►
From YouTube: CGD Seminar Series - Hillary Scannell
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
B
A
A
Okay,
I
think
we
can
get
started
hi
everyone
thanks
for
joining
us
for
the
cgd
seminar
this
morning.
I'm
pleased
to
introduce
dr
e,
dr
hillary
scannell,
from
columbia
university
as
our
speaker
today,
dr
scannell
is
a
postdoctoral
research
scientist
in
the
climate
data
science
lab,
which
is
part
of
the
lamont
dougherty
earth
observatory
at
columbia.
A
She
received
her
bachelor's
in
marine
science
and
her
master's
in
oceanography,
both
from
the
university
of
maine
and
her
phd
in
oceanography
from
the
university
of
washington
in
2020.
Her
dissertation
focused
on
the
spatiotemporal
evolution
of
marine
heat
waves
globally
and
also
worked
on
developing
a
novel
tracking
algorithm
to
characterize
dangerous
marine
heat
waves.
A
Hillary
is
interested
in
human
impacts
at
the
crossroads
of
the
information
age
and
the
anthropocene
for
research
on
complex
spatiotemporal,
variability
of
geophysical
fluids
and
their
impacts
leverages
and
innovates
new
data-driven
technologies
to
address
modern
environmental
challenges.
Today
she
will
speak
about
new
insights
into
the
spatiotemporal
connectivity
of
marine
heat
waves
globally.
Hillary
the
floor
is
yours,.
C
Thank
you,
katie.
Let's
share
my
screen.
C
All
right
so
hi
everyone
thanks
for
joining
and
the
invitation
to
speak
today.
I
first
want
to
acknowledge
my
co-authors
and
collaborators,
so
luanne
thompson.
She
was
actually
my
phd
advisor
and
david
john
gagne
who's
at
sizzle
and
then
dan
witt,
who
was
at
cgd,
and
I
was
at
nasa
and
this
work
as
katie
mentioned.
C
So
what
started
about
a
year
ago
when
I
was
a
graduate
visitor
during
the
advanced
study
program,
and
so
I
was
at
the
mesa
lab
about
two
weeks
before
it
all
shut
down
due
to
covid,
and
I
feel
very
fortunate
that
I
was
able
to
continue
this
work,
this
exciting
project
remotely
and
and
present
it
to
you
today.
C
C
This
is
of
the
sea
surface
temperature
anomaly
in
may
2015,
and
most
of
you
will
recognize
this
image
as
the
blob,
and
so
the
blob
took
us
by
storm,
essentially
in
2013,
and
it
persisted
for
multiple
years
through
at
least
2015
even
into
2016.,
and
this
was
such
an
extraordinary
event
because
it
really
put
marine
heat
waves
on
the
radar
of
a
lot
of
u.s
scientists.
C
So
this
wasn't
the
first
marine
heat
wave
ever
described
or
recorded,
but
it
certainly
was
a
very
important
event
for
its
impacts
on
the
marine
ecosystem
and
so
some
of
those
impacts.
Of
course
you
have
this
warm
water
blob
which
can
reduce
the
productivity
of
the
marine
ecosystem
in
this
region
and
so
with
a
reduction
in
productivity.
That
means
that
there's
less
food
available
for
animals
like
sea
lions,
sea
birds
to
survive,
and
so
it
was
happening.
C
We
saw
a
lot
of
stranded
sea
lion
pups
along
the
beach,
and
you
know
there
was
a
mass
die
off
of
common
mirrors,
like
the
one
shown
here
that
washed
ashore,
because
it
couldn't
find
food
that
it
needed
to
survive.
C
So,
as
a
result
of
this
marine
heat
wave,
a
lot
of
the
clamming
fisheries
were
closed
for
an
extended
period
of
time
on
the
west
coast,
and
so
that
also
led
to
economic
losses,
and
so
this
kind
of
drives
home.
Why
this
particular
event
was
so
important
to
understand
and
really
captured
a
lot
of
attention,
especially
from
the
media
when
you're
walking
the
beach
and
you
kind
of
see
this
response
in
the
ecosystem.
C
So
so
again,
that's
the
picture
kind
of
the
media
likes
to
paint
and
hold
on
to,
but
as
scientists
we
want
to
be
able
to
identify,
describe
and
eventually
predict
these
marine
heat
waves,
and
so
I'm
going
to
walk
through
how
we
define
them
and
it's
a
very
simplistic
walk
through.
So
sorry,
if
this
is
a
little
boring
to
you,
but
we'll
start
with
just
picking
out
a
point
in
the
northeast
pacific,
so
I'm
using
the
noaa
optimum
interpolation
sea
surface
temperature
data
set.
C
So
this
is
a
blend
of
in-situ
observations,
interpolated
to
a
grid,
and
here
I'm
showing
you
the
time
series
that's
available
so
about
late,
1981
through
2020..
So
you
have
this
really
nice
seasonal
cycle.
You
have
inter-annual
variability,
let's
just
zoom
in
on
this
five-year
period.
