Agroforestry Systems

, Volume 86, Issue 2, pp 141–157

Carbon stocks in coffee agroforests and mixed dry tropical forests in the western highlands of Guatemala


    • School of Public and Environmental AffairsIndiana University
  • Tom P. Evans
    • Department of GeographyIndiana University
  • Edwin Castellanos
    • Centro de Estudios Ambientales y de BiodiversidadUniversidad del Valle de Guatemala
  • J. C. Randolph
    • School of Public and Environmental AffairsIndiana University

DOI: 10.1007/s10457-012-9549-x

Cite this article as:
Schmitt-Harsh, M., Evans, T.P., Castellanos, E. et al. Agroforest Syst (2012) 86: 141. doi:10.1007/s10457-012-9549-x


Tree removal in Latin American coffee agroforestry systems has been widespread due to complex and interacting factors that include fluctuating international markets, government-supported agricultural policies, and climate change. Despite shade tree removal and land conversion risks, there is currently no widespread policy incentive encouraging the maintenance of shade trees for the benefit of carbon sequestration. In facilitation of such incentives, an understanding of the capacity of coffee agroforests to store carbon relative to tropical forests must be developed. Drawing on ecological inventories conducted in 2007 and 2010 in the Lake Atitlán region of Guatemala, this research examines the carbon pools of smallholder coffee agroforests (CAFs) as they compare to a mixed dry forest (MDF) system. Data from 61 plots, covering a total area of 2.24 ha, was used to assess the aboveground, coarse root, and soil carbon reservoirs of the two land-use systems. Results of this research demonstrate the total carbon stocks of CAFs to range from 74.0 to 259.0 Megagrams (Mg) C ha−1 with a mean of 127.6 ± 6.6 (SE) Mg C ha¹. The average carbon stocks of CAFs was significantly lower than estimated for the MDF (198.7 ± 32.1 Mg C ha−1); however, individual tree and soil pools were not significantly different suggesting that agroforest shade trees play an important role in facilitating carbon sequestration and soil conservation. This research demonstrates the need for conservation-based initiatives which recognize the carbon sequestration benefits of coffee agroforests alongside natural forest systems.


Shade-grown coffeeAgroforestryCarbonLand conversion


Coffee agroforestry has emerged as a promising land-use system for reducing or offsetting deforestation (Soto-Pinto et al. 2010; van Noordwijk et al. 2002), while at the same time sequestering carbon and contributing to climate change mitigation (Dossa et al. 2008; Soto-Pinto et al. 2010). Given enhanced carbon (C-) sequestration that occurs with tree planting and the practice of agroforestry farming, shade-grown coffee systems (“coffee agroforests”) have been recognized as viable afforestation and reforestation (A&R) strategies under the Clean Development Mechanism (CDM) of the Kyoto Protocol (IPCC 2000; UNFCCC 2006). However, there is little policy development encouraging the maintenance of current agroforestry systems for the benefit of C-sequestration. While programs such as REDD (the UN program for Reducing Emissions from Deforestation and Degradation) have been developed for tropical forests under threat, coffee agroforestry systems have yet to be included despite extensive deforestation, largely a result of a complex set of factors that include government-supported agricultural policies, international markets, and climate change (Ellis et al. 2010; Rice and Ward 1996).

Government-supported agricultural policies have historically played an integral role in influencing changes to- and modifications within- coffee production systems. For example, beginning in the mid-1970s, a push to “technify” or “modernize” the coffee sector emerged in northern Latin America mainly due to the spread of coffee leaf rust (Hemileia vastatrix) and its mitigation (Jha et al. 2011; Rice and Ward 1996). International efforts, largely financed by the United States Agency for International Development (USAID), led the way toward “technification” which predominantly involved the removal or reduction of shade cover, an increase in planting density of coffee bushes, and use of new high yielding varieties (Jha et al. 2011; Rice and Ward 1996). Even where coffee leaf rust was not expected to pose significant problems (e.g. higher elevation areas due to cooler temperatures), landscape-level transformations were widespread. On average, over 40 % of Latin American shaded coffee farms were “technified” or converted to sun coffee, a conversion of which has been likened to widespread deforestation of agricultural lands (Rice and Ward 1996).

More recently, record low international coffee prices between 1999 and 2003, in combination with repeated droughts, contributed to dramatic declines in employment in Central America within the coffee sector in a period known as the “coffee crisis” (Bacon 2005; Tucker et al. 2010). Coffee growers historically and currently face a number of uncertainties in the production of coffee, from the overproduction of lower-quality coffee which threatens market stability (Ponte 2002; Rice 2003), to climatic changes in Central America trending toward increased temperature (Magrin et al. 2007), reduced precipitation (Magrin et al. 2007; Neelin et al. 2006), and increased frequency of extreme storm events (Emanuel 2005; Webster et al. 2005). Those most vulnerable to these stressors are smallholder coffee producers given their economic disposition and lack of access to resources (Eakin et al. 2006), a combination of which may induce land cover conversion to maize, pasture, or other annual crop production (Ávalos-Sartio and Blackman 2010; Eakin et al. 2005, 2006; Ellis et al. 2010; Nestel 1995). Moreover, recent recommendations by policy-makers and donor agencies to coffee growers have suggested that areas less suited for- or less competitive in- coffee production should be switched to other crops (IADB et al. 2002; Tucker et al. 2010; Varangis et al. 2003). While intended to promote economic sustainability, such recommendations may also lead to land-cover change, with unintended ramifications for climate-change processes (e.g. carbon sequestration) and the provision of ecosystem services.

