Our seven mapped forest canopy traits proved
to be largely uncorrelated (Fig. 1 and fig. S9),
indicating the multidimensionality of derived
biogeographic information. These trait maps
make no a priori assumptions about forest type
or composition, instead expressing top-down,
remotely sensed variation in functional composition. Environmental modeling subsequently indicated that geologic substrate and elevation
[and thus temperature (23)] are the dominant
controls on canopy trait composition (Fig. 1).
Secondary drivers include topographic slope,
local hydrology, and solar insolation.
We used 301 forest inventory plots in the
Peruvian Andes and Amazon to draw connections
between mapped canopy traits and measured
floristic composition and community turnover
(figs. S14 and S15 and tables S2 to S6). There is
a highly significant relationship between canopy
functional composition, computed by principal
components analysis of the individual trait maps
(Fig. 1), and canopy floristic composition, cal-
culated from field inventory data (coefficient of
determination R2 = 0.43, P < 0.001) (fig. S16).
Building on this finding, we used cluster analysis
to integrate the seven mapped canopy traits into
36 forest functional classes (FFCs), each repre-
senting a common suite of functional properties
among coexisting species (Fig. 2). The seven traits
accounted for 78% of variation among FFCs (fig.
S15), and the FFCs mirrored field-based changes
in functional traits and floristic composition.
To support our forest functional conservation
assessment, we applied a hierarchical clustering
algorithm to the 36 FFCs to generate six forest
functional groups (FFGs) (Fig. 2C and figs. S10 to
S13). These FFGs were recognizable against known
regional topo-edaphic and geomorphological
features (fig. S12). FFG-1 covers a part of the
southern Amazonian lowlands that is centered
on the Fitzcarrald Arch geologic feature, and it
is characterized by high canopy foliar N and
low defense compound investment. FFG-2 extends over portions of the northern lowlands
on very low-fertility clay substrates harboring
canopies with low foliar P and Ca and high
phenol- and lignin-based defense investments.
FFG-3 is found in the lowland Amazonian flood-plains in areas of high rock-derived cation deposition. FFG-4 is located in P-rich colluvial deposition
zones at the base of the Andes. FFG-5 is found in
anoxic swamp regions of the Pastaza-Maranon
foreland basin, where canopy foliar N and P are
low and defenses are high. Last, FFG-6 covers the
submontane to montane reaches of the Andes,
where LMA and leaf water content are high, and
foliar N and Ca are low in concentration.
Applying government deforestation data (fig.
S17) to the FFG map, we found that the highest
deforestation rates of up to 7000 ha yr−1 occurred
in FFG-4 (fig. S18), which contains five FFCs
found in the high-fertility colluvial deposition
region at the base of the Andes. We also found
that FFG-1, which includes six FFCs in the southern lowland Peruvian Amazon, has undergone a
500% increase in deforestation since 2010. Cattle
ranching and illegal gold mining are the main
drivers of these forest losses (24, 25). Additional
forest functional groups have been lost at an average 1000 ha yr−1 since 2000.
Government land allocation data indicate that
petroleum oil exploration and logging together
threaten 33% of FFG-1 and 27% of FFG-3 in the
southern and northern lowlands, respectively
(Fig. 3, Table 1, and table S7). In contrast, 28 to
32% of FFG-2 and FFG-4 are threatened primarily by petroleum activities. Andean forests
in FFG-6 are proportionally less threatened (12%)
by these activities. However, these threat tallies
do not include the explosive rate of deforestation
associated with gold mining and oil palm expansion under way in FFG-1 and FFG-3, respectively.
Although these sources of loss contribute to the
deforestation results, government land-use allocation data do not exist for these activities to
support threat analyses.
Between 32 and 46% of each mapped FFG is
currently protected (Fig. 3, Table 1, and table S7),
of which about two-thirds are under government
Fig. 1. Seven forest canopy traits mapped throughout the Peruvian Andes-to-Amazon region
by using airborne imaging spectroscopy and modeling. The graph indicates the attribution of environmental factors and the total variance explained for each canopy trait mapped with airborne imaging
spectroscopy. LMA, leaf mass per unit area; hydro, hydrology; insol, solar insolation.