The following are available online at
www.mdpi.com/2072-4292/10/2/199. Figure S1: Histograms showing (A) the fraction of 4 m CAO VSWIR pixels within delineated crowns that passed through prescreening criteria in each collection site in Sabah, and (B) the fraction of tree crowns at each site that had suitable pixels following prescreening. Figure S2: (a) Examples of crown reflectance spectra derived from the mean number of pixels at each site (
Table 1) suitable for chemometric analysis following prescreening within each collection location. Zoom images of spectra are provided for (b) visible and (c) shortwave infrared regions to reveal subtle features associated with varying chemicals. Figure S3: Mean (solid line), standard deviation (dashed lines), and total range (gray area) of the CAO visible-to-shortwave infrared (VSWIR) crown spectra that passed the prescreening step (see Figure S1) for a collection site (Danum Valley;
Table 1). Figure S4: Linear regression analysis between model and lab-measured chemical traits and leaf mass per area (LMA) for crowns used in the canopy calibration model (70%), shown in grey, and the crowns held out for testing (30%), shown in black. The dashed line is the 1:1 line. Regression parameters are given in Table S4. Subfigures show (a) leaf mass per area (LMA), (b) nitrogen, (c) leaf water, (d) chlorophyll ab, (e) nonstructural carbohydrates (NSC), (f) δ
13C, (g) phosphorous, (h) calcium, (i) potassium, (j) magnesium, (k) boron, (l) iron, (m) total carbon, (n) lignin, (o) cellulose, (p) tannins, and (q) phenols. Figure S5: Comparison of selected PLSR weighting coefficients from the study in Sabah, Malaysia (red lines) and Peru in the Neotropics (black lines). Table S1: Location (Latitude, Longitude) and forest type for 13 field locations spanning a range of elevation and substrate types across Sabah, Malaysia. Table S2: Mean and range of foliar chemical traits and leaf mass per area sampled from collection locations in Kinabatangan, Danum Valley, and Sepilok in western Sabah, Malaysia. Table S2 cont.: Mean and range in foliar chemical traits and leaf mass per area from collection locations along an elevation gradient on Mt. Kinabalu in Sabah, Malaysia. The upper portion lists traits from sedimentary and granitic substrates. The lower portion lists traits from ultramafic substrates. * nd—no data were collected at these sites for these traits. Table S3: Principle components analysis results for 14 and 17 foliar traits measured at all locations (
n crowns = 424; spectra = 3993) or in the lowlands only (
n crowns = 182; spectra = 2430). All components except the final one (shown in italics) were significant in explaining a percentage of the total variation (right column) among the traits (Bartlett’s Chi-square test;
p < 0.01). Table S4: Parameters from linear regression analysis between model and lab-measured chemical traits and leaf mass per area (LMA) for crowns used in the canopy model development (70%) and the crowns held out for testing (30%). All models were significant
p < 0.001, except phenols where the significance value was 0.02. Graphical depictions of results are shown in Figure S4. Table S5: Correlation values (top half of table) among foliar traits measured in 424 tree crowns in Sabah, Malaysia. Significance values are shown in italics in the lower portion of the table. Table S6: Correlation values (top half of table) among weighting coefficients for 14 foliar traits retrieved from airborne spectroscopic measurements of 424 crowns in Sabah, Malaysia. Significance values are shown in italics in the lower portion of the table. Code S1. The location for the python code (ensemble_plrs.py). The code, which draws on elements of the autoPLS package from R, was created to develop and manage results of multiple individual PLSR models in a more computationally efficient way. The code can be found in a public github repository at
https://github.com/pgbrodrick/ensemblePLSR.git. The code requires an input csv file of coupled chemistry and spectral data from multiple crowns, as well as a settings file. Example can be found in the repository.