Impacts of intensifying or expanding cereal cropping in sub‐Saharan Africa on greenhouse gas emissions and food security

Cropping is responsible for substantial emissions of greenhouse gasses (GHGs) worldwide through the use of fertilizers and through expansion of agricultural land and associated carbon losses. Especially in sub‐Saharan Africa (SSA), GHG emissions from these processes might increase steeply in coming decades, due to tripling demand for food until 2050 to match the steep population growth. This study assesses the impact of achieving cereal self‐sufficiency by the year 2050 for 10 SSA countries on GHG emissions related to different scenarios of increasing cereal production, ranging from intensifying production to agricultural area expansion. We also assessed different nutrient management variants in the intensification. Our analysis revealed that irrespective of intensification or extensification, GHG emissions of the 10 countries jointly are at least 50% higher in 2050 than in 2015. Intensification will come, depending on the nutrient use efficiency achieved, with large increases in nutrient inputs and associated GHG emissions. However, matching food demand through conversion of forest and grasslands to cereal area likely results in much higher GHG emissions. Moreover, many countries lack enough suitable land for cereal expansion to match food demand. In addition, we analysed the uncertainty in our GHG estimates and found that it is caused primarily by uncertainty in the IPCC Tier 1 coefficient for direct N2O emissions, and by the agronomic nitrogen use efficiency (N‐AE). In conclusion, intensification scenarios are clearly superior to expansion scenarios in terms of climate change mitigation, but only if current N‐AE is increased to levels commonly achieved in, for example, the United States, and which have been demonstrated to be feasible in some locations in SSA. As such, intensifying cereal production with good agronomy and nutrient management is essential to moderate inevitable increases in GHG emissions. Sustainably increasing crop production in SSA is therefore a daunting challenge in the coming decades.


| INTRODUC TI ON
Globally, agriculture is estimated to be responsible for substantial emissions of greenhouse gasses (GHGs), that is, ca. 12% directly, for example, through methane emissions from livestock and rice production and N 2 O emissions through the use of fertilizers (Barker et al., 2007;Smith et al., 2007;Vermeulen, Campbell, & Ingram, 2012) and another ca. 15% indirectly through land use conversion to increase agricultural production (Barker et al., 2007;Van der Werf et al., 2009;Vermeulen et al., 2012). Currently, agricultural production and associated agricultural GHG emissions are relatively low in sub-Saharan Africa (SSA) in comparison to other parts of the world (FAO, 2019). However, GHG emissions from agriculture are expected to increase in this region, as food production needs to rise in the coming decades to keep up with the strongly growing food demands (Smith et al., 2007). Until the year 2050, the cereal demand is projected to more than triple relative to 2010 due to population increase and dietary changes . SSA has already seen a continuous increase in emissions from agriculturedriven deforestation between 1990 and 2015 (Carter et al., 2017), as agriculture is considered the dominant driver of deforestation (Curtis, Slay, Harris, Tyukavina, & Hansen, 2018). The central aim of the Paris COP21 Agreement-adopted by 195 nations-is to keep global warming below 2°C and to pursue efforts to stay within 1.5°C above pre-industrial levels (IPCC, 2018), while the sustainable development goals emphasize that climate change mitigation will have to go hand in hand with achieving food security (United Nations, 2016). This will be challenging as it has been argued that for SSA, the effects on food security of stringent climate change mitigation measures may be larger than the effects of climate change itself (Hasegawa et al., 2018).
Over the past decades, food production in SSA has been increased by significant area expansion and slowly increasing yields (FAO, 2019). This led to large increases in GHG emissions, mainly due to the land use change (Bennetzen, Smith, & Porter, 2016). It is often stated that intensification on existing cropland is the preferred way to go in terms of biodiversity loss and GHG emissions (Cassman, 1999;Cassman, Dobermann, Walters, & Yang, 2003), as intensified agriculture has in general the lowest emission per unit of product produced (Bennetzen et al., 2016). More specifically, sustainable intensification of crop production by narrowing the yield gap, that is, the gap between actual farmers' yields and potential yield (Van Ittersum et al., 2013), is proposed. Nutrient limitation is amongst the main causes of yield gaps in SSA (Kassie et al., 2014;Sanchez, 2015).
Current nutrient inputs are low in SSA (FAO, 2019), and intensification will therefore require substantial increases in (nitrogen) fertilizer input (ten Berge et al., 2019). Nitrogen is, in quantitative terms, the most important crop nutrient and its production and use is strongly related to emissions of the GHGs CO 2 and N 2 O. As a result of this increased fertilizer input, GHG emissions will increase irrespective of whether mineral or organic fertilizers are used (Palm et al., 2010).
Today's nutrient use efficiencies in SSA remain low, potentially aggravating emissions from increased fertilizer use (Reay et al., 2012).
A relevant question is thus how GHG emissions are related to different scenarios of increasing food production, ranging from intensifying production to agricultural area expansion. We address this question by assessing GHG emissions from cultivation of five main cereals in SSA (i.e. maize, millet, rice, sorghum and wheat), considering different combinations of intensification and crop area expansion to achieve cereal self-sufficiency by the year 2050. Ten countries across SSA are included in this analysis (i.e. Burkina Faso, Ethiopia, Ghana, Kenya, Mali, Niger, Nigeria, Tanzania, Uganda and Zambia), representing 52% of the population and 58% of the cropland area in SSA (FAO, 2019). In addition, for maize cultivation, the most important cereal in SSA, we assess (a) the influence of agronomic nitrogen use efficiency (N-AE, i.e. extra grain yield per kg of N applied) on GHG emissions, where N-AE expresses the level of agronomic management; (b) the optimal balance between intensification and crop area expansion with respect to GHG emissions; and (c) how these analyses are affected by uncertainties in modelling parameters.

