Modelling avocado-driven deforestation in Michoacán, Mexico

We began by delimiting Michoacán's 'Avocado Belt' based on production and spatial contiguity (figure 1). Of the 65 included municipalities, 52 are the top avocado producing municipalities in the state, and the 13 remaining municipalities complete a contiguous geographic region. This area contains approximately 1356 km² of avocado, 590 km² of which was planted between 1992 and 2017 (based on INEGI agricultural classification maps). The belt spans 18.7° N and 20.1° N latitude, and 100.1° W and 102.9° W longitude, and constitutes a cohesive ecological area in terms of temperature, precipitation, biogeography, vegetation types, soil groups, and terrain. Compared to the state as a whole, the Avocado Belt has lower average temperatures and higher precipitation. Because of this, the area is dominated by temperate forest types, including coniferous and oak forests, which differentiates it from the rest of the state, where tropical forests predominate. In terms of terrain, the Avocado Belt falls within the Trans-Mexican Volcanic Belt, giving it generally higher elevations than the rest of the state. This also impacts soils, as the highest domination of volcanic soils, mainly Andosols, in Michoacán occur within the Avocado Belt.

Expansion of new avocado into forests was modeled in two steps, one aspatial and the other spatially explicit. The aspatial component estimates the total land area of avocado expansion projected to occur by the year 2050 in Michoacán. The second component spatially allocates that amount across the landscape of the Avocado Belt.

To estimate the amount of land that could undergo land change for avocado expansion, data were obtained from Mexico's Agrifood and Fisheries Information Service (SIAP, in Spanish) on (a) area of avocado harvested in Mexico, (b) area of avocado harvested in Michoacán, and (c) area of avocado planted in Michoacán. All data range from 1980 to 2019. Linear trendlines were calculated for each dataset that were then used to forecast the additional area of avocado planted in Michoacán out to 2050 to create three scenarios of avocado expansion (figure 2).

Zoom In Zoom Out Reset image size

A fourth scenario was crafted from information in the FAO report of 'Medium-term outlook for global production and export and trade in bananas and Tropical Fruit' (FAO 2020). This report's projections about production and export growth for avocado, as well as tropical fruits more generally, range between 1% and 3% annual growth, regarding measures of land area, production output, and export volumes. Considering these data, a fourth scenario was calculated using a 1.5% annual growth rate for area planted in Michoacán as a conservative middle-ground between the different FAO projections.

The trends associated with the different scenarios described above create an envelope of potential avocado expansion, ranging from high to low expansion rates. All models are remarkably close in their predictions. Nonetheless, we take the ensemble average growth rate, rounded to 1000 km², to simulate the avocado planted area in 2050, a common practice in modeling (Gregory et al 2001, Dormann et al 2018), because it typically captures the overall trend while attenuating some of the noise of individual forecasts.

To allocate where the additional deforestation associated with future avocado expansion is most likely to occur, we calculated the deforestation probability based on actual deforestation attributed to avocado plantations from the period between 1992 and 2017. We used a spatial Bayesian probit model according to Smith and LeSage (2004) and Arima (2016) and included the explanatory variables in table 1. Cells were 100 × 100 m, and the model accounted for spatial autocorrelation between groups of 1 km² neighboring cells. Three climate scenarios were used to project avocado expansion into the future. Expansion was predicted using present climate variables, as well as two IPCC representative concentration pathways for climate change—a conservative and a worst-case scenario (RCP 2.6 and 8.5, respectively). The details of this model are described in supplementary information section 2 (available online at stacks.iop.org/ERL/17/034015/mmedia). The model overall performance assessment is described in the SI document, section 5. In addition to statistical significance, we also highlight in the Results section the practical significance of the variables of interest by measuring the so-called average partial effect (See SI document, section 6).

