Study area

The study was conducted on four farms in the region of Huatusco, state of Veracruz, between May 2012 and April 2013 (Figure 1). To reduce environmental variability, farms were selected according to their similarity in structure and management, as diversified or simple poly-cropping (^{Hernández-Martínez et al., 2009}), the degree of soil erodibility (factor K), annual rainfall, and slope length and steepness (Factor LS) (Table 1).

Figure 1 Location of the studied farms in central Veracruz. ■ farm; ○ weather station; EMH: weather station HUCH Huatusco. 

Table 1 Environmental characteristics of the studied farms. 

The data used to characterize the soil and calculate erodibility (according to ^{Römkens et al., 1997}) were obtained in October and November of 2011 from two topsoil samples (0-20 cm) obtained near the measuring plots on each farm. The climate is type (A)C(m), warm with abundant summer rains (^{Soto and García, 1989}), and the dominant soil is Andic acrisol (^{Geissert and Ibañez, 2008}). Daily and historic precipitation data, during the study period were obtained from the HUCH Huatusco weather station (19° 10’ 29.2” N, 96° 57’ 44.8” W and 1426 m of altitude) (Figure 1), of INIFAP-SAGARPA ( http://clima.inifap.gob.mx/redinifap/est.aspx?est=35873). Total annual precipitation for each farm was estimated by spatial interpolation in ArcGis 10.1 using the precipitation recorded by the nine closest weather stations (Amatitla, El Coyol, Coscomatepec, Elotepec, Tenampa, Ixhuatlán del Café, Totutla, Chapingo Regional University Center Huatusco and Santa María Tatetla) from the same network.

Experimental design and treatments

On each farm, three blocks were established, each consisting of three rectangular 8×4 m plots m (the longer sides parallel to the slope). A chapeo, herbicide, or control treatment was assigned randomly to each plot (Figure 2).

Figure 2 Experimental design of blocks and plots on farms. Treatments: T=control, H=herbicide, CH=chapeo. Black dots in each plot represent the stakes for measuring erosion. 

Chapeo in May, July and October 2012 consisted of cutting weeds with a machete to 10 cm above ground level. Herbicide was applied once in July with a manual backpack sprayer in the same manner on the four farms. Granulated Rival® glyphosate (74.7 % glyphosate) was used (100 g dissolved in 30 L water, resulting in a dosage of 521 L ha^-1). The control consisted of cutting weeds with a brush cutter or machete 5 cm from the ground according to each grower’s current practice.

Measurement of soil loss

To measure soil loss, 30 cm long, 0.5 cm diameter metal stakes treated with anticorrosive paint were used. A washer was soldered onto each stake 8 cm below the top edge. In each plot seven stakes, distributed to include herbaceous plant cover, litter, and bare soil, were hammered into the plot until the washer was at soil level. The stakes were installed in early May 2012 and were removed at the end of April 2013. The distance between the soil surface and the washer was measured on the downslope side of the stake with a digital Vernier with a precision of ±0.2 mm. In April 2013, a total of 63 registers of erosion were recorded, 21 in each treatment. Mean erosion per plot (E) was expressed in negative numbers (soil loss) and mean sedimentation (S) in positive numbers (soil accumulation). Net erosion (EN) refers to the difference between erosion and sedimentation, expressed in mm or in Mg ha^-1 and was calculated for each plot as EN=E+S, a formula equivalent to that of ^{Pizarro and Cuitiño (2002}) when negative values of E are used with the following equalities:

The conversion in units of soil loss (Mg ha^-1) was performed using bulk density (DA):

In each plot, the degree of ground cover (weeds, litter or bare soil) was determined using a visual percentage estimator (^{Folk, 1951}) in March 2013. Tree cover was estimated at its peak in July 2012 using a spherical concave densitometer at a height of 1.20 m from the ground. For bush cover, the density of coffee plants per plot was calculated (number of plants per 32 m²).

Analysis of the relationship between plant cover and erosion

The effect of the types of ground cover and the treatments (chapeo, herbicide, and control) on net erosion was determined by adjusting a mixed linear model using maximum likelihood. The fixed variables of the model included the treatments, herbaceous plant cover (%), litter (%), and bare soil (%) recorded in March 2013 (the date on which there was greatest contrast among the treatments (with herbicide, chapeo and control), density of coffee plants per plot, and tree canopy cover. The random effect factors corresponded to the blocks of plots nested within farm.

The spatial structure of the sampling design did not allow us to perform a simple ANOVA since none of the observations contributed with a complete degree of freedom and random factors were included in the analysis. To find the minimal adequate model, a simplification process (Stepwise) was carried out, making comparisons between the models that progressively included and excluded the different fixed variables and their second order interactions. This was done to optimize the selection of the variables that explained the most variation and that were orthogonal, thus reducing redundancy in the final model (^{Crawley, 2007}). To compare the nested models in the process of simplification, the parsimony criterion was used by comparing the Akaike Information Coefficient (AIC), for which the difference of each AIC (ΔCIA) relative to the maximum model was calculated. The selected model was that with the lowest number of explanatory variables and a ΔCIA lower than or equal to six, relative to the model with the highest ΔCIA (^{Cayuela, 2012}; ^{Richards, 2015}). Data were analyzed using the R statistical language and environment (Studio 0.98.994, R Development Core Team, https://www.r-project.org).

