Chapter 5. Discussion

A comparison of macrolevel methodologies: 1990 and 2003

The macrolevel results show that Kenya has the greatest nutrient depletion for N and K, followed by Ghana and Mali (Table 42). For P, Ghana and Mali show slightly negative balances, while the P balance for Kenya is neutral. One reason may be that farmers in Kenya applied 30 000 tonnes P with mineral fertilizer (IN1), which is 15 times that applied in Ghana and 5 times that applied in Mali. However, P depletion may also be underestimated in the erosion flow, as total P is derived from available P. In the volcanic soils of Kenya, there can be little available P when total P is large as most P is bound strongly to P-fixing soil particles. Table 42 also shows that results of the continental study by Stoorvogel and Smaling (1990) are reasonably in line with those of the present study, particularly for Ghana and Mali. Nutrient depletion in Kenya is less severe in the present study than that predicted by Stoorvogel and Smaling for 2000.

The macrolevel calculation procedure presented in Chapter 3 underwent a number of important methodological improvements compared with the original continental study by Stoorvogel and Smaling. First, the methodology was spatially explicit. This made it possible to take the spatial variation of soils and climate into account. It also provided the possibility to indicate areas with varying degress of nutrient depletion within the country. The procedures to calculate the nutrient flows also underwent significant improvement (Table 43). Finally, the soil nutrient stocks were quantified for each soil unit instead of only three soil fertility classes.

Comparing macro - meso and micro - meso levels

This study introduced the mesolevel in order to add value to the approaches that exist at national and farm level. The hypothesis was that the mesolevel could offer a suitable entry point for policy-makers and private-sector intervention, where macrolevel and microlevel are not appropriate for policy-making at subnational level. Within such a mesolevel system, the commercial component functions as the engine of the farming system and allows for intensification and expansion. This cash component can function as a driver for soil fertility management.

TABLE 42
Nutrient depletion, comparison between present study (2003) and Stoorvogel and Smaling (1990)

* Stoorvogel and Smaling (1990).

TABLE 43
Improvements in calculation procedure compared with the 1990 study

This study has shown that it is possible to construct a proper mesolevel nutrient balance provided sufficient data is available. The mesolevel results provide information that cannot be deduced from macrolevel and microlevel studies. For example, in Mali, the national nutrient balance per crop provides no intervention points for the CMDT and the cotton farmers as the scale is too coarse for targeted policies and institutional facilitation. At the microlevel, the village studies show relevant differences as to nutrient management, based on population density. For the CMDT, it is important to know about such differences and their causes. However, at the same time, its business-like structure forces it to act on the basis of subnational production targets and averages.

The above also holds for the Kenyan AEZ2, where differences in nutrient balances between farms were large (Figure 29). For this case, differences between AEZs are, in spite of farm-to-farm variation, more meaningful as entry-points for policy-makers and other stakeholders working at the district level. For intervention, it is also important to know that the relative depletion is much greater in AEZ1 than in the other zones (Figure 28). In Ghana, the macrolevel nutrient balance shows values for the different crops in different grid cells, and shows that cocoa is not causing greater nutrient depletion than other crops. However, what it does not show is the development of cocoa production in the two districts, and how this affects the soil fertility level. For this, and for actions to be proposed, mesolevel nutrient balances are needed. The figure for the total balance in this case is less important than differences between crops and individual flows that have a price (IN1 and IN2, OUT1 and OUT2).

Table 44 compares the mesolevel nutrient balances with the macrolevel balances of the study areas. It was not possible to make this comparison for the tea-coffee-dairy zone of Embu District because this study area was too small to determine a realistic nutrient balance at the macrolevel, i.e. the resolution at the macrolevel is too low. The table shows that the ‘macro’ figures for Ghana are considerably more negative than the ‘meso’ figures. This is because at the macrolevel the area under cocoa in the study areas is underestimated and too large an area is assigned to crops that have more negative nutrient balances. The mesolevel corrects this ‘error’, which stems from the macrolevel land-use map approach (Chapter 3). The macrolevel results for Mali agree reasonably well with those at the mesolevel, because of the similarity in land use at both levels (Table 45).

