Agroforestry is not yet widespread in the study area, nor was it prominent in the optimised portfolio under baseline conditions. Here the similarity between the optimal portfolios for risk-averse farmers and the current land-use composition in Tortí (see Bray–Curtis values in Fig.
2) speaks for the plausibility of our model results. Given the Panamanian Government’s policy to increase agroforestry practices in rural areas (MiAmbiente
2019), it is vital to understand the factors that could help facilitate a transition from conventional to more tree-based farming systems among smallholders. Our modelling approach is well suited to this task, because it allows us to look beyond the current land-use composition to investigate theoretically optimal land allocations under different environmental or socio-economic conditions. This scenario analysis allows us to explore the factors that may promote or hinder the selection of agroforestry within a diversified land-use portfolio: an analysis that may prove extremely difficult when relying on empiric methods alone.
Targeting Agroforestry: the Role of Farmer Priorities, Preferences and Attitudes toward Risk
Our model may help to understand the types of farmers for whom agroforestry may be most attractive, helping to target extension programs accordingly. For example, our “Prioritising individual objectives” scenario revealed large shares of agroforestry in the portfolios optimised for risk neutral and moderately risk-averse farmers who prioritise long-term income (quantified through NPV) over the other socio-economic objectives. This suggests that alley cropping and silvopasture may be attractive options for farmers who are more focused on longer-term profit but also more willing to accept risk. NPV could be an especially pertinent indicator for wealthy farmers, who may not depend as much on frequent and regular cash income from pastures or annual crops (Knoke et al.
2020b). The promotion of these agroforestry systems could therefore be targeted towards profit-oriented farmers managing larger farms, who have diversified income sources, including off-farm earnings, that help buffer financial risks (Bowman and Zilberman
2013).
Relying on NPV alone as a selling point for agroforestry, however, may limit the widespread adoption in regions where profit-oriented farmers are the exception rather than the rule. This may be the case in our study area. For instance, Gosling et al. (
2020a) found that the shorter-term goals of maintaining liquidity and meeting subsistence needs (as opposed to long-term profit) could best explain farmers’ current land-use decisions in Tortí. Other studies in the tropics have also found that smallholder farmers tend to prioritise immediate needs related to cash flow and food security over long-term goals of profit maximisation (Affholder et al.
2010; Umar
2013). It is therefore vital to explore the conditions under which agroforestry can help achieve a broader set of farm-level goals.
It is promising that accounting for farmers’ stated land-use preferences as an additional indicator in the multi-criteria model (Fig.
S3) enhanced the share of silvopasture in the optimised portfolio, because it suggests that this agroforestry system is compatible with farmers’ cultural values. In contrast, the lack of alley cropping in this portfolio implies that the silvoarable system may be less socially acceptable for farmers (despite being more profitable and less labour intensive than silvopasture). Cultural values can be important barriers or drivers of agroforestry adoption (Rahman et al.
2017; Tsonkova et al.
2014). Therefore, we would recommend developing and promoting silvopastoral (rather than silvoarable) systems in the study area, to better align with the cultural preferences of local farmers, recognising the importance of cattle for farmers’ livelihoods as a form of insurance and personal savings (Peterson St-Laurent et al.
2013). Nonetheless, demonstration farms that showcase alley cropping systems may help raise awareness and technical knowledge of this form of agroforestry among local farmers, which over time could foster greater acceptance of tree–crop systems within the farming community.
Farmers’ individual attitudes towards risk, however, will also influence the relative attractiveness of the two agroforestry options. In general, the highest shares of agroforestry occurred in portfolios optimised for a highly risk-averse farmer. This highlights the advantage of agroforestry as a diversification strategy to reduce risk (Baker et al.
2017; Lin
2011; Waldron et al.
