ArticlePDF Available

Abstract and Figures

Land surface phenology (LSP) characterizes episodes of greening and browning of the vegetated land surface from remote sensing imagery. LSP is of interest for quantification and monitoring of crop yield, wildfire fuel accumulation, vegetation condition, ecosystem response and resilience to climate variability and change. Deriving LSP represents an effort for end users and existing global products may not accommodate conditions in Australia, a country with a dry climate and high rainfall variability. To fill this information gap we developed the Australian LSP Product in contribution to AusCover/Terrestrial Ecosystem Research Network (TERN). We describe the product's algorithm and information content consisting of metrics that characterize LSP greening and browning episodes of the vegetated land surface. Our product allows tracking LSP metrics over time and thereby quantifying inter- and intraannual variability across Australia. We demonstrate the metrics' response to ENSO-driven climate variability. Lastly, we discuss known limitations of the current product and future development plans.
Content may be subject to copyright.
Aspatiallyexplicitlandsurfacephenologydataproductforscience,
monitoring and natural resources management applications
Mark Broich
a
,
b
,
*
, Alfredo Huete
b
, Matt Paget
c
, Xuanlong Ma
b
, Mirela Tulbure
a
,
Natalia Restrepo Coupe
b
, Bradley Evans
d
, Jason Beringer
e
, Rakhesh Devadas
b
,
Kevin Davies
b
, Alex Held
c
a
School of Biological, Earth and Environmental Sciences, University of New South Wales, Sydney, NSW 2052, Australia
b
AusCover TERN Sydney Node, Plant Functional Biology and Climate Change Cluster, University of Technology, Sydney, NSW 2007, Australia
c
AusCover TERN, CSIRO Marine and Atmospheric Research, Pye Laboratory, ACT 2600, Australia
d
Department of Biological Sciences, Macquarie University, NSW 2109, Australia
e
School of Earth and Environment, University of Western Australia, WA 6009, Australia
article info
Article history:
Received 25 August 2014
Received in revised form
14 November 2014
Accepted 16 November 2014
Available online
Keywords:
Remote sensing
MODIS
AusCover TERN
Climate variability
Time-series
abstract
Land surface phenology (LSP) characterizes episodes of greening and browning of the vegetated land
surface from remote sensing imagery. LSP is of interest for quantication and monitoring of crop yield,
wildre fuel accumulation, vegetation condition, ecosystem response and resilience to climate variability
and change. Deriving LSP represents an effort for end users and existing global products may not
accommodate conditions in Australia, a country with a dry climate and high rainfall variability. To ll this
information gap we developed the Australian LSP Product in contribution to AusCover/Terrestrial
Ecosystem Research Network (TERN).
We describe the product's algorithm and information content consisting of metrics that characterize
LSP greening and browning episodes of the vegetated land surface. Our product allows tracking LSP
metrics over time and thereby quantifying inter- and intraannual variability across Australia. We
demonstrate the metrics' response to ENSO-driven climate variability. Lastly, we discuss known limi-
tations of the current product and future development plans.
©2014 Elsevier Ltd. All rights reserved.
1. Introduction
With the launch of the rst MODerate-resolution Imaging
Spectroradiometer (MODIS) instrument in late 1999, a new set of
data products for continuous land surface monitoring at global to
regional scale became available to the user community (Justice
et al., 2002). Pre-processed data products are distributed to the
user community free of charge via the internet (https://lpdaac.usgs.
gov/). One of these products, the vegetation index product
(MOD13) was designed to monitor vegetation greenness across
large areas (Huete et al., 2002; Solano et al., 2012). Despite its
valuable information content, users and practitioners from outside
of the remote sensing community often nd it time consuming to
derive information for their specic applications.
This work extracts value added information for a range of users
by characterizing land surface phenological episodes of greening
and browning from time series of MOD13 for Australia in the form
of LSP metrics. Land surface phenology (LSP) is dened as the
seasonal pattern of variation in vegetated land surfaces observed
from remote sensing(Friedl et al., 2006) as opposed to vegetation
phenology, which describes in-situ observations of life cycle events
of individual plants or species such as bud break, owering, or leaf
senescence (Henebry and de Beurs, 2013). LSP depicts spatio-
temporal patterns of, for example, the timing of vegetation
growth, senescence, and dormancy at seasonal and interannual
time scales in a spatially aggregated form (e.g. pixel size of metres
to km depending on the input data source) (Friedl et al., 2006).
LSP is a function of short and long term changes in climate and
other factors such as reoccurring res, anthropogenic management
(e.g. agriculture), land and soil degradation, biodiversity loss, plant
pests and diseases and invariant factors such as topography and
soils (Bajocco et al., 2010; Bisigato et al., 2013; IPCC, 2007; Mackey
*Corresponding author. School of Biological, Earth and Environmental Sciences,
University of New South Wales, Sydney, NSW 2052, Australia.
E-mail address: mark.broich@gmail.com (M. Broich).
Contents lists available at ScienceDirect
Environmental Modelling & Software
journal homepage: www.elsevier.com/locate/envsoft
http://dx.doi.org/10.1016/j.envsoft.2014.11.017
1364-8152/©2014 Elsevier Ltd. All rights reserved.
Environmental Modelling & Software 64 (2015) 191e204
et al., 2012; Myneni et al., 1997; Schwartz and Hanes, 2010; White
et al., 1996, 1997; Zhang et al., 2003; Zhang et al., 2012). As such, LSP
provides valuable information to scientists and practitioners
interested in vegetation and ecosystem response to climate vari-
ability such as drought and land use change. Specic applications of
LSP information include for example the quantication of crop
yields, wildre fuel accumulation, ecosystem resilience and health
and land surface modelling (Eklundh and J
onsson, 2010; Friedl
et al., 2006; Liang and Schwartz, 2009; Pe~
nuelas et al., 2009;
Schwartz, 2013).
The global MODIS Land Cover Dynamics Product (MCD12Q2
unofcially known as the global phenology product) (Gray, 2012;
Ganguly et al., 2010) was developed to characterize the LSP of
global ecosystems, such as the timing of peak greenness and the
length of the growing season. LSP information retrieval works best
for regions with well-dened growing seasons, such as the mid and
high latitudes of the Northern Hemisphere where LSP cycles
reoccur annually. The algorithm producing the MOD12Q2 Product
has been purposely designed to be conservative and not produce
results if data are missing during transition periods or when the
Enhanced Vegetation Index (EVI) amplitude is very low. Therefore,
retrieving LSP in arid, evergreen or cloudy environments represents
an ongoing challenge (Ganguly et al., 2010). Studies focussing on
water-limited environments have been based on algorithms
developed specically for continental (Broich et al., 2014) to
regional study areas (Brown and de Beurs, 2008; Walker et al.,
2014).
For Australia, Zhang et al. (2006) found no well-dened spatial
pattern of LSP for arid and semi-arid areas, using the MOD12Q2
global LSP product. The absence of clear patterns may be related to
climatic variability causing high LSP variability or indicate that a
product specically designed for Australia's conditions is required.
In this work we therefore develop a product to characterize LSP
greening and browning episodesspecically for Australia.
Australian ecosystems are subject to high interannual variability
in rainfall and large areas are therefore covered by grassland and
drought-adapted vegetation (Australian Bureau of Meteorology,
2014a;Khan, 2008; Nicholls, 1991; Nicholls et al., 1997). Woody
vegetation in Australia is predominantly sclerophyll evergreen
shrubs and trees that form an open or sparse canopy over a her-
baceous understory (Australian Government Department of
Sustainability Environment Water Populations and Communities,
2013; Donohue et al., 2009; Eamus et al., 2013). Less interannual
variability in LSP occurs over Australia's vast cropping areas where
LSP is primarily a function of management. The Australian land-
scapes are also subject to climatic variability induced by the El Ni ~
no
Southern Oscillation (ENSO), evident in the transition from the
droughts of the past decade (2001e2009) to the La Ni~
na-forced
oods across large areas of Eastern Australia in 2010 (Heberger,
2011).
This extensive rainfall led to a high regional vegetation pro-
duction in 2011 that has inuenced the global carbon cycle through
a large interannual variability in the global signal of CO
2
concen-
tration (Poulter et al., 2014). LSP products can help elucidate these
features. Other large-scale drivers of rainfall variability in Australia
include the Indian Ocean Dipole, the MaddeneJulian oscillation,
the Southern Annular Mode and the Inter-decadal Pacic Oscilla-
tion (Risbey et al., 2009).
The objective of this paper is to introduce the Australian Land
Surface Phenology Product to the science, monitoring and natural
resources management user communities.
Specically this paper aims to:
1) Quantify LSP episodes of greening and browning from MODIS
EVI time series across Australia, including the temporally vari-
able, small amplitude LSP episodes that may be expected over
such semi-arid and arid environments.
2) Describe the product's algorithm and the information content
consisting of metrics that characterize LSP episodes of greening
and browning of the vegetated land surface.
3) Demonstrate the LSP episodes metrics' response to ENSO-driven
climatic variability.
4) Discuss known limitations of the current product and future
development plans.
2. Methods
2.1. Charactering LSP episode patterns from remotely sensed vegetation index time
series
The majority of studies use time series of spectral vegetation indices derived
from optical imagery. These indices are based on the difference between the red
region of the electromagnetic spectrum where green vegetation strongly absorbs
and the near infrared region of the electromagnetic spectrum where green vege-
tation strongly reects (Henebry and de Beurs, 2013). LSP studies commonly use
per-pixel index time series such as the Normalized Difference Vegetation Index
(NDVI) or the Enhanced Vegetation Index (EVI) that measure changes in vegetation
Dataset and data availability
Name of the Dataset Australian Land Surface Phenology
Product Version 1
Creator Climate Change Cluster, University of Technology
Sydney, Australia
Contact Mark Broich, mark.broich@gmail.com
Type of the dataset Spatio-temporal dataset in geolocated
TIFF format
Input data MOD13C2 product EVI 0.05-degree 16 day input
data from 18 Feb 2000 to 22 Apr 2013.
Coverage and information content The Australian Land
Surface Phenology
Product Version 1
provides annual land
surface phenology
metrics for all of Australia
2000e2013.
List of output data Metrics for each land surface phenology
episode: The rst and second minimum,
peak, start and end points, length and
amplitude as well as the episode
integrated greenness between the
episode's start and end.
Coordinate system Geographic, WGS 84 eepsg 4326. The
pixel size is 0.05-degrees. Spatial extent:
9.5
!
Se45
!
S and 112
!
Ee154
!
E.
Release year 2013
Data Access The data product is staged and can be
downloaded free of charge from the AusCover
TERN webpage
1
:http://www.auscover.org.au.
Licenses Creative Common Attribution (CC-BY) 3.0: Users of
the Australian Land Surface Phenology Product
agree to cite this manuscript. A Digital Object
Identier (DOI) for the Australian Phenology
Product is pending.
1
The Australian Land Surface Phenology Product is scheduled to permanently
migrate to the Australian Research Data Storage Infrastructure (RDSI) that is funded
through the Australian Government's Super Science Initiative and sourced from the
Education Investment Fund (EIF).
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204192
greenness as input data (de Beurs and Henebry, 2010; Eklundh and J
onsson, 2010;
Liang and Schwartz, 2009; Reed et al., 2003; Tan et al., 2011; Vrieling et al., 2011;
Zhang et al., 2003).
EVI is a red, near infrared and blue reectance optimized index with additional
parameters to adjust for atmospheric contamination and soil background (Huete
et al., 2002) and has been found to correlate with sub-pixel leaf chlorophyll con-
tent and leaf area index (Huete et al., 2014). The EVI data from the MODIS in-
struments mounted on the EOS TERRA and AQUA satellite platforms (launched in
1999 and 2002, respectively) have been the primary data source for LSP character-
izations since 2000. LSP is often characterized by tting mathematical curves such as
double logistic models (Beck et al., 2006; Eklundh and J
onsson, 2010;Zhang et al.,
2003, 2006), asymmetric Gaussian models (Beck et al., 2006; Eklundh and
J
onsson, 2010) or Fourier and Wavelet transformations (Moody and Johnson,
2001) to vegetation index greenness time series.
A limiting factor in the application of existing techniques on the Australian
continent is that the majority of LSP algorithms have been developed to characterize
ecosystems in the mid and high latitudes of the Northern Hemisphere where LSP
cycles reoccur annually (de Beurs and Henebry, 2010; Eklundh and J
onsson, 2010).