TABLE 44
Comparison of the nutrient balance between macrolevel and mesolevel

TABLE 45
Comparison of harvested areas for Koutiala Region

Unlike the macrolevel, the mesolevel analysis can also take specific management decisions and physiographic differences into account. This is the added value compared with the macrolevel approach. This also explains the differences between the results at the macrolevel and the mesolevel. For example, at the mesolevel in Ghana the small amount of erosion for land under cocoa was included in the OUT5 model, whereas erosion at the macrolevel is not land-use specific, adding to the overestimation of erosion and, hence, nutrient depletion for cocoa.

Microlevel studies provide a picture of the variation within a mesolevel unit. It is possible to include relevant management factors, and monitoring can check whether changes in nutrient management have a bearing on the nutrient balance and farm income. At the mesolevel, interventions can be targeted on the basis of microlevel variation by categorizing individual farms and villages into, for example, ‘good’, ‘moderate’, and ‘poor’ nutrient managers. Figure 29 illustrates the differences between farms in Embu District. Interventions can then be tailored to the diversity and dynamics in nutrient management as observed in the area. The villages studied in Mali provide evidence of the importance of population density and available land area as drivers of cotton production and farm nutrient management. M’Peresso, with a higher land pressure, has more livestock, uses more crop residues and less mineral fertilizer, and has a lower soil fertility. Noyaradougou, with a lower land pressure, has fewer livestock, uses fewer crop residues and more mineral fertilizer, which results in a more positive nutrient balance. For the CMDT, at the mesolevel, this could lead to a dual business approach, i.e. for areas under pressure and those under less pressure.

TABLE 46
Mesolevel nutrient balances

Comparing the mesolevel between countries

The nutrient balance at the mesolevel shows large differences between the three countries (Table 46). These differences stem from different physiographic circumstances and management. The cocoa-based farming system in Ghana receives very few inputs, which results in low yields and low but steady soil nutrient depletion. On the other hand, the cotton-based system in Mali is a relatively high-input system with little nutrient depletion, or even a positive balance in the case of P.

In all three farming systems, the cash crops are less nutrient depleting, or even have a positive nutrient balance, whereas the food crops have more negative nutrient balances. The reasons for the smaller depletion vary. Cocoa in Ghana receives very few inputs, but also has small outputs as a result of low crop production. In addition, cocoa is a shaded perennial crop, so leaching and erosion are minimal. Coffee and tea in Kenya receive more fertilizers than do food crops, but leaching and erosion remain important outflows. Cotton in Mali functions as the engine of the farming system. Relatively large inputs give a relatively good cotton production, whereas the food crops benefit in the next crop cycle from residual nutrients from fertilizer. In addition to the positive effect of the cash crops on the nutrient balance, they should also have a positive effect on farmers’ income. However, this depends considerably on: world market supply and prices; market imperfections as a result of protection or buyers’ quality requirements; and the national infrastructure and efficiency in product handling.

Data problems

Macrolevel

Macrolevel spatial data were amply available as a consequence of the many global and continental maps compiled in recent years. The following continental data were available: soil map of the world (FAO/UNESCO, 1997), land cover map (USGS et al., 2000), DEM from the HYDRO1k geographic database (USGS, 1998), rainfall map (Leemans and Cramer, 1991), irrigation map (Döll and Siebert, 1999), cattle density map and small-ruminant density map (FAO, 2001a), poultry density map and a Harmattan deposition map. These data enabled spatially explicit calculations of most nutrient inflows and outflows. Only for mineral fertilizers (IN1), and to a lesser extent crop products (OUT1) and crop residues (OUT2), was it necessary to divide national values evenly over the country because no spatial data were available.

The data were of differing quality. For example, the DEM has a resolution of 1 km, is based on satellite measurements and can be considered as high quality. The land cover map also has a resolution of 1 km and is based on satellite images of 1992 - 93, but these images had to be classified and the resulting maps are disputable. The classified maps should be checked locally in the field. However, this has not been done on a large scale. These maps are based on primary data, but other maps, e.g. the livestock-density maps, are derived from secondary sources, such as climate, soil and human population (FAO, 2001a). The Harmattan deposition map is the most inaccurate map because it is based on limited data from literature, which have been interpolated according to global wind patterns. However, these maps and data are probably the best available.

The accuracy of the tabular data also differs. FAOSTAT data rely on national statistics of differing quality. The quality of the statistics for Ghana and Mali is medium and for Kenya low according to FAO (http://faostat.fao.org/abcdq/about.htmx). All the crop data were based on sample surveys and not on total census data. Livestock data were normally better registered and available as total census data.

