Application of the Savitzky-Golay Filter to
Land Cover Classification Using Temporal
MODIS Vegetation Indices
So-Ra Kim, Anup K. Prasad, Hesham El-Askary, Woo-Kyun Lee, Doo-Ahn Kwak, Seung-Ho Lee, and Menas Kafatos
Abstract
In this study, the Savitzky-Golay filter was applied to smooth
observed unnatural variations in the temporal profiles of
the Normalized Difference Vegetation Index (
NDVI
) and the
Enhanced Vegetation Index (
EVI
) time series from the MOD-
erate Resolution Imaging Spectroradiometer (
MODIS
). We
computed two sets of land cover classifications based on
the
NDVI
and EVI time series before and after applying the
Savitzky-Golay filter. The resulting classification from the
filtered versions of the vegetation indices showed a substan-
tial improvement in accuracy when compared to the classi-
fications from the unfiltered versions. The classification by
the
EVIsg
had the highest K
(0.72) for all classes compared to
those of the
EVI
(0.67),
NDVI
(0.63), and
NDVIsg
(0.62). There-
fore, we conclude that the
EVIsg
is best suited for land cover
classification compared to the other data sets in this study.
Introduction
Land cover data provide key environmental information for
many scientific and policy applications, including resource
management. They play a pivotal role in evaluating ecosys-
tem processes and human activities (Cihlar, 2000; Homer
et
al.
, 2004; Marland
et al.
, 2003). Remote sensing has recently
become an important tool for preparing land use maps to
support a wide range of environmental research and planning
activities. Moreover, classification of spectral images has be-
come a particularly useful application for deriving land cover
maps (Heinl
et al.
, 2009; Herold
et al.
, 2008; Xia
et al.
, 2008).
Conventional procedures for land cover assessment generally
use one or more image sources such as those obtained from
Landsat, the Advanced Spaceborne Thermal Emission and Re-
flection Radiometer (
ASTER
), the Advanced Very High Resolu-
tion Radiometer (
AVHRR
), and other remotely sensed data (Bakr
et al.
, 2010; Samaniego and Schulz, 2009). However, such
remotely sensed data have some limitations such as spatial
and temporal resolutions, availability of data, and overall cost
considerations (Lunetta
et al
., 2006; Wardlow
et al
., 2007).
The MODerate Resolution Imaging Spectroradiometer
(
MODIS
) onboard the Terra and Aqua satellites, offers an
opportunity for large-scale land cover characterization.
MODIS
provides high-quality regional as well as global coverage with
high temporal (daily, 8-day, 16-day, and monthly compos-
ites) and intermediate spatial (250 m) resolution (Justice and
Townshend, 2002; Knight and Lunetta, 2006; Lunetta
et al
.,
2006). Wessels
et al.
(2004) found that typical land cover
classes such as agriculture regions, deciduous/evergreen for-
ests, and grassland areas could be successfully mapped using
MODIS
250 m data. Dash
et al.
(2007) used two operational
Medium Resolution Imaging Spectrometer (
MERIS
)-derived
vegetation indices (at 300 m spatial resolution), forming the
MERIS
global vegetation index (
MGVI
) and the
MERIS
terrestrial
chlorophyll index (
MTCI
), for land cover classification and
mapping, and achieved good accuracy (73.2 percent). They
also found a high degree of inter-class separability that varied
seasonally, resulting in higher accuracy in certain periods.
Knight and Lunetta (2003) suggested that the minimum
mapping unit should be close to the native resolution of the
sensor, since the resampling process to create coarser resolu-
tion data also increases the associated errors.
Moreover, time-series signatures derived from repeated
sampling of the study area throughout the year based on
satellite observations are increasingly being used by agricul-
tural scientists, ecologists, and environmentalists (Friedl
et
al
., 2002; Doraiswamy
et al
., 2005; Knight and Lunetta, 2006;
Oliveira
et al.
, 2010; Sulla-Menashe
et al
., 2011). In some cas-
es, the temporal signature proved to be more important than
spectral information from multiple bands or image texture for
identification of specific forest types, such as semi-deciduous
Atlantic forest (Carvalho
et al
., 2004). National Oceanic and
Atmospheric Administration (
NOAA
)/
AVHRR
time-series data
So-Ra Kim and Woo-Kyun Lee are with Korea University, Di-
vision of Environmental Science and Ecological Engineering,
Korea University, Seoul 136-713, Republic of Korea (leewk@
Korea.ac.kr).
Anup K. Prasad and Menas Kafatos are with Chapman Uni-
versity, School of Earth and Environmental Sciences, Schmid
College of Science, Chapman University, Orange, CA 92866,
and Center of Excellence in Earth Observing, Chapman Uni-
versity, Orange, CA, 92866.
Hesham El-Askary is with Chapman University, Alexandria
University, School of Earth and Environmental Sciences,
Schmid College of Science, Chapman University, Orange CA
92866; Center of Excellence in Earth Observing, Chapman
University, Orange, CA, 92866; and Department of Environ-
mental Sciences, Faculty of Science, Alexandria University,
Moharem Bek, Alexandria, 21522, Egypt.
Doo-Ahn Kwak is with Korea University,
GIS
-RS Center for
Environmental Resources, Korea University, Seoul 136-713,
Republic of Korea.
Seung-Ho Lee is with Korea Forest Research Institute,
Division of Forest Economics & Management, Korea Forest
Research Institute, Seoul, 130-712, Republic of Korea.
Photogrammetric Engineering & Remote Sensing
Vol. 80, No. 7, July 2014, pp. 000–000.
0099-1112/14/8007–000
© 2014 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.80.7.000
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2014
675