2016). Across the different scenarios we found that land-use portfolios optimised for risk-averse farmers generally contained more silvopasture than alley cropping. This suggests that silvopasture may be the better option for avoiding underperformance of the socio-economic objectives under uncertainty, because it holds relatively low risks. Silvopasture offers the security of annual income from cattle sales, for which yields and prices are typically stable (Connelly and Shapiro
2006), with the bonus of additional income from cedar at the end of the rotation. In contrast, alley cropping cannot guarantee an annual income because shading restricts maize cultivation from year 3 onwards. Instead, the bulk of revenue flows rely on timber prices at three points of time (the two thinnings and final harvest), which makes it inherently risky. Paul et al. (
2017) also report elevated risk levels for alley cropping compared to monoculture crops. Therefore, alley cropping may be less compatible with risk-averse decision-making.
The Effect of Labour, Budget and Land Constraints
Despite farmers’ preference for silvopasture (Gosling et al.
2020a), this agroforestry system is not common practice in the study area. This may reveal a conflict between the land-use systems that farmers wish to have, and those that they are able to implement given their hard economic constraints (Gosling et al.
2020b; Tschakert et al.
2007). Expanding on previous studies (Gosling et al.
2020a,
b), we explore the role of such farm-level restrictions on the optimal land-use composition by imposing fixed limits for labour demand and investment budgets in the optimisation model.
As expected, we found that “Labour and Investment constraints” reduce the share of agroforestry selected in the optimal portfolio. This aligns with other studies that found investment costs and labour demand to be barriers to agroforestry adoption in Latin America (e.g., Calle et al.
2009; Dagang and Nair
2003; Frey et al.
2012b). We found that silvopasture persists in the optimal portfolio when restricting investment costs, but is quickly replaced with conventional pasture when imposing labour constraints, suggesting that labour demand may pose the bigger barrier to silvopasture adoption.
In our model, the relative increase in labour demand when selecting silvopasture over conventional pasture is greater than the relative increase in investment costs, meaning the agroforestry system is hit harder by labour constraints. In practice, labour constraints may also be harder to overcome than capital constraints for farmers in the study area. It is common for farmers in Tortí to take out a loan to buy cattle when establishing conventional pasture systems (Peterson St-Laurent et al.
2013); the additional capital needed to establish trees for silvopastoral systems may be attainable through such loans, offering a means to overcome investment constraints. Meeting the additional labour requirement for silvopasture, however, may be more problematic, especially in tight labour markets (Baker et al.
2017; Pichón
1997). Labour shortages could be exacerbated by a hollowing of the forest frontier, which Sloan (
2008) has already observed in eastern Panama: this is a phenomenon where the population density of a deforested landscape declines as extensive farming practices increase. Peterson St-Laurent et al. (
2013) also report strong out-migration in eastern Panama as young people move to cities. In the face of tight labour markets it may therefore be necessary to adapt silvopastoral systems to better meet the needs of farmers constrained by labour shortages. This could be done by improving economies of scope, for example, through the use of multi-purpose trees where pruning could be combined with fodder production (Reyes Cáceres
2018). Such economies of scope are already a key advantage of the alley cropping system, in which trees and crops are weeded simultaneously (Paul et al.
2017).
Farmers’ land-use decisions will also be constrained by site conditions, which will influence the relative attractiveness of agroforestry. For example, simulating “Lower crop yields” increased the share of silvopasture selected in the optimal portfolio of a risk-averse decision-maker. This suggests that silvopasture may be a more attractive land-use option for farmers with less productive land (on which it is not possible to cultivate high yielding crops). These findings align with bio-economic studies that suggest agroforestry may be more advantageous on poorer growing sites (Crestani et al.
2017; Tsonkova et al.
2014). Moreover, the results underline the general importance of land condition (i.e., soil type and quality) for influencing the uptake of agroforestry and agricultural innovations (Pannell et al.
2014; Pattanayak et al.
2003).
Subsidies and Timber Prices to Promote Agroforestry Adoption
We found that the selection of agroforestry in the optimal portfolio was most responsive to a potential “Agroforestry subsidy” (lowering investment costs) and “Higher timber prices”. This suggests that cost-sharing arrangements could be an effective strategy to boost agroforestry adoption in the study area. For example, providing farmers with free tree seedlings and tree guards resulted in a 5 and 20% share of alley cropping and silvopasture in the optimal portfolio. Given its higher labour demand compared to conventional pasture, greater adoption of silvopasture could generate employment opportunities in the region if farmers hire day workers to assist with tree planting and pruning (Frey et al.