The quality of regression models, used to calculate leaching and gaseous losses, is always subject to debate. This is because such models are based on a limited data set, and they are not intended for use outside their own boundaries. The boundaries for the N-leaching regression model are: 40 - 2 000 mm rainfall, 3 - 54 percent clay content and 0.25 - 2 m layer thickness, based on 100 measurements (De Willigen, 2000). For the K-leaching regression model, only 26 measurements were available (which limited the borders): 1.3 - 8.1 cmol/kg for the CEC, 211 - 2 420 mm for rainfall, and 0 - 273 kg/ha for the amount of fertilizer.

Mesolevel

Data availability at the mesolevel was much lower than at the macrolevel because not all data are collected at district or province level. Moreover, the area was often too large for a representative farm survey. Spatial data at a representative scale were not available or were difficult to obtain. Most soil data were derived from national maps with scales of 1:250 000 (Ghana) or 1:1 000 000 (Kenya). These did not provide sufficient detail at the mesolevel or there were no related quantitative soil properties.

Land-use maps are generally not available at the mesolevel, making it difficult to use the procedure developed for the macrolevel. Moreover, this procedure is based on crop suitabilities. At the mesolevel, other factors such as management and land-use history are more important and they are necessary in order to make the level useful. The macrolevel grid resolution of 1 km would also be too coarse for the mesolevel. Aerial photographs and satellite imagery would be very useful, but are not always available.

Of all the data required, data on fertilizer use were the most difficult to obtain. Often, only recommended fertilizer rates were available (Kenya). Such rates are often greater than the actual application rates. Alternatively, it is possible to downscale national statistics, as was done for Ghana. In Mali, survey data were available, but their validity for the whole CMDT region was unknown because the distribution of the sample points was unknown. The quality of the data was not always reliable. Errors emerged by comparing such data with those obtained at the macrolevel. For example, the calculated yield, based on production and harvested area, should not be very different from the FAOSTAT data. This type of error occurred frequently and usually stemmed from typing errors, e.g. 12 000 kg/ha instead of 1 200 kg/ha maize.

Microlevel

The microlevel analysis used only farm survey data, combined with regression models. The quality of the data was normally good, but the representativeness of selected farms for the mesolevel was not easy to establish. Only a selection of the farms was surveyed, while variation is greatest at the microlevel. For example, the variability of soil properties between the different farms in one AEZ was larger than the variation between the different AEZs, as resulted from the VARINUTS study in Embu District (Table 34; SC-DLO et al., 2000). Spatial data are often available at the microlevel or are relatively easy to create. With the help of the farmer, it is possible to make a soil or crop map quickly because the area is small. However, surveying and monitoring a representative sample in a mesolevel unit involves costs: researcher and farmer time and money.

Modelling problems

Land-use map

The land-use map was based on crop suitabilities. This means that crops with the greatest requirements were allocated to the best locations and that crops with lower requirements were used to fill up the remaining grid cells. This ideal situation may differ considerably from reality because of socio-economic, demographic and political factors. While a map based on remote-sensing data would be better, such maps are often unavailable. Where they are available, classification of the images is difficult and often subjective. Moreover, it is necessary to perform many field checks in order to ensure an accurate and reliable land-use map. The regional distribution of the crops is more important than the accuracy of the individual grid cell. This is because of the subsequent aggregation of the grid cells before presenting the final results. The quality of the land-use map depends on the quality of the input data, which vary considerably from country to country. New initiatives, such as Africover (http://www.africover.org), offer alternatives for the land cover map, which is the most important input. The multipurpose land cover database is based on better and more recent satellite images, has been checked thoroughly in the field and has a more specific legend. However, it was not available for the present study.

The land-use map procedure included a 5-km grid in order to prevent large land units, which would otherwise appear in areas with little variation in altitude, climate, land cover and soils. This meant that the largest possible land unit was 2 500 ha. This approach cannot simulate multiple cropping systems very well because each land unit can only be allocated to one crop. However, aggregation of the final results does give an average for multiple cropping systems. Crops with smaller harvested areas might not receive a proper distribution because of the large grid size of the land unit. While the use of a higher resolution, e.g. 1 km, can improve the distribution, this might make the processing time too lengthy because of the large data sets. Another problem arises for countries with considerable variation in topography, such as Kenya. The global data sets describing climate and growing periods are too approximate for such a country. The simulated land-use pattern for Kenya is quite angular at the eastern side and coincides exactly with the rainfall and growing-period map. Moreover, potato is shown as growing at too low altitudes in this area because temperature data have been interpolated without taking altitude into account.