To facilitate development of large-scale policy prescriptions (e.g. REDD) or smaller-scale certification programs oriented around C-sequestration in coffee agroforestry systems, an understanding of the utility of coffee agroforests in sequestering carbon relative to tropical forests must be developed. To-date, few empirical, peer-reviewed investigations of carbon dynamics within coffee-forest landscapes have been conducted. Several studies have instead focused on examining biodiversity within coffee-forest landscapes, with results demonstrating that shade tree presence can reduce fragmentation and the isolation of forests, enhance connectivity, and support biodiversity comparable to natural forests (Bandeira et al. 2005; Bhagwat et al. 2008; Greenberg et al. 1997; Jha and Dick 2008; Jha and Vandermeer 2010; Perfecto and Vandermeer 2002; Perfecto et al. 1996, 2007; Soto-Pinto et al. 2001).

Research directly related to carbon (C-) storage finds that coffee agroforests store substantially more carbon than sun-grown coffee, maize, and other traditional agricultural systems (Ávila et al. 2001; Dossa et al. 2008; Harmand et al. 2006; Soto-Pinto et al. 2010; van Noordwijk et al. 2002); however, the comparability of C-storage within coffee agroforests and forests remains little studied. To our knowledge, the most comprehensive effort to-date in comparing carbon stocks within shade-grown coffee systems and “natural” forests was conducted in Sumatra, Indonesia by van Noordwijk et al. (2002). This research demonstrated that remnant secondary forests stored substantially more carbon (262 Megagrams C ha−1), on average, than shade-grown coffee systems (82 Mg C ha−1) and sun-grown coffee systems (52 Mg C ha−1). However, the authors also found deforestation in the region to result in carbon losses that were partially offset by gains associated with conversion of sun-coffee to shade-coffee, a result in support of coffee agroforests serving to offset or reduce deforestation (van Noordwijk et al. 2002).

With coffee occupying over 9.8 million ha of land worldwide (FAO 2009), and employing 20–25 million families in the production process, many of whom live in rural and impoverished communities (Jha et al. 2011; Lewin et al. 2004; Oxfam 2001), a better understanding of the capacity of coffee agroforests to store carbon relative to tropical forests must be developed. This research helps to fill this gap by comparing the carbon stocks of densely shaded smallholder coffee agroforests to a mixed dry tropical forest, results of which speak to the development of forest-based conservation initiatives such as REDD for shade-grown coffee systems. Ecological assessments of aboveground, coarse root, and soil carbon stocks were from a focused study area in the western highlands of Guatemala enabling a robust comparison of the two land-use systems. In estimating aboveground carbon, this study utilized published species-specific and generic allometric models, and examined the sensitivity of aboveground carbon estimates as a function of allometry.

Materials and methods

Study location and land-use systems

This study was carried out in two locations in the Department of Sololá, between the UTM northings of 1,606,500 and 1,648,000 m and eastings of 661,000 and 708,000 m (Fig. 1). The Department lies in the Sierra Madre Volcanoes region in the western highlands of Guatemala, covering an area of 1170 km2. The study region is topographically heterogeneous with elevations ranging from 628 to 3524 m ASL, and slopes ranging from 0° to 75°. Annual rainfall and temperature averages 2504 mm and 18–24 °C though there is high variability associated with altitudinal gradients (FAO 2002). In general, precipitation increases and temperature decreases with higher elevations, and a pronounced dry season occurs from November to April. Soils in the study area are predominantly mollic andisols of volcanic origin (MAGA 2002; Simmons et al. 1959).
Fig. 1

The Department of Sololá, Guatemala. Data were collected in the municipalities of San Juan, San Pedro, Santa Clara, and Santa Lucía Utatlán. Coffee agroforest plots were located in the vicinity of the hollow star whereas forest plots were located near the black star

The regional landscape is a mosaic of primary and secondary forests, shade-grown coffee, pasture, agriculture, and rural settlements. Farming of coffee, maize, and beans sustains a majority of the population; however, a national and international tourism industry is growing as Lake Atitlán is the third most visited tourist destination in the country (Balloffet and Martin 2007). The majority of coffee growers in the region are smallholders who farm no greater than 3 ha each, but the area also supports some large private landholders who produce coffee and other agricultural products for export.

Two land-use systems were selected for study: smallholder coffee agroforests (CAFs) and a mixed dry forest (MDF). The CAFs, located west of Lake Atitlán (see Fig. 1) are densely shaded polyculture systems in which the canopy is comprised mostly of planted timber, fruit, and leguminous species. The area has been in coffee for over 40 years and is composed of heterogeneous smallholders1 covering a collective area of approximately 500 ha.2 Shade management (e.g. tree species composition, shade tree density) varies slightly across the study region (see “Summary of tree metrics” section); however, the primary protective and productive functions of the CAFs remain similar—that of soil conservation, coffee production, and the production of non-coffee goods associated with shade trees (e.g. fuelwood, medicine, fruit).

Approximately five kilometers north of the coffee agroforest sites lies the MDF, Cerro Chuiraxamoló, a municipal park of 184 ha comprised mostly of alder, oak, and pine species. Following the Holdridge system of life zone classification, Cerro Chuiraxamoló is a lower montane moist forest. However, as with many montane areas there are precipitation gradients in localized areas and in this forest, a mixed dry forest regime predominates. While other forest systems are present in the region, Cerro Chuiraxamoló was selected for this research given its close proximity to the coffee agroforest sites, and its similarity in soil type and condition (e.g. texture, clay properties) (MAGA 2002; Simmons et al. 1959). Furthermore, both land-use systems lie north of Volcán San Pedro and receive less rainfall (~1100 mm annually) than sites south of the volcano (Tucker et al. 2010) making comparisons more viable.

Mixed forests have historically been used as sources of fuelwood and timber for nearby communities in the Lake Atitlán region, and are now found in small patches confined to remote and inaccessible areas (CEAB, unpublished). For Cerro Chuiraxamoló, timber extraction by the nearby community of Santa Clara was common in the 1980s and 1990s, but at the turn of the century, the community decided to protect the forest for its spiritual and ecological value. The park is now protected by a municipal decree and the area is located within Guatemala’s National System of Protected Areas (Sistema Guatemalteco de Áreas Protegidas, SIGAP) (Balloffet and Martin 2007). The forest resembles a secondary forest of approximately 40 years of age, but a considerable number of large, old trees are present. Human use is mainly for recreational purposes and the forest is considered to be in better-than-average condition compared to other forests in the region (CEAB, unpublished).

