Eugene D. Ungar
The integration of wheat and sheep production has been examined in the field ever a number of years at the Migda experimental station in the northern Negev. Research at Migda has aimed at determining the potential primary and secondary production in such an environment and in designing farming systems that could be implemented widely in the region. These systems would aim to provide a more stable income than the pure wheat production system that currently predominates.
In conjunction with research at Migda, a modelling study was conducted to examine the management problems involved in operating intensive agropastoral systems in a semi-arid environment. The aim was to develop a procedure for improving decision making, with an emphasis on management options created by integration with wheat production.
The system studied is lamb production from a flock of sheep, of constant animal. number from year to year, reproducing once a year at fixed dates. The flock is sedentary, grazing a rain-fed area (individual use) consisting of annual vegetation (all species of similar growth and palatability) in a semi-arid, winter-rainfall zone with mild to cool winters. The pasture is fertilized and the animals are supplemented to 'optimal' production. The economic environment is characterised by a high meat: grain price ratio. There is no drinking water limitation. Most importantly, the pastoral component is integrated with small grain production (wheat).
The approach adopted was to develop a series of optimisation algorithms for individual management decisions that can be usefully examined in isolation. The system model is built around these algorithms, rather than constructing a single biological simulator for the entire system with a large set of management control parameters. In this problem-oriented approach, the algorithms were developed independently with little emphasis on consistency of resolution or precision between them. Each optimisation algorithm is based on relatively simple biological formulations that seek to incorporate the essential dynamics of the relevant subsystem.
Emphasis was placed on using the models (1) to derive practical guidelines and rules for optimal management that would obviate the need for a model by the decision-maker and (2) to indicate not only the optimal management decision but also the cost of poor decision-making resulting from indifferent management, inadequate model formulation or poor field monitoring for parameterisation purposes.
By linking the individual decision algorithms to a relatively detailed biological simulation framework, system response to structural management decisions, such as stocking rate and land allocation, can be examined. In addition, using a higher-resolution whole-system simulation framework enables the optimality of solutions derived from simpler models to be evaluated.
Under deferred-grazing management, the flock is generally maintained in a holding paddock on supplementary feeds. The cost of feeding can be high since this period usually coincides with pregnancy or early lactation. A significant reduction in these feed costs can be achieved by grazing green wheat during some part of the deferment period. Experiments at Migda indicate that there is a period of at least six weeks after emergence of wheat during which defoliation does not cause a significant reduction in grain yield. Beyond this period, defoliation reduces grain yield, the effect on yield increasing with lateness and severity of defoliation. Thus, if grazing is restricted to the no-damage period, the problem can be formulated as minimising supplementation costs over the possible wheat grazing period. In a target-oriented approach to animal (ewe) feeding, this is equivalent to maximising herbage utilisation over the same period. The problem is therefore to determine the day of flock entry to the wheat that will maximise cumulative herbage consumption.
The problem has been defined in terms that are closely related to the management decision itself rather than in terms of maximising overall system profitability. In the latter case, an optimal solution may well have been found using a total-system simulator. However, the solution would most probably have appeared arbitrary in the sense that the underlying, reasoning or principle that determines the solution may have remained obscured in the sheer impenetrability so characteristic of total-system models. It is suggested that the insight into the essential dynamics of a problem that can be gained from analysis with simple models is more important than any claim to increased accuracy of solution that the 'big' model can provide. Perhaps the view should be added that simpler does not necessarily mean less explanatory or mechanistic.
The simple model used to locate the optimal day for stock entry is based on an exponential growth function (realistic for the early growth stage) and a ramp consumption function. All sections of the study connected with grazing dynamics were strongly based on the analyses of Noy-Meir (1975a; 1975bs; 1978a: 1978b). Table 1 briefly describes the model and lists the parameters required and a guess at the ease of monitoring them on-farm. Clearly, if the algorithm were to be applied at a specific location, a number of parameters would have to be based on other, possibly very different, locations. The relative growth rate (g) and stocking rate (H) are two parameters most likely to vary between locations and so these were chosen as criteria for exploring the response space. Figure 1 shows the response surface of mean daily intake rate per animal during the grazing period (for the optimal entry day) to stocking rate and relative growth rate. A sensitivity analysis was carried out in which the mean daily intake rate per animal was calculated for a fixed entry day of 25 days after emergence for the same H–g space. The shaded area in Figure 1 represents zones of this space for which an entry day of 25 days after emergence yields an intake rate with 10% of the optimised value.
