(tables S4 and S5). Factor 1 explains 23% of the
variance and contains MAP, vegetation cover,
erosion rate (factor loadings of 0.9, 0.6, and
0.5, respectively). We interpreted factor 1 as
representing the combined interactions of
vegetation, precipitation, and erosion. Factor 2
(20% of the variance) contains vegetation cover,
MAT (factor loading >0.6), and acid volcanic
rocks (“va”in table S4, factor loading 0.7). We
interpreted factor 2 as representing interac-
tions between vegetation, temperature, and sub-
strate composition. Both the factor and Pearson
statistical analyses showed no correlation or
covariance of erosion rates with quartz content
or rock type (tables S4 to S6 and figs. S1 and S6).
We report our results for the Pearson cor-
relation coefficients (R) averaged overncatch-
ments within 2° latitudinal increments (Fig. 2,
D and E). The PearsonRvalues are a metric
for the degree of linear dependence between
individual parameters and the erosion rate (e.g.,
fig. S9). We chose the 2° increments as the
minimum spatial scale over which a sufficient
number of catchments (n> 5) are available
for analysis (e.g., fig. S9 and table S5). The cor-relation coefficients are reported for a range of
values in each bin using a Monte Carlo analysis
of the 2srange of uncertainties in a catchment
(Figs. 1 and 2). We classified the absolute values
of the correlation coefficients into regions with
very weak (0.00 to 0.19), weak (0.20 to 0.39),
moderate (0.40 to 0.59), strong (0.60 to 0.79),
and very strong (0.80 to 1.0)Rvalues ( 23 ).
We found that precipitation had a clear lati-
tudinal gradient, but the correlation coefficients
between precipitation and erosion rate do
not follow this gradient. Instead, they oscillate
between very weak to moderate (Fig. 2D). Simi-
larly, no clear systematic latitudinal variation
in the correlation coefficients existed between
slope and erosion rates (except between 6°S
and 12°S). By contrast, vegetation-erosion cor-
relations show a pattern of latitudinal variations
to the north and south of the arid region (region
A, Fig. 2E, 18°S to 32°S). More specifically, in
the arid (<100 mm/year) and sparsely vegetated
region A (vegetation cover <20%), the vegetation-
erosion correlation indicated a very weak to
moderate negative relationship. To the north
and south of region A, the correlation betweenlatitudinal variations in erosion rates along this
gradient.
We measured cosmogenic radionuclide con-
centrations of 10 Be ( 12 ) from 12 samples and
combined this data with 74 published samples
from Peru and Chile (Fig. 1, A and B, and tables
S1 and S2). The 86 catchments are adjacent
to a similar tectonic plate boundary with 13
(broadly defined) catchment lithologies, includ-
ing Oligo-Miocene, Plio-Pleistocene volcano-
clastic deposits and ignimbrites, Jurassic and
Cretaceous sedimentary rocks, Paleozoic and
Cretaceous granodiorites, and Precambrian
gneiss ( 21 ). The total lithological-weighted
quartz content for each catchment varies be-
tween 15 to 49% (fig. S2). We used 10 Be concen-
trations to recalculate erosion rates using the
same sea level high-latitude production rates
and production-rate scaling ( 22 ) (fig. S1A). We
determined from MODIS, TRMM, CHELSA,
WolrdClim, SRTM, and GLiM datasets (Figs. 1C
and 2 and fig. S3) the 2 s range in vegetation
cover, mean annual precipitation (MAP) and
temperature (MAT), mean solar radiation
(MSR), catchment-averaged slope, local relief,
and lithologic quartz content for each catch-
ment. We found no large differences between
the TRMM, CHELSA, and WorldClim (MAT
and MAP values) datasets ( 23 ) (tables S1 to S6).
The catchment-averaged erosion rates var-
ied between 1.4 and 150 m per million years
(m/Myr) (solid line, Fig. 1B). Starting in the
north (6°S to 12°S), erosion rates displayed
increasing values between 0 and 150 m/Myr.
From 12°S to 20°S, the erosion rates decreased
(150 to 0 m/Myr). The lowest erosion rates were
located between 20°S to 30°S (0 to 50 m/Myr).
In the south (30°S to 36°S), erosion rates
ranged between 0 to 140 m/Myr and showed
increasing values from 30°S to 33.5°S and de-
creasing values from 33.5°S to 36°S. In general,
the quartz content for each catchment shows
no latitudinally dependent variation (fig. S2C).
Vegetation cover and MAP were the highest in
the north from 6°S to 10°S (50 to 85% and 200
to 700 mm/year, respectively; solid lines in
Fig. 2, A and B). Vegetation cover and MAP then
decreased to a minimum (5% and <50 mm/
year) at the latitudes of the Atacama Desert
(20°S to 30°S). Further south (30°S to 36°S),
the vegetation cover and MAP increased to a
southern maximum (40% and ~700 mm/year).
Catchment-averaged slopes had increasing
values up to 30° from 6°S to 12°S. Slopes
gradually decreased toward the south (12°S to
20°S) and varied between 25° to 10 °. The lowest
slopes (5° to 10°) were situated between 20°S to
30°S and increased to the south (30°S to 36°S)
up to 28° (solid line, Fig. 2C).
Results from a multivariate factor analysis
indicated that catchment erosion rate, vegeta-
tion cover, slope, MAT, MAP, MSR, local relief,
quartz content, and lithology result in four
factors that can explain 62% of the variance
SCIENCE 20 MARCH 2020•VOL 367 ISSUE 6484^1359
Fig. 1. Topography, erosion rates, and vegetation types along the western Andean margin.(A) Topographic
map showing the catchment-averaged erosion rate sample locations of river sediments from the Andean Western
Cordillera. Black dots indicate published^10 Be concentrations. Red dots are new data presented in this study
(tables S1 and S2). (B) Calculated catchment-averaged erosion rates (m/Myr) with 1suncertainty versus latitude
(°S). The black line represents the three-point moving average. All catchment-averaged erosion rates were
calculated with the same procedure (materials and methods). (C) Percent vegetation type versus latitude
across a 100-km-wide latitudinal profile in the Andean Western Cordillera. Gray lines represent barren or sparsely
vegetated areas. Black lines denote shrublands. Green lines represent grasslands, and purple lines indicate
woody savannas. Blue lines represent mixed forests and orange lines, evergreen forest. Values are derived from
MODIS 2012 vegetation continuous field data.RESEARCH | REPORTS