Erosion rates

Rainfall was high during the study year (2392 mm), with 22.2 % more precipitation than the historical mean in Huatusco (1938 mm) (Figure 3). Total rainfall was distributed over 209 days. Most of the rainfall was concentrated in the months of June to August, when there were 13 days with more than 50 mm of rain. When the intervals of precipitation and erosivity of ^{Pérez-Nieto et al. (2005}) were applied to registers of the Huatusco weather station, 51 days between June and October were detected with more than 12.7 mm of rain (24.4 % of the total), which had an erosive effect.

Figure 3 Monthly precipitation recorded in the INIFAP HUCH Huatusco weather station located in the municipality of Huatusco, Veracruz. 

Average net erosion in the control plots was -3.5 mm year^-1, equivalent to soil loss of -33.5 Mg ha^-1 year^-1. The high variability (standard error -38.6 Mg ha^-1 year^-1) (Table 2) corroborates that erosion has multiple causes related to the complexity of the interactions among variables (^{Boix-Fayos et al., 2006}). The runoff responsible for the erosion was mostly laminar (sheet erosion) and occasionally in small furrows (rill erosion).

Table 2 Net erosion (negative) and net sedimentation (positive) by farm and by treatment with the respective standard error. 

Of the farms with net erosion, HU03 had the lowest rate despite having a higher degree of soil erodibility than the others (Table 1). Farm HU16 had the highest erosion rate, possibly related to a steeper slope. On HU08 there was net sedimentation, resulting from transit of soil particles and aggregates and their accumulation in the plots. Previous studies in shade coffee plantations never reported net sedimentation because the measurements were made in runoff plots, from which an overall value for soil loss was obtained so that there was no possibility of quantifying accumulation within the plots. Farm HU07 had a high rate of erosion, as did HU16, which had similar conditions of precipitation and erodibility despite a less steep slope.

Erosion rates on farms HU07 and HU16 surpassed the maximum tolerance limit of 12.5 Mg ha^-1 year^-1 established by the US Soil Conservation Service (^{Soil Survey Division Staff, 1993}) for deep soils that are less susceptible to erosion. However, it is preferable to fix tolerance limits based on sustainable crop production, which takes into account conservation of organic matter and nutrients (^{Young, 1989}). In this regard, using the degrees of erosion in sun and shade coffee production systems established by Columbia’s Centro Nacional de Investigaciones de Café (Cenicafé) (^{Gómez-Aristizábal and Rivera-Posada, 1993}), farms HU07 and HU16 surpassed the maximum tolerance limit of 3 Mg ha^-1 year^-1, and erosion was very severe (>30 Mg ha^-1 year^-1). In contrast, the degree of erosion was very low (1.0-3.0 Mg ha^-1 year^-1) on HU03 and natural (<1.0 Mg ha^-1 year^-1) on HU08. These results show that soil loss by erosion in shade coffee plantations is not always below 5 Mg ha^-1 year^-1, as reported by ^{Ataroff and Monasterio (1997}), ^{Arellano (2000)} and ^{Pérez-Nieto et al. (2005}). The high values on HU07 and HU16 may be overestimates since measurements were taken on the slope directly below the stake. Here, water runoff around the stake increases turbulence even with 5 mm thick stakes. However, taking measurement downslope from stakes is a procedure that should adequately simulate the erosion process that occurs on farms located on sloping terrains where, according to our own observations, erosion often begins at the base of the coffee plant trunks. The possible overestimation of erosion with the stake technique, relative to measurements from sediment collectors, is a methodological problem that has yet to be solved (^{Marín and Desir, 2003}).

Effects of plant covers and their management on erosion

The mixed linear model revealed that, of the possible ground covers (weeds, litter, bare soil), only weeds had an effect on erosion rate. The second fixed explanatory variable was tree cover (% canopy). The final fixed factor model was:

Net erosion (mm)=19.56+(0.06*% weed cover)+(-0.28*% canopy)

The Akaike Information Coefficient was 236.68, with a ΔCIA of 3.57, relative to the model with the lowest AIC and -17.84, relative to the maximum AIC (Table 3). The estimated variance from random factors was 1.592 (with CI_95% 0.32-0.82) and residual 2.52 (with CI_95% 3.04-5.28).

Table 3 Mixed effect models that explain erosion and that, in the process of model simplification had a ΔCIA (difference from the minimum AIC) below or equal to 6. 

^†Only fixed variables of the contrasted models are mentioned: Treatment (Tr), weed cover (Ch), litter (H), bare soil (Sd) and coffee plant density (Dc) in March 2013, canopy cover (Cd) in July 2012. The random factors remained constant and were Farm nested in Block. ^¶Models were selected for their better explanatory power, their lower number of fixed variables and ΔCIA below 6. ^§Model that during the model simplification process had the minimum AIC.