On the basis of this analysis one might tentatively conclude that, for a specific site, a single robust solution for stock entry day can be derived given reasonable parameter estimation. The robustness, or otherwise, of the system as a whole to this management decision will be discussed later.
The management decision of when to commence grazing of green pasture is a major determinant of pasture dynamics. The optimal deferment period is that which enables maximum herbage utilisation. Utilisation can be defined in terms of green-herbage consumption and dry-herbage consumption weighted according to their relative nutritive values. In addition, in the integrated agropastoral system, the lower requirement for dry-pasture herbage due to the availability of wheat aftermath can be taken into account.
Table 1. Model to optimise early-season grazing of wheat: description, assumptions and required parameters.
Problem: Find entry day to wheat (a) such that cumulative herbage consumption between decision time and final wheat grazing day is maximised. Model:
Maximise
Herbage growth rate (logistic): Herbage consumption rate (negative exponential): C = H min(s(V - Vr), cs) ., Important auxiliary assumptions: Constant animal performance (=target - oriented approach), therefore constant energy requirements which must be met, therefore cost of deferment does not effect solution. Interactions between time of stock entry, sward canopy structure and herbage quality are ignored. Required parameters Ease of on-farm monitoring* |
|
| Number of ewes/hectare wheat (H) | 1 |
| Number of days since emergence | 2 |
| Final wheat grazing day (x) | 2 |
| Residual ungrazable biomass (Vr) | 5 |
| Current biomass (V) | 3 |
| Relative growth rate (g) | 4 |
| Grazing efficiency parameter (s) | 5 |
| Satiation intake rate of wheat (cs) | 5 |
* 0 = easy/readily available information, 5 = extremely difficult.
The simple model used to determine the optimal stock-entry biomass or day is based on a logistic growth function and a negative exponential consumption function. Table 2 briefly describes the model and required parameters.
Figure 1 . Results of the early-season wheat grazing algorithm. Contour map of the mean daily herbage intake rate (kg/animal/day) as a function of relative growth rate of wheat per day and stocking rate (animals/ha wheat) for the period from the optimal stock entry day to the end of the early-season grazing period
Table 2. Model to optimise green-pasture grazing deferment: description, assumptions and required parameters.
Problem: Find entry day (a) to green pasture such that cumulative herbage utilisation between decision time and expected start of next green season is maximised. Model:
Maximise
Herbage growth rate (logistic): Green herbage consumption rate (negative exponential): C = cs ( 1 - exp( - (V - Vr) / (Vs - Vr) Important auxiliary assumptions: Constant animal performance (=target-oriented approach), therefore constant energy requirements which must be met, therefore cost of deferment does not effect solution. Required parameters Ease of on-farm monitoring* | |
| Cumulative green pasture consumption (C) | computed |
| Time of end of green season (long-term average) (x) | 2 |
| Dry pasture biomass remaining at the end of the green season (V1) | computed |
| Peak undisturbed pasture/wheat biomass ( long-term average) (Vm) | 5 |
| Fraction of peak wheat biomass that remains available for grazing after harvest (long-term average) (l - I) | 4 |
| Ration of wheat to pasture area (R) | 1 |
| No. of days from end of green season to start of the next (long-term average) | 2 |
| No of ewes/hectare pasture (H) | 1 |
| Satiation intake rate at green/dry pasture (cs) | 5 |
| Relative value of dry: green pasture (F) | 5 |
| Relative growth rate (long-term average) (g) | 5 |
| Current biomass | 3 |
| Residual ungrazable biomass (Vr) | 5 |
| Biomass for intake satiation (Vs) | 5 |
* 0 = easy or readily available information, 5 = extremely difficult
It seems reasonable to venture, as a crude generalisation, that the number of parameters in a model stands in inverse relation to the experimental or analytic effort invested in establishing their values. In this study, it was possible to devote considerable effort to establishing the intake function and long-term average parameter values for the logistic pasture growth function. In the latter case, a relatively high resolution model was used to simulate 20 growing seasons and the simulated results were analysed statistically. Sometimes, though, the simple model becomes so succinct and abstract that the parameters become meta-parameters and not directly measurable. This objection can be countered by the fact that this implied distinction between theoretical construct and observable data has yet to be successfully argued by philosophers of language.