To estimate loss of different forest types, the model was run with 10 000 Monte Carlo simulations, where the probability of change to avocado production, assigned to each cell by the probit model, was compared against a randomly generated number between 0 and 1 taken from a uniform distribution. Deforestation was allocated to a non-deforested cell only if the calculated probability of conversion to avocado was greater than the randomly generated number. Each simulation of deforestation was then compared to forest type layers from the 2017 INEGI vegetation maps to calculate how much forest loss occurred for each forest type. This method allows an element of stochasticity and results in a distribution of 10 000 estimations of loss for each type of forest. The same technique was applied to estimate encroachment on protected areas and for how expansion will occur on different soil types.

This Monte Carlo approach was also used to query the physical environmental characteristics of where avocado is most likely to expand, including elevation, slope, mean annual temperature, and annual precipitation. For each simulation, the minimum and maximum values for these variables in areas of avocado conversion were extracted, resulting in a distribution of these values across 10 000 simulations (e.g. for 10 000 simulations of land change to avocado, 10 000 minimum elevations values of the converted land were extracted). The distributions were then compared to the actual minimum and maximum values for elevation, slope, temperature, and precipitation from the observed avocado expansion between 1992 and 2017 using the Mann-Whitney U test for non-parametric data. The purpose of these tests is to understand if these values in our projections for expansion between 2017 and 2050 differ significantly from the observed values from expansion between 1992 and 2017, signifying a change in the ecological range where avocado is planted.

The dependent and independent variables were compiled from multiple different datasets in order to create variable raster layers with a 100 m resolution for the entire study area. Biophysical variables (elevation, slope, soil, vegetation, climate) are included because they impact suitability for avocado growth. The most suitable conditions for avocado production occur between 1200 and 2500 m a.s.l. (Barsimantov and Navia Antezana 2012), in areas with the presence of Andosols, mean annual temperature between 12 °C and 33 °C (Whiley and Winston 1987), enough moisture (via rainfall or irrigation) to meet high water demand of avocado trees, and flat slopes to avoid the construction of terraces. The distance variables also affect likelihood of avocado expansion due to the structure of the supply chain. Avocado producers sell their fruit on the tree to packing houses, who hire harvesters to collect the fruit and transport it to packing houses. As such, accessibility (i.e. distance to roads, settlements) and proximity to packing houses affect the price packing houses offer to producers, and thus the profitability of an orchard. Finally, land tenure (i.e. communally managed, known as ejidos, or protected areas) is included as it also affects grower decision to expand avocado. The variables used in the model (and their sources) are listed in table S1, and the process for creating these layers is described in more detail in the SI document, section 1. To evaluate the performance of our approach for including regional spatial correlation, we also estimated a standard (non-spatial) probit regression and compared its prediction accuracy to our spatial probit model (SI document, section 9).

The three scenarios of expansion based on trends in SIAP data resulted in estimates of 977 km², 1017 km², and 991 km² of expansion, respectively, between 2017 and 2050. The middle-range scenario of 1.5% annual growth, based on the FAO report on tropical fruits, resulted in 1073 km² of expansion (figure 2). The four estimates of expansion are very similar, with an average of 1014 km². We therefore used a rounded value of 1000 km², a 74% increase since 2017, (or 100 000 cells of 100 × 100 m) to represent likely expansion to 2050. This area was allocated spatially according to the probability model derived from the spatial probit regression across the Avocado Belt, which is 24 674 km² in total area, with 1356 km² of avocado groves existing in 2017 (INEGI).

3.2.1. Spatial probit regression

Of 26 variables evaluated, only four were found to be not significant statistically (see SI table S2). The distance variables (to roads, packing plants, local villages, and agricultural areas) had the expected negative sign with the exception of distance to cities, which indicated a higher probability of avocado away from cities. This result is robust across specifications (SI 8 and SI 9) and is likely due to higher land prices near cities from competing alternative uses (e.g. residential, fields for vegetable gardens and dairy farming). Properties near cities tend to be smaller and more fragmented (Fujita 1989), which also increases production costs due to scale. The probability drops substantially as distance to roads and to all localities increases. It is important to mention that these distances will change over time, which is not accounted for in model projections. A 1 km increase in distance to roads and locality is associated with a drop of 0.008 and 0.006 in the probability of avocado respectively (table S4). This represents a change of 21% and 16% with respect to the naïve probability of 0.0361 (i.e. number of cells classified as avocado over all cells in our study area). Figure 3 shows the spatial distribution of the cells with the highest probabilities of change to avocado orchard up to 1000 km².