Erosion modelling

The results of erosion modelling with the LAPSUS model require careful interpretation. The model was developed at the watershed level and is now used at the national level. The spatial resolution used for this study was 1 km, while the original DEM resolution was 25 m. This resulted in an incorrect representation of the topography at the watershed scale because of the levelling out of features such as small valleys. Therefore, a 1-km grid is representative for landscape scale at national and continental level. Other processes, such as river incisions, become more important at this higher scale level, while processes such as re-sedimentation and tillage erosion are lost at the macrolevel (Okoth, 2003). Nonetheless, the results were promising and the erosion-sedimentation patterns were simulated correctly, according to the field observations. A problem is verification of the results at this national-scale level. At a watershed level, one can try to measure erosion for the whole area, a difficult undertaking in itself. It is impossible to measure erosion at a national level, one can only collect as many data as possible and make an estimate for the whole country. This means that until now the only way to verify the model has been through expert knowledge.

Soil depth and the soil erodibility factor (K factor) are derived from estimates and scarce literature data. Although not a measurable variable, the K factor is used as a calibration parameter. Therefore, the absolute value of the K factor is rather subjective. However, the relative difference between each soil type is the most important aspect. Management is one of the main factors affecting erosion on agricultural land. However, it is not possible to incorporate this factor at this macrolevel because each land use is treated the same for the whole country.

In view of the above considerations, the resulting grid cells should be aggregated to a larger cell size, e.g. 10 km. The model cannot predict exactly the amount of erosion within a specific grid cell, but it can provide a good estimate at the regional scale. Nevertheless, this is currently the most useful model for quantitative erosion estimates at national and regional scale. The model determines erosion and sedimentation at the landscape scale and simulates natural runoff patterns. The dynamic character of the model generates a more realistic erosion-sedimentation pattern. It is necessary to conduct further research in order to improve the verification of these models and there needs to be a greater focus on the scale-level discussions.

Nutrient-balance calculations

This study used the program MS Access to calculate the nutrient balance at the macrolevel and the program ArcView to create the maps. A database program is most appropriate for handling large amounts of different data and making simple calculations. MS Access has the option to create user-friendly input forms and is also widely available for future users in developing countries. However, the use of MS Access has some drawbacks. Its database structure means that it treats all grids in the same way. Thus, it is difficult to make exceptions. MS Access is not powerful enough for countries with many different grid cells because the program has a limited processing capacity. In this study, it was not possible to automate the procedure entirely because queries had to be converted to tables for the ultimate summation of all flows.

The study used the program MS Excel for the microlevel and mesolevel calculations. This was because: (i) the calculation did not involve a large amount of data; and (ii) it was necessary to make more exceptions for specific crop management. Management is much more important at the microlevel and mesolevel, which results in many exceptions. Therefore, manual input is more appropriate as this facilitates small adaptations. A drawback of spreadsheets is data management; overview is lost easily when many data are involved. Database programs are easier for data management and help reduce the number of errors that are introduced.

Shortcomings and caveats

On 17 and 18 February 2003, a workshop took place in Nairobi. This workshop brought together a group of soil fertility experts and mesolevel stakeholders to discuss the results of this study. The objectives were: (i) to jointly review the study; (ii) to share experiences on the macrolevel, mesolevel and microlevel; and (iii) to reach conclusions on the suitability of the approach for the normative programme of FAO. Twenty-six participants working in small groups discussed the study at each scale level and presented related research. The general opinion of the workshop was that nutrient balances are a useful tool for specific users at the different scale levels. The following sections provide more detail on specific comments and recommendations that emerged at the workshop.