Crops can also be expected to have an annual LSP cycle within a specic time
window. Yet, LSP greening and browning episodes in arid or semi-arid ecosystems
may not follow a predictable annual pattern and different algorithms may be needed
to characterize vegetation dynamics for such regions (Brown and de Beurs, 2008;
Tan et al., 2011). Advances of the global MODIS Land Cover Dynamics Product use
an algorithm that accommodates unevenly distributed growing seasons as they
occur for example in the deserts of the Southwest USA (Tan et al., 2011). Besides
temporal variability, the low seasonal amplitude of remotely sensed greenness in
semi-arid and arid ecosystems further complicates the characterization of LSP epi-
sodes (Ganguly et al., 2010; Ma et al., 2013; Walker et al., 2014, 2012). Based on the
MODIS Quality Assessment tools we found that LSP was not retrieved for more than
30e50% of the semi-arid and arid areas of Australia by the MODIS Land Cover Dy-
namics Product (MCD12Q2; Land Data Operational Product Evaluation, 2012)
possibly due to the low seasonal amplitude of remotely sensed greenness (Ganguly
et al., 2010).
Here we extend the concept of LSP to constitute an indeterminate period of
greening as indicated by EVI that may be expected over semi-arid and arid envi-
ronments. We therefore did not use the classical denition of a LSP period of
greening and browning as an annual or seasonal cycle of reoccurring events that
begins and ends within a calendar year, as commonly observed in the mid and high
latitudes of the Northern Hemisphere, where low winter temperatures typically halt
vegetation growth. We instead dened a LSP episodeas a period of EVI-measured
greening and browning that may occur at any time of the year, extend across the end
of a year, skip a year (not occur for one or multiple years) or occur more than once a
year. Multiple LSP episodes within a year can occur in the form of double cropping in
agricultural areas or be caused by aseasonal rain events in water-limited
environments.
Based on per-pixel greenness time series measured by MODIS EVI, we modelled
LSP episode curves and their key properties in the form of LSP episode metrics
including: the rst and second minimum point, peak, start and end of episode;
length of episode, and the amplitude of the episode. We also integrated EVI under
the curve between the start and end of the episode as a proxy of net primary pro-
ductivity (Ponce Campos et al., 2013; Zhang et al., 2013; see applicability restrictions
in the discussion section). The rst and second minimum point, peak, start and end
of episode points have time and magnitude dimensions. The concept of L SP curve
metrics is shown in Fig. 1 and the metrics are listed and dened in Table 1. By
tracking the derived LSP episode metrics over time, users can quantify the intra-and
interannual variability of the LSP episodes in time and space. This allows the analysis
of variability of LSP metrics that can then be related to climate variability, climate
change, and land use and management practices.
2.2. Model algorithm
To characterize LSP across Australia, we developed an internally consistent al-
gorithm, which was subsequently applied to every MODIS pixel from 2000 until
2013.
We generated the rst in a series of multi-resolution Australian Land Surface
Phenology Products, by applying the algorithm to the MODIS MOD13C2 Product
with a 0.05-degree spatial and 16-day temporal resolution (Huete et al., 1999).
LSP is also a function of spatial resolution (de Beurs et al., 20 09) and in this work
we used 0.05-degree data that attenuates noise that occurs at ner resolutions
(Solano et al., 2012).
2
We sourced input EVI data from 18 Feb 2000 to 22 Apr 2013
from the Land Processes Distributed Active ArchiveCenter (LP DAAC: https://lpdaac.
usgs.gov/). Filtering input data via pixel-level QA has been recommended for time-
series analysis to screen out data that is unsuitable and ensures that the quality of
the utilized data remains consistent and reliable (Roy et al., 2002). Therefore the rst
step was to screen the EVI greenness data using the quality assurance ags (QA ags)
provided with the MOD13C2 Product. We discarded observations with VI usefulness
>10(keeping the highest 2 of 11 qualitylevels), or Aerosol Quantity climatologyor
high, or Mixed Clouds present, or water in the Land/Water Flag.
Generally, the proportion of pixels discarded via QA ltering was low, as would
be expected for a study area with limited cloud cover and snow contamination. The
median percent of gap lled observations per-pixel across Australia was 0.7%, with
10 and 90 percent quartiles of 0.45% and 9.5%, respectively.
We then gap-lled the low quality obs ervations screened in the previous step
in per-pixel 2000 to 2013 EVI time series using cub ic spline interpolation
(Dougherty et al., 1989). As a last pre-processing step, we used a SavitzkyeGolay
smoothing lter (Savitzky and Golay, 1964) with a win dow size of 15 time steps and
a second order polynomial to reduce noise in the gap lled time series. We then
determined the locat ions of LSP episodes for each pixel's time se ries. We located
the episodes' posit ions by identifying mi nimum and peak point s of the gap lled
and smoothed time seri es using a moving window of nine time steps for both the
minimum and maximum p oints. A sequence of mini mumemaximumeminimum
points with a differenc e >0.01 EVI was identied as an episode of greening and
browning based on the statistics reported by the MODIS Land Validation Team
(MODIS land team 2014) and veried by visual assessment at our 36 test sites
(described in the next sec tion).
Using the identied points for each episode, we constrained the interval for the
mathematical function t. We tted a 7-parameter double logistic curve to each
episode in the per-pixel time series.
The equation of the 7-parameter double logistic curve is:
EVIðtÞ¼V
min
a
þV
max
&V
min
a
1þexp!
T
mida
&t
S
a
"&V
max
&V
min
b
1þexp!
T
midb
&t
S
b
"(1)
Where: V
min
a
is equal to the rst minimum point; V
min
b
is equal to the second
minimum point; V
max
is the peak point of the double logistic model, and S
a
and S
a
are
the time scale parameters on the ascending and descending side of the curve,
respectively. The parameters T
mid
a
and T
mid
b
are equal to ðV
min
a
þV
max
Þ=2 and
ðV
min
b
þV
max
Þ=2, respectively.
In a preliminary investigation we found that minimum points on either side of
LSP episodes can have different magnitude across the interior of Australia. Hence, we
chose the 7-parameter double logistic model as it accommodated differences in
minimum point magnitudes.
We tted a number of curves equal to the number of identied LSP episodes for a
given per-pixel trajectory to accommodate, for example, double cropping or fallow
elds within a given year. To identify the start and end of episode points we
employed a generic uniform amplitude threshold. We dened the start of the
episode as the point when the pixel's greenness reached 20% of its peak value and
the end of the episode when the greenness had decreased 80% from its peak, con-
forming with the relative threshold used in other studies (Delbart et al., 2005;
Eklundh and J
onsson, 2010; Jones et al., 2011; Tan et al., 2011). In a nal step we
quantied the length of the episode, amplitude of the episode and episode inte-
grated greenness.
2.3. Sample and test sites for algorithm evaluation
We evaluated the algorithm'sability to capture the spatio-temporal variabilityof
LSP episodes across the country using a set of 36 sites (single pixel time series)
Fig. 1. Conceptual illustration of the LSP curve metrics derived in this work. The
derived point metrics are: the rst and second minimum point (Min1 and Min2); the
peak point (Peak); the start and end of the episode point (Soe and Eoe) as 20% of the
difference between Peak and the respective Min point. The other metrics are the length
of the episode (Loe) as the number of days between Soe and Eoe, the amplitude of the
episode (Amp), and the episode integrated greenness (Eig) between the start and the
end of the episode points (the hatched area in the curve).
2
Implications of this spatial resolution are discussed in Section 4.2.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 193
distributed across Australia (Fig. 2). Twenty-one of the sites are the locations of eddy
covariance ux towers of the OzFlux network representing different land and
vegetation cover (http://www.ozux.org.au/). We selected 15 additional test sites to
represent climatic conditions, land and vegetation cover mostly in the interior of
Australia as well as for regions that are not covered by OzFlux sites. Table 2 provides
the site names, locations, major vegetation classes according to Australia's National
Vegetation Information System (NVIS), based on a combination of vegetation
structure and oristics (Department of the Environment and Water Resources,
2007), average annual rainfall amounts and rainfall variability.
To demonstrate the algorithm and provide an insight into the variability of
Australian LSP, we illustrate the EVI time series and LSP metrics of four example
0.05-degree pixels covering four OzFlux tower sites that represent different vege-
tation cover and climatic conditions: The Alice Springs ux tower represents semi-
arid mulga (Acacia) woodland of the centre of Australia. The Riggs Creek
site represents dry land agriculture (predominantly pasture) in the Southeast
of Australia. The Wallaby Creek site represents wet temperate Eucalypt forest in the
Southeast of Australia that burned in a re in 2009. The Howard Springs site rep-
resents wet tropical open woodland savanna of Northern Australia (Fig. 2).
We illustrate the steps of our algorithm using the highly variable EVI trajectory
of the Alice Springs ux site in Central Australia. Fig. 3A shows the EVI trajectory
after screening low quality observations, lling of observation gaps and smoothing
of the time series, and identication of the minimum and maximum points of the
trajectory to identify LSP episodes. Fig. 3B shows the result of a sequence of double
logistic curves tted to the agged minimum and maximum points to mathemati-
cally characterize the identied LSP episodes and the identied start and end of
episode points for each episode within the 14 year time series.
2.4. Model performance evaluation
According to the recent work by Bennett et al. (2013), the effective use of
environmental models for management and decision-making requires that con-
dence in model performance is established. To establish condence and document
limitations we characterized the performance of our LSP model using both con-
current comparison and direct value comparison methods following Bennett et al.
(2013).
Our modelling aim was to derive LSP episodes and episodes' metrics from
MODIS EVI time series. To evaluate our model's performance we used a set of
evaluation metrics in concurrent comparison at the 36 sites. We evaluated our
modelled curve t against the input EVI time series after screening out low quality
observations. We also used direct value comparison of the start and end of episode
timing derived from the EVI time series and start and end of episode timing derived
from gross primary productivity (GPP) time series at the Howard Springs eddy
covariance ux tower site (Beringer et al., 2014).
For concurrent comparison we calculated y-residual metrics of the LSP episodes
modelled via subsequent 7-parameter double logistic curves and the raw EVI time
series after screening out low quality observations for each of our 36 test sites across
the time series.
Specically, we calculated the mean error (bias), mean absoluteerror (MAE), and
root mean square error (RMSE). We calculated the bias to quantify if our model t
was on average overestimatingor underestimating EVI. Positive and negative errors
may cancel each other in the bias calculation. To prevent cancellation and topresent
a set of complementary evaluation metrics we also calculated the MAE and the
RMSE, which emphasize small and large deviation events, respectively. We analysed
these metrics in the context of the LSP amplitude at each site. We approximated LSP
amplitude independent of our algorithm as the inter-quartile range (IQR), the dif-
ference between the 90% and the 10% quartile. As a last step in our concurrent
comparison, we visually evaluated residual plots for each of the 36 test sites to
detect autocorrelation patterns in the residuals.
We also conducted a direct value comparison of our EVI LSP start and end of the
episodes timing with the start and end of episodes timing derived from GPP time
series derived from the Howard Springs ux site. Fluxes have been measured at the
Howard Springs site since 2001 (Beringer et al., 2014). As input we used daily GPP
modelled using the dynamic integrated gap lling and partitioning for Ozux
(DINGO) system (Beringer et al., 2007). We calculated the median GPP per 16-day
interval to match the temporal resolution of the EVI data, the input to our
phenology product. We applied the same algorithm developed for the EVI and
described in Section 2.2, to the GPP time series to derive start and end of episodes
timing of GPP episodes for the Howard Springs ux site.
3. Results
3.1. Illustration of EVI time series and LSP metrics of selected sites
Our model shows that Australia's LSP is highly variable both for
a specic site interannually and across the continent (Figs. 3 and 4).
Table 1
List of LSP curve metrics, their denition, use, and measurement units.
a
Name Abbreviation Denition juse Units
First minimum point Min1 The boundary between the current episode and the previous episode. j
Separation of previous and current episode
Time: Day of year in a given year Magnitude: EVI
Start of episode Soe The start of the greening phase jIdentication of the start of greening Time: Day of year in a given year Magnitude: EVI
Peak of episode Peak The episode peak and point between the greening and the browning phase j
Identication of the peak of greening
Time: Day of year in a given year Magnitude: EVI
End of episode Eoe The end of the browning phase jIdentication of the end of browning Time: Day of year in a given year Magnitude: EVI
Second minimum
point
Min2 The boundary between the current and the following episode j
Separation of current and following episode
Time: Day of year in a given year Magnitude: EVI
Length of episode Loe The time between the start and end of episode points j
Measuring the length of an episode
Number of days
Amplitude of episode Amp The difference between Peak and average Start and End of episode:
Peak e((Start of episode + End of episode)/2) jMeasuring the
amplitude of greenup
EVI
Episode integrated
greenness
Eig The integrated daily EVI under the curve between the start and
end of the episode time ja proxy of productivity
(see applicability restrictions in the discussion section)
Episode integrated EVI
a
Measurement units are time (day of year in a given year) and EVI (greenness). EVI (greenness) provides an approximation of sub-pixel chlorophyll content and leaf area
index (Huete et al., 2014).