2012a). Establishment grants for silvopasture could help farmers finance this additional labour. While the legal framework for such incentives exists, they are yet to be consistently implemented in the study area.
In our scenario testing, we found that moderate increases in timber prices could lead to substantial shares of agroforestry being selected in a land-use portfolio that balances trade-offs between the five socio-economic objectives. For example, a 30% increase in teak price would result in a 18% share of alley cropping in the optimal portfolio, while a 30% increase in the cedar price would lead to a 33% silvopasture share. We also found that a small (10%) increase in the teak price could favour the selection of silvopasture in the portfolio. As the rising teak price makes alley cropping and plantation more profitable, the underperformance of pasture in terms of NPV becomes too great and it is first replaced with silvopasture and then by alley cropping and teak plantation in the optimal portfolio as the teak price continues to increase (Supplementary Fig.
S4d).
Timber prices strongly depend on market factors, and are thereby harder to engineer through government programs. However, the Panamanian Government’s recently legislated tax exemptions for timber grown in agroforestry systems (Law 69, 2017) could increase revenues from timber sales. Such tax incentives could particularly benefit the selection of alley cropping, which would become more competitive against pure teak plantation. This assumes, however, that farmers are earning enough to pay income tax, which may not be the case for many farm households (Díaz et al.
2012). Alternatively, farmer training programs on tree management (e.g., pruning and pest control techniques) could improve silvicultural practices, helping farmers to produce higher quality timber and hence obtain higher prices. Training programs and certification schemes could also help farmers build their capacity to access markets and obtain price premiums (Holmes et al.
2017; Somarriba et al.
2012). Nonetheless, when considering current timber prices (baseline scenario), only very small shares of agroforestry were included in the optimal portfolio. This could signal that further development of timber markets is a prerequisite for widespread adoption of timber-based land-use systems among smallholder farmers in the study area.
Limitations of Modelling Approach and Research Outlook
Our study is a rare example of a multi-criteria evaluation of agroforestry that takes a portfolio approach to account for the effects of land-use diversification and uncertainty on farmers’ land-use decisions. However, we acknowledge limitations of our study, which could be addressed in future research.
First, we rely on static modelling approaches in both the land-use and multi-criteria models. For instance, the land-use model ignores adverse environmental effects such as soil depletion over time (Janssen and van Ittersum
2007). This may overestimate the productivity of conventional land uses, and hence downplay drivers of agroforestry adoption. Future studies could therefore integrate production decay functions (e.g., following Sanchez
1976) to better account for the effect of nutrient depletion and soil structural changes on crop yields. Similarly, the multi-criteria model identifies theoretically optimal land allocations, but not how these could be achieved over time. Using a more dynamic optimisation approach, such as the one Knoke et al. (
2020a) recently developed to investigate smallholders’ deforestation decisions in Ecuador, would allow us to simulate farmers’ land-use decisions in smaller time steps. This would allow for staggered planting of trees, which might be a more feasible path for smallholders to adopt agroforestry (Bertomeu and Giménez
2006). A dynamic approach may also help to account for the option value of agroforestry systems and their conventional counterparts, an aspect which is overlooked in this study. In our land-use model, the timing of timber harvesting is fixed: this fails to capture the flexibility that a farmer has to postpone harvest if timber prices are unfavourable (Frey et al.
2013).
Second, our robust optimisation model is not spatially explicit. The model identifies what portions of a hypothetical farm could be allocated to each land-use option, but does not specify the exact location or arrangement of these land-use options (Bertomeu and Giménez
2006). This approach implicitly assumes homogeneous site conditions. Therefore, our multi-criteria model ignores the potential influence that farmers’ existing land use as well as variation in soil quality, slope and distance from the farm homestead may have on their land-use decisions, including their adoption of agroforestry (Bannister and Nair
2003; Pannell et al.
2014; Pattanayak et al.