CAFs sample design and field inventory

Given the heterogeneous management of the 500-ha coffee agroforest footprint, we expected biomass to vary as a function of coffee planting age and management practices. In order to capture this diversity, a 2009 ASTER3 satellite image was processed to calculate the Normalized Difference Vegetation Index (NDVI)4 which is correlated with aboveground biomass. From the calculated NDVI values, sample plot locations were randomly generated using ArcGIS 9.3 software. Points were generated in low- and high-NDVI pixels to minimize plot selection biases, and in relatively homogeneous 3 × 3 pixel areas to account for spatial error correspondence between the image and geo-located field plot locations. One hundred sample plot locations were generated and served as the basis for locating plots in the field using a Garmin Oregon 400t GPS receiver.

Vegetation inventories and soil samples were collected in thirty-eight 400 m² (20 m × 20 m) square plots in CAFs. Trees and saplings with diameter at breast height (DBH; taken at a height of 1.37 m) ≥2.5 cm were recorded in these plots. When buttresses and other irregularities were present at 1.37 m, we measured 50 cm above the protuberances (Condit 1998). Trees were defined as having DBH ≥10 cm and saplings as having DBH between 2.5 and 9.9 cm. Tree and sapling heights (H) were recorded using a Suunto PM-5/360 PC clinometer. The densities of trees and saplings were determined by counting the number of stems present at 1.37 m height and dividing by the total area sampled (400 m2). The basal area of shade tree species was summed and averaged for all CAF plots, and plot-level tree species composition was examined in terms of tree species richness.

Within each 400 m2 plot, a 25 m² (5 m × 5 m) subplot was selected to measure coffee trees. Because coffee trees were frequently pruned, two stem diameter measurements were taken for each coffee tree at 1 cm and at 15 cm from the ground. All coffee trees sampled were in their productive years and were analyzed separately from shade trees and saplings to minimize biases associated with density and diversity calculations. Coffee tree densities were estimated by counting the number of stems present at 15 cm height and dividing by the total area sampled (25 m²).

Soil samples were collected in each plot to determine bulk density and organic carbon. Each 400 m2 plot was subdivided into four 10 m × 10 m quadrats and soil samples were collected at quadrat center points and aggregated for further processing (described in “Soil sampling and analysis” section). Additional information including slope aspect and gradient, canopy cover, and evidence of natural or anthropogenic disturbance was reported for each plot. Canopy cover was assessed using a Canon EOS Rebel digital camera and 18-mm focal length lens. Eight canopy cover photographs (two per 10 m × 10 m quadrat) were taken upward to the canopy from a leveled platform at the center of each quadrat. The digital images were analyzed using ERDAS Imagine software, and a supervised classification yielded the percent of total pixels captured in the image as vegetation, tree bole, sky, and cloud. Canopy cover assessments were used alongside ecological inventories to characterize the management style of CAFs as defined in the literature (Jha et al. 2011; Moguel and Toledo 1999; Philpott et al. 2008).

MDF sample design and field inventory

The MDF, Cerro Chuiraxamoló, was sampled by collaborators at the Centro de Estudios Ambientales y de Biodiversidad (CEAB) in June 2007 as part of the International Forestry Resources and Institutions (IFRI) program (IFRI 2007). Using the IFRI protocol, twenty-three nested circular plots were randomly established in Cerro Chuiraxamoló. Circular plots were deemed preferential in both land-use systems given their low edge:area ratio; however, creating concentric circular plots in the CAFs was not feasible5 and to maximize efficiency and reduce potential sampling biases within each land-use system, we employed a nested square- and circular-plot design in CAFs and the MDF, respectively. We expect the varied plot designs to have minimal effect on carbon stock estimates given efforts to reduce sampling biases and maintain similar plot-level surface areas in each land-use system.

For each nested plot in the MDF, a 314 m² plot (10-m radius) was used to measure trees with DBH ≥10 cm, and a 28 m² subplot (3-m radius) was used for saplings with DBH between 2.5 and 9.9 cm. Soil samples were taken randomly in a 1 m² subplot (0.6-m radius) to determine bulk density and organic carbon (described in “Soil sampling and analysis” section). Additional information including slope aspect and gradient, canopy cover, and evidence of natural or anthropogenic disturbance was reported for each plot.

Similar to data processing in the CAFs, the densities of trees and saplings were determined by counting the number of stems present at 1.37 m height and dividing by the total area sampled (314 m2 for trees; 28 m2 for saplings). The basal area of tree species was summed and averaged for all MDF plots, and plot-level tree species composition was examined in terms of tree species richness.

Aboveground and belowground carbon estimation

For both land-use systems, allometric models were used to estimate aboveground and coarse root carbon stocks of all measured trees, saplings, and coffee trees. Each of the allometric models utilized incorporated DBH measurements in the determination of aboveground biomass (Table 1). In the case of Musa paradisiacal (banana), Pinus spp. (pine), Quercus spp. (oak), Coffea arabica (coffee), Grevillea robusta (the Australian exotic “lacewood”), and species in the genus Inga, species-specific allometric models were employed (Table 1). Four of these models were developed locally in the western highlands of Guatemala (Castellanos et al. 2010; Castellanos et al. 2011). For all other trees, an allometric model developed by Sandra Brown and colleagues for dry tropical forests (900–1500 mm), published in Pearson et al. (2005), was employed. This model was deemed appropriate because it more closely corresponded to the species-specific allometry developed locally for Inga species (Fig. 2a) (Castellanos et al. 2010) and Quercus species (Fig. 2b) (Castellanos et al. 2011) than allometry developed for moist tropical forest systems (Pearson et al. 2005) and agroforestry systems (Segura et al. 2006).
Table 1

Regression equations used to estimate total aboveground biomass

Species group

Allometric model



DBH range (cm)