Figure 2 shows the relationship between the herbage consumption objective function and length of deferment for various stocking rates. Note the increasing sensitivity to decision-making with increasing stocking rate. One way of analysing the cost of poor parameter estimation is to map out the response surface of the decrease in objective function to actual and estimated parameter values. An example is shown in Figure 3 for the relative growth rate parameter. To construct this response surface, the optimal entry day is calculated using an estimated g value in the optimisation algorithm. This deferment period is then implemented in the deferment model using a wide range of actual g values. The consumption objective function is calculated for each actual value taken and the reduction in cumulative consumption caused by a suboptimal deferment period is computed. The response surface of forfeited utilisation is not exactly intuitively obvious. However, with a model of such simplicity, it is not a major undertaking to 'take it apart' and find out why such surfaces are obtained.
Figure 2. Results of the green pasture grazing deferment algorithm: the relationship between normalised herbage consumption and the length of the grazing deferment period, for various stocking rates
Figure 3. Results of the green pasture grazing deferment algorithm: the response surface of forfeited intake (kg/ha) resulting from poor parameter estimation to actual and estimated relative growth rate (g), at a stocking rate of 5 ewes/ha of pasture
The management decision regarding supplementary and complete ration feeding of lambs essentially consists of whether or not to provide feed and at what rate. The choice of feed is not considered here: it is assumed that a high-energy and high-protein concentrate feed is available. Since only the maintenance and liveweight change functions are involved in the growing lamb, the problem of lamb feeding is amenable to optimisation.
The system, as defined in the introduction, is limited in the amount of product rather than by the production rate. The product of the number of lambs born, the maximum saleweight and the price of meat constitutes an income ceiling that cannot be exceeded. Hence, annual profit is maximised by maximising profit per unit output rather than per unit time and the optimal feeding level is that which minimises the cost per unit liveweight gain. The fact that time itself may represent a cost in terms of interest and risk does not alter the underlying approach. Such factors can be incorporated into the computation of cost per unit liveweight gain.
Figure 4 outlines the main elements of the feeding optimisation algorithm. Analytic treatment of the problem can yield useful formulae for certain restricted cases only.
Figure 4. Main elements of the lamb feeding optimisation algorithm
The general relationship between cost per unit liveweight gain and supplementation rate is for the cost to decline rapidly with increasing supplementation rate down to the minimum point, and to subsequently increase relatively slowly. If the animal can put on weight at a reasonable rate in the absence of supplementation, then the function is at the minimum at zero supplementation and the cost per unit gain increases over the entire supplementation range.
The problem of monitoring for parameterisation purposes is severe, though perhaps it is misleading to apply the term 'monitoring' to parameters that are (wishfully) treated as universal constants. Even so, there remain items such as lamb liveweight and pasture intake. An appealing solution to that problem is to simulate what you cannot measure, which has the additional advantage of greatly simplifying model validation. Given these limitations, it becomes even more imperative to extract rules of operation which can be implemented relatively easily. Towards this end, the following conclusions were drawn on the basis of extensive runs of the feed optimisation algorithm.
