Vincent de Leijster

95 Ecosystem services trajectories in coffee agroforestry in Colombia over 40 years 5 where CP is obtainable coffee productivity, NPS the number of productive shoots, NF the number of fruits per node, NFN the number of fruiting nodes per plant, and DeadB the number of dead branches. Further, we included physical coffee bean quality as a proxy for coffee quality (section 5.2.3.1). At the farm-level, we recorded coffee yield as farmer’s self- reported coffee yield as another proxy for quantity (section 5.2.3.2). Additionally, we included timber volume as an indication for potential timber production. To this purpose, we calculated the timber volume of the two timber tree species that were provided in the reforestation projects: C. alliodora and E. grandis (only present on one farm). The timber volume (TV) of C. alliodora in coffee farms was calculated using the allometric equation of Somarriba and Beer (1987): TV C. alliodora = -0.0176 + 0.000034 · DBH 2 · H – 0.000086 · DBH 2 + 0.00336 · H (equation 10) where DBH is diameter at breast height and H the tree height. The timber volume of E. grandis in coffee plantations was calculated using the allometric equation of FNC - Cenficafé (2006): TV E.grandis = exp(-8.988 + 0.847 · ln(DBH 2 · H)) (equation 11) 5.2.5 Data analysis 5.2.5.1 Effect of time since agroforestry on canopy cover and ecosystem services We investigated how farm vegetation characteristics varied with ‘time since agroforestry’, and therefore we excluded the monoculture coffee farms (Table 5-1). We first tested whether the development of canopy cover, canopy height, tree density, basal area, and tree species richness followed an asymptotic shape. We used a nonlinear Michaelis-Menten model, expressed generally as: Y = (equation 12) where Y is any of the vegetation characteristics variables, TSA is time since agroforestry, β max is the asymptote at which Y is saturated, β 1/2 is Y at half its asymptote. If the asymptote model showed poor fit, we tested the fit of a linear relationship by using ordinary linear regression (OLR), except for tree species richness for which we used a generalized linear model (GLM) with a Poisson distribution.