Ecological approach to agricultural production and ecosystem theory: The amoeba approach

G. Pastore and M. Giampietro

Instituto Nazionale della Nutrizione, Rome, Italy

Abstract

The paper argues that agriculture operates on the interface of two complex, hierarchically organised systems: the socio-economic system and the ecosystem. So in any defined farming system one will always find legitimate and contrasting perspectives with regard to the effects of changes in the system, and the effects are not likely to result in absolute improvement for all stakeholders. A methodological tool, the amoeba multi-dimensional reading, is described to characterise farming system performance in an integrated way taking at various scales according to different perspectives.

Introduction

Agriculture operates on the interface of two complex, hierarchically organised systems: the socio-economic system and the ecosystem (Hart 1984; Lowrance et al 1986; Conway 1987; Ikerd 1993; Giampietro 1994a; Giampietro 1994b; Wolf and Allen 1995; Giampietro 1997a). This implies that in any analysis of a defined farming system one will always find legitimate and contrasting perspectives with regard to the effects of changes in the system. For example, increasing return for farmers (e.g. intensification of crop production) is coupled to more stress on ecological systems (e.g. loss of biodiversity and soil erosion). Similarly, improvements for certain social groups (e.g. lower retail price of food for consumers) tend to represent a set-back for others (e.g. lower revenues for farmers).

Thus, changes in agriculture, induced either by new policies or by technical innovations, are unlikely to result in absolute improvements for all stakeholders and social actors involved, nor in absolute improvements on all the scales (soil, farm fields, watershed, regional, global) on which the (side-)effects of agricultural production can be described. Hence, a 'correct' assessment of agricultural performance should best be based on an analysis of trade-offs that reflect the various perspectives, both positive and negative, with regard to the effects that a proposed technological or policy change will induce on the various scales and actors involved.

In this paper we present a methodological tool, the amoeba multi-dimensional reading, that can be used to characterise farming system performance in an integrated way on various scales and according to various perspectives.

Theoretical basis of integrated assessments based on a multi-dimensional description

Some basic concepts

Agricultural systems are complex systems made of many different components which operate in parallel on different space-time scales. Examples of such components are soil micro-organisms, populations of selected plant species in crop fields, individual farmers, farmer households, rural communities, local economies, local agro-ecosystems, watersheds, regional economies, biosphere processes stabilising bio-geo-chemical cycles of water and nutrients, and socio-economic processes stabilising the boundary conditions of farming activities. In addition to being hierarchically organised on several scales, ecological and human systems are made up of 'holons' (term introduced by Koestler 1969), that are a whole made of smaller parts (e.g. a human being is made of organs, tissues, cells, molecules etc) and at the same time they form a part of some greater whole (an individual human being is part of a household, a community, a country, the global economy).

Understanding the holarchic structure of agricultural systems is a fundamental prerequisite for a sound analysis of their performance. Agricultural research tends to be plagued by systematic errors in the choice of the hierarchical level of analysis and investigation. In fact, analyses performed at a certain level (e.g. compatibility of crop production techniques with soil health) not necessarily provide sound information on what goes on at other levels (e.g. compatibility of the production technique with expected farmer income in a defined rural community).

The amoeba reading

The basic idea of the amoeba reading is to provide a graphic representation of system performance as assessed over a certain number of aspects/qualities that cannot be expressed as a function of the others. In this way, it is possible to have an overall assessment by a visual recognition of the existing difference between the profile of expected (or acceptable) values and the profile of actual values over families of indicators of performance referring to non-comparable qualities. This method is used, for example, in marketing to obtain an overall assessment of consumer satisfaction with regard to different aspects of a product (e.g. for a car: price, driving characteristics, design, gasoline consumption, assistance and services, reliability, choice of colours). Wide differences between expected and actual values indicate lack of consumer satisfaction. Areas of the graph in which the gap between expectation and actual performance is particularly wide indicate priorities in terms of intervention. Such a reading is illustrated in Figure 1 and represents a metaphor of the need of multiple reading for sustainability. The particular product considered will not be 'sustainable' in the market place if it completely fails on one or more of the qualities affecting consumer choice.

Figure 1. Example of amoeba reading: performance of a car.