2003). Thus, caution is needed when generalising the model results to farms with highly heterogeneous soils and/or contrasting topography, both within and outside of the study area.
Third, we integrated tree–crop and tree–pasture interactions in our land-use model through plausible assumptions (Paul et al.
2015), rather than detailed biophysical modelling. Our projected tree growth and crop yields were comparable to those simulated for the study area using the tree–crop model WaNuLCAS (Paul et al.
2017), while the economic coefficients for pasture-based systems reflect the lower, but very stable economic returns of cattle grazing in Panama (Connelly and Shapiro
2006). Nevertheless, the modelling approach could be enhanced by integrating biophysical modelling to simulate tree, crop and pasture growth in monoculture and agroforestry systems (e.g., using WaNuLCAS, Santos Martin and van Noordwijk
2011). Such modelling could be particularly useful for evaluating different layouts of agroforestry systems, for example, to identify the most promising systems for field trials. Ultimately, such local field experiments are essential to obtain empiric data, which remains the best foundation for land-use planning (Reith et al.
2020).
In presenting our study, we recognise the usefulness, but also limits, of models as decision support tools. Our modelling approach explores theoretically optimal land allocations for achieving a particular outcome under a certain set of assumptions. We do not intend to prescribe exact farm compositions that farmers in the study area should adhere to. Instead, we seek to explore the conditions under which agroforestry might be a desirable complement to help farmers reduce trade-offs between socio-economic objectives. The decision of whether or not to adopt a given land-use system rests with the farmer, and will depend on his or her objectives and constraints (Janssen and van Ittersum
2007; Pannell et al.
2006). Our study therefore does not seek to develop a decision support tool for farmers, but is rather targeted at researchers and political decision-makers. For researchers our modelling approach may help to identify the agroforestry systems and conditions under which more detailed field trials are most warranted, because the systems show a high probability of being of interest to farmers. For policy-makers, such approaches can help to identify the circumstances under which promoting agroforestry appears to be promising without generating conflicts with farmers’ goals.
However, as with any decision support tool, we acknowledge a potential gap between the results of our theoretical model and the reality of farmers’ decision-making (McCown
2001). Such gaps between theory and practice may stem from potential biases and uncertainties in model input data. Although we actively account for such uncertainty by implementing a form of robust optimisation (Doole
2012; Knoke et al.
2015), field experiments remain crucial to deliver reliable empiric data. The gap between theory and practice may also stem from the assumptions and limitations of the multi-criteria model, which cannot capture all aspects influencing farmers’ decisions. For example, in the scenario analysis we alter one aspect at a time to understand how this affects the share of agroforestry selected in the optimal portfolio. In reality, however, such aspects will be changing simultaneously, potentially leading to complex interactions that we do not account for. With these limitations in mind, care is needed when generalising our results to other areas: the more the region differs to the biophysical and socio-economic conditions of Tortí, the greater the gap is likely to be between our theoretical and the actually optimal land allocations. However, we again emphasise that we do not seek to give exact land-use recommendations for this study site, but rather demonstrate how such an approach may inform future research and policy design.
Finally, we see potential to further develop our approach through participatory and collaborative modelling. Indeed, greater farmer interaction is likely to help narrow the gap between scientific theory and real-world practice (Janssen and van Ittersum
2007; McCown
2001). For example, farmers could help to validate input data, based on their local knowledge and experience. Moreover, as simple, stylised land-use portfolios, we believe the output of the multi-criteria model could be readily interpreted and evaluated by smallholder farmers. Discussing model results with farmers in the study area could help to validate and improve the model, for example, by changing objectives or adding additional constraints to better match the local situation (Groot et al.
2012). Optimised portfolios might also provide a good starting point for stakeholder discussions as part of participatory land-use planning (Le Gal et al.
2013). For this type of landscape-scale planning the multi-criteria model could easily integrate ecological indicators (either based on expert opinion, e.g. Reith et al.
2020, or modelled and measured data, e.g. Knoke et al.
2020a), to derive the optimal land-use compositions for achieving a wider range of ecosystem services.