| Four scenarios which achieve cereal selfsufficiency in 2050
In this study, we used four scenarios to assess GHG emissions as affected by agricultural intensification or area extension for org; Grassini et al., 2015; as also used in van Ittersum et al. (2016).

| Cereal yields and cereal areas required for self-sufficiency
For each of the four scenarios, we take the view that individual countries aim for self-sufficiency in cereal production in 2050. While we acknowledge that full self-sufficiency of cereals is generally not an explicit aim, it is generally agreed that substantial dependence on food imports is only possible if economic development is sufficient to afford them, while economic development of low-income countries to support such imports requires a strong agricultural development (Chang, 2012;Johnston & Mellor, 1961). Here, we briefly describe how calculations regarding self-sufficiency were performed; a complete overview can be found in van Ittersum et al. (2016).
Self-sufficiency is calculated as the ratio between domestic cereal production and cereal demand and is equal to 1 to obtain full self-sufficiency. Cereal demands for 2050 were derived from recent population projections (medium fertility variant of the UN population projections; United-Nations, 2015), and per capita consumption (Robinson et al., 2015). The predicted per capita cereal consumption includes direct human consumption of cereals, but also cereals as used for animal feed and other purposes like bioenergy and brewing.
Predicted increased consumption per capita in Robinson et al. (2015) is similar to other forecast studies such as that of Alexandratos and Bruinsma (2012). In all forecast studies, predicted increase in cereal demand for 2050 compared to 2015 is mainly determined by the population growth in SSA, and to less extent to the increased per capita consumption of livestock products (Alexandratos & Bruinsma, 2012;van Ittersum et al., 2016). We expressed production and demand data at standard moisture content (maize: 15.5%; rice, sorghum, millet: 14%; wheat: 13.5%).
Each scenario has a different intensification level, with corresponding cereal yield and necessity to expand the production area to achieve self-sufficiency. We assumed that higher levels of intensification come with higher cereal yields per hectare, and therefore, less area needs to be converted to agricultural land to obtain full selfsufficiency. In our scenarios, cereal area expansion occurs through conversion of present grassland and/or forest (in proportion of current availability per country), and it is assumed that the productivity of newly converted land is equal to that of existing cereal land. Data on current areas of cereals, grassland, and forest were obtained from The potential area available for expansion of crop production was taken from Chamberlin, Jayne, and Headey (2014), who considered per country the land area suitable for cropland expansion as land which is currently not cultivated, not part of a national park or other gazetted area, not of a low-yield potential and with a low population density. The potential cereal area for expansion was the land area suitable for cropland expansion multiplied by the current share of cereal land in the total cropland.