Dry tropical forest (900–1500 mm)

AGB = 0.2035*DBH2.3196




Pearson et al. (2005)a

Quercus species

AGB = 0.1773*DBH2.2846




Castellanos et al. (2011)

Musa paradisiacal

AGB = 0.0303*DBH2.1345




van Noordwijk et al. (2002)

Inga species

AGB = 0.01513*DBH3.0054




Castellanos et al. (2010)

Grevillea robusta

AGB = 0.09517*DBH2.47013




Castellanos et al. (2010)

Pinus species

AGB = exp(−1.170 + 2.119*ln(DBH))




Brown (1997)

Coffea arabica

AGB = 0.1955*DBH1.648




Castellanos et al. (2010)

AGB aboveground biomass, DBH diameter at breast height [tree DBH (cm) is measured at 1.37 m above ground, coffee tree DBH (cm) is measured at 0.15 m above ground]

aEquation is cited in Pearson et al. (2005) as “Brown, unpublished”. It is the updated version of the dry tropical forest equation published in Brown (1997)
Fig. 2

Aboveground biomass (AGB) estimates of aInga species in CAFs and bQuercus species in the MDF from different assumed allometric relationships. Species-specific models derived locally (a,e) were plotted and compared to generic allometric models (b,c,d) to determine the most appropriate allometric function to employ in estimating AGB for each land-use system. Superscripts take the form: AGB = 0.01513*DBH3.0054 (Castellanos et al. 2010); AGB = 0.2035*DBH2.3196 (Pearson et al. 2005); c AGB = exp(−2.289 + 2.649ln(DBH) − 0.021ln(DBH2)) (Pearson et al. 2005); dAGB = 0.14655*DBH2.223 (Segura et al. 2006); and AGB = 0.1773*DBH2.2846 (Castellanos et al. 2011)

Figure 2 demonstrates the allometric model developed for moist tropical forests to overestimate the biomass of Inga species (Fig. 2a) and Quercus species (Fig. 2b), particularly for large trees, indicating its ill suitability for estimating AGB in both land-use systems. Furthermore, the generic allometric model developed for agroforestry systems underestimates the biomass of Inga species when compared to the locally derived species-specific allometry (Fig. 2a). For both Inga and Quercus species, the generic allometric relationship that most closely corresponds to the species-specific allometric functions is that derived for dry tropical forests (Pearson et al. 2005) (Fig. 2).

For all trees, saplings, and coffee trees, the carbon present in living biomass was assumed to be 50 % (Brown 1997; Penman et al. 2003). The biomass of coarse roots was estimated by applying an allometric model developed by Cairns et al. (1997) (r² = 0.84) (Eq. 1 below). Application of this allometry has been widely utilized (Brown 2002a, b; Kirby and Potvin 2007; Malhi et al. 2009; Nakakaawa et al. 2009; Pearson et al. 2005; Soto-Pinto et al. 2010) and is an accepted methodology within the IPCC’s Land Use, Land-Use Change and Forestry program (Penman et al. 2003). Fine root biomass was not assessed for this study.
$$ Y =\exp \, \left( { - 1.0587 + 0.8836 \, \left( {\ln \, AGB} \right)} \right) $$
where Y root biomass density (Mg ha−1) and AGB aboveground biomass density (Mg ha−1)

Soil sampling and analysis

In each plot and land-use system, soil samples were collected to determine bulk density and organic carbon. Because most marked differences in soil organic carbon following land conversion (e.g. forest to agriculture, pasture to agriculture, agriculture to pasture) occurs in the upper soil layer (Fearnside and Barbosa 1998; Powers 2004; Rhoades et al. 2000; Schlesinger 1986; 1990; Wang et al. 1999), focus was given to soil layers to 10-cm depth. Soil samples were collected at each plot location using a 10-cm metal cylinder of known volume (188 cm3) with a driving tool (Rowell 1994). All soil samples were air-dried at room temperature and sub-samples from each plot location were oven-dried at 105 °C to determine residual moisture and correct all calculations to a dry basis. Bulk density was calculated as the mass of oven-dry soil divided by the core volume.

To determine carbon content, soil samples were passed through a 2-mm sieve, oven-dried at 105 °C, and analyzed for C and N within 2–3 months after collection. Soil carbon concentrations (percent soil C) were determined at the CEAB laboratory in Guatemala City by combustion on an automated Flash EA 1112 NC soil analyzer (by CE Elantech). Soil organic carbon content (SOC) was calculated using bulk density and sampling depth (Eq. 2).
$$ SOC =CC * D * BD $$
where SOC soil organic carbon from 0 to 10 cm (Mg ha−1), CC organic carbon content in the soil (g/g), D sampling depth (cm), and BD bulk density (g cm−3).

Statistical analyses and allometric model congruity

Soil bulk density, tree metrics (e.g. density, basal area, species richness), and total carbon stocks (AGB C + coarse root C + SOC) of each land-use system were calculated and compared statistically. Tests for homogeneity of variance in AGB demonstrated heteroscedasticity so we assumed a non-parametric dataset and used Mann–Whitney U tests for all statistical operations. All comparisons between datasets were conducted at the 95 % confidence level.

