Download Supplementary

Remote Sens. 2016, 8, 161, doi:10.3390/rs8020161
S1 of S8
Supplementary Materials: Tree Species Abundance
Predictions in a Tropical Agricultural Landscape
with a Supervised Classification Model and
Imbalanced Data
Sarah J. Graves, Gregory P. Asner, Roberta E. Martin, Christopher B. Anderson,
Matthew S. Colgan, Leila Kalantari and Stephanie A. Bohlman
Figure S1. Rank abundance curves for 8 species classification studies. The number of samples for each
study was summarized from published data on the number of samples used to train the classification
model. Feret and Asner 2012 [7] sampled 50 pixels per class to train the model. Studies are listed in
Table 1 of the main text.
Figure S2. Overall accuracy of multiple SVM classifications with changes in the number of classified
species. Overall accuracy as measured by the number of species (a) and as measured by the minimum
crown number across all species (b). Points represent the model-level overall accuracy for 8 separate
SVM classifications.
Remote Sens. 2016, 8, 161, doi:10.3390/rs8020161
S2 of S8
Figure S3. Linear regression between the number of pixels per species and the prediction bias for 10
model iterations. The shaded region is the 95% confidence interval around the mean.
Table S1. Summary of datasets for SVM model variations to test strategies to account for data
imbalance.
Field Sample
Full
Even
Imbalanced
Weighted
Number of
Species
20
20
20
20
Number of Crowns
Per Species
20–116
20
9–52
9–52
Total Crowns
Weighting
890
400
400
400
No
No
No
Yes
Remote Sens. 2016, 8, 161, doi:10.3390/rs8020161
S3 of S8
Figure S4. Subset of segmentation and shared edge calculations of the lidar canopy height model. The
amount of shared polygon edge was calculated. Crowns with 65% or less of shared crown edge were
considered to be agricultural trees. Field delineated crowns are shown in black outlines. Polygons are
colored by the percent of shared edge. The SVM model was applied to these polygons to produce a
landscape species prediction map.
Figure S5. Median accuracy confusion matrix (accuracy measured with overall accuracy) selected
from 30 model iterations. The number and colors correspond to the total number of crowns. Correct
classifications are seen on the diagonal and misclassifications are the off-diagonal. This confusion
matrix was used to calculate the error-adjusted area of predicted species areas.
Remote Sens. 2016, 8, 161, doi:10.3390/rs8020161
S4 of S8
Figure S6. Plot of the species-class size and the prediction bias for 15 African savanna species-classes.
Data is from Colgan et al. [4]. Line shows a linear model between class size and prediction bias with
a 95% confidence interval around the mean.
References
1.
2.
3.
4.
5.
6.
7.
8.
Cho, M.A.; Mathieu, R.; Asner, G.P.; Naidoo, L.; Aardt, J. Van; Ramoelo, A.; Debba, P.; Wessels, K.; Main, R.;
Smit, I.P. J.; Erasmus, B. Mapping tree species composition in South African savannas using an integrated
airborne spectral and LiDAR system. Remote Sens. Environ. 2012, 125, 214–226.
Clark, M.L.; Roberts, D.A.; Clark, D.B. Hyperspectral discrimination of tropical rain forest tree species at
leaf to crown scales. Remote Sens. Environ. 2005, 96, 375–398.
Clark, M.L.; Roberts, D.A. Species-level differences in hyperspectral metrics among tropical rainforest trees
as determined by a tree-based classifier. Remote Sens. 2012, 4, 1820–1855.
Colgan, M.S.; Baldeck, C.A.; Féret, J.-B.; Asner, G.P. Mapping savanna tree species at ecosystem scales using
support vector machine classification and BRDF correction on airborne hyperspectral and LiDAR data.
Remote Sens. 2012, 4, 3462–3480.
Dalponte, M.; Bruzzone, L.; Gianelle, D. Tree species classification in the Southern Alps based on the fusion
of very high geometrical resolution multispectral/hyperspectral images and LiDAR data. Remote Sens.
Environ. 2012, 123, 258–270.
Féret, J.B.; Asner, G.P. Tree species discrimination in tropical forests using airborne imaging spectroscopy.
IEEE Trans. Geosci. Remote Sens. 2012, 51, 1–12.
Féret, J.B.; Asner, G.P. Semi-supervised methods to identify individual crowns of lowland tropical canopy
species using imaging spectroscopy and LiDAR. Remote Sens. 2012, 4, 2457–2476.
Jones, T.G.; Coops, N.C.; Sharma, T. Assessing the utility of airborne hyperspectral and LiDAR data for
species distribution mapping in the coastal Pacific Northwest, Canada. Remote Sens. Environ. 2010, 114,
2841–2852.
© 2016 by the authors; licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons by Attribution
(CC-BY) license (http://creativecommons.org/licenses/by/4.0/).