Validation

A major point raised at the workshop concerned the lack of sufficient validation and the considerable uncertainties attached to the different nutrient flows. Large-scale and data-demanding studies are difficult to validate because of the large areas and the large amounts of different data. This makes validation in the field very difficult and expensive. At the macrolevel, it is not possible to validate all the nutrient flows because this would require a massive number of samples. It might be possible to validate each nutrient flow at the microlevel, but these validations would then need scaling up to the mesolevel and the macrolevel. Other large-scale studies in the context of climate change and biodiversity research have similar, inherent validation problems. Although experiments are a relatively simple way of validating some nutrient flows, such as leaching, other flows, such as erosion or mineral fertilizer application, are much more difficult to validate. As it is almost impossible to validate the whole nutrient balance, one can choose to validate only those specific flows that are considered most important. For example, one can measure erosion in the field where this is one of the main losses according to the nutrient balance. These field observations and measurements should be performed according to a sound sampling scheme. Connecting the validations of process research, e.g. studies of N₂O losses, to system research, such as this study, is both practical and feasible.

Gaps

Although the nutrient balance includes the most important nutrient flows, it fails to take some aspects into account. At the macrolevel, it does not incorporate large-scale processes such as forest burning and river-basin sediment transport. At the livestock level, it does include urine specifically although its nutrient content is quite different from that of dung. In addition, nutrient losses from urine are very large because of leaching and volatilization. Some other aspects, although not directly linked with the nutrient balance, can be of importance for the functioning of the whole agro-ecosystem. For example, belowground biodiversity has a direct effect on soil structure and the release of nutrients from organic material. Offsite effects, such as sedimentation into reservoirs and excessive nitrate leaching to groundwater, can also be related to the nutrient balance. Depending on the definition of the system, transnational imports and exports of products can be important flows in the nutrient balance, e.g. export of cash crops and import of fertilizers. Economic dynamics, such as the withdrawal of subsidies or trade liberalization effects, provide the all-important context that needs to be known before suggesting any improved nutrient management. Finally, it may be necessary to examine nutrients other than N, P and K, such as calcium and sulphur, or organic carbon to link up with carbon sequestration research groups.

Specific problems for each scale level

In nutrient-balance calculations, each scale level has its own specific problems. At the macrolevel, the most important problems are: data quality; map interpretation; resolution differences; and groundtruthing. Intensive field checks in accordance with a sound sampling scheme can provide a partial solution. Soil properties and nutrient stocks might also be collected with new techniques for rapid estimation by reflectance spectroscopy (Shepherd and Walsh, 2002). At the mesolevel, the main problems are: lack of spatial data; incorporation of different management systems; and the absence of socio-economic explanatory factors, e.g. credit facilities and marketing. Spatial data will be increasingly available in the future. A classified satellite image and a DEM will improve the mesolevel nutrient balance significantly. At the microlevel, much research has already been done. The NUTMON-toolbox is a useful application, which also includes the monetary part (www.nutmon.org). The issues at this level are: how to deal with diversity between and within farms; how to incorporate INM and integrated soil fertility management (ISFM) techniques; and how to scale up results. Possible options are: stratification in sampling methods; INM techniques in farmer field schools; and the use of GIS for upscaling.

Presentation of outcomes

Model results expressed in terms of kilograms of nutrient per hectare are not very meaningful for policy-makers. They prefer outcomes expressed in terms of yield loss or in monetary values. The nutrient balance should have links to other tools and data in order to make it more useful. Combining a simple soil fertility/crop production model, such as the quantitative evaluation of the fertility of tropical soils or QUEFTS (Janssen et al., 1990), with the nutrient balance makes it possible to express nutrient depletion in terms of yield loss. Other attractive indicators to possibly attach to the nutrient flows and balance are the nutritive value of diets, food and cash needs, and equity indicators. Other options are to make use of decision-support systems and scenario studies. One way of making the nutrient-balance model more interactive is to link it to a model such as that of the conversion of land use and its effects (CLUE) (Veldkamp and Fresco, 1996), which simulates land-use changes and its effects. It is also possible to combine the results with other GIS data, such as food security or poverty maps.

Usefulness for policy-makers

It is important that policy-makers be aware of any gaps in the nutrient balance, so they know what the limitations of the nutrient balance model are. This raises the question of whether present outputs can serve as tools for policy-makers or whether further research is required. The nutrient-balance model proved to be a useful indicator for informed policy-makers, but the results as presented so far offer no entry points for intervention. The model raises awareness of soil fertility problems, indicates areas with nutrient depletion or accumulation, and gives a quantified picture of the nutrient flows at the macrolevel. At the mesolevel, it is possible to: (i) identify specific constraints; (ii) use quantified nutrient flows for planning purposes; and (iii) extrapolate results to other similar areas. Furthermore, outcomes might convince policy-makers to make action plans to improve soil fertility.