The management problem of lamb rearing consists of selecting a rearing pathway that maximises profits. The rearing pathway is a nutritional time course where nutrition is determined by the physical location of the lamb in the system, whether or not the lamb is sucking, and the supplementary feeding regime. In an agropastoral system, seven nutritional locations can be defined: holding paddock, pasture, wheat (each sucking or weaned), and fattening unit (weaners only). In the expanded system, there may also be a special-purpose pasture for lamb fattening, as a forward creep or for weaners. (The fattening unit and holding paddock are nutritionally equivalent.)
The rearing options can be defined in a lamb-movement matrix which specifies the possible flow links between each. of the rearing locations. A standard configuration is shown in Table 3. Having defined the optimal supplementation level for any given location on the basis of minimum cost per unit liveweight gain, the same principle can be extended to selecting between locations. The optimal location at any point in time is the one that provides the lowest cost per unit liveweight gain. Thus the lamb-movement algorithm predicts lamb performance for each possible alternative, as defined by the movement matrix, and does so at the optimal supplementation level for each one. Lambs are moved to the location that provides the lowest cost per unit gain. Using this approach, it is not necessary to arbitrarily set criteria for weaning, supplementation or lamb sale.
Table 3. The standard configuration of the lamb-movement matrix — a definition of the possible flow links between nutritional locations of a lamb in an agro-pastoral system.
To | ||||||
From |
Holding Sucking |
paddock Weaned |
Pasture/ Sucking |
wheat Weaned |
Spec.- purpose pasture weaners |
Fattening unit Weaners |
Holding paddock | ||||||
– sucking |
1 |
1 |
1 |
1 |
1 |
1 |
– weaned |
0 |
1 |
0 |
1 |
1 |
1 |
Pasture or wheat | ||||||
– sucking |
1 |
0 |
1 |
1 |
1 |
1 |
– weaned |
0 |
0 |
0 |
1 |
1 |
1 |
Special purpose | ||||||
Pasture |
0 |
0 |
0 |
0 |
1 |
1 |
Fattening unit |
0 |
0 |
0 |
1 |
1 |
1 |
1 = possible flow.
0 = non-feasible or restricted flow
This crude short-term optimisation approach is inadequate if the response surface of profit to rearing pathway has local optima that represent a significantly lower total income to the global optimum. This would mean that there are circumstances in which it might be more profitable to suffer poor economic performance in the short term in order to follow a pathway of high income accumulation later. Such a possibility is not taken into account in this approach. However, the lamb-movement matrix enables the optimality of the solution under the standard matrix setting to be checked by simply using the matrix to force alternative rearing pathways. Testing of the algorithm requires a total system model with the necessary interfacing with pasture production and ewe performance. Results from the algorithm will be discussed later.
The various short-term decision algorithms, including a number not described above, were incorporated as subroutines into an overall biological simulation framework. Primary production (pasture, wheat, special-purpose pasture) was based on ARID CROP (van Keulen, 1975). Secondary production of ewes and lambs was based on ARC equations (ARC, 1980), some functions based on Migda data and a few speculative hypotheses regarding the response of ewe lactation to level of nutrition. The primary and secondary production sections here interfaced by a low-resolution intake subroutine. (A detailed description of the model is given in Ungar, 1984.) The model was implemented using 21 years of climatic data for Migda in the northern Negev. The standard run refers to a land allocation of 50% natural pasture, 50% wheat (no special-purpose pasture), a stocking rate of 10 ewes/ha of pasture and a meat grain price ratio of 10:1. Ewe prolificacy is low-to-medium relative to the potential in the field.
The following results were obtained at the system level regarding the short-term management decisions discussed earlier.
Early-season grazing of green wheat: Despite the seemingly significant saving in feed costs that should derive from utilising early-season green wheat, there are compensatory effects operating at the system level that reduce the opportunity cost of this management option (Table 4). Firstly, by not grazing the wheat the ewe spends more time in the holding paddock but the total seasonal energy requirements of the ewe are thereby reduced. Secondly, early-season grazing reduces the peak biomass and hence quantity of wheat aftermath for dry-season grazing. Though the straw baling decision algorithm was not described above, Table 4 shows the response of the system to various management options regarding aftermath utilisation. The (normal' straw-baling criterion is simply to bale everything in excess of the expected requirement until ploughing time, while the remainder is left in the field to be grazed. Unless the manager makes obviously stupid decisions, he is generally operating in an insensitive zone of response space.