| Minimum N input requirement (high-efficiency variant)
In the high-efficiency variant, N input for each of the scenarios (S1-S4) is estimated by the minimum nutrient input approach. It is postulated that the annual application rates of macronutrients (N, P, K) should at least be equal to total nutrient uptake in the aboveground crop biomass (grain and stover) of a given target yield, Y T .
The nutrient input requirements for a given Y T are calculated from: (a) coefficients to express physiological efficiency (kg grain per kg uptake), uptake efficiency (kg uptake per kg applied) and agronomic efficiency (kg grain per kg applied) of each nutrient (Table S2)

| N application rate under current mean N-AE in SSA (low-efficiency variant)
In the low-efficiency variant, N input is estimated by the short-term nutrient input requirement, and assuming an initial N-AE of 14.3 kg/kg. This value corresponds with the mean value currently found for maize in on-farm field trials or on-station experiments in SSA (ten Berge et al., 2019). Thus, our N input requirement first increases linearly with target yield according to the SSA-mean N-AE, and then increases more steeply for relative target yields exceeding 0.62. This variant does not assume a steady-state equilibrium of soil N but accounts, instead, for current soil N supply. This method is therefore more compatible for short-term assessments. Soil N uptake was estimated in two steps: (a) extrapolation of current actual N inputs and yields with N-AE (14.3 kg/kg) to obtain yield at zero N input; (b) from this yield level at zero N input, the crop N uptake from soil was calculated using the physiological efficiency (Section 2.4).

| GHG emissions
Total GHG emissions from cereal production consist of emissions from land (forest or grassland) conversion to cereal area, the use of fertilizer, the production of mineral fertilizer and from flooding of rice fields. GHG emissions from removal of crop residues are not included. All emissions were converted to CO 2 equivalents. For GHG emission calculations and parameter values, we used the IPCC tier 1 approach, unless it is specifically indicated that another data source was used (Tables S1 and S2).

| CO 2 emission from land use change
Emission from land use change from either forest area or grassland to cereal cropland in the year 2050 is the total emission due to change in soil organic carbon (SOC) content, removal of forest biomass, and/or removal of grassland biomass. Data on the fraction of forest and grassland area in the specific country are obtained from FAOSTAT (FAO, 2019). We assume that the land use change from 2015 until 2050 is linear, and that the default time period for transition between equilibrium SOC values is 20 years (IPCC, 2006b). This means that the total forest or grassland area which needs to be converted to cropland to obtain full self-sufficiency in 2050 is equally distributed over the years.
The CO 2 emission from removal of forest is the aboveground carbon content of the forest times the land use change area (discounted over 20 years) and the proportion of forest of the total forest and grassland area in the specific country. We obtained aboveground forest biomass (AGB) per country by combining a forest cover map (forest defined as more than 10% tree cover; Hansen et al., 2013) with a biomass map (Zarin et al., 2016). Both maps have a resolution of 30 m, and are dated circa 2000. Average AGB was converted to aboveground forest carbon content by using a conversion factor of 0.5.
The CO 2 emission from the removal of grassland is similarly defined, namely the aboveground biomass of the grassland times the land use change area (discounted over 20 years) and the proportion of grassland of the total forest and grassland area in the specific country.
The CO 2 emission from SOC loss due to land use change depends on the SOC before conversion minus SOC after conversion. SOC before conversion is the total carbon content from the forest minus the aboveground carbon content. The total carbon content of the forest was obtained from conversion of the total biomass content of the forest using a conversion factor of 0.5. The total biomass map was derived from the AGB density map by applying the equation: Total biomass = AGB + 0.489AGB 0.89 (Saatchi et al., 2011). SOC stock after land use conversion was obtained by multiplying the SOC stock before conversion with the relative stock change factors for cropland for land use, tillage and inputs used (Tables S1 and S2).

| N 2 O emission from fertilizer use
Total emission from nitrogen input is composed of direct N 2 O-N emission from applied fertilizer, indirect N 2 O emission through NH 3 and NO x volatilization and indirect N 2 O-N emission from leaching and run-off. For brevity, we assumed that all nitrogen inputs come from mineral fertilizer only (see Section 2.3).
The direct N 2 O-N emission from fertilizer application was estimated as the mineral fertilizer N applied multiplied by the emission factor for direct N 2 O emission. The indirect emission of N 2 O-N by volatilization of N as NH 3 and NO x , was estimated as the mineral fertilizer N applied multiplied by the fraction of NH 3 and NO x volatilized and the emission factor for N volatilization. The indirect emission of N 2 O-N by leaching and run-off from land of N was estimated as the mineral fertilizer N applied multiplied by the fraction of N leached and run-off and the emission factor for leaching and run-off (Table S1).