Zoom In Zoom Out Reset image size

Within the Avocado Belt, the coefficient for elevation is positive and its quadratic term negative, indicating that the probability of avocado plantation increases as elevation rises to 1538 m (inflection point) when probability starts to decline. This is consistent with the literature that describes a relationship between avocado production and elevation (Barsimantov and Navia Antezana 2012, Dubrovina and Bautista 2014). However, the average partial effect of elevation is quite small, (table S4). Precipitation and temperature also behave similarly, with inflection points around 1707 mm yr⁻¹ and 24.9 °C but partial effects are negligible.

These results may indicate the role of technological adoptions that have allowed farmers to overcome certain environmental limitations, including better-adapted varieties and use of irrigation, or alternatively, it may indicate exceptionally strong economic incentives to grow avocado, even on lands deemed agronomically inferior (Dubrovina and Bautista 2014).

All of the soil group variables that were included in the analysis had positive coefficients, with Andosol having the largest magnitude coefficient with an average partial effect of 0.02% or 60% of the naïve probability. This result is in agreement with other studies which find that the volcanic Andosols are ideal for avocado growth (Dubrovina and Bautista 2014), and it means that spatial patterns of future avocado expansion can be partially predicted by presence of Andosols, a well-structured and well-drained volcanic soil relatively rich in nutrients (IUSS Working Group WRB 2014).

In terms of land tenure, ejido lands are less likely to have avocados. This may be because, although ejido lands are legally alienable, many are still communally governed, meaning group approval is needed for transfer or sale. The variable 'protected areas' was not statistically significant in the spatial model, suggesting that protected areas had no effect on preventing avocado conversion. This is likely because the largest protected areas in the region allow for sustainable land use within their buffer zones. Our data also show over 2 k ha of avocados inside protected areas.

Avocado plantations are highly spatially correlated as indicated by the spatial regional effect parameter ρ = 0.79. In all, the model performs well and correctly predicts 81% of the existing avocados on a cell-by-cell basis and more than 99% of the non-avocado cells, a performance much superior than the standard probit model (SI document, sections 5 and 9 respectively).

3.2.2. Different climate scenarios

When compared to probabilities under current climate conditions, probabilities of avocado expansion decrease in the central-western portion of the belt and increase in the eastern side of the region, near the Monarch Reserve, for both RCPs. Larger increases in probability are observed under RCP 2.6 (figure S6(a)) than under RCP 8.5 (figure S6(b)). This is likely due to more pronounced changes in precipitation and milder increases in temperatures for the former scenario. A correlational analysis shows that change in probability of avocado expansion is positively correlated to change in precipitation and negatively correlated with change in temperature (figure S9). In other words, places with higher future precipitation will have higher probabilities of land change for avocado, while places that will have higher increases in temperature will have lower probabilities of avocado. Our models predict that scenario RCP2.6 will increase the probability of avocados over an area of 335 k ha when compared to the current climate, but 1.16 million ha will have lower probabilities across the Avocado Belt. Scenario RCP8.5 is even less favourable to avocados; only 238 k ha will observe increases whereas probabilities will decline in 1.25 million ha. These results are in agreement with recent models that predict a reduction in the potential area for avocados in Mexico under climate change (Charre-Medellín et al 2021).

Because of this dynamic, our simulations of avocado expansion under climate change show less expansion in pine and pine-oak forests and more expansion into all other vegetation types than if expansion happened under current climate conditions (figure 4).

