17 LS, and 16CC. Among the CC tree species, 1 was a pioneer,
7 were intermediate succession, and 8 were LS species (Table 1).
For a summary of characteristics of agroforests see Table 2.
Diversity in forest sites was higher than in coffee farms (average
Shannon diversity on farms = 1.52 and in forests = 2.43, p-value
< 0.001); the same trend is true for tree richness (average species
richness on farms = 7.7 and in forests = 18, p-value < 0.001).
Shannon diversity was a positive predictor for the proportions
of CC and LS trees in farms, also when disaggregated by previous
land use (see Table S1).
4.2. Binomial logistic regression
Shade tree abundance and the proportion of Inga spp. trees
were found to be negatively associated with the presence of CC
trees; while basal area and forest as a previous land use were
positively associated. Forest as a previous land use was the
strongest positive driver, while proportion of Inga spp. trees was
the strongest negative driver of CC trees. All explanatory variables
for CC trees were found to be statistically significant (p-value
< 0.05; see Fig. 1 for standardized parameter coefficients and 95%
and 68% confidence intervals).
The most important positive driver of trees of LS was forest as
previous land use; the most important negative driver was the
proportion of Inga spp. trees, followed by shade tree abundance. All
explanatory variables for LS trees are statistically significant (pvalue
< 0.05), except basal area (p-value > 0.05; see Fig. 1 for
standardized parameter coefficients and 95% and 68% confidence
intervals).
For farms established on lands that were once fallow, all variables
were statistically significant (p-value < 0.05), except basal area as an
explanatory variable for the proportion of LS trees (p = 0.07) and
fallow age as explanatory variable for the proportion of CC trees
(p = 0.06). For both the proportion of CC and LS trees, shade tree
abundance and the proportion of Inga spp. trees were negative
drivers, while basal area was a positive driver for the proportion of
trees of CC. The magnitude of the effect of the proportion of Inga spp.
trees as an explanatory variable was much larger among farms
established on fallow than those established on forest. Fallow age
was a positive predictor for the proportion of LS trees (p = 0.02). For
coffee farms established in forests, the proportion of Inga spp. trees
had statistically significant a negative impact on both the
proportion of CC and LS trees; the other variables did not have
statistically significant effects (see Fig.1 for standardized parameter
coefficients and 95% and 68% confidence intervals).
We found that the proportion of CC trees in forests (mean =
25.3%, range: 10–44%) is higher than in farms established on forest
(mean = 10.7%, range: 0-33%; p-value < 0.01). The difference
between the proportions of LS trees in forests (mean = 18.2% range