| CO 2 emission from production of mineral fertilizer
The CO 2 emission from the production of mineral fertilizer was calculated as the amount of mineral fertilizer of a specific type multiplied by the emission factor for that specific type of fertilizer. We took world average emission factors for all fertilizer types (Table S2).
The types of fertilizer used are based on FAOSTAT (FAO, 2019), and were assumed to remain the same in 2050 compared to 2015.

| CH 4 emission from rice fields
The CH 4 emission from rice cultivation was calculated as the default world CH 4 emission constant multiplied by a scaling factor for either irrigated or rainfed cultivation (Table S2).

| Negative emissions
We assumed that cereal area is converted to either forest and/ or grassland area when the land area required to obtain full selfsufficiency of cereals in 2050 is less than the current cereal area (country by scenario). This means uptake of CO 2 (indicated with negative emissions) due to change in SOC content, sequestration of carbon in forest biomass, and/or sequestration of carbon grassland. Recovery of SOC of forest until equilibrium takes also 20 years similar to what was assumed for SOC breakdown (Guo & Gifford, 2002), but for SOC of grassland, it was assumed to take 53 years (Guo & Gifford, 2002). Recovery of grass biomass was assumed to take 20 years, and natural regeneration of forest 100 years, but carbon increment is fastest in first 20 years and differs per climate type (Albanito et al., 2016).

| Uncertainty analysis
For each scenario, total GHG emissions were computed based on estimates of the model parameters. Therefore, any uncertainty in the model parameter values leads to uncertainty in the predicted emissions. To assess this prediction uncertainty, we performed an uncertainty analysis.
The uncertainty of the model parameter values is expressed in terms of a probability distribution for each parameter (Table S1). To assess the resulting uncertainty of model predictions, we drew 1,000 samples from parameter space using a replicated Latin hypercube design (Pleming & Manteufel, 2005). For each sample, we computed the resulting model predictions. These predictions were summarized in terms of the mean and a tolerance interval containing 95% of all model predictions.

| Sensitivity analysis
We performed a sensitivity analysis based on the same samples that were used for the uncertainty analysis. Sensitivity analysis is aimed at decomposing the uncertainty of model predictions into terms that are attributed to the model parameters. Thus, sensitivity analysis helps us to identify influential parameters.
The decomposition is based on a regression function with the model parameters as independent variables and the model predictions as dependent variable (Jansen, WaH, & Daamen, 1994). The regression function was acceptable when a fit of R 2 > 0.9 was ob-  (Jansen et al., 1994).

| GHG emissions from intensification and land use conversion-high-efficiency variant
Analysis revealed that cereal intensification to 80% yield gap closure (Scenario 4) will require an enormous increase in nitrogen (N) application per hectare of at least 16 times the current use in SSA (Table 1). Current N input is very low (< 10 kg N/ha) (Table 1), and therefore does not even compensate for N offtake in actual yields. We estimate that just to sustain current yields, N input per hectare has to increase by almost a factor 4 (Scenario 1).
The increase in N application as a result of cereal intensification through yield gap closure will come with substantial GHG emissions ( Figure 1). The fertilizer-induced emission (grey bars in Figure 1) consists for the largest part of direct N 2 O emission from soils and CO 2 emission due to the production of fertilizer, respectively, on average across the four scenarios 45% and 41%. Indirect N 2 O emission through leaching and run-off, and from NH 3 and NO x volatilization accounted, respectively, for 10% and 4%. Area expansion results in large GHG emissions especially due to the removal of C from standing biomass of forest, which accounted on average across the scenarios for 63% of the total emissions from land use change (green bars in Figure 1). CH 4 emission from rice is the largest emitter of GHGs per unit area compared to the other cereal crops, followed by maize, wheat, sorghum and millet; therefore, for countries with rice cultivation, this is a substantial part of the total emissions (yellow bars in Figure 1).   Nevertheless, for most countries, Scenario 4 still requires land use conversion to meet self-sufficiency in 2050 (Table S3). The results of the analysis at the national level also reveal that matching food demand through land use conversion, instead of intensification, is not always an option as land area required for such expansion is not available in most SSA countries (Table 1; Table S3