In the field of natural resources management, the amoeba approach has been proposed by Brink et al (1991) as a tool for dealing with the multi-dimensionality of environmental stress assessment by using different indicators of ecological stress referring to events occurring on different space-time scales. In our model, the graphic representation of the system is simply based on a division of the plane of a 'radar diagram' into four quadrants, each describing a distinct view on the system (Figure 2). Within each quadrant, a number of axes referring to different indicators of performance are then drawn. The choice of both quadrant and axis is arbitrary and is made according to the system's characteristics which are considered relevant for the analysis. With regard to agricultural sustainability, aspects to be considered are generally related to the existing relation of farming to its socio-economic context, such as economic development and openness of the food system, and to its ecological context, such as type of ecosystem and demographic pressure.

In the example provided in Figure 2, the agricultural system is described by quadrants that refer to the following four aspects of performance: the farmers' view (upper left quadrant), the national (or regional) economy (lower left quadrant), extent of environmental loading (upper right quadrant), and the ecological footprint of the food production system (lower right quadrant). The latter is a measure of the extent to which a steady-state description of the agricultural system misses relevant information and hence this quadrant somehow accounts for the fact that today almost no agricultural system is in reality in steady-state (inputs and outputs are increasingly based on stock depletion and filling of sinks).

Figure 2. Amoeba reading applied to sustainable agriculture.

At this point we have a tool that describes the effect of changes in the system in parallel on different hierarchical levels (space-time scales) and according to a given selection of perspectives (those forming the basis for the selection of indicators in the four quadrants). However, in order to move from this multi-dimensional representation to a discussion of trade-offs in decision making in agriculture (e.g. to evaluate possible scenarios and/or use multicriteria methods of evaluation) three additional steps are needed.

1. The selection of indicators to describe the effects of a particular change (the states of the system) on different levels and according to different perspectives must be validated by involving social groups and agents whose views are considered in the analysis (who decides what is the best selection of indicators of performance?).

2. The various 'readings' of the states of the system on different scales are to be bridged by a model of analysis that links events at one level (described on a certain space-time scale) to events occurring at other levels (described on different space-time scales).

3. Models used to link changes on one level to changes on other levels must be critically appraised by the different stakeholders that are supposed to use them in the discussion of scenarios with the help of scientists from various different scientific backgrounds (who decides how good are the proposed models?).

Clearly, the success of such an integrated assessment completely relies on the 'political will' to involve straight from the outset as many stakeholders as possible in the process of decision-making using participatory techniques, and on the 'political integrity' to respect the indications obtained by such an integrated assessment.

Establishing bridges among levels

The graphic representation in Figure 2 provides a parallel description of states of the system as 'seen' and 'recorded' at different scales on different hierarchical levels. It should be noted, however, that the values of the variables used as indicators of performance are not independent of each other both within and across quadrants. Many of these variables can be linked through equations of congruence across levels referring to biophysical throughputs of agriculture (Giampietro 1997a; Pastore et al 1998; Giampietro and Pastore 1999). For example, technical coefficients (throughputs per hectare of land and per hour of labour and output/input ratios) and market variables (sale prices, structure of costs, and taxes and subsidies) define a direct link among many of the variables considered in the amoeba reading.

Passage from individuals to types to cross levels

Some aspects of the amoeba reading deserve particular attention. When the system is read at the household level, the quadrants describing the effects of farmer's choice on the environment (e.g. environmental loading) and on the socio-economic context (e.g. food surplus produced and its related cost) refer exclusively to the specific and limited space-time scale at which the individual farm household is defined and described (e.g. a 200 ha farm in the USA over a period of 1 year). To assess the effect of farmers' choices on a larger scale (e.g. regional or national), one needs to aggregate the effects induced by all the individual farm households operating in the area (region/country). Given that the effects of individual farm households are not necessarily uniform, one must first define the various 'farm types' that can be distinguished and then, using curves of distribution, obtain an aggregated effect over the given set of farm types.

To obtain a set of characteristics that can be used to define a farm type, start by analysing the constraints affecting farmers' options, as determined by internal links among the variables on the amoeba (for more details see Pastore et al 1998). Our steps in this process are as follows:

• choice of the set of indicators of performance to be used which determine the skeleton of indicators for the amoeba

• definition of a viability domain for each indicator (range of values within which the farm is supposed to operate)

• definition of the preferences of farmers in relation to different indicators, that is, establishing a preference relation among the different areas within the viability domain

• characterisation of the farm type in terms of selected combinations of farming techniques that tend to saturate to the maximum extent the existing resources (land, labour time and financial capital) available to the farm household considered given a set of objectives defined in step 3.