Fig. 2. Locations of the 0.05-degree pixels over 21 OzFlux ux tower sites (black tri-
angles) and 15 additional sites (yellow solid circles) used for algorithm evaluation. The
background map shows the average peak EVI (average EVI value at the LSP episodes'
peak). The ve sites marked by large circles are analysed later in the text. Full site
names of abbreviations used here can be found in Table 2. (For interpretation of the
references to colour in this gure legend, the reader is referred to the web version of
this article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204194
For example, at the Alice Springs site, the timing, length, amplitude
and magnitude of LSP episodes vary interannually and no episodes
were detected in 2008 and 2009. At Riggs Creek, a dry land agri-
culture (predominantly pasture) site, the observed episodes
showed high amplitude and interannual variability with smaller
peaks and integrals in 2002 and 2006. The evergreen wet
temperate eucalypt forest at the Wallaby Creek site had LSP epi-
sodes with high minimum and peak point magnitude and conse-
quently small amplitude. Following a re in 2009, greenness
decreased and the regeneration period can be seen with subse-
quent cycles of increasing peak and low point magnitude. While the
regeneration period was well characterized, the quantied pre-re
LSP appeared noisy, likely as a function of limited amplitude in the
EVI time series and a relatively high level of high frequency noise. In
contrast, LSP episodes at the Howard Springs wet tropical savanna
site followed a relatively regular pattern with moderately high
minimum points and amplitude of episodes that start at the end of
the year and end during the rst quarter of the following year.
While the timing and magnitude of LSP episodes and their metrics
may vary slightly, all LSP metrics occurred every year at the Howard
Springs site.
LSP episodes of interior Australia can be highly variable and
non-annual. Using an additional sample site in the interior of
Southeast Australia featuring sparse natural vegetation cover
(Fig. 5), we illustrate an example of LSP episodes as they occurred
during the 2000e2013 time interval in water-limited areas of
Australia. In the EVI prole shown in Fig. 5, we see an episode that
started in 2001 and ended in 2002, with no new episode until 20 04.
In 2007 an episode started, peaked and ended. Also in 2007 a
second episode started, peaked in 2008 and ended in 2009. The two
episodes in 2010 had higher peaks compared to the other identied
episodes within the time series.
3.2. Model performance evaluation
The bias, MAE and RMSE values were low relative to the sites'
inter-quartile range (IQR; as a proxy of each site's LSP amplitude)
for most sites (Fig. 6;Table 3). Further, the residual plots (not
shown) did only indicate a minor level of residual autocorrelation
relative to the LSP amplitude as approximated by the IQR.
The results of our direct value comparison of our EVI LSP start
and end of LSP episode timing with the start and end of episode
Table 2
Names, locations, major NVIS classes, average annual rainfall amounts and rainfall variabilities for the 36 sites shown in Fig. 2. Example sites in Figs. 2e5are underlined.
Site name Ozux
site
Abbreviation
as in Fig. 2
Lat Long NVIS major class (Department of the
Environment and Water Resources, 2007)
Average annual rainfall
[mm] (Australian Bureau
of Meteorology, 2014b)
a
Rainfall
variability
class (Australian
Bureau of
Meteorology,
2014b)
b
Nullarbor NU &30.275 127.175 Chenopod shrublands,
samphire shrublands and forblands
200 5
Great Blight Desert GBD &29.125 133.075 Acacia open woodlands 200 4
Interior Southeast Australia IEA &29.425 144.225 Acacia shrublands 300 4
East of Shark bay ESB &24.475 116.325 Acacia shrublands 300 5
Alice Springs x AS &22.275 133.225 Acacia shrublands 400 5
Lake Eyre LE &27.425 137.225 Hummock grasslands 200 6
Central Western Australia CW &24.125 124.175 Hummock grasslands 300 4
Simpson Desert SD &20.475 124.025 Hummock grasslands 400 3
Hamersley x HA &22.275 115.725 Hummock grasslands 400 5
Queensland Tussock QTU &21.225 143.075 Tussock grasslands 500 4
Sturt Plains x SP &17.175 133.375 Tussock grasslands 600 3
Northwest Queensland NWQ &19.525 140.025 Tussock grasslands 600 4
Great Western Woodlands x GWWF &31.925 120.075 Low closed forests and tall closed shrublands 400 2
Calperum x CP &34.025 140.375 Mallee woodlands and shrublands 300 2
West Australian wheat belt WAW &32.125 117.425 Cleared, non-native vegetation, buildings 400 2
Irrigated cropping IC &35.275 145.275 Cleared, non-native vegetation, buildings 400 3
Riggs Creek x RC &36.625 145.575 Cleared, non-native vegetation, buildings 800 2
Otway x OT &38.525 142.825 Cleared, non-native vegetation, buildings 1000 1
Cumberland Plain x CU &33.725 150.725 Cleared, non-native vegetation, buildings 1000 2
Arcturus x AR &23.875 149.275 Eucalypt open forests 800 3
Wombat x WO &37.425 144.075 Eucalypt open forests 1000 2
Wallaby Creek xWC &37.425 145.175 Eucalypt open forests 1200 2
Nimmo x NI &36.225 148.575 Eucalypt open forests 1600 2
Tumbarumba x TU &35.675 148.175 Eucalypt open forests 1600 2
Samford x SA &27.425 152.825 Eucalypt open forests 1600 2
Great Western Woodlands GWW &30.225 120.625 Eucalypt woodlands 300 3
Daly River Pasture x DRP &14.075 131.375 Eucalypt woodlands 1200 2
West of North Queensland WNQ &16.275 142.475 Eucalypt woodlands 1200 3
Howard Springs x HO &12.475 131.175 Eucalypt woodlands 1600 2
Dampier peninsula DP &15.125 125.725 Eucalypt woodlands 1600 3
Gingin x GG &31.375 115.725 Other forests and woodlands 800 1
Dry River x DR &15.275 132.375 Tropical Eucalypt woodlands/grasslands 1000 2
Dargo x DA &37.125 147.175 Other grasslands, herblands,
sedgelands and rushlands
2000 2
Northwest Tasmania NWT &41.225 145.175 Rainforest and vine thickets 2000 1
Cape Tribulation x CT &16.125 145.375 Rainforest and vine thickets 8000 2
Daintree x DT &16.225 145.425 Rainforest and vine thickets 8000 2
a
Calculated as 30 year average.
b
Variability classes are provided as 30-year average. Higher rainfall variability classes indicate higher variability.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 195
Fig. 3. Algorithm steps applied to the 14-year MODIS EVI time series (MOD13C2 single 0.05-degree pixel) for the Alice Springs ux site representing semi-arid mulga (Acacia)
woodland of the centre of Australia. A: EVI time series after screening out low quality observations (brown circles), EVI time series after gap lling and smoothing (blue hollow
circles), and agged minimum and peak of episode points (green diamonds). B) Curves tted as 7-parameter double logistic functions (red squares), and identied start and end of
episode points (yellow circles) delineating the characterized LSP episodes. The timing, length, amplitude, and magnitudes of the LSP episodes at the Alice Springs ux site vary
interannually. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
Fig. 4. Examples of temporal variability of the characterized LSP episodes across 14-years for three OzFlux sites at Riggs Creek, Wallaby Creek, and Howard Springs that represent
dry land agriculture, wet temperate eucalypt forest and wet tropical open woodland savanna, respectively. Brown circles represent MODIS EVI time series after screening out low
quality observations, blue hollow circles represent the EVI time series after gap lling and smoothing, red squares represent the tted 7-parameter double logistic functions and
yellow circles represent the identied start and end of episode points. A LSP episode does not occur in every year and incomplete episodes take place in some years. (For inter-
pretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204196
timing derived from GPP time series modelled for the Howard
Springs ux site are shown in Table 4. The site showed one episode
per year with the Soe in the second half of the year and the Eoe in
the rst half of the following year. For the time overlap between the
EVI and GPP time series we identied 11 LSP episodes. The median
difference between the GPP and EVI-derived Soe points was one 16-
day time steps. The median difference between the GPP and EVI-
derived Eoe points was two 16-day time steps. The GPP-derived
Soe and Eoe points consistently preceded the EVI-derived Soe
and Eoe points. The maximum difference between the GPP-based
and EVI-based Soe and Eoe points was three 16-day time steps
following the GPP-based estimate.
3.3. Start and end of episode timing as examples of highly variable
LSP
We found the LSP of Australia to be highly variable across space
and time. To provide a rst insight into this variability we illus-
trated the timing of the LSP start and end of episode dates across
ENSO stages over the 14-year time series.
Fig. 7 shows the median of the rst yearly start and end of ep-
isodes' timing metric in La Ni~
na, neutral and El Ni~
no years
(Australian Bureau of Meteorology, 2014c) demonstrating the
average response of the LSP metrics to ENSO-induced climate
variability. The gure also shows the difference between the 25%
and 75% quartiles of the start of episode metric and the difference
between the 25% and 75% quartiles of the end of episode metric
across the entire time series illustrating high variability in episodes'
timing mostly over the interior of Australia. For most areas, one LSP
episode per year is the norm whereas a second episode, which
occurs primarily in the Northern (and thereby warmer) parts of the
Southeast and Southwest Australian wheat belts (Fig. 8), may be
associated with double cropping.
3.4. Illustration of all LSP metrics for a dry year and a wet year case
We illustrate and contrast the LSP metrics across all of Australia
for the 2002 El Ni~
no dry year and the 2010 La Ni~
na wet year
(Australian Bureau of Meteorology, 2014c; Heberger, 2011). We
hereby demonstrate the product's information content and the LSP
episode metrics' response to ENSO-driven climate variability
(Heberger, 2011). For the dry year case (2002) we show the rst
minimum point, start of episode point, peak of episode point, end
of episode point, second minimum point, amplitude of the episode,
length of the episode, and episode integrated greenness (Fig. 9).
The point metrics have a magnitude and a time dimension.
The peak magnitude and episode integrated greenness in 2002
were largest along the North, East and Southeast Australian coasts,
in the Southwest of the continent as well as Tasmania (panel in the
centre of Fig. 9). The highest minimum point magnitudes (paler
grey areas) occur along the Eastern coast and across most of
Tasmania.
Areas shown in white, which are mostly located in the interior of
Australia, mark pixels where a given metric did not occur in 2002.
These areas need to be viewed in the context of the LSP episode as,
for example, years with no phenological episode peak may repre-
sent situations where the LSP trajectory is tracking up towards a
peak in the following year, or browning down from a peak in the
previous year. For example, in Fig. 5 we saw that an episode ended
in 2002 but no new episode started and no peak occurred that year.
Fig. 5. Characterized LSP episodes for the Interior Southeastern Australia site derived from 14-years of MODIS EVI data after screening out low quality observations (brown circles),
EVI time series after gap lling and smoothing (blue hollow circles), tting 7-parameter double logistic functions (red squares) and start and end of episode points (yellow circles)
delineating the characterized LSP episodes. A LSP episode does not occur in every year. In some years, part of episodes or incomplete episodes are identied. (For interpretation of
the references to colour in this gure legend, the reader is referred to the web version of this article.)
Fig. 6. RMSE (grey squares) of the y-residual metrics of the LSP episodes modelled via subsequent 7-parameter double logistic curves and the raw EVI time series after screening out
low quality observations for the 36 test sites (for acronym description, please see Table 3) across the time series. The inter-quartile range (IQR ¼90% quartile - 10% quartile shownas
black diamonds) is provided as a proxy of the LSP amplitude to put the RMSE magnitude into context.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 197
In 2008, an episode peaked but the associated minimum points
occurred in 2007 and 2009. This example illustrates that the white
areas in Fig. 9 are an expression of LSP where episodes do not start
and end within a year or where episodes may not occur in a given
year.
During the 2010 La Ni~
na wet year case (Fig. 10), metrics such as
the minimum and peak points as well as the episode integrated
greenness were higher compared to the 2002 El Ni~
no dry year case,
an expected result given that vegetation responds quickly to
change in rainfall. In 2010 La Ni~
na-induced rainfall caused wide-
spread ooding over Eastern Australia (Australian Bureau of
Meteorology, 2014c; Heberger, 2011). In general, for most pixels
across Australia, episodes started and ended in 2010 while in 2002
few areas showed an episode start, which we attribute to the dry
conditions. In Fig. 5, showing the sample trajectory for Interior
Southeastern Australia, an episode peak in early 2010 was detected.
This peak was followed by an end of episode, a minimum point and
a start of episode point in mid-2010 (Fig. 5). These dynamics are
shown in Fig. 10, which reveals that every metric except for a sec-
ond minimum point, which occurred in 2010 in the Interior
Southeastern site (Fig. 10 blue dot (in the web version)). For this
area of Australia, which is covered by sparse Mulga (Acacia aneura)
woodlands with a tussock grass understory (Department of the
Environment and Water Resources, 2007), the Australian Water
Availability Project (Raupach et al., 2009, 2012) shows high percent
rainfall ranks early and in the second half of 2010. The high
magnitude of the LSP peaks in 2010 and early 2011 and their dense
temporal spacing (Fig. 5) likely reects the strong and rapid
response of the grass understory to rainfall.