| Robustness of scenario results for maize
In the previous section, we used the high-efficiency variant to estimate

| Maize intensification and agronomic N use efficiencies
The effects of initial N-AE on the outcomes were further explored for maize, by using a range of initial N-AE values instead of the fixed value of 14.3 kg grain yield kg/N as assumed in the low-efficiency variant. Figure 3 shows for maize at which initial N-AE value intensification is still superior to land use conversion in terms of GHG emissions. As already noted, with current nutrient management (i.e. an initial N-AE of 14.3 kg grain yield/kg N), Scenario 3 results in less GHG emission than Scenario 4, and we estimate that it would be most optimal in terms of minimum GHG emissions to intensify maize production until ca. 60% of Y w , leading to a total GHG emissions of 92 Mton CO 2 eqv. (Figure 3b). When initial N-AE exceeds 20 kg grain yield/kg N, Scenario 4 results in less GHG emission than Scenario 3 (Figure 3a). With an initial N-AE of 30 kg grain yield kg/N, intensifying maize production until ca. 70% of Y w would be most optimal and emissions would be reduced to 50 Mton CO 2 eqv. Note that at an initial N-AE of 30, Scenarios 3 and 4 are clearly superior to Scenarios 1 and 2 (Figure 3a).
F I G U R E 2 Total greenhouse gas (GHG) emission from maize production in 10 sub-Saharan Africa countries with its uncertainty, and the contribution of different input parameter categories to this uncertainty. Total GHG emission in 2050 for the different intensification scenarios (panels a, b), and for 2015 (dashed lines). Error bars represent the standard deviation. Contribution of all categories of input parameters to the total uncertainty (panels c, d) as represented by the error bars in panels (a)

| D ISCUSS I ON
This study provided insight into the consequences for GHG emissions of achieving future cereal self-sufficiency in SSA through scenarios with different levels of intensification and/or area expansion and different nutrient management variants. Evidently, irrespective of the investigated scenarios, GHG emissions in SSA will increase towards 2050 compared to present values, due to the tripling demand for cereals. We showed that for the 10 studied countries jointly GHG emissions from cereal cropping can increase up to 500% in 2050 compared to 2015 for scenarios in which area expansion is a main pathway to increase production (which has been the case in recent decades). Note that this assumes high nutrient use efficiency; a low nutrient use efficiency would lead to even larger increases. Such increases would have a large impact on the total GHG emissions from SSA, as cereal cropping alone would then already increase the total GHG emissions from SSA by 20% (CAIT, 2017).
We also show that intensification of cereal production with efficient use of fertilizers will moderate the increase in GHG emissions, although it requires a large increase in nutrient inputs. In this study, we only investigated the role of agricultural production in mitigation of GHG emissions, but additional mitigation benefits could be gained from the whole value chain, for example, by improved waste management, and more efficient distribution and transportation.

| Nutrient inputs
There is currently extensive soil nutrient depletion, which is general practice in SSA (Giller, Witter, Corbeels, & Tittonell, 2009), due to low nutrient inputs. We showed that a large increase in nitrogen (N) application is required to sustain current yields. Intensifying cereal production will require even more N application, thereby reaching application levels which are similar to European Union average values (Van Grinsven et al., 2012).
Irrespective of the scenario chosen, increased N input should always come with efficient nutrient management, as inefficient nutrient management (i.e. low values of agronomic N use efficiency, N-AE) results in high emissions of GHGs (plus other types of nutrient losses) from fertilizer use independent of the intensification level. The N-AE is of key importance in determining whether intensification is more favourable than area expansion for climate change mitigation and how much. If current N-AE of 14 kg grain yield kg/N would be increased to 30 kg grain yield kg/N, this would on average across the scenarios already result in a reduction of 26% in GHG emissions. We suggest that an N-AE of 30 kg grain yield kg/N can be well achieved in SSA, as in some locations, it is already obtained (ten Berge et al., 2019), and it is also a common efficiency achieved in for instance the United States (Ciampitti & Vyn, 2012). Yet, such enhanced use efficiency requires substantial improvements in current management practices, including good seed quality of the right crop cultivars, good planting densities, balanced crop nutrition, integrated soil fertility management (Vanlauwe et al., 2010) and improvements in controls of weeds, pests and diseases. There is thus a need for farmers to adopt new strategies, but adoption of these measures is currently already difficult for smallholder farmers and might become more difficult in the future as more climate variability is expected (Burke & Lobell, 2010). This points at the need to invest in research and development on nutrient management to go hand in hand with good agronomy to enhance the nutrient use efficiency of fertilizers.

