Second, a careful parameterization is needed to ensure bet-
ter image classification accuracy by random forests. At least a
moderate number of trees (several dozen) should be used for
random forests in order to generate stable overall classifica-
tion accuracy. However, large tree numbers (such as hundreds
or thousands) are not recommended here, since they would
not help further improve the classification accuracy after
random forests become stable; instead, the possible benefit
of using more trees could be overshadowed with a higher
computational cost, particularly when very large image da-
tasets are being processed. Although a small feature number
(i.e., square root of the entire feature number) may favor the
classifier’s performance, a relatively large feature number
coupled with a moderate number of trees should be used to
ensure better overall and categorical classification accuracies
by random forests.
Last, the classification accuracy of random forests can
be greatly affected by the level of spectral complexity with
respect to specific land cover classes. Spectrally homogenous
categories tend to be classified with much higher thematic
map accuracies, while heterogeneous classes tend to have
relatively lower accuracies. A more careful parameterization
is needed to empower random forests in classifying some
spectrally complex land cover classes.
Although our study has some major merits, there are
several potential limitations. Like many other comparative
studies, this study is based on a single data set, a moderate
resolution multispectral image from the Landsat
OLI
sensor,
which might limit the extrapolation of the sensitivity analy-
sis, reported here. Further research may need to consider
different types of data with various quality and completeness
and test the sensitivity of random forests in classifying finer
levels of land-cover types in different environmental settings
varying in landscape complexity.
Acknowledgments
The research was partially supported by the Florida State
University through a Multidiscipline Support Grant and Natu-
ral Science Foundation of China through the grant “A Study
on Environmental Impacts of Urban Landscape Changes and
Optimized Ecological Modeling” (
ID
41230633). Comments
from three anonymous reviewers helped improve the scholar-
ly quality of this paper. The authors also would like to thank
Dr. Russell G. Congalton, Editor-in-Chief of
PE&RS
, for his
critical comments and valuable help.
References
Adam, E., O. Mutanga, J. Odindi, and E.M. Abdel-Rahman, 2014.
Land-use/cover classification in a heterogeneous coastal
landscape using RapidEye imagery: Evaluating the performance
of random forest and support vector machines classifiers,
International Journal of Remote Sensing
, 35(10):3440–3458.
Amit, Y., and D. Geman, 1997. Shape quantization and recognition
with randomized trees,
Neural Computation
, 9(7):1545–1588.
Anderson, J.R., E.E. Hardy, J.T. Roach, and R.E. Witmer, 1976.
A Land
Use and Land Cover Classification System for Use with Remote
Sensor Data
, USGS Professional Paper 964, Sioux Falls, South
Dakota.
Atlanta Regional Commission (ARC), 2014.URL:
atlantaregional.com
, Atlanta, Georgia (last date accessed: 25
April 2016).
Biau, G., 2012. Analysis of a random forests model,
The Journal of
Machine Learning Research
, 13:1063–1095.
Breiman, L., 1996. Bagging predictors,
Machine Learning
, 26(2):123–140.
Breiman, L., 1999. Using adaptive bagging to debias regressions,
Technical Report 547
, Statistics Department, University of
California-Berkeley.
Breiman, L., 2001. Random forests,
Machine Learning
, 45(1): 5–32.
Breiman, L., and A. Cutler, 2004. Random forest. URL:
stat.berkeley.edu/~breiman/RandomForests/
, University of
California-Berkeley, California (last date accessed: 25 April
2016).
Chan, J.C.W., and D. Paelinckx, 2008. Evaluation of random forest
and adaboost tree-based ensemble classification and spectral
band selection for ecotope mapping using airborne hyperspectral
imagery,
Remote Sensing of Environment
, 112(6):2999–3011.
Clark, M.L., T.M. Aide, and G. Riner, 2012. Land change for all
municipalities in Latin America and the Caribbean assessed
from 250-m MODIS imagery (2001-2010),
Remote Sensing of
Environment
, 126:84–103.
Congalton, R., 1991. A review of assessing the accuracy of
classification of remotely sensed data,
Remote Sensing of
Environment
, 37:35–46.
Congalton, R.G., and K. Green, 2009.
Assessing the Accuracy of
Remotely Sensed Data: Principles and Practices
, Second edition,
CRC press, 183 p.
Criminisi, A., J. Shotton, and E. Konukoglu, 2011. Decision forests for
classification, regression, density estimation, manifold learning
and semi-supervised learning,
Microsoft Research Cambridge,
Technical Report MSRTR-2011-114
, 5(6):12.
DeFries, R.S., and J.C.W. Chan, 2000. Multiple criteria for evaluating
machine learning algorithms for land cover classification from
satellite data,
Remote Sensing of Environment
, 74(3):503–515.
Dietterich, T.G., 1998. An experimental comparison of three methods
for constructing ensembles of decision trees: Bagging, Boosting,
and Randomization,
Machine Learning
, 40:139–158.
Dietterich, T.G., 2000. Ensemble methods in machine learning,
Multiple Classifier Systems
, 1857:1–15.
ENVI, 2009.
Atmospheric Correction Module: QUAC and FLAASH
User’s Guide
, URL:
/
envi/Flaash_Module.pdf
(last date accessed: 25 April 2016).
Feller, W., 1968.
An Introduction to Probability Theory and Its
Application
, Third edition, Vol. 1, Wiley, New York, 509 p.
Ghimire, B., J. Rogan, V.R. Galiano, P. Panday, and N. Neeti, 2012. An
evaluation of bagging, boosting, and random forests for land-
cover classification in Cape Cod, Massachusetts, USA,
GIScience
and Remote Sensing
, 49(5):623–643.
Ghosh, A., R. Sharma, and P.K. Joshi, 2014. Random forest
classification of urban landscape using Landsat archive and
ancillary data: Combining seasonal maps with decision level
fusion,
Applied Geography
, 48:31–41.
Gislason, P.O., J.A. Benediktsson, and J.R. Sveinsson, 2006. Random
forests for land cover classification,
Pattern Recognition Letters
,
27(4):294–300.
Grinand, C., F. Rakotomalala, V. Gond, R. Vaudry, M. Bernoux, and
G. Vieilledent, 2013.Estimating deforestation in tropical humid
and dry forests in Madagascar from 2000 to 2010 using multi-
date Landsat satellite images and the random forests classifier,
Remote Sensing of Environment
, 139:68–80.
Guo, L., N. Chehata, C. Mallet, and S. Boukir, 2011. Relevance
of airborne lidar and multispectral image data for urban
scene classification using Random Forests,
ISPRS Journal of
Photogrammetry and Remote Sensing
, 66(1):56–66.
Hall, M., .E. Frank, G. Holmes, B. Pfahringer, P. Reutemann, and
I.H. Witten,2009. The WEKA data mining software: An update;
SIGKDD Explorations
, Volume 11, Issue 1.URL:
.
cs.waikato.ac.nz/ml/weka/index.html
, Machine Learning Group
at the University of Waikato, Hamilton, New Zealand (last data
accessed: 25 April 2016).
Ham, J., Y.C. Chen, M.M. Crawford, and J. Ghosh, 2005.
Investigation of the random forest framework for classification
of hyperspectral data,
IEEE Transactions on Geoscience and
Remote Sensing
, 43(3):492–501.
Hamza, M., and D. Larocque, 2005. An empirical comparison of
ensemble methods based on classification trees,
Journal of
Statistical Computation and Simulation
, 75(8):629–643.
416
June 2016
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