While we present results using the best known allometric models for our study region, we acknowledge that other allometric models have been used to estimate AGB in CAFs. To determine the variability of aboveground carbon estimates as a function of allometry, we recalculated the carbon stocks of trees, saplings, and coffee trees using other, commonly used allometric models developed specifically for CAFs. Table 2 outlines the alternative combinations of allometric models (“model set 1”; “model set 2”) employed to draw comparisons on aboveground carbon outcomes. Model set 1 employed allometry derived for shade trees in Nicaraguan coffee farms (Súarez Pascua 2002), and allometry derived for coffee trees in Indonesia (van Noordwijk et al. 2002). Model set 2 employed allometry derived for coffee trees and shade trees in Nicaraguan agroforestry systems (Segura et al. 2006) (Table 2). For both model sets, species-specific allometry derived for Musa paradisiacal (van Noordwijk et al. 2002) was utilized.
Table 2

Alternative combinations of allometric models for trees, saplings, and coffee trees employed to draw comparisons on biomass outcomes

Model set

Aboveground pool

Allometric model



Trees and saplings

log10AGB = −0.9578 + 2.3408(log10DBH)

Súarez Pascua (2002)


Coffee trees

AGB = 0.2811*DBH2.063

van Noordwijk et al. (2002)


Trees and saplings

log10AGB = −0.834 + 2.223(log10DBH)

Segura et al. (2006)


Coffee trees

log10AGB = −1.181 + 1.991(log10DBH)

Segura et al. (2006)

Models described for shade trees were used in sapling biomass calculations as well. In all model sets, species-specific allometry derived for Musa paradisiacal by van Noordwijk et al. (2002) was utilized

AGB aboveground biomass, DBH diameter at breast height [tree DBH (cm) is measured at 1.37 m above ground, coffee tree DBH (cm) is measured at 0.15 m above ground]


Summary of tree metrics

Twenty-four shade-tree species were identified in CAFs, including Persea americana, Grevillea robusta, Musa paradisiacal, Mangifera indica, Yucca guatemalensis, Trema micrantha, Cedrela odorata, and species in the Inga, Eucalyptus, Diospyros, and Quercus genera. In conversations with coffee growers, many of the shade-tree species inventoried were identified as important sources of fuelwood (e.g. Inga, Quercus, and Cedrela spp.), medicine (e.g. Eucalyptus spp.), or fruit for sale on the local market (e.g. Persea, Musa, and Mangifera spp.). Tree species richness at the plot-level was low (\( \bar{x} \) = 2.66 ± 0.166) and was highly skewed in favor of Persea americana and Inga species. These two tree species collectively accounted for 86 % of the total basal area of shade trees (≥10 cm DBH) in CAFs. Plot-level tree species composition, together with supervised classifications of digital photographs demonstrating the canopy cover of all plots to average 66 %, coincides with characterizations of traditional and commercial polyculture management systems described in the literature (Moguel and Toledo 1999; Philpott et al. 2008).

Within the MDF, 48 species were identified, including Alnus arguta, Alnus jorullensis, Billia hippocastamun, Ostria carpinus, Pinus maximinoi, Quercus flagellifera, and Quercus sapotifolia. Tree species richness at the plot-level averaged 4.17 ± 0.46 and was skewed in favor of species in the Quercus and Alnus genera, which together accounted for 54 % of the total basal area of trees ≥10 cm DBH. Species in the Quercus genera figured prominently in the understory layer, alongside species in the Ardisia, Cornus, and Ostria genera.

In comparing tree metrics associated with both land-use systems, tree density was significantly lower in CAF plots (218 stems ha−1) than in MDF plots (539 stems ha−1), on average (Table 3; Fig. 3). The sapling density in CAF plots was also significantly lower (Table 3), in part because coffee trees occupied understory that would otherwise be occupied by saplings. When incorporating coffee trees in the sapling density calculation, CAF sapling density equaled 10,025 ± 672 stems per ha, or three times greater than the sapling density in the MDF plots.
Table 3

Soil bulk density and tree metrics for coffee agroforests (CAFs) and the mixed dry forest (MDF)







Tree density (ha−1)

217.76 ± 14.01

538.63 ± 57.23




Sapling density (ha−1)

140.79 ± 27.35

3307.45 ± 506.18




Coffee density (ha−1)

9884.21 ± 661.55





Tree DBH (cm)

26.26 ± 0.78

20.61 ± 0.75




Tree H (m)

11.30 ± 0.23

11.46 ± 0.31




Tree basal area (m2 ha−1)

15.23 ± 1.10

27.18 ± 5.16




Soil bulk density (g cm−3)

0.80 ± 0.02

0.38 ± 0.02




The values are presented as the mean ± 1 standard error

DBH diameter at breast height [tree DBH (cm) is measured at 1.37 m above ground, coffee tree DBH (cm) is measured at 0.15 m above ground]; H height, Z Mann–Whitney U test statistic, P P value
Fig. 3

Distribution of carbon stocks and stem density among 10-cm diameter classes in a coffee agroforests and b the mixed dry forest. Carbon stocks (bars) are presented as the mean per hectare ±1 standard error. Tree density (points) is the mean number of stems per hectare ±1 standard error

Shade trees in CAFs had a significantly larger mean DBH (26.3 ± 0.8 cm) than trees in the MDF (20.6 ± 0.8 cm) (Table 3). This difference largely resulted from the high proportion of small- to medium-sized (10–30 cm) trees (n = 343; 88 % of all trees inventoried) relative to large trees in the MDF plots (Fig. 3). Stems ≤30 cm DBH were frequent in CAFs as well, but their proportion relative to the total inventory was lower (70 %).

Soil bulk density in CAFs and the MDF ranged from 0.5 to 1.0 g cm−3, and 0.2–0.55 g cm−3, respectively. While bulk density values in CAF plots were generally higher than in the MDF, both land-use systems had bulk density values averaging less than 1.0 g cm−3 (Table 3) suggesting high soil porosity and low compaction in the two land-use systems (FAO 2006).