For most pixels no second minimum point occurred within a
twelve month period (Fig. 8). The majority of exceptions is pixels
where a second episode started within a twelve-month period
and are over agricultural areas associated with double cropping
(Fig. 8).
To further illustrate the contrast between the metrics of the
2002 dry and the 2010 wet year case shown in Figs. 9 and 10 above,
we provide a set of difference images in Fig. 11. As a consequence of
high rainfall amounts over large areas of Eastern Australia, the
timing of the rst minimum, start of episode and end of episode
point occurred earlier in the 2010 wet year as compared to the 2002
dry year. Conversely, the peak point occurred later in 2010 than in
2002. Over Eastern Australia, the 2010 greenness magnitude at all
of the point metrics as well as the amplitude and the episode in-
tegrated greenness metrics were higher and the length of the
episode metric was longer than the equivalent of these metrics in
2002. Note that only areas where a metric pair showed values in
2002 and 2010 are shown in Fig. 11. Given that the interior of
Australia is rainfall limited,
3
not every metric occurs in every year,
especially in dry years. Spatially continuous LSP metrics were
detected along coastal areas. The differences in metric values rep-
resenting the dry and wet year case are large, specically in the East
where positive rainfall anomalies occurred in 2010 (Australian
Bureau of Meteorology, 2014c).
4. Discussion
4.1. Key points regarding the utility of the Australian Land Surface
Phenology Product
First, LSP episodes are subject to high interannual variability
especially for interior Australia. Second, across the rainfall-limited
interior of Australia, episodes may not occur every year. In some
years an episode may end without a new episode starting or an
episode may start after one or multiple years where no episode
occurred. Consequently, not every metric occurs every year as a
function of variable LSP episodes. This non-occurrence of metrics is
an important feature in the analysis of LSP. The non-occurrence of
metrics likely reects periods that are too dry for vegetation
growth. Third, the metrics are sensitive to the effect of the 2002 El
Ni~
no and the 2010 La Ni~
na, demonstrating the inuence of climatic
variability on LSP across Australia. Large and small variability of
metric values as well as the non-occurrence of metrics is relevant
for studies aiming to understand ecosystems and their response to
climate variability.
Based on these ndings, we suggest that users of the Australian
LSP product take advantage of the rich information content rep-
resented by analysing the set of LSP metrics in a temporally
continuous fashion that allows the characterization of individual
Table 3
Bias, MAE and RMSE of the y-residual metrics of the LSP episodes modelled via
subsequent 7-parameter double logistic curves and the raw EVI time series after
screening out low quality observations for each of our 36 test sites. Inter-quartile
range (IQR ¼90% quartile e10% quartile) is provided as a proxy of the LSP ampli-
tude to put the residual metrics into context.
Site name Abbreviation
as in Fig. 2
Bias MAE RMSE Inter-quartile
range
(IQR ¼90%
quartile e10%
quartile)
Nullarbor NU 0.000 0.01 0.02 0.21
Great Blight Desert GBD 0.000 0.03 0.04 0.44
Interior Southeast
Australia
IEA &0.002 0.02 0.03 0.40
East of Shark bay ESB 0.000 0.02 0.03 0.21
Alice Springs AS 0.000 0.01 0.01 0.11
Lake Eyre LE &0.001 0.03 0.04 0.52
Central Western
Australia
CW 0.000 0.00 0.01 0.09
Simpson Desert SD 0.000 0.00 0.01 0.07
Hamersley HA 0.000 0.01 0.01 0.13
Queensland Tussock QTU &0.001 0.02 0.03 0.36
Sturt Plains SP &0.002 0.01 0.02 0.25
Northwest
Queensland
NWQ 0.000 0.02 0.04 0.34
Great Western
Woodlands
GWWF 0.000 0.00 0.01 0.06
Calperum CP 0.000 0.01 0.01 0.21
West Australian
wheat belt
WAW 0.000 0.02 0.03 0.29
Irrigated cropping IC &0.003 0.02 0.03 0.23
Riggs Creek RC 0.001 0.01 0.02 0.25
Otway OT &0.002 0.02 0.04 0.22
Cumberland Plain CU 0.000 0.00 0.01 0.07
Arcturus AR 0.001 0.01 0.02 0.30
Wombat WO &0.001 0.01 0.01 0.09
Wallaby Creek WC 0.000 0.01 0.01 0.08
Nimmo NI &0.001 0.01 0.03 0.17
Tumbarumba TU 0.002 0.01 0.02 0.22
Samford SA &0.001 0.01 0.02 0.18
Great Western
Woodlands
GWW 0.000 0.00 0.01 0.07
Daly River Pasture DRP 0.000 0.01 0.02 0.19
West of North
Queensland
WNQ 0.001 0.01 0.02 0.27
Howard Springs HO &0.005 0.02 0.04 0.26
Dampier peninsula DP 0.000 0.02 0.04 0.16
Gingin GG 0.000 0.01 0.02 0.22
Dry River DR 0.000 0.01 0.01 0.10
Dargo DA &0.001 0.01 0.01 0.13
Northwest Tasmania NWT 0.001 0.03 0.04 0.52
Cape Tribulation CT 0.001 0.02 0.03 0.25
Daintree DT 0.000 0.00 0.01 0.10
3
See map of average annual, seasonal and monthly rainfall by the Australian
Government Bureau of Meteorology: http://www.bom.gov.au/jsp/ncc/climate_
averages/rainfall/index.jsp.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204198
episodes. The demonstrated sensitivity of the metrics to dry and
wet years is expected to allow value added information extraction,
for example, for crop yield quantication and re fuel accumulation
studies as well as questions requiring information about vegetation
response to disturbances and land use change. A more in depth
analysis of LSP response to decadal climate variability across
Australia can be found in Broich et al. (2014).
4.2. Implication of the current product's spatial and temporal
resolution
The Australian Land Surface Phenology Product Version 1 is
currently provided at 0.05-degree spatial resolution and the met-
rics are organized by calendar year.
The input data consist of 16-day EVI composites at a spatial
resolution of 0.05-degree.
The spatial and the temporal resolution have certain implica-
tions. The 16-day EVI compositing method results in less noisy time
series with less temporal resolution compared to daily observations
or 8-day composites. The 16-day EVI composites are generated
using the constrained view maximum value (CV-MVC) compositing
algorithm that selects the maximum EVI observed within a view
angle range of ±30
!
over a 16 day interval. The compositing
method aims to reduce the noise induced by undetected clouds,
cloud shadows and large, off-nadir view angles and favour high
quality observations recorded within the interval (Huete et al.,
1999; Solano et al., 2012).
The series of 0.05-degree MODIS climate modelling grid (CMG)
products is derived as an aggregate of higher spatial resolution
versions of the respective products. As a consequence of the spatial
aggregation procedure, the CMG products are likely less inuenced
by bias in individual higher resolution pixels caused by undetected
clouds and view angle variation. Yet, this also implies that the
current product does not resolve ne spatial detail in vegetation
cover such as riparian vegetation dynamics in a desert environment
where the fraction of soil reection exceeds chlorophyll absorption
by vegetation.
The identied LSP is a function of the spatial resolution of the
input data and studies such as de Beurs et al. (2009) have shown
the value of investigating LSP trends at multiple spatial resolutions.
Besides reducing the noise in time series, the 16-day composite
may also not resolve important details and key transition dates of
the LSP episodes (Zhang et al., 2009). To further investigate both of
these points, and to enable investigation at improved spatial detail,
future versions of the Australian Land Surface Phenology Product
will use higher spatial and temporal resolution input datasets such
as the 500-m spatial resolution, 16-day MOD13A1 Product. Using
the 500-m spatial resolution daily MOD09GA data is also under
consideration. Future studies should also aim to characterize the
extent to which biases may occur due to Bidirectional Reectance
Distribution Function (BRDF) effects in sparse vegetation. The
development of a quality assurance model that quanties the un-
certainty of LSP episodes and LSP metrics is also planned for future
versions of the product.
We have currently organized the product's LSP metrics by cal-
endar year even though LSP episodes for certain areas of Australia
occurred at any time of the year, extend across the end of a year,
skip a year (did not occur for one or multiple years) or occurred
Table 4
Direct value comparison of Soe and Eoe derived from GPP and MODIS EVI at the Howard Springs site.The rst two rows of the table show the difference between GPP-derived
and EVI-derived Soe and Eoe timing as number of 16-day time steps with positive numbers referring to EVI leading GPP and negative numbers referring toEVI following GPP.
The bottom four rows show the estimated Soe and Eoe expressed in day of year. The median of the difference between estimates and the median of the estimated day of year is
also provided.
2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 Median
Difference in 16-day time steps Soe GPP eSoe EVI 1 0 &20&2&10&3&1&1&1e&1
Eoe GPP eEoe EVI e&1&2&1&3&2&1&3&3&2&2&1&2
Estimated day of year Soe GPP 273 273 289 289 257 273 273 273 273 257 241 e273
Soe EVI 257 273 321 289 289 289 273 321 289 273 257 e289
Eoe GPP e161 161 177 145 177 161 129 145 145 129 161 161
Eoe EVI e177 193 193 193 209 177 177 193 177 161 177 177
Fig. 7. Median day of year of the rst start of the LSP episode (Soe_1; top row) and rst end of the LSP episode (Eoe_1; bottom row) for La Ni~
na years, neutralyears and El Ni~
no
years (Australian Bureau of Meteorology, 2014c). The difference in number of days between the 75% quartile and the 25% quartile of the rst start of the LSP episodes' across the time
series as well the difference in number of days between the 75% quartile and 25% quartile of the rst end of the LSP episodes' across the time series is shown in the right column.
(For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 199
more than once a year. The implications are twofold: 1) There is no
ideal 12 month interval for organizing the data and more impor-
tantly 2) the derived LSP metrics are best analysed in a temporally
continuous manner.
4.3. LSP episodes of vegetation greening and browning in the
context of vegetation productivity
The LSP metrics derived in the Australian Land Surface
Phenology Product are key variables that characterize LSP episodes
of vegetation greening and browning. EVI provides an approxi-
mation of sub-pixel chlorophyll content and leaf area index (Huete
et al., 2014). In temperature-limited ecosystems, low winter tem-
peratures reset the LSP episode (Nemani et al., 2003; Richardson
et al., 2013). The clear interruption of plant growth allows the use
of EVI integrated over a LSP episode as an approximation of net
primary productivity (Ponce Campos et al., 2013; Zhang et al.,
2013). Because EVI can be considered as a surrogate for the frac-
tion of absorbed photosynthetically active radiation, EVI integrated
over a LSP episode is near linearly related to vegetation primary
production and is a strong indicator for the yields of annual crops
(Xin et al., 2015).
However, using the episode integrated greenness metric (Eig) to
estimate productivity is more complex if sub-pixel vegetation cover
is sparse or if multi-layer (herbaceous and woody) or evergreen
woody vegetation occurs within a pixel. For example, over sparse
herbaceous vegetation, pixels with different sub-pixel vegetation
cover and spectral properties of the soil may have an identical Eig
metric value.
Besides sparse vegetation cover, sub-pixel vegetation cover
consisting of mixed crops also complicates the use of integrated
greenness metric to estimate productivity (Xin et al., 2013, 2015).
Further, large areas of Australia are covered by savanna and open
Fig. 8. Number of years within the 14-year time series where two peaks occurred.
Double cropping is more commonly practiced in the northern parts of the Southeast
and Southwest Australian wheat belts (cropping and pasture areas according to
Lymburner et al. (2011) are shown with a blue outline). (For interpretation of the
references to colour in this gure legend, the reader is referred to the web version of
this article.)
Fig. 9. Illustration of the thirteen LSP metrics in the 20 02 El Ni~
no dry year case. The metrics shown in the rst row are the timing (t) of the rst minimum point (Min_1), start of
episode (Soe_1), peak of episode point (Peak_1), end of episode point (Eoe_1), and second minimum point (Min_2). The metrics shown in the second row are the magnitude (vi) of
the rst minimum point (Min_1), start of episode (Soe_1), peak of episode point (Peak_1), end of episode point (Eoe_1), and second minimum point (Min_2). The metrics shown in
the third row are the amplitude (Amp_1), length of episode (Loe_1) and episode integrated greenness (Eig_1) of the rst episode that started in 2002. The blue circle represents the
location of the LSP time series shown in Fig. 5. Pixels where a given metric did not occur in 2002 are displayed in white. For details on the naming convention of the metrics and
organization of the data please refer to Figs. 12 and 13 and Table 5 in the Appendix. (For interpretation of the references to colour in this gure legend, the reader is referred to the
web version of this article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204200
temperate eucalypt forest that consist of an evergreen woody and
seasonally green herbaceous layer. The section of the integral below
the minimum or start and end of episode points could be used to
characterize the evergreen contribution of the woody layer and
separate it from the grassy understory in Australia's widespread
savannas and open forests (Donohue et al., 2009). Yet the magni-
tude of the minimum or start and end of episode changes during
wet and dry years, thus complicating the disaggregation of the
woody and the grassy layer signal. Separating the LSP signal into
herbaceous and evergreen woody fractions is the subject of
ongoing work (Guerschman et al., 2009, 2012; Lymburner et al.,
2011) and would be useful for applications such as re fuel load
estimation.