The model estimated an average erosion in the study plots of 19.56 mm (CI_95% 1.5-37.7 mm). With more weed cover, there was less erosion, and in some cases even sedimentation (Figure 4), although herbaceous plant cover was a fairly poor explanatory variable (slope 0.06; CI_95% -0.01-0.12). Canopy cover had greater influence but opposite to that of weed cover (Figure 5), with more tree cover leading to more erosion (slope -2.28; with CI_95% -0.28-0.49), perhaps due in part to the increase in the erosive effect of the larger drops that run off leaves versus those falling as precipitation (^{Nanko et al., 2004}) and the negative effect of shade on weed cover.

Figure 4 Effect of weed cover on net erosion. The continuous line indicates values estimated from the mixed linear model. Each dot represents mean net erosion (negative) of seven stakes per plot. 

Figure 5 Effect of shade on net erosion. The continuous line indicates values estimated from the mixed linear model. Each dot represents mean net erosion (negative) of seven stakes per plot. 

The combined effect of tree, bush, herbaceous plant, and litter cover on variation in soil loss is complex. On a shade coffee farm in north-central state of Veracruz, ^{Pérez-Nieto et al. (2012}) determined that only bush and litter cover had a reducing effect on soil loss. Their analysis was done using individual cover types, and soil loss was measured in microcatchments with a surface runoff sampler and no replication. Their results differ from the erosion model obtained in our study, which showed that only the combined effect of tree and weed cover had an influence on erosion. Our model indicates that a tree canopy cover of ≤60 % would not cause erosion even when the herbaceous plant stratum is absent. In this case, it would make no difference in erosion whether weed control was done with chapeos or with herbicide. However, under an 80 % canopy a 40-50 % increase in weed cover would reduce erosion four fold (from 0.76 mm year^-1 to 0.18 mm year^-1), and with 60 % cover erosion should stop altogether. Under these two scenarios, it would be recommendable to maintain or increase weed cover.

The mixed linear analysis did not detect statistically significant differences between net erosion values and the herbicide, chapeo, or control treatments. Thus, conducting more tests of means was not necessary since this factor did not help to explain the variability in net erosion. The herbicide treatment, in particular, had no conclusive effect on net erosion after one year of measurement. After application in July, weeds died within a week with plants continuing to lose leaves in September. By December they had begun to grow again, and in March 2013, just before removing the stakes, weed cover was on average 18 % in herbicide treated plots, 45 % in chapeo plots, and 42 % in the control plots. By the end of the year, and compared with the controls, the herbicide application had produced an increase in net erosion only in HU08 plots (from 2.5 mm to -5.3 mm). On the other farms, there was a decrease in erosion (HU07 and HU16) and even some sedimentation (HU03) (Table 2). At this latter site, transport of particles and aggregates by runoff was stopped by dead shoots of weeds scattered over the surface of the plot, thus favoring sedimentation. These conclusions are similar to those reported by ^{Carvalho et al. (2007}) for sun coffee plantations with different treatments of herbaceous plant cover. The sites treated with herbicide had intermediate erosion values, compared to those of weed cover management and bare soil, due to the protective effect of the dead plants on the soil surface. In a shade plantation in Costa Rica, it took nine applications of glyphosate (Roundup 35,6L) over two years for soil loss to reach 0.33 Mg ha^-1 year^-1 (^{Gómez, 2005}). Thus, repeated applications of herbicides during the same year on farms in Veracruz should also increase erosion rates beyond those observed in our study.

The results obtained in our study are valid only for the environmental and management conditions of the farms studied: annual precipitation of 1750 to 1850 mm, soil erodibility K factor 0.05 to 0.10, slope LS factor 9.04 to 11.41, canopy cover >60 %, coffee plant density 1900 to 3300 ha^-1, and litter cover ≥90 %. Nevertheless, the study contributes to knowledge of erosion in multi-stratified shade coffee plantations by demonstrating spatial variation in erosion is due to erosive and sedimentation processes and the influence of management of plant cover at the soil level. However, improving our understanding of the causes and magnitude of erosion with the stake technique would require extending the study to a period of three to five years since soil erosion also varies over time (^{Afandi et al., 2002} ; ^{Verbist et al., 2010}). Additionally, a larger number of farms and plots should be included to encompass broader environmental variation since the lithological and edaphic sensitivity of a site is even more important to erosion than the age of the coffee plantation or the presence of shade trees (^{Verbist et al., 2010}). We should also note that measuring soil erosion with stakes is a procedure recommended at the plot scale when laminar processes and diffuse runoff dominate. However, this method may not be adequate under conditions of rill erosion or when extrapolating measurements to the watershed scale where erosion in gullies and landslides occur. In any event, management of plant cover is the simplest means of controlling erosion in shade coffee plantations on slopes. Therefore, integrated weed management (^{Quiroz-Marín and Hincapie-Gómez, 2007}), as well as the careful selection of tree, bush and herbaceous species (^{Sidle et al., 2006}), is recommended.