Zoom In Zoom Out Reset image size

3.2.3. Types of forest loss and protected area encroachment

3.2.4. Avocado range expansion

Understanding the types of forest that are most vulnerable to loss from avocado expansion can improve understanding of the biodiversity threats of increasing avocado production and target responses or preventive actions. According to our modeling, pine-oak forests are vulnerable to the largest absolute area loss from expansion in the Avocado Belt of Michoacán. These forests are generally made up of Pinus and Quercus species, but can also have species of associated genera, such as Alnus, Crataegus, Clethra, and Arbutus (Cruz Angón et al 2019). In addition to their biodiversity value, pine-oak forests have been found to have higher carbon storage capacity than other anthropogenic land uses in the region (like avocado orchards and other agriculture) (Ordóñez et al 2008). Loss of pine-oak forests would result in a net source of carbon into the atmosphere, a concern for global climate change.

Beyond pine-oak forests, loss of mesophilic montane forests and oyamel fir forests is of national and global concern. Although the total area of mesophilic montane forest projected to be lost is much lower than that of pine-oak forest, such loss represents a high percentage of this already rare forest type. Mesophilic montane forests are highly biodiverse but are rare and fragmented within Michoacán. As such, any loss would have a big proportional impact on the ecosystem type and the species the rely on it for habitat. This forest type includes tree species of genera such as Alnus, Clethra, Pinus, Quercus, Styrax, and Symplocos, with numerous epiphytes (Cruz Angón et al 2019). Oyamel fir forest is similarly rare within Michoacán, as it is restricted to small areas with high altitudes (2600–3500 m asl) and high humidity. These forests are dominated by Abies religiosa, and they are of particular conservation importance since they serve as the southern migration and winter hibernation habitat for monarch butterflies (Danaus plexippus) (Cruz Angón et al 2019).

Loss of oyamel fir forest is linked to the model results showing encroachment of avocado expansion into protected areas. The results show that most of the projected encroachment would happen on federal protected areas, including into the margins of the Monarch Biosphere Reserve (MBR). This reserve is already designated as being of 'significant concern' by the IUCN World Heritage Outlook because of the loss and degradation of forests in both the buffer and core zones (IUCN World Heritage Outlook 2020). Recent studies confirm avocado expansion into the buffer zone of the MBR between 2006 and 2018 and warn of the risk of future expansion in the Reserve (Sáenz-Ceja and Pérez-Salicrup 2021). The IUCN report does not identify avocado as a specific risk, and instead describes the vulnerability of the reserve to logging, poorly managed tourism and livestock grazing. Vulnerability to avocado expansion puts further pressure on this already threatened reserve and must be considered by development agencies promoting avocado production. Yet, some critics question the idea of strictly protected natural areas, arguing that the designation of a core area free from human disturbance undermines sustainable community forest management and leaves the fir forests even more vulnerable to exploitation (Gonzalez-Duarte 2021).

In addition to types of forest loss and encroachment into protected areas, the expansion of avocado production onto higher slopes risks increased soil erosion, an already existing threat (Dubrovina and Bautista 2014). Moreover, increased proportion of growth on Leptosols and Regosols would increase risk of soil erosion because of these soil groups' lower water infiltration capacity (IUSS Working Group WRB 2014). Degradation of soil poses even more risk for existing vegetation and water resources in these ecosystems, which in turn, further threatens the biodiversity within these habitats.

The widening crop niche of avocado production does not only pose risks of ecosystem degradation; it also increases risks for growers, as many of these marginal lands have lower suitability for avocado growth. As avocado is projected to expand into higher and lower elevations, higher and lower temperatures, lower precipitation, and less suitable soil groups, production may be more vulnearable to weather fluctuations. This in turn may lead to increased application of inputs such as fertilizer, pesticides, herbicides, and irrigation, furthering impacting the surrounding ecosystem. Water scarcity and chemical pollution of groundwater are already problems in the state (Borrego and Allende 2021).