Carbon stocks

The total amount of carbon stored in CAFs ranged from 74.0 to 259.0 Mg ha−1 with a mean of 127.6 ± 6.6 Mg ha−1 (Table 4). Trees contained the greatest amount of carbon followed by soil (to 10 cm depth) (Table 4; Fig. 4). On average, 47 % of the carbon stored in CAF plots was found in tree biomass, 30 % in soils, 13 % in coarse roots, and 10 % in coffee tree biomass (Fig. 4). Saplings stored less than 1 % of the total carbon stocks of CAFs as they were largely replaced by coffee trees.
Table 4

Summary by land-use system of the aboveground, belowground, and soil carbon pools


Carbon (Mg ha−1)

Mann–Whitney U test






Aboveground pools

 Trees (≥10 cm DBH)

59.44 ± 5.48

114.64 ± 27.71




 Saplings (2.5–9.9 cm DBH)

0.84 ± 0.16

13.71 ± 2.91




 Coffee trees

12.90 ± 0.88

0 ± 0




 Total aboveground carbon

73.18 ± 5.30

128.35 ± 27.28




Belowground pools

 Roots (trees, saplings, coffee trees)

16.21 ± 0.98

25.21 ± 4.45




 Soil (0–10 cm)

38.24 ± 1.54

45.09 ± 2.71




 Total belowground carbon

54.45 ± 1.88

70.30 ± 5.36




 Total estimated carbon

127.62 ± 6.57

198.65 ± 32.08




The units for all values are Mg ha−1 and are presented as the mean ± 1 standard error

DBH diameter at breast height [tree DBH (cm) is measured at 1.37 m above ground, coffee tree DBH (cm) is measured at 0.15 m above ground), Z Mann–Whitney U test statistic, PP value, NS not significant
Fig. 4

Total carbon stocks (Mg ha−1) of coffee agroforests and the mixed dry forest by carbon pool

The total carbon stored in the MDF ranged from 60.6 to 807.7 Mg ha−1 with a mean of 198.7 ± 32.1 Mg ha−1 (Table 4). The distribution of carbon in MDF plots followed a similar trend to CAF plots, with 58 % of carbon held in tree biomass, 23 % in soils, 13 % in coarse roots, and 7 % in sapling biomass (Fig. 4). One outlier plot existed within the MDF in which the C-storage value (807.7 Mg ha−1) was more than twice the second-highest value (366.0 Mg ha−1). This plot contained four trees ≥70 cm DBH. While no other MDF plot was characterized by an abundance of large trees, other studies in Guatemala have reported similar, high variability in carbon stocks within disturbed forests in the region, mainly due to the occasional presence of a few, very large trees (Castellanos et al. 2007). From conversations with Santa Clara community members, large trees were expected to be found intermittently in Cerro Chuiraxamoló because they are left undisturbed for spiritual reasons. As such, we retained this plot in our analyses of carbon stocks.

When all aboveground, belowground, and soil components were included, the MDF plots had significantly more carbon than CAF plots at the 95 % confidence level (Table 4). In comparing individual carbon stock pools, the MDF had more carbon in tree biomass (p = 0.057), sapling biomass (p = 0.000), root biomass (p = 0.046), and soil organic matter (p = 0.059) than estimated for CAFs (Table 4). However, neither tree biomass nor soil organic carbon was significantly different between the two land-use systems at the 95 % confidence level.

Separating trees into 10-cm size-classes (Fig. 3) demonstrated that large trees play a substantial role in determining variations in aboveground carbon between the two land-use systems. Stems ≥100 cm DBH were absent in CAFs and while infrequent in the MDF, they greatly contributed to its total carbon stock (Fig. 3). For example, three trees ≥100 cm DBH contributed 22 % to the total tree carbon stock, a proportion roughly matched by the 343 trees with DBH between 10 and 30 cm (29 %).

Congruity of allometric models for CAFs

Evaluations of alternative allometric models (described in Table 2) demonstrate high levels of variability in aboveground carbon estimates as a function of different assumed allometric relationships (Table 5). Two important results are evident from comparisons of aboveground carbon stocks following application of alternative allometric models (Table 5). First, tree (≥10 cm DBH) biomass estimates derived from allometry developed specifically for agroforestry systems (Segura et al. 2006; Súarez Pascua 2002) were lower than estimates derived in this study using species-specific allometry and allometry developed for dry tropical forests (as described in Table 1). The difference in tree carbon stock estimates is largely reflected in the form of the regression curve for trees in dry tropical forests (Pearson et al. 2005) as opposed to agroforestry systems (Segura et al. 2006; Súarez Pascua 2002). The allometric relationship between DBH and biomass for dry tropical forests rises more steeply at large diameters than is evident for agroforestry systems (Fig. 2a), thereby producing higher estimates of aboveground biomass for sites with large trees.
Table 5

Summary of the aboveground carbon pools for coffee agroforests using alternative sets of allometric models described in Table 2


Carbon (Mg ha−1)

This study

Model set 1

Model set 2

Aboveground pools

 Trees (≥10 cm DBH)

59.44 ± 5.48

36.87 ± 3.29

31.39 ± 2.62

 Saplings (2.5–9.9 cm DBH)

0.84 ± 0.16

0.54 ± 0.10

0.51 ± 0.09

 Coffee trees

12.90 ± 0.88

40.07 ± 3.25

8.19 ± 0.64

Total aboveground carbon

73.18 ± 5.30

77.48 ± 4.10

40.09 ± 2.53

The units for all values are Mg ha−1 and are presented as the mean ± 1 standard error

DBH diameter at breast height (tree DBH [cm] is measured at 1.37 m above ground, coffee tree DBH [cm] is measured at 0.15 m above ground)

A second finding from Table 5 demonstrates high variability of coffee tree carbon stocks associated with allometry. Using a locally-derived allometric model (Castellanos et al. 2010), this study estimated coffee carbon biomass as 12.9 Mg C ha−1 (Table 4). In comparison, allometry developed by Segura et al. (2006) in Nicaragua and van Noordwijk et al. (2002) in Indonesia resulted in an average coffee biomass of 8.2 and 40.1 Mg C ha−1, respectively (Table 5). Use of van Noordwijk et al. (2002) allometry led to a three-fold increase in coffee tree biomass (Table 5) and a 26 % gain in total C-storage (assuming all else equal from Table 4). This gain resulted in CAF and MDF plots not being significantly different in terms of their total C-storage (Z = −0.223, p = 0.823, data not shown).


