Resource management

Economic exchanges

Land-use dynamics

Role-playing games

Theoretical models

Didactic models

Coupled models

All models

FloAgri: simulations and main results

Bommel, P., Sist, P., Piketty, M-G., Bendahan, A. B.

Back to the introduction page of the model. Back to model description

Each agent owns a single lot of 100 ha and it is not allowed to move to another farm nor to expand by buying new lots. Each lot has a similar vegetation cover. The agents vary only on their specific activities (FM, PFAC and/or LR compliance). For standard simulations, each family has 3 workers and 3 children. It starts with 7 200 R$ (members × initial cash/person = 6 × 1 200 R$) and 468 available workdays (workers × work days per season = 3×156). In its standard version, the model is deterministic: no randomness is involved during the simulations. Thus, from a given initial state, it will always produce the same output for each strategy (Note that a stochastic version -with randomness- will be used in the analysis in order to test the impact of sickness probability or when specifying a new initial land cover).

Fig. 1: Example of land cover evolution for a standard simulation for 6 different farms

Starting from the same state, the land cover at the end of a simulation (40 years) is quite different depending on the performed scenario. After 12 years, the conventional breeder (StandStrat) for example has converted all his land into pasture (this period is called installation phase). In contrast, the agents in charge of FM as well as the agents who are artificially obliged to respect the law, have maintained their LR (80 ha). Their installation phase is much shorter (5 years).

By looking at the agents’ incomes, as noticed previously, the output data cannot be rigorously compared with real incomes as the agents keep their standard of living whatever their savings. What is more relevant is to compare the economic results between the agents in order to rank the incomes’ levels according to the scenarios.

Scenarios results

The impacts of FM and PFAC for agents respecting their Legal Reserve

Given the fact that the agents FmStrat, Fm+Pfac, StandLR and PfacLR do all respect their LR, the efficiencies are compared on the same area available to cultivate (20 ha). Indeed, at the end of the third year, each agent has deforested his authorized surface. The rest remains forest and only the agents FmStrat and Fm+Pfac can use it to harvest timber during the FM cycles.

Fig. 2: Evolution of the annual incomes of 4 agents respecting their LR, during a standard simulation of 40 years.

The standard breeder obliged to preserve his LR (StandLR) without any alternative activity, has an averaged income of 13 000 R$/year for three workers (calculated from year 10 to 40, i.e. after the installation phase), that is below the minimum salary (the Brazilian minimum salary in 2009 was 465 R$/month, see Portalbrasil, 2009). During the 5 first years, the family revenues originate mainly from annual crops production and from timber sales when converting forest into pasture. During this installation phase, the smallholder invests intensively in livestock by buying cows and implanting new pastures. When the authorized 20 ha are fully covered with pasture and because the smallholder cannot buy new land, he manages his herd and sells out the remaining available working days of the household. The regular fluctuations of the livestock income are due to the sale of the cull cows and the purchase of heifer calves.
FM increases the incomes of agents FmStrat and Fm+Pfac every 30 years. The first growth peak is smaller than the second one because the agents invest in livestock: while implanting new pastures and buying many cows, they transfer a high part of the first FM profits into livestock capital.
Between the timber’s harvesting periods, incomes of agents FmStrat and Fm+Pfac drop to a value closed to that of agent StandLR. Nevertheless, even during this forest regeneration phase, FM has a positive effect: mean income during the renewal phase is approximately 365 R$/month/worker for StandLR, while it reaches 556 R$/month/worker for FmStrat and Fm+Pfac, which is 52% better. Such difference is linked with livestock and pasture quality: short of savings, agent StandLR does not succeed in maintaining all the pastures and the cattle that gradually degrade and as a result, become less productive (eq. 1). Thus, the mean monthly income of livestock for agent StandLR is 344 R$ when it is more than twice better (797 R$) for agent FmStrat. In this scenario, FM incomes provide some financial resources allowing the smallholder to maintain the productivity of livestock and pasture.

Agent PfacLR starting phase (10 first years) is as difficult as that of agent StandLR. But, by adopting permanent crops, this agent succeeds in gradually recovering the degraded pastures and thus is able to reach higher cattle productivity. This agent also increases the household incomes by harvesting permanent crops. In counterpart, as PFAC needs more labor for implantation and maintenance, the agent sells fewer working days. Nevertheless, after 12 years, the annual income becomes higher than that of StandLR, and it is equivalent to FmStrat and Fm+Pfac after 20 years. PFAC provides positive effects on family’s income but at mid-term (15 years).

