASA and the soil properties from the WISE and the HWSD, could vary of up to 60% when considering the spatial and soil properties data. When considering national scale studies, SOC estimates vary also (Table 6). For instance, SOC assessment of Congo was estimated 3300 Tg of C by Schwartz and Namri (2002) while 9300 Tg of C were es-15 timated in this study for 0-100 cm soil layers, respectively. But Schwartz and Namri (2002) based their calculation on only ca. 90% of the country area due to lacking information for specific landscape units, mainly in Northern areas, and concerning mostly hydromorphic soils and in a less extent peats; both soil types usually with high level of OC. Moreover Schwartz and Namri (2002) recognized that small areas of Podzols with 20 humic horizon containing considerable quantities of OC (up to 120 kg m −2 ) could not be taken into account in their established procedure of calculation. SOC estimates of Benin ranged 251-260 and 526-543 Tg C in Volkoff et al. (1999) while the estimates of this study were 374 and 687 Tg C for 0-30 and 0-100 cm soil layers, respectively. In addition, average SOC density was estimated to be 3.5 kg C m −2 in this study while for 0-100 cm layer for Benin and 181% for Congo. The difference between estimates could be explained by (1) imprecision in the boundaries of the territory that lead to an overestimation of the area of about 12.4%, (2) lacking information for specific soil type, and (3) imprecision of SOC content which increased with soil depth. We observed that SOC estimates for Kenya (Batjes, 1996) were quite similar (% of error less than 5%) 5 which could be explained by the same origin for soil characteristics. Reliability of the information presented here is variable. While some parts of the world are highly reliable (Southern Africa, Latin America and the Caribbean, Central and Eastern Europe), other part of the world are considered less reliable (North America, Australia, West Africa and South Asia). Most of the improvement of the compre-10 hension of the soil compartment at global scale could be attributed to the development of the SOTER database that already covers Africa: Northern Africa, Southern Africa, Kenya and Central Africa. However, in order to improve the estimates of the soil biophysical properties and the impact of anthropic activities, the data on spatial resolution of the soil map units and the land use units, the proportion of soil types within on soil 15 map units, the soil properties and the stock change factors for different management activities have to be improved.

Conclusions
There are uncertainties in any estimate of soil carbon stocks at national and continental scale. The SOC estimates using the Harmonized World Soil Database clearly 20 overestimated the SOC for the different layers when comparing with other national studies. When comparing different soil databases and soil maps it appeared that SOC estimates ranged 133 420-184 116 Tg for 0-10 cm soil layer of Africa. On the other hand, a variation of about ±30% was reported when comparing the soil properties or the spatial data of the different databases. With an estimate of 166 397 Tg for the 25 0-100 cm, soil appeared to contain 68% of the terrestrial C pool of Africa. Most of the CO 2 emissions in Africa is due to land use change that affect both aboveground and belowground compartments. It is difficult to estimate the amount of C that could be 811 released from the soil as no data exist on SOC before conversion to agriculture and the impact of anthropic activities on the different soil types is not known. Adoption of recommended management practices can result in important C sequestration particularly in degraded/marginal soils. Soil degradation and food security are among major challenges in sub-Saharan Africa. Commoditisation of SOC through trading C credits 5 under the Kyoto's CDM could improve livelihoods and provide incentives for exchanging soil quality and restoring degraded soils and ecosystems.