Carbon storage in coffee agroforests

The CAFs included in this study stored substantial amounts of carbon in aboveground biomass and soil. Our mean carbon stock values are comparable to carbon values of 120.9–195.0 Mg C ha−1 for shade-grown coffee systems in Costa Rica7 (Ávila et al. 2001). Other studies have shown carbon stocks of shade-grown coffee systems to equal 82 Mg C ha−1 in Indonesia (van Noordwijk et al. 2002), 82 Mg C ha−1 in Togo (Dossa et al. 2008), and 167.4–213.8 Mg C ha−1 in Mexico (Soto-Pinto et al. 2010).8 The wide range in carbon stock values stems from differences in agrosilvicultural management, ancillary factors (e.g. soil condition, climate, system age, land-use history), and allometric model selection, all of which influence the development of large-scale policy prescriptions aimed at C-sequestration in agroforestry systems.

Agrosilvicultural management, in particular shade tree management, plays an integral role in determining the C-sequestration potential of coffee agroforests. In our CAF plots, total carbon stocks were heavily dependent on non-coffee biomass with trees, saplings, and their associated root biomass storing 57 % of the total carbon. This finding corresponds to research demonstrating the presence of shade trees to confer additional carbon benefits not found in unshaded coffee systems, maize cropping, or other annual crop system (Ávila et al. 2001; Dossa et al. 2008; Harmand et al. 2006; Soto-Pinto et al. 2010). However, not all shaded coffee plantations are equal, nor equally beneficial in terms of carbon sequestration. Worldwide, coffee is grown along a strong gradient from traditional “rustic” (highly shaded, less managed) to sun coffee (Moguel and Toledo 1999; Philpott et al. 2008), variations of which affect carbon accumulation and the provision of other ecosystem services. In Guatemala, an estimated 98 % of the country’s coffee grows beneath a canopy of shade (ANACAFÉ (Asociación Nacional del Café de Guatemala) 2008); however, high variability exists regarding the diversity of shade trees planted and the total percent shade cover. For example, Guatemala’s Antigua region tends to have a monoculture canopy of Grevillea robusta, the Huehuetenango region tends to have a diverse canopy dominated by Inga species, and the San Marcos region tends to have little to no shade in which shade trees are pruned into low-stature cover (Jha et al. 2011).

While our CAF plots are characterized as polyculture coffee agroforests, a high level of diversity existed with regards to shade tree species composition and density, and shade-tree management practices. For example, plots where carbon stocks exceeded 200 Mg C ha−1 (n = 2) were characterized by few (<5), but large trees, and carbon-rich soil relative to other agroforest and forest plots (>60 Mg C h−1 in 0–10 cm depth). In contrast, plots where carbon stocks were less than 90 Mg C ha−1 (n = 4) were often characterized by lower soil organic carbon (<35 Mg C ha−1) and many (≥10) medium-sized trees. We expect carbon stock variability to result primarily from agrosilvicultural management; however, we recognize that environmental factors (e.g. altitude, climate, local disease problems) and social processes and structures also play a significant role in determining carbon accumulation. For example, elevation and climate influence shade cover and associated C-storage in Chiapas, Mexico, with coffee agroforests in high tropical agro-climatic zones (≥1000 m ASL) storing more carbon than in intermediate and low agro-climatic zones (Soto-Pinto et al. 2010).

Though environmental and social factors affect the C-storage capacity of a system, allometric models (assuming biometric sampling rather than micrometeorological sampling) ultimately determine the carbon “value” of a system, and play a significant role in the development of conservation-based initiatives. Overall, allometric models are one of the weakest links in the chain of calculations to estimate carbon stored in a system (Nair et al. 2009) with errors inherent in the model(s) and associated with model selection contributing to uncertainty in carbon estimates (Chave et al. 2003; Ketterings et al. 2001; Kirby and Potvin 2007; Segura et al. 2006). Results of the model congruity exercise (Table 5) demonstrated the average carbon stocks of trees (≥10 cm DBH) and coffee trees to vary by up to 45 and 80 %, respectively, within CAFs when calculated with the alternative allometric models.

A particularly challenging issue in studying agroforestry systems is the absence of species-specific allometric models for shade trees, as most models have been developed for trees in primary or secondary forests. Because the size of individual tree canopies, tree-management practices, and crown architecture differ considerably by forest type (Nair et al. 2009), species-specific allometry is needed to improve the precision of carbon estimates. Allometry is particularly important for agroforest shade trees and coffee trees given pruning and removal of branches for firewood. Because tree management practices differ considerably among and within countries, allometric models developed for a country or set of conditions different from the system(s) under study will introduce biases.

To enable the development of conservation-based initiatives oriented around C-sequestration in coffee agroforests, species-specific allometry that accounts for local climatic and edaphic conditions, as well as common tree management practices must be developed. Furthermore, more destructive sampling, particularly for large trees, is needed to improve the range of applicable diameters in allometric functions. The largest trees inventoried in this study were 95.8 cm DBH and 130 cm DBH in CAFs and the MDF, respectively, while the generic allometric model applied to estimate biomass (Pearson et al. 2005) was viable for trees less than 63 cm DBH. We expect, therefore, that the biomass of large trees may be overestimated. This may be particularly true for CAFs, where an additional source of error stems from tree pruning which is not accounted for in allometric functions.

A limitation of this research is the exclusion of dead organic matter, litter, fine roots, and soil >10 cm depth which likely leads to the underestimation of total C-storage within CAFs and the MDF. In particular, we expect the C-storage capacity of soils in both land-use systems to be higher than estimated in this study given our sampling depth of 0–10 cm. In general, the amount of carbon in soil usually exceeds that of carbon in living vegetation (Post and Kwon 2000) and sampling at sub-10 cm depths will provide a more comprehensive picture of the total C-storage capacity of each land-use system, an area to be explored in future work.