The following diagram shows the annual incomes of the agents according to the structure of the family. The same former simulations have been run but for a family of 8 members with successively 1 to 8 workers.

Fig. 3: Annual incomes of the agents according to the number of workers.

Not surprisingly, whatever the agent strategy, annual income increases as the family has more workers. This increase is mostly due to higher number of working days available to be sold. However, for the breeders who do not perform FM (agents StandLR and PfacLR), an income shift is observed when the workers are more than four: above this number, the family is able to maintain and clean up all its pastures, thus the difference of annual income with agents performing FM is low and solely due to the timber harvest peaks. Below four workers, the family lacks labor force to maintain all its crops and pastures which productivity decreases. We can thus conclude that, when subcontracted, FM is mainly useful for families with few workers as it helps to compensate the lack of manpower.

The cost of staying into legality

This section assesses the cost of staying into the legality by preserving the LR. As stated by the old Forest Code but also by the new code in some sensitive areas, the law required the landowners in the Amazon to conserve 80% of the farmland in forest. The following simulations compare agents that respect their LR (agents FmStrat, Fm+Pfac and StandLR) with those who do not (StandStrat and PfacStrat). Figure 4 shows the evolution of the agents’ annual income. The installation phase (time to cut off the forest) is much longer for agents StandStrat and PfacStrat (12 years) whereas it is 5 years for agents FmStrat, D, StandLR, PfacLR. During this phase, they face low income (short farm production and investment in livestock). But after this period, when they have completely turned their farm into pasture, agents StandStrat and PfacStrat succeed in increasing their savings more rapidly than the other agents who kept their LR. Such shift is linked with the livestock incomes on 100 ha instead of 20 ha that do not require much manpower for its maintenance (300 working days/year for 100 ha of pastures with cattle).

Fig. 4: Evolution of the annual incomes during a standard simulation of 40 years, according to the 4 agent types

When the standard breeder get 100 ha of pastures and because he cannot buy new land, his activity consists mainly in managing his herd then selling the extra labor force. For these specific simulations, FM does not compensate the breeding loose due to pasture limitation. Excluding the installation phase, the monthly income of agent StandStrat is about 4 250 R$, whereas it is only 1 670 R$ for a breeder doing FM, i.e. 60 % lower.
Agent PfacStrat has similar results than StandStrat. As he needs more manpower to implant and maintain PFAC, his income is a little lower than agent StandStrat’s. Unlike agent PfacLR on 20 ha, PFAC has no positive effect when using the 100 ha, even after a long period.

The new Brazilian Forestry Code allows, in certain zones (consolidated zones, defined by the ZEE plan), to use and deforest until 50% of the farm:

Fig. 5: The annual incomes of the agents when LR is set to 50%.

When authorized to cultivate half of their lot, StandStrat and PfacStrat have the same incomes than previously (obviously, because they do not care about LR). But a change of LR ratio increases the monthly incomes of the agents that keep their LR: from 1 090 to 2 500 R$ for agent StandLR (+ 130 %); 3 035 R$ (+81 %) and 2 680 R$ (+74 %) for FmStrat and Fm+Pfac. Compared to agent StandLR, the FM has lower impact on incomes, except for the first years of installation.
Thus, economically speaking, a reduction of LR to 50% would better help the small farmers than applying FM on 80 ha. On the other hand, if LR were complied (that is seldom the case), it would decrease the forest size from 80% to 50 %.

Portion of property initially deforested

To ensure a good understanding of the model, it is also necessary to initialize the simulations from initial states with less virgin forest, as it is the case nowadays for a majority of smallholders in Pará and Mato Grosso. Modifying the initial deforestation part, from 0% to 100%, has resulted in the following output. For each new initial state of the farms, deforested plots are randomly covered by annual crop, pasture or fallow, with a random degradation level. In the cases where the deforestation is greater than the authorized portion (20%), the agents that respect the law (FM and LR strategies) will not use a part of their land (covered by fallows and old crops) so that forests can regenerate. But, as forest regeneration is a very slow process, the simulations show the full restoration of the LR after a long period (about 38 years according to initial degradation level). In that case, the agents undertaking FM can harvest timber even if their forests are below 80 ha (which is something that is not authorized in practice). For each value of the initial deforestation portion parameter, 100 simulations are repeated in order to normalize the effect of the randomness.

Fig. 6: Mean annual incomes according to initial deforestation.