Impact

Any assessment of the impact of a negative nutrient balance needs to consider the actual soil fertility, i.e. the nutrient stocks. A negative nutrient balance on a rich soil will not affect yield in the short term, while crop yield on a poor soil may decline each year as a result of nutrient depletion. At some stage, a negative nutrient balance may no longer affect production in marginal areas. This may be the case when yields depend on natural inputs, such as IN3, to make up for any minimal losses.

The declines in yield and N stocks are most significant in the initial decades and then they begin to level off. Figure 30 shows that yields on a rich soil remain unaffected for a much longer time. Hence, one could say that nutrient depletion often does not manifest itself clearly, but problems are likely to occur for the ‘future generations’ of the Brundtland definition (Brundtland, 1987).

In this example, the yield decline is related with the decline in N stocks. For a better simulation, a crop growth model should be linked to the nutrient depletion model. QUEFTS (Janssen et al., 1990) may be suitable as it is relatively simple to use and takes the interaction between N, P and K into account.

It is also possible to express declines in yield and nutrient stocks in economic terms. In this case, yield decline is a private (farmer) cost, whereas the decline in nutrient stocks is a social cost. The nutrient stock estimates in this study (Annex 6) were based on data sets from the 1970s and 1980s. This may imply that the actual stocks are probably lower.

Farmers will normally adjust their management when they experience yield decline. Where they do not have the means to increase fertilizer or manure use, they may try to make better use of low-cost integrated nutrient management technologies (e.g. tree-crop mixtures, green manures, rock phosphate, soil and water conservation, niche management, benefiting from soil microvariability).

Depletion of the natural N stock of a soil causes a decline in the efficiency of fertilizer use. As N nutrient depletion progresses over time, N fertilizer application needs to increase to maintain yields, i.e. the ratio kg N fertilizer/kg Y increases. It is assumed that the opportunity cost of decreasing fertilizer productivity over time represents the value of soil N decline.

A dynamic model was used to estimate the costs of nutrient depletion. The model was linked to a biophysical model that simulated the effects of nutrient balance changes in the farming systems of Ghana (2), Tanzania and Mali. Seasonal N stocks in the soils were used to estimate yield changes over time, yields being a dependent variable of nitrogen availability (Figure 30). When N removal became equivalent to N input, yields were assumed to stabilize (at a low level) for the remaining seasons in the period of 20 years. The crop and site-specific fertilizer use efficiency decline was estimated using an e-log function, i.e. NFP = b × e^-t where: NFP = nitrogen fertilizer productivity, and t = time. The integral of the function over 20 years estimated the decline in N fertilizer use efficiency as the result of soil N removal:

Costs were computed for every crop in each farming system and expressed per hectare of crop. These costs were adjusted for the area cropped within the farming system to arrive at a “weighted” cost. The total costs of N nutrient depletion were computed by the sum of two components: the opportunity cost of production losses due to yield decline, potential yields were assumed to grow at 1 percent per year, and the costs of the permanent impact of reduced N fertilizer productivity within each farming system as a proxy for the decline in soil N stocks (Table 47). The net present values per hectare of “farming system” were discounted at 10 percent per annum.

The result suggests that the cost of N fertilizer productivity is higher than the cost of yield reduction in the Embu and Koutiala cropping systems. The opportunity cost of production losses due to yield decline constitutes foregone farm income; the decline in fertilizer use productivity is a directly incurred production cost by the farmers. Addressing the nutrient exchange capacity of soils will not only protect farmers’ future income, it will also prevent an adverse effect on the competitiveness of the production systems caused by higher fertilization cost. The latter is of particular relevance for farmers operating in an increasingly integrated global market. The cost of addressing soil fertility decline constitutes an attractive intervention in the districts in Ghana and Kenya and to a lesser extent in Mali. The cost involved could vary between US$70/ha in Mali and US$1 500/ ha in Kenya, with a possible share to be financed by farmers between 10 and 70 percent.

TABLE 47
Nutrient depletion benefits and cost, US$/ha