Theoretically, dense evergreen forest vegetation in areas such
as the humid tropical areas of Northern Queensland or the forests
of Western Tasmania and higher elevation forests in the Great
Dividing Range should be productive throughout the year.
In these areas, however, undetected sub-pixel cloud cover may
cause a noisy EVI trajectory and increase the uncertainty of the
retrieved LSP metric. Previous studies have documented LSP epi-
sodes for humid tropical areas (e.g. Huete et al., 2006) as multiyear
average EVI time series. For most of Australia, persistent cloud
cover does not occur (Roy et al., 2006), making the above-
mentioned noise inuence a localized effect. The development of a
quality assurance model that quanties the uncertainty is planned
for future versions of the Australian Land Surface Phenology
Product.
4.4. Model performance evaluation
Our characterization of the LSP curve via the 7-parameter
double logistic function was satisfactory for most sites, including
semi-arid and arid sites with high interannual LSP variability, as
shown in the concurrent comparison evaluation. However, certain
sites require further algorithm adjustment and methodological
development. Limitations were mostly identied over forested
evergreen sites due to the limited amplitude of EVI-derived LSP
episodes and the presence of high frequency noise likely due to
unscreened clouds. Examples are the Daintree site (rainforest and
vine thickets NVIS major class), the Tumburumba site and the
Wallaby Creek (Fig. 4) site before the re (both eucalypt open for-
ests according to NVIS major classes).
We also conducted a direct value comparison of our EVI LSP
start and end of the episode timing with the start and end of
episode derived from GPP at the Howard Springs ux site. We
used the same algorithm to derive Soe and Eoe from the EVI and
GPP data. In general, the GPP-derived Soe and Eoe timing esti-
mates compare well with the estimates we derived from EVI. We
found that the GPP-derived estimate precedes the EVI-derived
estimate, a pattern that has also been identied by Ma et al.
(2013). EVI is a spectral index that is sensitive to chlorophyll
content and leaf area index (Huete et al., 2014) and the EVI tra-
jectory can differ from ux tower-modelled GPP time series (Ma
et al., 2013). Ma et al. (2013) also provides a discussion and ref-
erences of possible reasons why GPP responds more rapidly to
Fig. 10. The thirteen LSP metrics in the 2010 La Ni~
na wet year case. The metrics shown in the rst row are the timing (t) of the rst minimum point (Min_1), start of episode (Soe_1),
peak of episode point (Peak_1), end of episode point (Eoe_1), and second minimum point (Min_2). The metrics shown in the second row are the magnitude (vi) of the rst minimum
point (Min_1), start of episode (Soe_1), peak of episode point (Peak_1), end of episode point (Eoe_1), and second minimum point (Min_2). The metrics shown in the third row are
the amplitude (Amp_1), length of episode (Loe_1) and episode integrated greenness (Eig_1) of the rst episode that started in 2002. The blue circle represents the location of the LSP
time series shown in Fig. 5. Pixels where a given metric did not occur in 2010 are displayed in white. For details on the naming convention of the metrics and organization of the
data please refer to Figs. 12 and 13 and Table 5 in the Appendix. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this
article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 201
changing conditions compared to LAI and chlorophyll concentra-
tion approximated by EVI.
The difference between the EVI and GPP Soe and Eoe
found here may also be related to the 16-day time step that
we used in this rst version of the Australian Land Surface
Phenology Product to preserve the temporal resolution of the
input EVI data.
Detailed validation of the LSP episodesmetrics is important
and the subject of ongoing research. A range of complementary
validation methods has been discussed in the literature (Liang and
Schwartz, 2009; Restrepo-Coupe et al., 2013; Richardson et al.,
2007). Methods include time series of ux tower productivity
measurements as attempted by Ma et al. (2013). Other LSP vali-
dation methods include photosynthetically active radiation and
short wave radiation sensor measurements, and multispectral
image time series recorded by phenocameras mounted on ux
towers as well as citizen science accessed via crowd sourcing
(Restrepo-Coupe et al., 2013). The coarse spatial scale of MODIS
pixels relative to ground-based measurements complicates the
validation of MODIS-derived LSP and ner spatial resolution image
time series such as Landsat may be used as an intermediate step in
scaling ux tower measurements to MODIS pixels (Kovalskyy et al.,
2012). The establishment of an Australian phenocamera network is
underway.
5. Conclusion
The Australian Land Surface Phenology Product presented here
is a continent-specic, synoptic dataset that allows the quantitative
analysis of Australia's LSP derived from the MODIS Enhanced
Vegetation Index (EVI) satellite image data stream using a model-
ling algorithm designed to accommodate the high climatic and
resulting high LSP variability of Australia. We developed the
Australian Land Surface Phenology Product as a contribution to the
Australian Terrestrial Ecosystem Research Network (TERN) and
disseminate it online and free of charge through the AusCover data
portal.
This paper introduced the Australian Land Surface Phenology
Product to support studies and applications requiring a character-
ization of the timing and magnitude of episodes of vegetation
greening and browning and quantitative information of episodes
inter- and intraannual variability at the landscape to continental
scale from 2000 to 2013.
With the Australian Land Surface Phenology Product we provide
scientists and practitioners requiring a spatio-temporal character-
ization of vegetation dynamics with an accessible data set that
reduces the initial effort for quantitative spatio-temporal analysis
of crop yield, re fuel accumulation, inuence of land use, climate
uctuation and climate change on natural and cultivated vegetation
Fig. 11. Difference between the thirteen LSP metrics of a dry-year case (2002) and a wet-year case (2010) shown in Figs. 9 and 10. The metrics shown in the rst row are the
difference in timing (t) of the rst minimum point (Min_1), start of episode (Soe_1), peak of episode point (Peak_1), end of episode point (Eoe_1), and second minimum point
(Min_2). The metrics shown in the second row are the difference in magnitude (vi) of the rst minimum point (Min_1), start of episode (Soe_1), peak of episode point (Peak_1), end
of episode point (Eoe_1), and second minimum point (Min_2). The metrics shown in the third row are the difference in amplitude (Amp_1), length of episode (Loe_1) and episode
integrated greenness (Eig_1) of the rst episode that started in 2002. The rst and second minimum points as well as the start of episode, peak and end of episode points have a
difference in time and difference in magnitude dimension. Red and blue areas in the time metrics mark location where the metric occurred earlier and later, respectively, in 2010
compared to 2002. Red and blue areas in the magnitude, amplitude length of episode and episode integrated greenness domain mark locations where the 2002 metrics were lower
(shorter) and higher (longer) than the in 2010 metrics, respectively. Black colour marks areas where a metric did not occur in 2002 or 2010. Not every metric is expected to occur
every year in the water limited interior of Australia. (For interpretation of the references to colour in this gure legend, the reader is referred to the web version of this article.)
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204202
across large areas using remote sensing time series data. Such
analysis can include the investigation of short term LSP variability
and possibly change trends related to natural or anthropogenic
factors.
In this manuscript we provide users with the information
required for product utilization including specication of input
data, the underlying processing algorithm, data access, data
structure and known limitations. We demonstrated the character-
ization of LSP episodes and derived episode metrics for sample sites
representing different vegetation cover and climatic conditions as
well as an initial investigation of the response of the LSP metrics to
rainfall variation in an ENSO-induced wet and a dry year across
Australia. The intra- and interannual variability of LSP episodes
across Australia can be high and has been characterized here for the
rst time. Not every LSP metric occurs in every year. Both the high
interannual variability of LSP episodes and the non-occurrence of
metrics in certain years and areas are a function of rainfall vari-
ability predominantly in the interior of Australia.
Future releases of the Australian Land Surface Phenology Prod-
uct will include multi-resolution versions of the product including
a 0.05-degree and a 500-m spatial resolution version based on 16-
day data. These multi-resolution versions will help satisfy the
needs of a wider user community by overcoming some of the
limitations associated with the current spatial resolution of 0.05-
degree. Forthcoming releases may also use heritage and new
remote sensing data streams such as Advanced Very High Resolu-
tion Radiometer (AVHRR), MEdium Resolution Imaging Spectrom-
eter (MERIS), Visible Infrared Imaging Radiometer Suite (VIIRS),
and Sentinel 3 Ocean Land Colour Instrument (OLCI) to provide
longer time series and further advance the characterization of LSP
episodes and the quantication of their variability in support of
Australian's cross disciplinary scientic applications and land re-
sources monitoring and management. The methodology for
deriving LSP from satellite image data time series developed here
can also be applied to other semi-arid and arid zones of the globe
such as for example the deserts and semi deserts of South America,
Asia and Africa.
Acknowledgements
The development of the Australian Land Surface Phenology
Product was funded by the AusCover Facility of the Australian
Terrestrial Ecosystem Research Network (TERN) (http://www.tern.
org.au) and supported by the ARC-DP1115479 grant entitled
Integrating remote sensing, landscape ux measurements, and
phenology to understand the impacts of climate change on
Australian landscapes(Huete, CI). Calculations were performed on
the University of Technology, Sydney eResearch high performance
computing facility. Beringer is funded under an ARC FT (FT1110602)
and savanna data collection was supported by DP0772981 and
DP130101566. Support for collection, development and archiving
was provided through TERN's AusCover facility.
Appendix A. Supplementary data
Supplementary data related to this article can be found at http://
dx.doi.org/10.1016/j.envsoft.2014.11.017.
References
Australian Bureau of Meteorology, 2014a. Climate of Australia. Australian Govern-
ment Bureau of Meteorology, Australian Government, Canberra, ACT.
Australian Bureau of Meteorology, 2014b. Climate Data Online: Average Annual,
Seasonal and Monthly Rainfall (mm) and Rainfall Variability (Index of Vari-
ability). Australian Government, Canberra, ACT.
Australian Bureau of Meteorology, 2014c. El Ni~
no eDetailed Australian Analysis and
La Ni~
na eDetailed Australian Analysis. Australian Government, Canberra, ACT.
Australian Government Department of Sustainability Environment Water Pop-
ulations and Communities, 2013. Australia's Ecoregions.
Bajocco, S., Rosati, L., Ricotta, C., 2010. Knowing re incidence through fuel
phenology: a remotely sensed approach. Ecol. Model. 221 (1), 59e66.
Beck, P.S., Atzberger, C., Høgda, K.A., Johansen, B., Skidmore, A.K., 2006. Improved
monitoring of vegetation dynamics at very high latitudes: a new method using
MODIS NDVI. Remote Sens. Environ. 100 (3), 321e334.
Bennett, N.D.,Croke, B.F.W.,Guariso, G., Guillaume, J.H.A., Hamilton,S.H., Jakeman, A.J.,
Marsili-Libelli, S., Newham, L.T.H., Norton, J.P., Perrin, C., Pierce, S.A., Robson, B.,
Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013. Characterising perfor-
mance of environmental models. Environ. Model. Softw. 40 (0), 1e20.
Beringer, J., Hutley, L.B., Tapper, N.J., Cernusak, L.A., 2007. Savanna res and their
impact on net ecosystem productivity in North Australia. Glob. Change Biol. 13
(5), 990e1004.
Beringer, J., Hutley, L., Abramson, D., Arndt, S., Briggs, P., Bristow, M., Canadell, J.,
Cernusak, L., Eamus, D., Evans, B., Fest, B., Goergen, K., Grover, S., Hacker, J.,
Haverd, V., Kanniah, K., Livesley, S.J., Lynch, A., Maier, S., Moore, C., Raupach, M.,
Russell-Smith,J., Scheiter,S., Tapper,N., Uotila, P.,2014. Fire inAustralian savannas:
from leaf to landscape. Glob. Change Biol. http://dx.doi.org/10.1111/gcb.12686.
de Beurs, K.M., Henebry, G.M., 2010. Spatio-temporal statistical methods for
modelling land surface phenology. In: Hudson, I.L., Keatley, M.R. (Eds.),
Phenological Research. Springer Netherlands, Dordrecht, pp. 177e208.
de Beurs, K.M., Wright, C.K., Henebry, G.M., 2009. Dual scale trend analysis for
evaluating climatic and anthropogenic effects on the vegetated land surface in
Russia and Kazakhstan. Environ. Res. Lett. 4 (4), 045012.
Bisigato, A.J., Campanella, M.V., Pazos, G.E., 2013. Plant phenology as affected by
land degradation in the arid Patagonian Monte, Argentina: a multivariate
approach. J. Arid Environ. 91 (0), 79e87.