The range of annual mean temperature where avocado expansion occurred between 1992 and 2017 was 11.2 °C–24.3 °C. The mean minimum and maximum values for mean annual temperature projected for avocado production in 2050 are 7.1 °C and 28.5 °C, respectively. This projection pushes avocado production beyond previously established optimal range for productivity. Previous studies place the temperature range for avocado at 12 °C–33 °C (Whiley and Winston 1987), although the ideal range is 20 °C–25 °C. Above temperatures of 28 °C, avocado trees can lose flowers (Lovatt 1990).

Our results show that projected avocado expansion may occur in areas of lower annual precipitation. This would likely increase irrigation demand in these areas, as avocados are a water-intensive crop (Quiroz Rivera 2019, Charre-Medellín et al 2021). This potential impact of expansion raises concern since issues of water scarcity have been the cause of conflict over water allocation in the region in the past (Alarcón-Cháires 2018). Use of more marginal habitats requires more investment for irrigation and other more costly landscape management approaches.

Avocado expansion in the past was partially enabled by the presence of Andosol soils, but dwindling availability means expansion will increasingly occur in different soil groups. Andosols, Luvisols and Phaeozems have been found to be suitable for avocado production, with the first being by far the most suitable due to their deep profiles, high capacity for water infiltration and water retention (Dubrovina and Bautista 2014). Luvisols, while still relatively suitable for avocado, have a higher clay content and thus require the construction of berms within orchards to allow for proper root development, making the cost of production of avocado on Luvisols higher than that for Andosols. Beyond these three aforementioned soils groups, Cambisols, Regosols, Leptosols, and Vertisols have low suitability for avocado (Dubrovina and Bautista 2014) and thus may considerably impact an orchard's productivity.

In addition to environmental concerns, avocado expansion may turn out to be a risky investment. Although avocado is currently highly profitable (Borrego and Allende 2021, Denvir et al 2022), production in less than ideal environments is likely to be more risky and therefore more vulnerable to negative market shocks or abnormal weather events. Compounding matters further, our climate change simulations indicate a net reduction in the suitability for avocados in Michoacán, as indicated by an overall reduction in spatial probabilities. Currently, Michoacán holds exclusive access to the US market, but negotiations are under way to allow other regions of Mexico to export to the US. For instance, the state of Jalisco has won approval and will soon ship fresh avocados to the US. Moreover, there are reports of neighboring states smuggling avocados into Michoacán in order to export them to the U.S. This activity is illegal and under-studied and no reliable data exist on the quantities involved. For that reason, this aspect of illegal indirect land use change is not included in this analysis but could be a focus of future research. With more competition, profitability of avocados in Michoacán may decline. From a societal perspective, questions remain if the financial benefits of this risky investment in avocado expansion are worth its environmental costs.

Avocado expansion is just one of the various threats to forests in Michoacán. Future research could consider additional socioeconomic and political factors that will direct future expansion of avocado and how they interact with other threats, such as cattle ranching, illegal logging, berry production, and urban settlement. Variables like market prices and grower access to capital are likely important in deciding which growers participate in the expansion of avocado, which were not evaluated here. Moreover, Michoacán is currently the only state in Mexico that is authorized to export avocados to the United States, which is by far Mexico's largest export market. It is reasonable to ask how the spatial distribution of avocado expansion will be affected once avocado exports to the U.S. from other Mexican states, such as Jalisco, are allowed. Would such a situation relieve pressure from forests in Michoacán? Or would it cause similar problems in other states, where growing conditions are not as ideal as Michoacán? An approach such as outlined by Denvir et al (2022), which calls for targeted sustainability action at different parts of the supply chain, would allow for also addressing the sustainability concerns invoked by the roles of distributors and wholesalers, who in conjunction with growers and consumers, affect the social and environmental costs of a globally sought-after commodity. This general approach could prove useful for assessing the socioenvironmental consequences of many other food commodities.