The analysis shows that the incomes of the agents StandStrat and PfacStrat, who do not respect the LR constraints, decrease rapidly with the initial degradation rate. This is because the farmer spends time and money to maintain degraded pastures that are basically unprofitable owing to their poor productivity. As the beginning of the settlement is very sensitive, loosing time and money for land recuperation during the first stage holds back the development of the agents. When 40% of the farm is deforested, the standard breeder has the same income than StandLR, i.e. he is mainly a farm laborer.
For the agent FmStrat, in addition to spend energy to maintain unprofitable pastures on 20 ha, the lack of preserved forest reduces the FM profits and has more negative effect on the incomes compared to Fm+Pfac, StandLR and PfacLR.
The signatures of the agents doing PFAC (PfacStrat, Fm+Pfac, PfacLR) are dissimilar: Fm+Pfac and PfacLR that work on 20 ha are not affected by this parameter. Because they do not spend much in livestock, they are able to recover their degraded pastures and manage a small area whatever the initial degradation level. Fm+Pfac has the best income when initial deforestation is higher than 25%. On the opposite, PfacStrat has bad resistance to initial degradation, mainly for low deforestation rate. This is because he spends a lot of energy to implant permanent crops while investing at the same time in livestock and expanding on the forest. So, when starting from degraded land, PFCA presents positive effects but solely when the farmer does not invest heavily in livestock at the same time. Depending on the degraded surface, the period to get PFAC positive effects can be long. By comparing for example the dynamics of StandStrat and PfacStrat from a 20% degraded farm, the mean income of StandStrat over the 40 years’ simulation is higher than of PfacStrat. But PfacStrat has much better incomes solely after a long period (20 years, not visible on figure 6).

FM, PFAC and resilience  

Analyzing the effect of hazard on smallholders is important since the family workforce plays a key role for the household strategies. For the previous analysis, the sickness probability was set to zero but the available statistics show that, in the Amazonian pioneer fronts, hazard is greater and the risk of becoming seriously ill or being injured is higher than elsewhere. With this regard, IBGE (2008) estimates that a person stays in hospital once every 10 years (5 % /season/person).
For the current analysis, the sickness option has been activated. In the next graph, the sickness probability has been tested from 0% to 12% per season (analyzing upper probability is not useful since the available labor force is almost zero). At each season, every member of the family has a random chance to be sick and to be unable to work during this time. Due to stochastic events, the simulations are repeated 100 times for each parameter new value.

Fig.7: Annual incomes of the agents according to sick probability per member. 0.05 is the standard risk from literature.

Obviously, all the agents loose income at the extent that hazard risks increase. Interestingly, agents StandStrat and PfacStrat that do not respect their LR are more sensitive to hazard. They can hardly expand and maintain their pastures and cattle when they lose manpower. In contrast, agents managing their forests (FmStrat and Fm+Pfac) tend to be more resistant to hazards. This because of the extra revenues originated from FM that make them much less affected by a temporary loss of labor. As PFAC requires additional investments in money and manpower, PfacStrat and PfacLR are more sensitive to hazards. FM compensates this sensitivity in the case of agent Fm+Pfac. Figure 8 shows the distribution of incomes when hazard probability is set to its standard value (5 % per season/person). In addition, it compares the effect of LR authorized level for the two scenarios 80 % and 50 %.

Fig. 8: Annual incomes when LR is set to 50% or 80%, with 5% risk probability. Sorted distribution of the incomes for each simulation (left chart). Averaged incomes over 100 simulations and standard deviation (right chart).

The bar chart (right) presents the averaged income over the 100 simulations of each agent. The agents StandStrat, FmStratLR: 50% and Fm+PfacLR: 50% have similar mean incomes (about 30 000 R$/year). But the standard deviations are quite different: StandStrat has irregular incomes when FmStrat and Fm+Pfac are much more regular. This irregularity of incomes is more visible on the left chart: here, each point is the annual income of an agent for one simulation. By sorting these values for 100 simulations, the curves show how these incomes are distributed. For standard breeder exploiting 100 ha, the chances to succeed are spread from highest to lowest incomes (high variance): in more than 20 % of the simulations, StandStrat fails to manage his farm. When an accident or a disease occurs in the installation phase, the consequences strongly impact this agent’s economic situation and his farm productivity. The agent is more able to resist to hazards when his expansion phase is complete. These results are worst for agent PfacStrat who fails in 60% of the cases: this agent is more sensitive to hazard during installation phase. As PFAC needs investments, it is only profitable in the long term when the family is strongly installed.