Table 4. Comparison of alternative strategies of wheat utilisation in feeding the ewe. Alternatives are arranged in descending order of profitability.
Early wheat grazing |
Wheat aftermath grazing |
Straw baling* | Gross Margin |
No |
Yes |
Max |
109.9 |
Yes |
Yes |
Max |
106.6 |
No |
No |
Rega |
105.2 |
Yes |
No |
Regb |
103.0 |
Yes |
Yes |
Regc |
97.7 |
No |
Yes |
Reg |
93.1 |
Yes |
Yes |
No |
82.1 |
No |
Yes |
No |
71.6 |
Yes |
No |
No |
68.1 |
No |
No |
No |
56.3 |
*Max = always bale if biomass > 1000 kg/ha, Reg = normal straw baling criterion.
a. Equivalent to No No Max.
b. Equivalent to Yes No Max.
c. Standard run.
Gross margin is mean for 21 years, in $/ha.
Green-pasture grazing deferment: The optimality of the optimisation algorithm for grazing deferment was examined using the relatively complex system model. The optimal entry biomass (OEV) supplied by the deferment algorithm was multiplied by a factor ranging from zero (no deferment) to 1.2, and the adjusted value used to determine stock entry in the system model. The results for 21-year runs are summarised in Table 5. The gross margin (GM) response to deferment is quite flat over a broad range at the standard stocking rate. Factors such as the relative timing of lambing and break of season, changes in the lamb-rearing pathway with changes in OEV and even the length of the management decision time step can contribute to the buffering capacity of the system.
Table 5. Analysis of sensitivity to the length of the grazing deferment period. Deferment is defined in terms of the optimal entry biomass (OEV) as determined by the simple grazing deferment model.
OEV |
Gross margin |
0 |
86.3 |
0.1 |
92.8 |
0.2 |
98.3 |
0.3 |
103.0 |
0.4 |
104.8 |
0.5 |
106.0 |
0.6 |
106.7 |
0.7 |
103.3 |
0.8 |
102.0 |
0.9 |
99.5 |
1.0* |
97.7 |
1.1 |
96.8 |
1.2 |
95.3 |
Gross margin is mean for 21 years, in $/ha.
*Standard run.
The mean GM for the non-adjusted OEV was only 8% below the maximum value obtained. However, there is really no justification to assume that the system-model optimum is necessarily more accurate than that generated by the deferment algorithm. Rather, such a test of optimality lends confidence to the approach of utilising a low-resolution algorithm for decision making.
Lamb feeding and rearing: The approach taken to lamb rearing via use of the lamb-movement matrix opens up a large range of rearing permutations. Despite this, the algorithm almost always selected fairly conventional patterns and, in so doing, offered a straightforward explanation as to why such pathways are indeed preferable.
The overall performance of the rearing algorithm can be summarised as follows:
Figure 5. Bar chart of weaning frequency according to lamb weight and age, based on 189 seasons (nine sets of 21-year runs) using various combinations of stocking rate and land allocation
By forcing the selection of alternative pathways, the lamb-movement matrix can be used to evaluate the optimality of a given pathway. An example of such an exercise is shown in Figure 6 in which four alternative rearing pathways were examined for the 1976/77 (poor rainfall distribution) season, at a stocking rate of 7.5 ewes/ha of pasture. Clearly, the principle of short-term optimisation in lamb rearing does not always lead to a global optimum. However, in all cases examined, the default matrix setting yielded solutions close to the 'after-the-fact' global. optimum (found by trial-and-error). Here again, there are compensatory effects inherent in the biology of the system that tend to reduce the opportunity cost of suboptimal lamb rearing. This is evident to some extent in the supplementation values sham in Figure 6.