| Avoiding crop area expansion
Our study reveals that compared to scenarios in which area expansion is the main pathway to increase production (which has been the case in recent decades), intensification of cereal production with efficient use of fertilizers will lead to much lower GHG emissions and might conserve forest and/or permit reforestation.

| Methodological considerations and uncertainties
We showed how our results depend on various assumptions and uncertainty in parameters. The IPCC tier 1 estimate for direct N 2 O emissions, which is directly linked to mineral fertilizer N applied (IPCC, 2006a), contributed most to the uncertainty in the resulting GHG emissions of each scenario. Despite a recent meta-analysis of the N 2 O emission factor (Albanito et al., 2017), it is widely recognized that especially for Africa, a better estimate for the N 2 O emission factor is required, due to the limited availability of data in the region (Albanito et al., 2017;Reay et al., 2012). More attention should therefore be given in future research to obtain more precise estimates of this emission factor.
In this analysis, we assumed that mineral fertilizer is used to fulfil the nutrient input requirements, but other sources of nutrients can also be used, such as leguminous crops and animal manure. Per kg of N applied, animal manure results in similar direct N 2 O emissions (IPCC, 2006a), but have additional GHG emissions from amongst other storage and methane emissions from animals (Monteny, Bannink, & Chadwick, 2006), while no CO 2 emissions from fertilizer production.
The specific size of the effect of including animal manure on GHG emissions is therefore unknown, but generally manure is only sparsely available in most of SSA. In each of the four presented scenarios, the available amount of manure which can be used will be similar, thereby probably not changing the observed trends and our main conclusions.
Inclusion of leguminous crops in cropping systems can reduce the mineral fertilizer N requirement for the subsequent crop (Jensen et al., 2012). A meta-analysis for SSA revealed that this residual effect of legumes can result in 450-700 kg/ha extra maize yield (Franke, Van Den Brand, Vanlauwe, & Giller, 2017). This potentially lowers the input of mineral fertilizer by 0-51 kg N/ha resulting in 0.03-12.18 Mton CO 2 eqv. less total GHG emissions depending on the scenario. However, Palm et al. (2010) showed for two contrasting sites in SSA that GHG emissions per unit maize produced is lower if only mineral fertilizer is used in comparison to using only green manure. Apparently, the increased N 2 O emissions from legume residue incorporation outweigh the benefits of reduced needs for mineral fertilizer inputs.
In this study, we included current climate variability, but did not consider the implications of long-term climate change. In addition, we also did not take into account the adoption of technological and genetic improvements which may partly offset negative effects of climate change. Furthermore, until 2050, the projected effect of climate change is not only highly uncertain but also relatively small compared to the large yield gaps (see van Ittersum et al., 2016 for more details).
It seems likely that due to climate change, potential yields will be affected (varying between a slightly positive impact, up to 10%, in high elevation regions of east SSA to negative impact up to approximately 20% elsewhere in SSA; Niang et al., 2014;Porter et al., 2014), but how climate change will affect N-AE, and thus, N requirements remains unclear. If we assume climate change has no effect on N-AE, it will result in the need for more area expansion (assuming average yields will somewhat decrease), and thus, climate change will favour intensification scenarios rather than expansion scenarios in terms of GHG emissions. If N-AE is negatively affected by climate change, the sensitivity analysis of N-AE revealed that this will favour the expansion scenarios, but at the same time, climate change results in the need for more expansion because of lower yields. So, overall, we argue that short-term climate change is likely to have neutral to aggravating effects on relative advantages of intensification scenarios over expansion scenarios.

ACK N OWLED G EM ENTS
This work was implemented as part of the CGIAR Research Program on Climate Change, Agriculture and Food Security (CCAFS), which is carried out with support from CGIAR Fund Donors and through bilateral funding agreements. For details, please visit https ://ccafs. cgiar.org/donors. We also acknowledge a financial contribution of the International Fertilizer Association. IFA played no role in the collection, analysis or interpretation of data, in the writing, nor in the decision to publish. The views expressed in this document cannot be taken to reflect the official opinions of these organizations.