Different strategies adopted by farmers (e.g. maximisation of economic return or minimisation of risk) can be studied as movements of the system in different areas of the amoeba. The existence of 'internal constraints' (e.g. a farm household can not use more time, land or capital than it has available) implies that, given technical coefficients and structure of prices, costs and taxes, the possible choices for the farm household are limited. Studies of the nature of this limitation specifically address the peculiarity of research at farm level as compared to research at the plot level (see Pastore et al 1998 and Giampietro and Pastore 1999).

Each combination of techniques that satisfies the above-mentioned conditions of (1) saturating the existing budgets of land, labour time and capital, and (2) operating within the feasibility domain of the selected set of indicators of performance, represents a viable technical option for farmers. That is, such a combination is one possible state (a type) for the farm. Each farm-type defined in this way implies a certain combination of trade-offs (a defined profile of values on the amoeba) for the environment and the national economy.

An application of the multi-dimensional amoeba approach

First step: Selection of indicators of performance on different scales and related to different perspectives

A list of indicators of agricultural performance (and the range of their values) that can be used to reflect the various perspectives generated at the household level is shown in Table 1. Assessments of the performance of a farming system at this level can consider various objectives, such as minimisation of risk (e.g. safety from climatic, market and political disturbances), food security, maximisation of income and net disposable cash, and maximisation of the potentiality of the members of the farm household (e.g. better education, better communication and information processing, intensification of social and cultural events).

Several indicators assessing agricultural performance at the level of the national (or regional) economy are listed in Table 2. At this level, several goals should be considered, such as self-sufficiency in food production, minimisation of indirect costs of the food system, minimisation of the direct economic cost of food supply, and minimisation of gradients in economic development between rural and urban areas.

Indicators to monitor ecological impacts are presented in Table 3. The set of indicators should cover various distinct scales (e.g. global level, region, watershed, village, farm, field, and soil level). They might refer to (i) direct measurements of environmental loading (e.g. fertiliser and pesticides applied per hectare per year and per unit of crop output, pollutants discharged into the environment), (ii) alterations of natural configurations of matter and energy flows (e.g. thermodynamic indices of ecosystem), (iii) use of bio-indicators (e.g. key species providing information on the health of the natural system within which they operate; they can be vegetal associations, biodiversity of different taxa related to different scales such as protozoa and earthworms for the soil, arthropods for cropped areas, and birds and mammals for larger areas), and/or (iv) evaluation of landscape patterns (e.g. fractal dimension of agricultural landscape). Again an appropriate combination of these indicators depends on the scale and the type of information needed in the process of decision making.

Table 1. Indicators that can be used to assess material standard of living at household level.           

1. THT (Total Human Time) = Total number of individual × 8760 (hours in one year).

2. C = Total time (hours per year) allocated by the whole society to labour in primary sectors of economy.

3. Net disposable cash.

Finally, indicators can be used to describe the 'degree of freedom' of the considered agricultural system from local biophysical constraints (a measure of how misleading the description is of the system in steady-state). Some of these indicators are listed in Table 4 (e.g. dependence of agricultural output on stock depletion, filling of sinks and related accumulation of pollutants, and imports from distant ecosystems). Basically this assessment compares the ecological footprint (the demand for natural capital on which the present agricultural system performance is based) with the amount of natural capital available in the agro-ecosystem for a sustainable agriculture (without generating irreversible deterioration in ecological systems). Indicators in this quadrant often represent the extent of 'openness' or linearisation of matter and energy flows in the agro-ecosystem: The higher is the speed of throughputs on the farm, the higher is the linearisation of matter and energy flows in the agro-ecosystem, the more human choices are 'free' from local natural constraints (technological inputs short-cut the ecological system of feedback controls), and the more human activity is at risk of generating negative consequences for the ecosystem (Giampietro 1997b).

Table 2. Indicators that can be used to assess the performance of agricultural systems according to socio-economic context.

 1. (Total Human Time) = Total number of individual × 8760 (hours in one year).

2. C = Total time (hours per year) allocated by the whole society to labour in primary sectors of economy.

Second step: Defining feasibility domains for selected indicators

Having chosen the variables on different axes (distributed over different quadrants), one must define a range of 'feasible' values for each indicator (the gray area on each axis indicated in Figure 2 within which light gray means 'good', dark gray means 'bad'). Within the 'feasibility domain' 'target values' may be added to the graph (the dots in Figure 2) that reflect the goals expressed by the representatives of different perspectives.

