Ha Nguyen Phenology 2018 presentation on Melbourne pollen trends
Linking land use land cover
changes in pastures to timing
and trends of grass pollens in
Ha Thanh Nguyen
University of Technology Sydney
• Exotic grass species
produced large quantities of
pollens and are responsible
for the allergic respiratory
diseases in Australia.
• In Melbourne, Victoria, the
annual amount of grass
pollens has been decreasing
with accelerating date of
maximum daily pollen count.
• Hypothesis: land use change
affects the location and
strength of the pollen sources
´ Generate pollen seasonality metrics from Parkville, Melbourne, Victoria for
2016: start/ end dates, absolute peak and secondary peaks
´ Generate wind trajectories up to 72 hours prior to dates of absolute and
secondary peaks in pollen records (HYSPLIT)
´ Classify points along such trajectories by land cover (Google Earth):
pasture/ non pastures
´ Assess the percentage area that peak in EVI within 3 x 3 km2 surrounding
each point (MODIS Subset)
´ Extract time series of Enhanced Vegetation Index from Landsat and
analyze for past changes.
Methodology: combine wind back
trajectory with satellite remote sensing
The Hybrid Single Particle Lagrangian
Integrated Trajectory (HYSPLIT) model
´ uses observed/ calculated 3-dimensional meteorological fields (>=
0.5deg) to estimate the most likely central path over geographical areas
that provided air to a receptor at a given time.
´ starts a trajectory from a single location and height every 3 hours
backward in time and then sum the frequency that the trajectory passed
over a grid cell and then normalize by either the total number of
trajectories or endpoints.
• Points with higher trajectory
frequencies are expected to
contribute more to pollen
concentration within the time window
(24/ 72 hours) i.e. “potential point
• Wind trajectories are different on
different days, suggesting that the
potential point sources for daily counts
changed with the day/ episodes.
• Most of the potential point sources are
close to the trap, so we limited our
analysis to one Landsat scene that
encompassed the Parkville pollen trap.
• What is the dominant land
cover within an 3 x 3km2
area surrounding each
HYSPLIT point source?
• What is the vegetation
phenology within each 3 x
3km2 area? (MODIS
• To contribute to the pollen
count of a certain date, an
area must have flowered ,
and hence must have
peaked in greenness prior
to that date.
16-days 250m MODIS data
§ For each HYSPLIT point
source, percentage of the
3 x 3km2 area that has
peaked in EVI prior to a
date was assessed.
§ A point with higher area
percentage that has
peaked in EVI prior to a
date is expected to
contribute more to the
pollen count of that date.
´ Using wind back trajectories, Google Earth and MODIS, we have identified
potential point sources that (i) an air parcel most likely passed through prior
to reaching the Melbourne pollen trap and (ii) have peaked in EVI prior to a
high pollen episode.
´ Points with 0%, 50% and 100% peaking EVI in their surroundings were
selected and analyzed for past land cover changes using BFAST (Breaks For
Additive Season and Trend).
Our goal: link land use land cover
changes to timings and trends in grass
´ Decompose data into trend,
seasonality, and residuals.
´ Define breakpoints by directional or
magnitude change between new
data and historical data.
´ Make use of the entire time series. Only
requires the tweaking of one
parameter h, which is related to the
duration and frequency of the process
of interest and data availability.
One Landsat pixel: 30 x 30m
Slope_2 = slope of the most recent trend
100% peaking before 14-10-2016 50% peaking before 14-10-2016
0% peaking before 07-11-201650% peaking before 07-11-2016
13 Slope_2 = slope of the most recent trend
Validation and Future works
• Fine-tuning parameters of BFAST
• Grass observations from the Living
Atlas of Australia.
• Insight from agronomists and farmers
of changes in pastures.
Subtropical species -
Temperate species -
[Source : Medek et al., 2016]
´ In this study, we analyzed land use land cover changes at locations that (i)
an air parcel most likely has passed through prior to reaching the
Melbourne pollen trap and (ii) have peaked in EVI (and hence flowered)
prior to a high pollen episode.
´ The potential sources of grass pollens are different for different episodes of
high pollen counts throughout the pollen monitoring period of 2016. These
potential sources also differed in terms of their vegetation phenology.
´ At these potential point sources, the most popular land use change
trajectories are urbanization, conversion of pasture to other types of
vegetation and decrease in pasture greenness.