Figure 6. Comparison of four alternative lamb rearing pathways for the 1976/77 season
Is the agropastoral system robust, insensitive, sluggish to respond to management? (Can such assertions be extended to complex agricultural. systems generally?) Does management matter or is the multi-dimensional response space characterised by gently undulating plains? Perhaps looked at from a sufficient distance, any surface appears flat, and the perspective of the observer is what really counts. For the farmer, a small 'pot hole' might mean a discernable difference in his income — and sensitivities are generally high to reductions in income. For the planner, such variation may simply be noise.
In one sense, the economic performance of agropastoral systems must be robust by the mere fact that (a) they are practiced at all in one form or another under conditions of weather unpredictability, price uncertainty, and total absence of quantitative monitoring, and (b) if managers were confronted by precipitous cliffs on all sides, many of the questions examined in this study would have been solved (at least empirically) long ago and there would have been no need for the model.
On the other hand, there are clearly dangers in operating such systems. Firstly, the higher the system intensity (e.g. stocking rate, prolificacy) the less leeway does the manager have in decision making. Careful monitoring becomes more important and a greater responsiveness in management is called for. Secondly, it should be noted that the management model is very much self-correcting in that all short-term decisions are based on the state of the system at the decision time (as opposed to fixed strategies) and therefore respond to errors in previous decisions. Operating according to fixed strategies is likely to exhaust the compensation capacity of the system very quickly.
So the answer for the farmer might be, if management standards are low performance might be very erratic. If management standards are good, a high degree of robustness can be expected. Nevertheless, the difference in income between excellent and good management would seem to be very significant.
ARC (Agricultural Research Council). 1980. The nutrient requirements of ruminant livestock. HMSO, London, UK. 351 pp.
van Keulen H. 1975. Simulation of water use and herbage growth in arid regions. Simulation Monographs. PUDOC, Wageningen, The Netherlands. 176 pp.
Noy-Meir I. 1975a. Primary and secondary production in sedentary and nomadic grazing systems in the semi-arid region: Analysis and modelling. Research report to the Ford Foundation, Project 7/E–3. Department of Botany, the Hebrew University, Jerusalem, Israel.
Noy-Meir I. 1975b. Stability of grazing systems: an application of predator-prey graphs. J. Ecol. 63: 459–481.
Noy- Meir I. 1978a. Stability in simple grazing models: effects of explicit functions.J. Theor. Biol. 71:347–380.
Noy-Meir I. 1978b. Grazing and production in seasonal pastures: analysis of a simple model. J. Appl. Ecol. 15: 809–835.
Ungar E D. 1984. Simulation of grazing systems in semi-arid regions. Ph.D. thesis, The Hebrew University of Jerusalem, Israel.
Question – In the model constant prices are assumed. In a situation with fluctuating sales prices it could be profitable to alter the fattening period: take longer for the growing period, and supply less costly feed. Could you optimise under that assumption?
Answer – Not yet, because the assumption of constant prices is critical to this whole approach. With fluctuating prices adjustments must be made using common sense. I started working on this, trying to solve it analytically, but it is tricky.
Comment – You could check the effect of a different feed with a different price and feed value as long as it was constant for the period.
Reply – Yes, and the results did not change, except for a different level at the end. I also want to point out that the interest rate does not matter, i.e. the point of transfer from pasture to fattening units shifted by at most a few days, when ridiculously high interest rates were assumed. So these time-based costs, in a reasonable range, are not important.
Question – Is there a minimum sale weight target for those systems?
Answer – The target weight used is 45 kilograms. In the model the lambs are always sold at the upper sale weight limit, rather than as a result of the cost per unit liveweight gain exceeding the price of meat.
Comment – If there is some minimum, sale weight, then in moving a lamb through a certain pathway several options will disappear in time because they would not produce that minimum weight.