Table 3. Indicators that can be used to assess ecological impact of agriculture.       

Regarding the hierarchical levels distinguished on the socio-economic side (e.g. household, region, and country), the selection of both indicators and their feasibility domain is difficult because according to the specific situations considered many different social groups (e.g. ethnic minorities, future generations) could be included into the assessment inducing conflicting definitions of what is acceptable. The existing link across levels implies that the socio-economic context within which the farming system operates is affecting ranges of acceptable values on lower levels. For example, a return of one dollar per hour of farm labour would be a remarkable achievement for a Chinese farmer (operating in a country with a low average return of labour per hour), whereas such a return would render farming in the European Union not viable.

Third step: Assessing current situation on a multi-dimensional state space

In this step, the actual value of each indicator of performance in each of the four quadrants is recorded on the graph. This makes it possible to visualise the position of the actual values, e.g. are they inside or outside their feasibility domain, what is the distance from the edge of their domain, and what is the distance from their target? The multi-dimensional state space obtained at this point makes it possible to compare the current position of the system against the 'states' defined as target for policy implementation by stakeholders and against the 'feasibility domain' based on the underlying biophysical links across hierarchical levels. Wide differences between actual values and expected values (either target values or values which would be required by congruence of matter and energy flows across levels) can be assumed to indicate stress in both natural and socio-economic sub-systems and hence indicate need for intervention.

Table 4. Indicators that can be used to assess the degree of freedom of agricultural production from local biophysical constraints.     

Two examples of a multi-dimensional reading are provided in Figures 3 and 4 that refer to farming systems in developing and developed countries, respectively.

The amoeba reading shown in Figure 3 characterises the situation of a subsistence farming system operating without external inputs. When population pressure is moderate, ecological indicators of stress are within the acceptable range. Note, however, that the values of the set of indicators characterising material standard of living would be unacceptable according to Western standards. The net disposable cash generated per hour of labour time, average body mass, and other social indicators of development are far away from the viability domain at which rural households operate in developed countries.

Figure 3. Amoeba reading applied to a subsistence farming household.

Figure 4. Amoeba reading applied to a farming household in a developed country.

The reading shown in Figure 4 characterises the situation of farmers operating in developed countries. The values of indicators of development at the farm household level are worse than the corresponding national averages (rural communities are poorer than urban communities), although the gap is kept fairly narrow through national policies supporting the rural population. In absolute terms, the situation of farmers in developed countries is much better than that recorded for subsistence farming in Figure 3. However, the multi-dimensional analysis reveals the trade-offs implied by this positive achievement on the socio-economic side. Higher returns for humans in developed countries are paid for by a larger environmental impact of agriculture, and by a heavy dependence on stock depletion (e.g. fossil energy), and often by import of ecological activity from distant ecosystems (e.g. imported animal feed and other agricultural commodities in Europe).

A comparison of the two profiles in Figures 3 and 4 (the distribution of actual results over the feasibility domains, i.e. the gray areas) shows the unbalanced negotiation of contrasting goals (perspectives) referring to distinct levels when the farming system is operating under a different combination of socio-economic and ecological contexts. The ecological perspective tends to be the looser in intensive agriculture as soon as the demographic and socio-economic pressure rise (see Giampietro 1997a; Giampietro 1997b). This explains why traditional farmers undergo heavy stress when fast development of their socio-economic context makes their traditional techniques no longer viable.

Discussion on scenarios in terms of effects of changes across levels using a case study in China

In a four-year project studying agricultural intensification in a rural area of Hubei Province, PR China the following three main farm types could be distinguished.

• Farm type 1: Farmers that maximise the net disposable cash (NDC) through cultivation of cash crops and off-farm labour, even though this means taking risks and a heavy work load.

• Farm type 2: Farmers that minimise their risk by growing mainly subsistence crops and that maximise their leisure time (max THT/C) by avoiding off-farm jobs, even if this implies remaining behind in the fast process of modernisation of China (low NDC);

• Farm type 3: Farmers that operate minimising risk (relying on subsistence crop), but aiming at the same time for a high net disposable cash (off-farm jobs and cultivating whenever possible also cash crops). This choice is paid for by heavy work loads (low THT/C).

