A Novel Method for Separating Woody and
Herbaceous Time Series
Qiang Zhou, Shuguang Liu, and Michael J Hill
Abstract
Mapping the spatial distribution of woody and herbaceous
vegetation in high temporal resolution in savannas would be
beneficial for modeling interrelationships between trees and
grasses, and monitoring fuel loads and biomass for livestock.
In this study, we developed a frequency decomposition meth-
od to separate woody and herbaceous vegetation components
using Normalized Difference Vegetation Index (
NDVI
) time
series. The results were validated using fractional cover data
derived from high-resolution images. The validation revealed
a close relationship between our decomposed
NDVI
and cor-
responding fractional cover (R
2
= 0.55 and 0.64 for woody
and herbaceous components, respectively). We examined the
spatial and temporal patterns of the decomposed
NDVI
, where
woody and herbaceous
NDVI
showed different responses to
precipitation. The methods proposed in this study can be
used to separate the woody and herbaceous
NDVI
time series
as an alternative approach for monitoring woody and herba-
ceous vegetation interrelationships related to climatic drivers.
Introduction
Savannas provide essential ecosystem services to environ-
ments and societies, including supporting the subsistence
economy, climate regulation, biodiversity, water balance, and
hydrodynamics (Seghieri
et al.
1995; Guerschman
et al.
2009).
On the other hand, savannas are very dynamic ecosystems
under the influences of various forces, incl
variability and change, fire, herbivory, and
(Verbesselt
et al.
2006; Verbesselt
et al.
200
2010; Sankaran
et al.
2005; Bucini and Han
al.
2011; Gessner
et al.
2013). The mosaics of woody canopy
and herbaceous cover of savannas form a distinct vegetation
structure and, separately and together, they provide various
essential functions to the ecosystem (e.g., carbon sequestra-
tion, wildfire ignition, and biodiversity conservation) to
human livelihoods (e.g., grazing, fuelwood, and agriculture)
(Gessner
et al.
2013). Thus, mapping the spatial and tempo-
ral changes of the relative abundance and phenology of the
woody and herbaceous components in highly dynamic savan-
nas is critical for further understanding the impacts of climate
change, land use, and disturbances on ecosystem structures
and services including wildlife habitats, productivity, and
biogeochemical cycles (Kahiu and Hanan 2018). In addition,
understanding the dynamics of these structural variables
provides the basis for effective management decisions over
extensive and heterogeneous savanna regions (Knoop and
Walker 1985; Bucini and Hanan 2007; Sankaran
et al.
2008).
Separating the woody and herbaceous components over
large areas remains a challenge due to the similarity of the
seasonal variation between the two components. Remote
sensing technology has been the primary tool to investigate
phenological dynamics of evergreen (or semideciduous)
woody and seasonal herbaceous components (Lu
et al.
2003;
Donohue
et al.
2009; Helman
et al.
2015; Zhou
et al.
2016).
Various approaches have been proposed to decompose time
series of satellite-derived vegetation signals. For example, Lu
et al.
(2003) separated 10-day evergreen woody and seasonal
herbaceous time series in Australia using the Normalized Dif-
ference Vegetation Index (
NDVI
) from the Advanced Very High-
Resolution Radiometer (AVHRR) dataset, by applying empiri-
cal relationships between evergreen woody and herbaceous
seasonal variations. Helman
et al.
(2015) used the dry season
NDVI
to extract the evergreen woody component and then
used the wet season
NDVI
to estimate the seasonal herbaceous
component from the Moderate Resolution Imaging Spectrora-
diometer (
MODIS
). Both studies used a constant
NDVI
to reduce
the soil background influence. However, these methods
cannot distinguish between woody deciduous and seasonal
etation components. Woody vegetation with
of deciduousness is prevalent in southern
as such as the Mopane woodlands and Kalahai
woodlands. Moreover, the diverse species
of woody vegetation exhibit wide variation in intra-seasonal
phenology and canopy morphology and density, which makes
the separation from the herbaceous signal more challenging.
A range of techniques have been applied to extract vegeta-
tion phenology dynamics from remote sensing time series.
Seasonal and Trend decomposition using Loess (
STL
) (seasonal
decomposition of time series by LOcally wEighted Scatterplot
Smoothing (
LOESS
)) was used to separate seasonal component
from multi-year trends based on the
LOESS
methods (Cleveland
et al.
1990). The harmonic analysis of
NDVI
time series (Me-
nenti
et al.
1993) was used to smooth data (Jun
et al.
2004),
correlate responses with climatic variables (Lhermitte
et al.
2008; Roerink
et al.
2003; Bradley
et al.
2011), extract vegeta-
tion phenology information (Leinenkugel
et al.
2013), and clas-
sify vegetation types with distinct phenology (Jakubauskas
et
al.
2002; Yui
et al.
2004; Geerken
et al.
2005). In the harmonic
analysis, the first two harmonics were used mostly in vegeta-
tion phenology studies: The amplitude of the first harmonic
was associated with the dominance of annual vegetation
species, the amplitude of the second harmonic indicated the
Qiang Zhou and Michael J Hill are with the Department of
Earth System Science and Policy, University of North Dakota,
4149 University Avenue, Grand Forks, ND, 58202.
Shuguang Liu is with the National Engineering Laboratory
of Forest Ecology and Applied Technology in South China,
and Faculty of Life Science and Technology, Central South
University of Forestry and Technology, Changsha 410004,
Hunan, China.
Michael J Hill is also with the School of the Environment,
Flinders University, Bedford Park, SA, 5042, Australia.
Qiang Zhou is also with the U.S. Geological Survey Earth
Resources Observation and Science (EROS) Center, 47914
252nd Street, Sioux Falls, SD 57198.
Photogrammetric Engineering & Remote Sensing
Vol. 85, No. 7, July 2019, pp. 509–520.
0099-1112/19/509–520
© 2019 American Society for Photogrammetry
and Remote Sensing
doi: 10.14358/PERS.85.7.509
PHOTOGRAMMETRIC ENGINEERING & REMOTE SENSING
July 2019
509