Discussion : Optimal conditions versus hazard and degraded lands

As stated by K. Popper for theories, a model cannot be validated in the sense of having a complete confidence on its outputs. Like theories, a simulation model is inherently wrong: sometimes it can be corroborated by data but in most cases, it is refuted by empirical observations. Comparing the model outputs with data does not authorize to conclude its definitive validation; it is just a way to increase the likelihood of its results. As the alternative practices developed in the FloAgri project are recent and not broadly used in the Amazon, it is also hard to compare the model outputs with real data. But, while the lack of historical data prevents us to “validate” the model, its dynamics are coherent with observations and expert knowledge. Instead of comparing static data on each practice independently (performing a cost/production comparison is quite simple), the ABM helps to understand how various activities can interact. It helps to explain why, depending on the context, some practices can be efficient or either fail. Of course, it should be interesting to compare our results with an optimization model that aims in determining the best mixture of land use patterns. Nevertheless, it should be emphasized that our ABM does not seek to optimize the land use distribution. Since it focuses on the behavior of agents in a temporal dimension, it helps us understand the reasons for success or failure of a strategy in various situations.
Our results show that without any agricultural alternative, a small breeder compliant with LR has low income that comes mainly from the sale of his labor force. StandStrat change from 80% to 50% of LR would obviously help him in improving his living conditions.
The model shows that FM and PFAC are not miraculous solutions allowing a small holder to prosper while preserving his LR. Compared to a standard breeder artificially restricted on 20 ha (StandLR), the adoption of alternative practices can increase his income (twice his standard income in the best case). But what he earns from 20 ha of crops and pastures, plus the revenue he obtains occasionally from FM, remain lower than the income he gets from 100 ha of pasture.
The additional practices, however, may have positive effects when various types of hazard are introduced in the model. For example, the standard breeders are more sensible to sick probability than the agents performing FM. As the risk of being seriously sick or wound is higher in the Amazon, FM increases the resilience to hazards. The additional revenue generated by FM appears useful when the labor force is reduced because it helps in investing in productive livestock that do not require a lot of work.
PFAC has positive effects when a reasonable part of the farm is degraded. By using no-till techniques, it allows recovering some productive pastures when standard breeder can just maintain poorly profitable cattle. But PFAC is a risky activity because no-till farming needs competences, manpower and financial investment. The model shows that PFAC weakens the farmers when they are in installation phase: they cannot invest in such technique and livestock at once. By gradually recovering degraded pastures, PFAC can be seen as an intensification practice that improves cattle productivity on the long term.
As a result, when starting from a degraded farm and when taking into account hazard probabilities, the FM may offer equivalent mean income than for unrestricted standard breeder, but with higher stability. Finally, coupling FM and PFAC seems to be an interesting alternative that might make the farmers more resistant to hazards with higher income, while at the same time preserving the major part of their forest.

Conclusion

By implementing agricultural activities related to standard or specific strategies, an ABM offers the advantage of explaining how several actions can complement or compete. In contrast, a simple comparison of static economic returns between different activities does not inform about such complementarities. By targeting the model on the management of the labor force, the simulations allow us to explain how some actions may affect the achievement of others. Thus, our ABM helps to understand why some alternative practices can be efficient or either fail, depending on the context.
As several studies have already pointed out, succeeding in making small-scale forest management plan financially viable in the Amazon region remains a challenge and many barriers have still to be removed (Hajjar et al 2011, Drigo and al. 2012). It is considered relatively risky in the short-term to invest in forest management whereas investing in cattle is often considered as a safety option, at least in the short-term. The latter situation may inverse in the long-term related to efforts to decrease the many barriers threatening the long-term viability of forest management (Drigo and al. 2012).
However, standard breeding, which is the most common production system among smallholders in the Amazon, is also risky during the installation phase. When taking into account the current available cash and manpower of a family, our ABM shows that a breeder quickly meets difficulties that prevent him from developing as it would do under optimal conditions. The model reveals that the agents that prefer to pursue ranching activities by converting all their land to pasture, are economically vulnerable at the beginning and they can easily fail due to increasing risk and land degradation.
In contrast, the agents who are able to manage their forests tend to strengthen their resilience capacities against hazards mainly when considering sickness probability. When subcontracted, the FM cannot improve much the mean annual income of smallholders in the Amazon but it tends to stabilize this income and improves the household resilience in the long-term by helping to maintain pastures and livestock productivity.
Adding PFAC increases potential benefit for families to invest in FM underlying potential interesting complementarities. When few pastures are degraded, PFAC improves cattle productivity on the long-term. Nonetheless, as PFAC is an intensification technique, it cannot be performed while investing in livestock.

Back to the introduction page of the model.

Back to model description

Centre de coopération internationale en recherche agronomique pour le développement Informations légales © Copyright Cirad 2001-2015 cormas-webmaster@cirad.fr