These three solutions to saturate the land, time and capital resources available represent farm types that can be considered 'attractors' in the area studied given the existing socio-economic and ecological context and cultural identity of farmers. These three farm types imply different trade-offs in terms of performance of indicators reflecting other perspectives, or in other words, these three farm types have different shapes when described with the amoeba reading.

Farm type 1 implies higher income for farmers but at the same time a larger environmental load and total absence of rice surplus to feed the urban population of China (these farmers are net consumers of rice). This is evident when comparing the amoeba reading of farm type 1 (Figure 5b) with that of farm type 2 (Figure 5a). This implies that if farm type 1 would be the only one practised in entire rural China, the country would no longer be able to feed its urban (and rural) population without heavy reliance on import.

Similar multi-dimensional amoeba readings for farm types 2 and 3 are graphically illustrated and discussed in Pastore et al (1998) and Giampietro and Pastore (1999). This is to emphasise that each of these farming types defined at the household level can be linked to a certain pattern of landscape use (defined on the space scale of the farm) and to certain effects (when aggregated on a large scale) on the national economy. Latter effects are obtained by considering the distribution of the population of farm households over the possible set of farming types. Given a spatial distribution of rural villages in a determined area and assuming several different distributions of the population of rural households over the set of possible farming types simulate changes in landscape use and effects on the national economy can actually be generated by the (simulated) changes in the area. In this way, the effect of government policies or technological changes can be studied by simulating the effect that they will have on the distribution of households over possible farm types. Clearly, dramatic changes both in technology, farmers feelings, environmental settings, and governmental policies can scramble the existing picture by introducing new possible farm types, making existing ones obsolete, generating dramatic changes in the distribution of individual households over the accessible farm types.

Figure 5. Example of amoeba reading of two Chinese farming-types: Household optimising THT/C(5A) and household optimising NDC (5B).

To illustrate our approach to crossing scales, we refer to the amoeba readings presented in Figures 6a and 6b that describe two 'virtual villages' simulated on the basis of the information obtained from the amoeba reading of the farm types in rural China.

We assumed that the village described in Figure 6a is characterised by a majority of farmers that optimise the net disposable cash (this simulation is based on a distribution of 80% of farmers belonging to farm type 1; 10% to farm type 2; and 10% to farm type 3). The village described in Figure 6b is characterised by a majority of farm households practicing traditional agriculture, hence minimising risks and time allocated to work (simulation is based on a distribution of 80% of farm households belonging to farm type 2; 10% to farm type 1, and 10% to farm type 3).

Thus, the graphs in Figures 5a and b represent the household level and are based on real data, while those in Figures 6a and b represent the village level and are simulated. Note, the different space time scale of the amoeba readings: the scale of the village (Figures 6a and b) is larger, that is, it covers a larger area and is slower in reacting to changes. At both household and village levels, it is possible to obtain the amoeba reading either by simulation (defining distribution curves of lower-level elements over types) or by gathering real data. Doing both data collection and simulation at either one level is useful to validate the assumptions adopted in the simulation. For example, in our study in China we found that the geographic location of villages (access to market and off-farm jobs opportunity) was a significant factor affecting the distribution of farmers over the possible farm types. Similar hypotheses can be tested when considering population characteristics (age and sex structure, ethnic origin, level of education) as possible factors affecting the distribution of farmer households over the existing set of farm types.

Figure 6. Example of amoeba reading of two simulated Chinese villages: Virtual village A generated by a given curve of distribution of farming types (6A) and virtual village B generated by a different curve of distribution of farming types (6B).

Considering the amoeba readings in Figure 6, we see that the first virtual village (market-driven choice based on off-farm work and intensive production of cash crops) is the one that generates by far the highest environmental loading and is to the larger extent dependent on coal and oil for food production. From the national perspective, this village does not produce any surplus of rice, on the contrary, it erodes the rice surplus produced by near-by villages. As expected, however, what is detrimental to the environment and the food self-sufficiency of the country also has its positive side: a high net disposable cash for farmers. The productive pattern adopted by village 1 is therefore benign to the villagers and to the people of the close-by town that have access to cheap supply of fresh vegetables and other food. On the contrary, the second virtual village (Figure 6b) with the highest surplus of rice (good for self-sufficiency of China) generates a moderate environmental impact (good for the environment). This environmental benign solution is paid for in terms of low net disposable cash from agriculture. People living in village 2 are at risk of loosing contact with the dramatic socio-economic transformation which is taking place in China. A general amplification of village 2 type will imply locking a large part of the Chinese rural population into a situation of poverty and lack of modernisation.