Broich, M., Huete, A., Tulbure, M.G., Ma, X., Xin, Q., Paget, M., Restrepo-Coupe, N.,
Davies, K., Devadas, R., Held, A., 2014. Land surface phenological response to
decadal climate variability across Australia using satellite remote sensing.
Biogeosciences 11, 5181e5198. http://dx.doi.org/10.5194/bg-11-5181-2014.
Brown, M.E., de Beurs, K.M., 2008. Evaluation of multi-sensor semi-arid crop season
parameters based on NDVI and rainfall. Remote Sens. Environ. 112 (5),
2261e2271.
Delbart, N., Kergoat, L., Le Toan, T., Lhermitte, J., Picard, G., 2005. Determination of
phenological dates in boreal regions using normalized difference water index.
Remote Sens. Environ. 97 (1), 26e38.
Department of the Environment and Water Resources, 2007. Australia's Native
Vegetation: a Summary of Australia's Major Vegetation. Australian Government,
Canberra, ACT.
Donohue, R.J., McVicar, T.R., Roderick, M.L., 2009. Climate-related trends in
Australian vegetation cover as inferred from satellite observations, 1981e2006.
Glob. Change Biol. 15 (4), 1025e1039.
Dougherty, R.L., Edelman, A., Hyman, J.M., 1989. Nonnegativity-, monotonicity-, or
convexity-preserving cubic and quintic hermite interpolation. Math. Comput.
52 (186), 471e794.
Eamus, D., Cleverly, J., Boulain, N., Grant, N., Faux, R., Villalobos-Vega, R., 2013.
Carbon and water uxes in an arid-zone Acacia savanna woodland: an analyses
of seasonal patterns and responses to rainfall events. Agric. For. Meteorol.
182e183 (0), 225e238.
Eklundh, L., J
onsson, P., 2010. TIMESAT 3.0 Software Manual, pp. 1e74.
Friedl, M., Henebry, G., Reed, B., Huete, A., White, M., Morisette, J., Nemani, R.,
Zhang, X., Myneni, R., 20 06. Land Surface Phenology. A Community White Paper
Requested by NASA. April 10. http://cce.nasa.gov/mtg2008_ab_presentations/
Phenology_Friedl_whitepaper.pdf.
Ganguly, S., Friedl, M.A., Tan, B., Zhang, X., Verma, M., 2010. Remote sensing of
environment land surface phenology from MODIS: characterization of the
collection 5 global land cover dynamics product. Remote Sens. Environ.114 (8),
1805e1816.
Gray, J., 2012. User Guide for the MODIS Land Cover Dynamics Product (MCD12Q2).
http://www.bu.edu/lcsc/les/2012/08/MCD12Q2_UserGuide.pdf.
Guerschman, J.P., Hill, M.J., Renzullo, L.J., Barrett, D.J., Marks, A.S., Botha, E.J., 2009.
Estimating fractional cover of photosynthetic vegetation, non-photosynthetic
vegetation and bare soil in the Australian tropical savanna region upscaling
the EO-1 Hyperion and MODIS sensors. Remote Sens. Environ. 113 (5), 928e945.
Guerschman, J.P., Oyarzabal, M., Malthus, T., McVicar, T., Byrne, G., Randall, L.,
Stewart, J., 2012. Evaluation of the MODIS-based Vegetation Fractional Cover
Product. CSIRO, Canberra.
Heberger, M., 2011. Australia's millennium drought: impacts and responses. In:
Gleick, P.H. (Ed.), The World's Water. Island Press/Center for Resource Eco-
nomics, pp. 97e125.
Henebry, G.M., de Beurs, K.M., 2013. Remote sensing of land surface phenology: a
prospectus. In: Schwartz, M. (Ed.), Phenology: an Integrative Environmental
Science. Springer, Netherlands, pp. 385e412.
Huete, A.R., Justice, C.O., van Leeuwen, W., 1999. MODIS Vegetation Index (MOD13).
Algorithm Theoretical Basis Document.
Huete, A., Didan, K., Miura, T., Rodriguez, E.P., Gao, X., Ferreira, L.G., 2002. Overview
of the radiometric and biophysical performance of the MODIS vegetation
indices. Remote Sens. Environ. 83 (1e2), 195e213.
Huete, A.R., Didan, K., Shimabukuro, Y.E., Ratana, P., Saleska, S.R., Hutyra, L.R.,
Yang, W., Nemani, R.R., Myneni, R., 2006. Amazon rainforests green-up with
sunlight in dry season. Geophys. Res. Lett. 33 (6), 2e5.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204 203
Huete, A., Miura, T., Yoshioka, H., Ratana, P., Broich, M., 2014. Indices of vegetation
activity. In: Hanes, J. (Ed.), Biophysical Applications of Satellite Remote Sensing.
Springer, Berlin Heidelberg.
IPCC, 2007. Climate Change 2007: Impacts, Adaptation and Vulnerability. Contri-
bution of Working Group II to the Fourth Assessment Report of the Intergov-
ernmental Panel on Climate Change. Cambridge University Press, Cambridge.
Jones, M.O., Jones, L.A., Kimball, J.S., McDonald, K.C., 2011. Satellite passive micro-
wave remote sensing for monitoring global land surface phenology. Remote
Sens. Environ. 115 (4), 1102e1114.
Justice, C.O., Townshend, J.R.G., Vermote, E.F., Masuoka, E., Wolfe, R.E., Saleous, N.,
Roy, D.P., Morisette, J.T., 2002. An overview of MODIS land data processing and
product status. Remote Sens. Environ. 83 (1e2), 3e15.
Khan, S., 2008. Managing climate risks in Australia: options for water policy and
irrigation management. Aust. J. Exp. Agric. 48 (3), 265e273.
Kovalskyy, V., Roy, D.P., Zhang, X.Y., Ju, J., 2012. The suitability of multi-temporal
web-enabled landsat data NDVI for phenological monitoring ea comparison
with ux tower and MODIS NDVI. Remote Sens. Lett. 3 (4), 325e334.
Land Data Operational Product Evaluation, Facility, 2012. MODIS Land Global
Browse of MCD12Q2, 2001e2010, Collection 5.
Liang, L., Schwartz, M., 2009. Landscape phenology: an integrative approach to
seasonal vegetation dynamics. Landsc. Ecol. 24 (4), 465e472 .
Lymburner, L., Tan, P., Mueller, N., Thackway, R., Lewis, A., Thankappan, M.,
Randall, L., Islam, A., Senarath, U., 2011. The National Dynamic Land Cover
Dataset. Geoscience Australia, Symonston, Australia, 105e105.
Ma, X., Huete, A., Yu, Q., Coupe, N.R., Davies, K., Broich, M., Ratana, P., Beringer, J.,
Hutley, L.B., Cleverly, J., Boulain, N., Eamus, D., 2013. Spatial patterns and
temporal dynamics in savanna vegetation phenology across the North Austra-
lian tropical transect. Remote Sens. Environ. 139, 97e115 .
Mackey, B., Berry, S., Hugh, S., Ferrier, S., Harwood, T.D., Williams, K.J., 2012.
Ecosystem greenspots: identifying potential drought, re, and climate-change
micro-refuges. Ecol. Appl. 22 (6), 1852e1864.
MODIS land team, 2014. Status for: Vegetation Indices (MOD13). http://landval.gsfc.
nasa.gov/ProductStatus.php?ProductID¼MOD13.
Moody, A., Johnson, D.M., 2001. Land-surface phenologies from AVHRR using the
discrete Fourier transform. Remote Sens. Environ. 75 (3), 305e323.
Myneni, R.B., Keeling, C.D., Tucker, C.J., Asrar, G., Nemani, R.R., 1997. Increased plant
growth in the northern high latitudes from 1981 to 1991. Nature 386 (6626),
698e702.
Nemani, R.R., Keeling, C.D., Hashimoto, H., Jolly, W.M., Piper, S.C., Tucker, C.J.,
Myneni, R.B., Running, S.W., 2003. Climate-driven increases in global terrestrial
net primary production from 1982 to 1999. Science 300 (5625), 1560e1563.
Nicholls, N., 1991. The El Ni~
no/Southern oscillation and Australian vegetation.
Vegetatio 91 (1e2), 23e36.
Nicholls, N., Drosdowsky, W., Lavery, B., 1997. Australian rainfall variability and
change. Weather 52 (3), 66e72.
Pe~
nuelas, J., Rutishauser, T., Filella, I., 2009. Phenology feedbacks on climate change.
Science 324 (5929), 887e888.
Ponce Campos, G.E., Moran, M.S., Huete, A., Zhang, Y., Bresloff, C., Huxman, T.E.,
Eamus, D., Bosch, D.D., Buda, A.R., Gunter, S.A., Scalley, T.H., Kitchen, S.G.,
McClaran, M.P., McNab, W.H., Montoya, D.S., Morgan, J.A., Peters, D.P.C.,
Sadler, E.J., Seyfried, M.S., Starks, P.J., 2013. Ecosystem resilience despite large-
scale altered hydroclimatic conditions. Nature 494 (7437), 349e352.
Poulter, B., Frank, D., Ciais, P., Myneni, R.B., Andela, N., Bi, J., Broquet, J., Canadell, J.G.,
Chevallier, F., Liu, Y.Y., Running, S.W., Sitch, S., van der Werf, G.R., 2014.
Contribution of semi-arid ecosystems to interannual variability of the global
carbon cycle. Nature 509, 600e603.
Raupach, M.R., Briggs, P.R., Haverd, V., King, E.A., Paget, M., Trudinger, C.M., 2009.
Australian Water Availability Project (AWAP): CSIRO Marine and Atmospheric
Research Component: Final Report for Phase 3.
Raupach, M.R., Briggs, P.R., Haverd, V., King, E.A., Paget, M., Trudinger, C.M., 2012.
Australian Water Availability Project. CSIRO Marine and Atmospheric Research,
Canberra, Australia.
Reed, B.C., White, M., Brown, J.F., 2003. Remote sensing phenology. In:
Schwartz, M.D. (Ed.), Phenology: an Integrative Environmental Science.
Springer, Netherlands, pp. 365e381.
Restrepo-Coupe, N., Huete, A., Broich, M., Davies, K., 2013. Phenology Validation.
Terrestrial Ecosystem Research Network.
Richardson, A., Jenkins, J., Braswell, B., Hollinger, D., Ollinger, S., Smith, M.L., 2007.
Use of digital webcam images to track spring green-up in a deciduous broadleaf
forest. Oecologia 152 (2), 323e334.
Richardson, A.D., Keenan, T.F., Migliavacca, M., Ryu, Y., Sonnentag, O., Toomey, M.,
2013. Climate change, phenology, and phenological control of vegetation
feedbacks to the climate system. Agric. For. Meteorol. 169 (0), 156e173 .
Risbey, J.S., Pook, M.J., McIntosh, P.C., Wheeler, M.C., Hendon, H.H., 2009. On the
remote drivers of rainfall variability in Australia. Mon. Weather Rev. 137 (10),
3233e3253.
Roy, D.P., Borak, J.S., Devadiga, S., Wolfe, R.E., Zheng, M., Descloitres, J., 2002. The
MODIS land product quality assessment approach. Remote Sens. Environ. 83
(1e2), 62e76.
Roy, D.P., Lewis, P., Schaaf, C., Devadiga, S., Boschetti, L., 2006. The global impact of
clouds on the production of MODIS bidirectional reectance model-based
composites for terrestrial monitoring. IEEE Geosci. Remote Sens. Lett. 3 (4),
452e456.
Savitzky, A., Golay, M.J.E.,1964. Smoothing and differentiation of data by simplied
least squares procedures. Anal. Chem. 36 (8), 1627e1639.
Schwartz, M., 2013. Introduction. In: Schwartz, M. (Ed.), Phenology: an Integrative
Environmental Science. Springer, Netherlands, pp. 3e7.
Schwartz, M.D., Hanes, J.M., 2010. Continental-scale phenology: warming and
chilling. Int. J. Climatol. 30 (11), 1595e1598.
Solano, R., Didan, K., Jacobson, A., Huete, A., 2012. MODIS Vegetation Indices
(MOD13) C5 User's Guide. HTML Version 1.00 (May 27, 2010). Terrestrial
Biophysics and Remote Sensing Lab, The University of Arizona. http://vip.
arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_01_2012.pdf.
Tan, B., Morisette, J.T., Wolfe, R.E., Gao, F., Ederer, G.a., Nightingale, J., Pedelty, J.a.,
2011. An enhanced TIMESAT algorithm for estimating vegetation phenology
metrics from MODIS data. IEEE J. Sel. Top. Appl. Earth Observ. Remote Sens. 4
(2), 361e371.
Vrieling, A., de Beurs, K.M., Brown, M.E., 2011. Variability of African farming systems
from phenological analysis of NDVI time series. Clim. Change 109 (3e4),
455e477.