The potential for conservation-based carbon programs for coffee agroforests

Despite complexities in understanding carbon accumulation in coffee agroforestry systems, and the lack of standardized datasets with which to develop widespread policy programs (Nair et al. 2009), it is increasingly necessary to recognize the carbon stored in shade-grown coffee systems. Processes of land-use/cover change have been documented for many coffee-growing regions (Ávalos-Sartio and Blackman 2010; Eakin et al. 2005; 2006; Ellis et al. 2010; Nestel 1995), with market forces, governmental policies, and climate factors underpinning coffee abandonment and the expansion of unshaded agricultural systems (Ellis et al. 2010; Rice and Ward 1996).

Densely shaded coffee agroforests have demonstrable carbon benefits, and provide a number of other ecosystem services and functions (e.g. biodiversity, soil conservation, water regulation) similar to “natural” forest ecosystems. In this study, the average carbon stored in CAFs (127 Mg C ha−1) was significantly lower than estimated in the nearby MDF, Cerro Chuiraxamoló (198.7 Mg C ha−1); however, neither tree biomass nor soil organic carbon was significantly different between CAFs and the MDF. Furthermore, a recent report by Castellanos et al. (2007) found the carbon stored in 25 different pine-oak forests in the western highlands of Guatemala to average 129 Mg C ha−1, a finding that suggests our coffee agroforest sites sequester carbon similarly to other middle-elevation forests in the region.

While the development of coffee certification programs (e.g. biodiversity, organic, Fair Trade, bird friendly) has improved social and ecological development within the coffee sector (though arguably for a small subset of coffee producers) (Bacon 2005), such programs have not implicitly recognized the value of carbon sequestration. Furthermore, REDD programs to-date have largely focused on the carbon benefits conferred by improved tropical forest management while research and development in tropical agroforestry systems have garnered differentially less attention. Such focused conservation efforts fail to account for the vast tracts of land under agroforestry farming, the carbon sequestration potential of these lands, and their contribution to livelihoods. Tropical landscapes are increasingly characterized by patchworks of forest fragments in a matrix of agriculture (Perfecto et al. 2009), and while the matrix concept has been proposed as a new paradigm for biodiversity conservation efforts, it is applicable to climate change mitigation strategies as well. Institutional efforts have largely ignored the matrix in favor of focused attention on forest fragments; however, market volatility, climatic stressors, and expansion of “low quality matrices” (Perfecto et al. 2009) necessitates development of sound institutions that aim to maximize not only the economic benefits of coffee, but the ecological and social benefits as well.

The recent implementation of a “climate friendly” seal by Rainforest Alliance (RA 2011), and the potential inclusion of agroforests as REDD programs, may incentivize coffee growers to maintain or enhance C-sequestration in the face of external stressors. Such programs should aim to promote carbon benefits alongside other ecosystem services to prevent tradeoffs associated with enhancement of one service at the expense of another (Bennett et al. 2009; Robertson and Swinton 2005). Recent work by Raudsepp-Hearne et al. (2010) suggests ecosystem service “bundling” to determine potential tradeoffs among interacting ecosystem services that result from management interventions. While bundling interactions may complicate the design and implementation of broad comprehensive policy prescriptions (given variations in agrosilvicultural management, climate, and socio-economic context), bundling may also promote synergistic services and a greater understanding of the full agro-ecological system.


In recognition of the important role coffee agroforests may play in the global carbon cycle, quantifying and understanding the carbon profile of shade-grown coffee systems is critical to the development of climate change mitigation strategies. Results of this study demonstrate that both coffee agroforests and natural forests in the tropics store substantial amounts of carbon. As such, protection of both forest-based systems (“avoided deforestation”) will likely have a larger impact on landscape-level carbon stocks than focused attention on only one system. Because agroforestry systems are complex and highly variable across regions, greater research is needed for the development of CDM and carbon accounting systems. This study focused on understanding one type of agroforestry system, densely shaded polyculture coffee agroforests; however, future work will expand this research by quantifying the C-storage capacity of coffee systems as they relate to percent shade cover, pruning frequency, and agrosilvicultural management. Greater attention will also be given to soil carbon stocks which remain understudied despite their importance to carbon sequestration.


There is no record of the exact number of smallholders in the study region. To our knowledge, all smallholders in the region grow coffee under shade trees, though the extent and magnitude of shade cover varies.


Total area is an approximation based on aerial photographs and observations made during fieldwork.


The ASTER image was distributed by the Land Processes Distributed Active Archive Center (LP DAAC), located at the U.S. Geological Survey (USGS) Earth Resources Observation and Science (EROS) Center (


NDVI is the Normalized Difference Vegetation Index, calculated as (NIR − VIS)/(NIR + VIS) where NIR and VIS stand for the spectral reflectance measurements acquired in the near-infrared and visible (red) regions, respectively. Values close to one indicate higher densities of ‘greenness’ than values close to zero (Mather 2004).


Coffee tree density and the presence of low-lying coffee branches in the CAFs complicated the creation of circles, and without removing or breaking branches extensively, accurate plot boundaries were not easily attained.


All values are presented as the mean ± 1 standard error.


Carbon stock estimates by Ávila et al. (2001) include shade tree biomass, coffee tree biomass, and soil organic carbon from 0 to 25 cm depth. While soil sampling in our CAFs was less than that of Ávila et al. (2001), our average total carbon stocks are similar.


Soil sampling depths in Indonesia, Togo, and Mexico equaled 30 cm (van Noordwijk et al. 2002), 40 cm (Dossa et al. 2008), and 30 cm (Soto-Pinto et al. 2010), respectively.



This research was funded by the National Science Foundation’s Geography and Spatial Sciences Program (DDRI #0927491). Many thanks to the smallholder coffee growers in Guatemala who graciously allowed us to collect data on their land during harvesting season and to the community of Santa Clara for allowing access to their municipal park. We also thank Arturo Ujpán Mendoza at Ati’t Ala’ for his help collecting field data, Gabriela Alfaro at the Universidad del Valle de Guatemala for her help with data analysis, and Sarah Mincey at Indiana University for her comments and suggestions on the manuscript.

Copyright information

© Springer Science+Business Media B.V. 2012