Conclusions

The existence of different dimensions for the concept of performance of agricultural systems and the existence of different space-time scales at which agricultural processes can be described make it impossible to determine 'optimal' solutions (optimal for how long? and optimal for whom?). When dealing with sustainable development the relations of preference and indifference are not enough, because when an action is better than another one for some criteria, it is usually worse for others, so that many pairs of action remain incompatible with respect to a dominant relation. Moreover it is impossible to assess with a single type of description/analysis the effect of a particular combination of techniques of farming on all the chosen dimensions. By acknowledging the existing problems, we propose a model of analysis that:

1. does not claim to provide 'the correct' analysis of the system. It simply generates several sets of 'view dependent' representations of the reality. The peculiarity is that we acknowledge such a dependency from the beginning.

2. can be enriched by including new alternative sets of view-dependent representations and can be used as a tool for enhancing negotiation among groups with different views and interests about the performance of food systems.

3. it acknowledges that the goals related to the concept of sustainable development can not all be achieved at the same time, just by adopting a single 'silver bullet' technical solution. Decision-making implies finding compromise solutions among legitimate but contrasting views.

4. it enables to use information generated in different scientific fields (economics, sociology, agronomy, applied and theoretical ecology etc) and referring to descriptions of the system obtained on different space-time scales.

5. makes easier the discussion of possible scenarios by either fixing the values of variables describing the performance of agriculture on the socio-economic and ecological level and then check the values of technical coefficients and/or market variables that would be required to guarantee such a performance, or by estimating future technical coefficients and then calculate the possible consequences in terms of performance of agriculture in ecological and socio-economic terms. Studying possible effects that policies formulated at the country and regional level can have on the options available to farmers and on farmers' choice over available farm types.

6. this multi-dimensional approach forces to evaluate perspectives of levels (and/or social groups) that normally are not included in the 'traditional' analysis. In fact the environment (= future generations) has been until now the big looser in negotiations among contrasting views belonging to different levels. That is, including the perspective of the environment (= considering and assessing environmental costs at several space-time scales-soil, field, farms, watershed, regional) should become a mandatory step when formulating policies affecting the sustainable use of natural resources.

References

Brink B.J.E., Hosper S.H. and Colijn F. 1991. A quantitative method for description and assessment of ecosystems: The amoeba approach. Marine pollution bulletin 23:265–270.

Conway G.R. 1987. The properties of agro-ecosystems. Agricultural systems 24:95–117.

Giampietro M. 1994a. Sustainability and technological development in agriculture: A critical appraisal of genetic engineering. BioScience 44(10):677–689.

Giampietro M. 1994b. Using hierarchy theory to explore the concept of sustainable development. Futures 26(6):616–625.

Giampietro M. 1997a. Socioeconomic pressure, demographic pressure, environmental loading and technological changes in agriculture. Agriculture, ecosystems & environment 65:201–229.

Giampietro M. 1997b. Socioeconomic constraints to farming with biodiversity. Agriculture, ecosystems & environment 62:145–167.

Giampietro M. and Pastore G. 1999. Multi-dimensional reading of the dynamics of rural intensification in China. Critical reviews in plant sciences 18(3):299–330.

Hart R.D. 1984. The effect of interlevel hierarchical system communication on agricultural system input-output relationships. International Association for Ecology Series Study. Options Mediterraneennes Ciheam IAMZ-84-1.

Ikerd J.E. 1993. The need for a system approach to sustainable agriculture. Agriculture, ecosystems & environment 46:147–160.

Koestler A. 1969. Beyond atomism and holism—The concept of the holon. In: Koestler A. and Smythies J.R. (eds), Beyond reductionism. Hutchinson, London, UK. pp. 192–232.

Lowrance R., Hendrix P. and Odum E. 1986. A hierarchical approach to sustainable agriculture. American journal of alternative agriculture 1:169–173.

Pastore G., Giampietro M. and Li Ji. 1998. Understanding the dynamics of rural intensification in China: Land-time cross budget of five villages in Hubei Province. Critical reviews in plant sciences, special issue (in press).

Wolf S.A. and Allen T.F.H. 1995. Recasting alternative agriculture as a management model: The value of adept scaling. Ecological economics 12:5–12.