Walker, J.J., de Beurs, K.M., Wynne, R.H., Gao, F., 2012. Evaluation of landsat and
MODIS data fusion products for analysis of dryland forest phenology. Remote
Sens. Environ. 117 (0), 381e393.
Walker, J.J., de Beurs, K.M., Wynne, R.H., 2014. Dryland vegetation phenology across
an elevation gradient in Arizona, USA, investigated with fused MODIS and
landsat data. Remote Sens. Environ. 144 (0), 85e97.
White, J.D., Ryan, K.C., Key, C.C., Running, S.W., 1996. Remote sensing of forest re
feverity and vegetation recovery. Int. J. Wildland Fire 6 (3), 125e136.
White, M.A., Thornton, P.E., Running, S.W., 1997. A continental phenology model for
monitoring vegetation responses to interannual climatic variability. Glob. Bio-
geochem. Cycles 11 (2), 217e234.
Xin, Q., Gong, P., Yu, C., Broich, M., Myneni, R., Suyker, A., Yu, L., 2013. A production
efciency model-based method for satellite estimates of corn and soybean
yields in the Midwestern US. Remote Sens. 5 (11), 5926e5943.
Xin, Q., Broich, M., Suyker, A., Yu, L., Gong, P., 15 February 2015. Multi-scale eval-
uation of light use efciency in MODIS gross primary productivity for croplands
in the Midwestern United States. Agric. For. Meteorol. 201, 111e119.
Zhang, X., Friedl, M.A., Schaaf, C.B., 20 03. Monitoring vegetation phenology using
MODIS. Remote Sens. Environ. 84, 471e475.
Zhang, X., Friedl, M.A., Schaaf, C.B., 20 06. Global vegetation phenology from moderate
resolution imaging spectroradiometer (MODIS): evaluation of global patterns
and comparison with in situ measurements. J. Geophys. Res. 111 (G4), 1e14.
Zhang, X., Friedl, M.a., Schaaf, C.B., 20 09. Sensitivity of vegetation phenology
detection to the temporal resolution of satellite data. Int. J. Remote Sens. 30 (8),
2061e2074.
Zhang, X.Y., Friedl, M.A., Tan, B., 2012. Long-term detection of global vegetation
phenology from satellite instruments. In: Zhang, X. (Ed.), Phenology and
Climate. InTech.
Zhang, Y., Susan Moran, M., Nearing, M.A., Ponce Campos, G.E., Huete, A.R.,
Buda, A.R., Bosch, D.D., Gunter, S.A., Kitchen, S.G., Henry McNab, W.,
Morgan, J.A., McClaran, M.P., Montoya, D.S., Peters, D.P.C., Starks, P.J., 2013.
Extreme precipitation patterns and reductions of terrestrial ecosystem pro-
duction across biomes. J. Geophys. Res. Biogeosci. 118 (1), 148e157.
M. Broich et al. / Environmental Modelling & Software 64 (2015) 191e204204
... In this method, abrupt changes in trends can be related to vegetation perturbations, defining periods (between perturbations) for which the phenological signal is homogeneous and accounted for by the parametric function. For all parametric methods, the theoretical framework works by assuming that regular, annual curves are a good representation of the plant phenological cycle (e.g., [29,30]), which is not valid for all types of vegetation, for instance in arid or semi-arid ecosystems where green-ups of vegetation follow unpredictable non-seasonal rainfall [31][32][33]. Moreover, significantly different results can be obtained on phenometric estimations when using different interpolation techniques [23]. ...
Article
Full-text available
Citation: Chávez, R.O.; Estay, S.A.; Lastra, J.A.; Riquelme, C.G.; Olea, M.; Aguayo, J.; Decuyper, M. npphen: An R-Package for Detecting and Mapping Extreme Vegetation Anomalies Based on Remotely Sensed Phenological Variability. Remote Sens. 2023, 15, 73. Abstract: Monitoring vegetation disturbances using long remote sensing time series is crucial to support environmental management, biodiversity conservation, and adaptation strategies to climate change from global to local scales. However, it is difficult to assess whether a remotely detected vegetation disturbance is critical or not, since available operational remote sensing methods deliver only maps of the vegetation anomalies but not maps of how "uncommon" or "extreme" the detected anomalies are based on the available records of the reference period. In this technical note, we present a new release of the probabilistic method and its implementation, the npphen R package, designed to detect not only vegetation anomalies from remotely sensed vegetation indices, but also to quantify the position of the anomalous observations within the historical frequency distribution of the phenological annual records. This version of the R package includes two new key functions to detect and map extreme vegetation anomalies: ExtremeAnom and ExtremeAnoMap. The npphen package allows remote sensing users to detect vegetation changes for a wide range of ecosystems, taking advantage of the flexibility of kernel density estimations to account for any shape of annual phenology and its variability through time. It provides a uniform statistical framework to study all types of vegetation dynamics, contributing to global monitoring efforts such as the GEO-BON Essential Biodiversity Variables.
... Several studies in recent years demonstrated the sensitivity of different phenological parameters (i.e., SOS and LOS) of vegetation to the climate change response of forest vegetation [15][16][17][18]. Global and regional scale forest phenology is primarily derived from remotely sensed satellite data often called LSP that provides wall-to-wall coverage [12,[19][20][21]. ...
Article
Full-text available
Understanding ecosystem functional behaviour and its response to climate change necessitates a detailed understanding of vegetation phenology. The present study investigates the effect of an elevational gradient, temperature, and precipitation on the start of the season (SOS) and end of the season (EOS), in major forest types of the Kumaon region of the western Himalaya. The analysis made use of the Normalised Difference Vegetation Index (NDVI) time series that was observed by the optical datasets between the years 2001 and 2019. The relationship between vegetation growth stages (phenophases) and climatic variables was investigated as an interannual variation, variation along the elevation, and variation with latitude. The SOS indicates a delayed trend along the elevational gradient (EG) till mid-latitude and shows an advancing pattern thereafter. The highest rate of change for the SOS and EOS is 3.3 and 2.9 days per year in grassland (GL). The lowest rate of temporal change for SOS is 0.9 days per year in mixed forests and for EOS it is 1.2 days per year in evergreen needle-leaf forests (ENF). Similarly, the highest rate of change in SOS along the elevation gradient is 2.4 days/100 m in evergreen broadleaf forest (EBF) and the lowest is −0.7 days/100 m in savanna, and for EOS, the highest rate of change is 2.2 days/100 m in EBF and lowest is −0.9 days/100 m in GL. Winter warming and low winter precipitation push EOS days further. In the present study area, due to winter warming and summer dryness, despite a warming trend in springseason or springtime, onset of the vegetation growth cycle shows a delayed trend across the vegetation types. As vegetation phenology responds differently over heterogeneous mountain landscapes to climate change, a detailed local-level observational insight could improve our understanding of climate change mitigation and adaptation policies.
... The window size was set to 5 and the number of polynomials was set to 3. Several methods have been developed to detect land surface phenology based on vegetation indices [60][61][62][63]. A simple and effective dynamic threshold method was used to extract land surface phenology from the reconstructed NDPI time series (Figure 4) [64]. Pixels with an intra-year NDPI change of less than 0.1 were excluded, and these were assumed to be areas without significant seasonality, as the NDPI of vegetation with distinct phenological stages generally increases from very small values (close to 0) to above 0.4 in study area. ...
Article
Full-text available
In high-elevation mountains, seasonal snow cover affects land surface phenology and the functioning of the ecosystem. However, studies regarding the long-term effects of snow cover on phenological changes for high mountains are still limited. Our study is based on MODIS data from 2003 to 2021. First, the NDPI was calculated, time series were reconstructed, and an SG filter was used. Land surface phenology metrics were estimated based on the dynamic thresholding method. Then, snow seasonality metrics were also estimated based on snow seasonality extraction rules. Finally, correlation and significance between snow seasonality and land surface phenology metrics were tested. Changes were analyzed across elevation and vegetation types. Results showed that (1) the asymmetry in the significant correlation between the snow seasonality and land surface phenology metrics suggests that a more snow-prone non-growing season (earlier first snow, later snowmelt, longer snow season and more snow cover days) benefits a more flourishing vegetation growing season in the following year (earlier start and later end of growing season, longer growing season). (2) Vegetation phenology metrics above 3500 m is sensitive to the length of the snow season and the number of snow cover days. The effect of first snow day on vegetation phenology shifts around 3300 m. The later snowmelt favors earlier and longer vegetation growing season regardless of the elevation. (3) The sensitivity of land surface phenology metrics to snow seasonality varied among vegetation types. Grass and shrub are sensitive to last snow day, alpine vegetation to snow season length, desert to number of snow cover days, and forest to first snow day. In this study, we used a more reliable NDPI at high elevations and confirmed the past conclusions about the impact of snow seasonality metrics. We also described in detail the curves of snow seasonal metrics effects with elevation change. This study reveals the relationship between land surface phenology and snow seasonality in the Qilian Mountains and has important implications for quantifying the impact of climate change on ecosystems.
... A next step involves the calculation of the phenology indicators from the prepared cloud-free time series data, i.e., the LSP metrics. Numerous studies have dealt with the retrieval of phenological phases from remotely sensed data [55,[62][63][64][65]. LSP metrics quantification over croplands is widely used for yield estimation, or to improve management and timing of field works (planting, fertilizing, irrigating, crop protection or harvesting) [66,67]. ...
Chapter
National and International space agencies are determined to keep their fingers on the pulse of crop monitoring through Earth Observation (EO) satellites, which is typically tackled with optical imagery. In this regard, there has long been a trade-off between repetition time and spatial resolution. Another limitation of optical remotely sensed data is their typical discontinuity in time, caused by cloud cover or adverse atmospheric effects. Enduring clouds over agricultural fields can mask key stages of crop growth, leading to uncertainties in crop monitoring practices such as yield predictions. Gap-filling methods can provide a key solution for accurate crop phenology characterization. This chapter first provides a historical overview of EO missions dedicated to crop monitoring. Then, it addresses the rapidly evolving fields of gap-filling and land surface phenology (LSP) metrics calculation using a new in-house developed toolbox, DATimeS. These techniques have been put into practice for homogeneous and heterogeneous demonstration landscapes over the United States. Time series of Difference Vegetation Index (DVI) were processed from two EO data sources: high spatial resolution Sentinel-2 and, low spatial resolution MODIS data. LSP metrics such as start and end of season were calculated after gap filling processing at 1km resolution. Over the homogeneous area both S2 and MODIS are well able to capture the phenology trends of the dominant crop and LSP metrics were successfully mapped. Conversely, the MODIS dataset presented more difficulties than S2 to capture the phenology trend of winter wheat over heterogeneous landscape.
... Due to the increase in the world population, the need for food has increased, which makes agriculture an essential field as it is a significant origin for supplying food [1]. Modern farming is involved with using technology in agriculture can have a positive effect on the economy [2]. ...
... Thermal availability has a direct influence on plant phenology. Detecting changes in Phenology has become a recurrent theme, and precision agriculture has increasingly attracted the attention of farmers, governments, and researchers, since plant monitoring is one of the essential tools for the management and optimization of agricultural resources [3]. The complex interactions between plant development and the environment represent a significant source of uncertainty for growers. ...
Article
Full-text available
In the field of agricultural research, Machine Learning (ML) has been used to increase agricultural productivity and minimize its environmental impact, proving to be an essential technique to support decision making. Accurate harvest time prediction is a challenge for fruit production in a sustainable manner, which could eventually reduce food waste. Linear models have been used to estimate period duration; however, they present variability when used to estimate the chronological time of apple tree stages. This study proposes the PredHarv model, which is a machine learning model that uses Recurrent Neural Networks (RNN) to predict the start date of the apple harvest, given the weather conditions related to the temperature expected for the period. Predictions are made from the phenological phase of the beginning of flowering, using a multivariate approach, based on the time series of phenology and meteorological data. The computational model contributes to anticipating information about the harvest date, enabling the grower to better plan activities, avoiding costs, and consequently improving productivity. We developed a prototype of the model and performed experiments with real datasets from agricultural institutions. We evaluated the metrics, and the results obtained in evaluation scenarios demonstrate that the model is efficient, has good generalizability, and is capable of improving the accuracy of the prediction results.
... With global coverage and frequent repeat observations, satellite remote sensing technology is a particularly valuable tool in this respect, having proven an effective means of monitoring, mapping, and quantifying productivity and phenological changes in vegetation over time and across large scales (Briggs et al., 1997;Goodin and Henebry, 1997;Huete, 2016;Huang et al., 2018). Numerous studies have characterized vegetation phenology (i.e. the timing of recurring biological cycles, such as emergence of first leaf, senescence, growing season length) by exploiting time-series of remotely-sensed spectral vegetation indices (Zhang et al., 2006;Ganguly et al., 2010;Henebry and de Beurs, 2013;Broich et al., 2015;Wu et al., 2016;Richardson et al., 2018). Satellite data have also offered the opportunity to capture years with extreme high and low productivity, including dormant growing seasons for plants in environments with extreme climate variability like Australia (Broich et al., 2014). ...