Reply – That is not relevant as long as the animals grow. I work with minimum cost per unit liveweight gain. If time has a cost to it, that cost is included in the equation and converted to the cost per unit liveweigt gain. So, as long as the animals grow they do so at minimum cost until they reach the upper weight limit. An interesting result of the model is that if there is no fattening unit then the lambs are sold earlier, at about 30 kg of liveweight, which is what happened in the Bedouin systems before they started using large amounts of supplementary feed. The reason is that it is not profitable to supplement at pasture due to the higher maintenance requirements of the animal, leading to a cost per unit liveweight gain exceeding the price of meat. In a fattening unit the cost per unit liveweight gain, at the optimal supplementation level, is less than the price of meat and the profits increase by selling at a high liveweight. Having a fattening unit is thus critical to the profitability of the system.
Question – Is the cost per unit liveweight gain not time-dependent because of the maintenance requirements? Higher maintenance costs lead to longer fattening periods and hence to higher costs per unit liveweight gain.
Answer – Yes, therefore at growth rates close to zero the cost per unit liveweight gain increases dramatically. If the growth rate is 1 g/day the costs per kilo are tremendous. But if the system operates continuously at the minimum cost per unit liveweight gain while the animals are growing, it will lead to the maximum profit at the end.
Comment – But the total cost is changing, because both the average cost per unit liveweight gain and the number of days required to reach the target weight are changing.
Reply – That is right, but as long as at each point the gain is produced at minimum costs, the total costs will also be at a minimum. This was substantiated by, for example, forcing early weaning and transfer to the fattening unit onto the model. The target weight was reached much earlier then, but because of the high cost the profit was lower.
Question – Would your conclusion be that such agropastoral systems are more robust than pastures only because they allow the farmer to do almost anything?
Answer – Within limits yes, but the higher the system intensity, i.e. the higher the stocking rate, the more the cliff-edge syndrome is operating. Its position with respect to pastures only very much depends on the meat/grain price ratio. In the present agropastoral system there is a pastoral component and a wheat component. The presence of the wheat aftermath is a very important slack in the system. In the model all kinds of decision algorithm are included that help in deciding whether to bale that straw and feed it indoors or graze it. Again it appears that the trade-offs are very finely balanced, and that the result is insensitive to the decision.
Comment – You have missed a very important point: the inference is that the agropastoral system is more robust because its performance is less sensitive to the decisions of the manager.
Reply – Yes. However, the optimal allocation between wheat and pasture is very dependent on the meat/grain price ratio, as is the optimal strategic set up of the system in terms of prolificacy, stocking rate and land allocation. But all the response surfaces have very flat zones and usually the area of operation is exactly there. If, as a farmer, you take all the wrong decisions you are going to lose money and you are really swimming against the flow of the river. The model should point out which way the river is flowing, where the rapids are, where the cliff edges are. Subsequently the behaviour of the manager depends on what he wants: if he is willing to take high risks, there is a chance of making high profits. Generally it is better to operate in the safe zone, at a reasonable intensity.
Question– Is the greater biological stability not simply due to an overall reduction in stocking rates, due to the presence of that wheat area and to the use of concentrates?
Answer – No; on the contrary, the stocking rate per unit area of pasture at the optimal level would be higher in an integrated system than in a true pastoral system. For the sake of comparison it should be expressed per unit area pasture. The concentrate serves as an important buffer in this system.
Comment – The system may be robust but in some years the situation will be bad and in others it may be pretty good. As a manager, it is about the best you can do, but it does mat mean that because the system is robust in management it is that stable from year to year.
Answer – No, it is definitely not. On the contrary, the higher the system intensity, the higher the variability in income. An interesting result from the model was that, for the 20 years used in the study, an almost linear relation was found between rainfall and income. Repeating the analysis for a higher stocking rate resulted in a relation with a steeper slope, crossing the first one at about average rainfall. Hence; at a higher stocking rate in years with good rainfall income will be higher and in bad rainfall years, the losses will be greater, i.e. the variability is greater, with higher system intensity.
Comment – The same holds for biological productivity: increased system intensity increases variability.