Article
Species composition is a key determinant of grassland ecosystem function and resilience. Climate change is predicted to alter the distribution of cool season (C3) and warm season (C4) grasses, however, the lack of spatial distributions and temporal variations of grass functional type information severely limits our understanding of climate impacts on grasslands. This study classified C3 and C4 grasses per pixel according to the peak of growing season generated from Enhanced Vegetation Index time series. From 2003 to 2017, the C3-C4 composition of Australian rain-fed grasslands and pastures was mapped at 500 m resolution on an annual basis across a wide geographical range (10°S – 45°S), and revealed extreme inter-annual fluctuations. Over the 15-year period, the satellite-derived ratio of C4 to C3 grasses significantly increased (p < 0.05), indicating a long-term shift in community composition that was confirmed with 182,911 Atlas of Living Australia ground observations. The most pronounced changes occurred in mid-latitude transitional areas where C3 and C4 grasses co-dominate. Our climate analysis indicated the inter-annual fluctuations of C4/C3 grass ratios were significantly associated (p < 0.05) with warm/cool season rainfall ratios, and not with temperature or annual rainfall. This suggests that an increase in the warm/cool season rainfall ratio favors C4 grasses and a decrease in the warm/cool season rainfall ratio favors C3 grasses. Our findings reveal spatially-detailed dynamics of grasslands and demonstrate large-scale grassland compositional changes over 15 years. The grass composition maps should help improve ecological forecasting of grass distributions and enable researches on grassland ecosystem responses to climate change that are relevant to both adaptation of rangeland agricultural and fire management practices. Our study should also help predict grass distribution under future climate conditions, and assist in the accurate modelling of global water, carbon, and energy exchanges between the land surface and atmosphere.
Article
Full-text available
The state of Amapá within the Amazon biome has a high complexity of ecosystems formed by forests, savannas, seasonally flooded vegetation, mangroves, and different land uses. The present research aimed to map the vegetation from the phenological behavior of the Sentinel-1 time series, which has the advantage of not having atmospheric interference and cloud cover. Furthermore, the study compared three different sets of images (vertical–vertical co-polarization (VV) only, vertical–horizontal cross-polarization (VH) only, and both VV and VH) and different classifiers based on deep learning (long short-term memory (LSTM), Bidirectional LSTM (Bi-LSTM), Gated Recurrent Units (GRU), Bidirectional GRU (Bi-GRU)) and machine learning (Random Forest, Extreme Gradient Boosting (XGBoost), k-Nearest Neighbors, Support Vector Machines (SVMs), and Multilayer Perceptron). The time series englobed four years (2017–2020) with a 12-day revisit, totaling 122 images for each VV and VH polarization. The methodology presented the following steps: image pre-processing, temporal filtering using the Savitsky–Golay smoothing method, collection of samples considering 17 classes, classification using different methods and polarization datasets, and accuracy analysis. The combinations of the VV and VH pooled dataset with the Bidirectional Recurrent Neuron Networks methods led to the greatest F1 scores, Bi-GRU (93.53) and Bi-LSTM (93.29), followed by the other deep learning methods, GRU (93.30) and LSTM (93.15). Among machine learning, the two methods with the highest F1-score values were SVM (92.18) and XGBoost (91.98). Therefore, phenological variations based on long Synthetic Aperture Radar (SAR) time series allow the detailed representation of land cover/land use and water dynamics.
Article
Coal mining activities not only damage the ecosystems located inside mining areas, but also affect the buffer areas located within a certain radius of the mining activities. Effectively identifying the range disturbed by mining activities is beneficial for guiding ecological restoration and ecological assessments in mining areas. However, methods for quantitatively identifying the range of disturbances associated with coal mining activities are still lacking. In this study, the phenological disturbance range index (PDRI) based on remote sensing observed vegetation phenology was proposed. Normalized difference vegetation index (NDVI) time series data were collected, and land surface phenology parameters were estimated in the Pinshuo mining area on the Loess Plateau. The proportion of the phenology-climate coupling relationship destroyed by human activities was then calculated. The range of mining activity disturbances was then determined by mutually validating the function-fitting and trajectory-segmentation algorithms. The results show that the proposed PDRI can effectively identify the extent of disturbances resulting from mining activities. Clear differences among the PDRIs reflecting mining activity disturbances, urbanization disturbances and natural states were found. Taking the Pingshuo mine as an example, the range of disturbances was approximately 11 ∼ 12 km, and shaft mining was found to have a greater impact than open pit mining. The PDRI proposed herein solves the initial problem from a climate-response perspective and can be used as a standardized scheme to detect the extent of disturbances resulting from coal mining activities. This study can also provide a reference for policy formulation and ecological compensation range.
Chapter
This chapter gives a theoretical overview of various contact, proximal and remote monitoring solutions available for precision agriculture. Visual inspection of crop damage, which can be detected using these sensors, are introduced at first. Precision agriculture methodologies and sensors are reviewed with particular emphasis on variable rate fertilization. Different sensor platforms reviewed in the chapter ranged from drone images to tractor-mounted and hand-held devices, including the overview of autonomous platforms and robots in precision agriculture. After the theoretical overview a couple of use-cases are described to illustrate the most common practices of using proximal sensing sensors for precision agriculture. The use-case from Estonia demonstrates hand-held proximal sensor usage for variable rate fertilization. The use-cases from Lithuania illustrate field-scale monitoring and mapping of soil characteristics.KeywordsPrecision agricultureVariable rate fertilizationN-sensorN-testerSoil sensingCrop sensingRobot platformChlorophyllElectrical conductivity
Chapter
Full-text available
As this edition of The World’s Water goes to press in early 2011, eastern Australia is recovering from devastating floods that claimed more than 20 lives and destroyed hundreds of homes. The heavy rains of 2009 and 2010 that caused so much destruction also marked the end of Australia’s decade-long Millennium Drought. Beginning in about 1997, declines in rainfall and runoff had contributed to widespread crop failures, livestock losses, dust storms, and bushfires. Such are the vagaries of water on the continent with the world’s most uncertain and variable climate.
Article
Full-text available
Land surface phenological cycles of vegetation greening and browning are influenced by variability in climatic forcing. Quantitative spatial information on phenological cycles and their variability is important for agricultural applications, wildfire fuel accumulation, land management, land surface modeling, and climate change studies. Most phenology studies have focused on temperature-driven Northern Hemisphere systems, where phenology shows annually recurring patterns. However, precipitation-driven non-annual phenology of arid and semi-arid systems (i.e., drylands) received much less attention, despite the fact that they cover more than 30% of the global land surface. Here, we focused on Australia, a continent with one of the most variable rainfall climates in the world and vast areas of dryland systems, where a detailed phenological investigation and a characterization of the relationship between phenology and climate variability are missing. To fill this knowledge gap, we developed an algorithm to characterize phenological cycles, and analyzed geographic and climate-driven variability in phenology from 2000 to 2013, which included extreme drought and wet years. We linked derived phenological metrics to rainfall and the Southern Oscillation Index (SOI). We conducted a continent-wide investigation and a more detailed investigation over the Murray–Darling Basin (MDB), the primary agricultural area and largest river catchment of Australia. Results showed high inter- and intra-annual variability in phenological cycles across Australia. The peak of phenological cycles occurred not only during the austral summer, but also at any time of the year, and their timing varied by more than a month in the interior of the continent. The magnitude of the phenological cycle peak and the integrated greenness were most significantly correlated with monthly SOI within the preceding 12 months. Correlation patterns occurred primarily over northeastern Australia and within the MDB, predominantly over natural land cover and particularly in floodplain and wetland areas. Integrated greenness of the phenological cycles (surrogate of vegetation productivity) showed positive anomalies of more than 2 standard deviations over most of eastern Australia in 2009–2010, which coincided with the transition from the El Niño-induced decadal droughts to flooding caused by La Niña.
Article
Full-text available
[1] Precipitation regimes are predicted to shift to more extreme patterns that are characterized by more heavy rainfall events and longer dry intervals, yet their ecological impacts on vegetation production remain uncertain across biomes in natural climatic conditions. This in situ study investigated the effects of these climatic conditions on aboveground net primary production (ANPP) by combining a greenness index from satellite measurements and climatic records during 2000–2009 from 11 long-term experimental sites in multiple biomes and climates. Results showed that extreme precipitation patterns decreased the sensitivity of ANPP to total annual precipitation (PT) at the regional and decadal scales, leading to decreased rain use efficiency (RUE; by 20% on average) across biomes. Relative decreases in ANPP were greatest for arid grassland (16%) and Mediterranean forest (20%) and less for mesic grassland and temperate forest (3%). The cooccurrence of heavy rainfall events and longer dry intervals caused greater water stress conditions that resulted in reduced vegetation production. A new generalized model was developed using a function of both PT and an index of precipitation extremes and improved predictions of the sensitivity of ANPP to changes in precipitation patterns. Our results suggest that extreme precipitation patterns have substantially negative effects on vegetation production across biomes and are as important as PT. With predictions of more extreme weather events, forecasts of ecosystem production should consider these nonlinear responses to altered extreme precipitation patterns associated with climate change.
Article
The process of observing land surface phenology (or LSP) using remote sensing satellites is fundamentally different from ground level observation of phenophase transition of specific organisms. The scale disparity between the spatial extent of the organisms and the spatial resolution of the sensor leads to an ill-defined mixture of target and background or signal and noise. Much progress has been made in the monitoring and modeling of land surface phenologies over the past decade. The chapter first provides a brief overview of land surface phenology, starting with the Landsat 1 in 1972, and then proceeds to a survey of current LSP products. The problem of indistinct phenometrics in remote sensing data is considered and the alternative phenometrics derived from the convex quadratic model are presented with an application in the North American Great Plains using MODIS data from 2001 to 2012. The chapter concludes with a view forward to outstanding challenges for LSP research in the coming decade. © 2013 Springer Science+Business Media B.V. All rights are reserved.
Chapter
In this chapter we explain satellite-based vegetation indices (VIs) as dynamic spectral measures of vegetation activity. VIs are among the most widely used satellite products in monitoring ecosystems and agriculture, resource management, and estimations of many biophysical canopy properties. A theoretical basis for their formulation is presented and we describe how VIs are processed and composited from satellite imagery. Recent trends in their validation and quality assessment using in situ tower measurements are also discussed. Finally, a cross section of major findings involving the use of satellite VIs in ecological and climate science is presented and we conclude with research challenges and environmental issues that will drive future uses of satellite VIs.
Article
In attempting to analyze, on digital computers, data from basically continuous physical experiments, numerical methods of performing familiar operations must be developed. The operations of differentiation and filtering are especially important both as an end in themselves, and as a prelude to further treatment of the data. Numerical counterparts of analog devices that perform these operations, such as RC filters, are often considered. However, the method of least squares may be used without additional computational complexity and with considerable improvement in the information obtained. The least squares calculations may be carried out in the computer by convolution of the data points with properly chosen sets of integers. These sets of integers and their normalizing factors are described and their use is illustrated in spectroscopic applications. The computer programs required are relatively simple. Two examples are presented as subroutines in the FORTRAN language.
Article
Satellite remote sensing provides continuous observations of land surfaces, thereby offering opportunities for large-scale monitoring of terrestrial productivity. Production Efficiency Models (PEMs) have been widely used in satellite-based studies to simulate carbon exchanges between the atmosphere and ecosystems. However, model parameterization of the maximum light use efficiency ( ) varies considerably for croplands in agricultural studies at different scales. In this study, we evaluate cropland in the MODIS Gross Primary Productivity (GPP) model (MOD17) using in situ measurements and inventory datasets across the Midwestern US. The site-scale calibration using 28 site-years tower measurements derives values of 2.78 ± 0.48 gC MJ−1 (± standard deviation) for corn and 1.64 ± 0.23 gC MJ−1 for soybean. The calibrated models could account for approximately 60–80% of the variances of tower-based GPP. The regional-scale study using 4-year agricultural inventory data suggests comparable values of 2.48 ± 0.65 gC MJ−1 for corn and 1.18 ± 0.29 gC MJ−1 for soybean. Annual GPP derived from inventory data (1848.4 ± 298.1 gC m−2 y−1 for corn and 908.9 ± 166.3 gC m−2 y−1 for soybean) are consistent with modeled GPP (1887.8 ± 229.8 gC m−2 y−1 for corn and 849.1 ± 122.2 gC m−2 y−1 for soybean). Our results are in line with recent studies and imply that cropland GPP is largely underestimated in the MODIS GPP products for the Midwestern US. Our findings indicate that model parameters (primarily ) should be carefully recalibrated for regional studies and field-derived can be consistently applied to large-scale modeling as we did here for the Midwestern US.