This
is
when
the
blob
that
marine
heat
wave
in
this
pacific
occurred,
and
so
we're
going
to
look
at
this
section
of
time
to
describe
how
we
define
a
marine
heat
wave.
C
So
the
first
thing
we
do
once
we
have
this
data
set.
We
want
to
fit
the
mean
the
seasonal
cycle
and
the
trend
to
it.
So
that's
what
I'm
showing
here
by
this
orange
time
series,
and
so
this
orange
time
series
is
what
we'd
expect
the
temperatures
to
be
like.
According
to
this,
you
know
40-year
averaging
period
and
then
you
can
see
that
there
was
periods
where
the
temperature
was
above
this
orange
line,
and
so
those
are
times
when
it
was
warm.
C
So
we
can
subtract
the
temperature
from
that
fitted
reference
of
the
seasonal
cycle,
the
trend
and
the
mean
to
get
anomalies,
and
it's
not
surprising
that
this
five-year
period
was
warmer
than
average.
That's
when
we
had
that
marine
heat
wave,
but
not
all
of
those
positive
temperature
anomalies,
are
considered
a
marine
heat
wave.
So
we
have
different
thresholds.
C
We
use
to
determine
at
what
point
that
a
marine
heat
wave
occurs
and
the
most
common
threshold
that
is
used
is
looking
at
a
percentile
based
metric,
and
so
the
90th
percentile
is
one
that
is
often
referred
to.
C
So
I'll
show
that
here
as
this
thick
dashed
line,
and
so
when
the
temperature
normally
exceeds
that
it's
considered
a
marine
heat
wave
now
you
can
also
increase
that
threshold,
say
the
95th
to
the
99th
and
you'll
see
that
those
marine
heat
waves
you
know,
become
less
common
because
they're
not
surpassing
that
high
threshold
as
frequently
now,
of
course,
you
can
do
this
on
daily
data
as
well.
In
fact,
that's
more
preferred
and
with
daily
data
you
might
introduce
a
minimum
duration
criteria.
C
C
C
C
Okay
and
then
there's
another
proposed
way
of
classifying
marine
heat
waves,
and
that
is
assigning
them
a
category,
so
these
categories
are
determined
by
multiples
of
the
90th
percentile
difference.
C
So
this
type
of
analysis,
where
you're
computing
these
metrics
at
each
single
grid
point,
is
fairly
common
and
gives
you
a
nice
picture
of
what
the
spatial
distribution
looks
like
around
the
globe.
So
I'm
showing
you
a
series
of
maps
now
on
the
left.
This
is
the
and
these
maps
are
calculated
by
doing
exactly
the
method
I
showed
earlier,
where
at
each
grid
point
they're
calculating
okay.
What's
the
frequency
in
this
point?
What's
the
intensity?
What's
the
average
duration,
and
so
those
maps
on
the
left
are
showing
you
the
annual?
C
These
are
marine
heatwave
days
now
so
the
number
of
marine
heatwave
days
per
year.
So
you
can
see
that
there's
more
more
marine
heat
waves
in
the
central
equatorial
eastern
equatorial
pacific
and
more
at
high
latitudes.
C
The
middle
plot
on
the
left
is
now
the
average
marine
heat
wave
intensity
and
the
bottom
is
the
average
marine
heat
wave
duration
and
these
patterns
look
quite
familiar.
So
if
you're
looking
at
sst
variability
so
panel
e
over
here,
you
can
see
that
that
looks
a
lot
like
the
marine
heat
wave
intensity
and
so
in
the
bottom
panel,
the
duration
maps
pretty
well
on
to
the
memory
time
scale
of
ssts
in
the
ocean
and
so
because
we're
defining
marine
heat
waves
from
this
variability
in
the
sea
surface
temperature.
C
It's
not
surprising
that
these
patterns
come
out
to
you
all
right
so
so
this
is
a
nice
way
to
kind
of
look
at
discrete
events
that
have
occurred,
and
so
here
is
a
a
composite
image.
So
they're
averaging
the
marine
heat
wave
anomalies
for
different
events
and
they're
fining
these
events,
by
taking
the
average
sea
surface
temperature
anomaly
in
each
of
these
gridded
boxes
and
so
they're.
Finding
when
that
temperature,
the
anomaly
time
series
is
max
and
then
they're
grabbing
the
spatial
field
in
that
area.
C
So
essentially,
this
is
just
a
snapshot
of
several
different
marine
heat
waves
that
have
occurred
in
the
21st
century
that
you
know
had
impacts
and
were
well
described,
and
so
you
can
see
that
a
lot
of
them
tend
to
be
coastal
and
they're
they're
quite
large.
So
this
kind
of
gives
you
a
sense
of
the
scale
of
these
events
and
their
locations.
C
Okay,
so
I
wanted
to
ask:
what
does
this
characterization
miss?
Looking
at?
You
know
a
single
point
in
drawing
out
these
marine
heat
wave
statistics,
and
one
way
to
kind
of
draw
you
towards
that
answer
is
to
examine
this
animation,
showing
the
marine
heat
wave
anomalies
evolving
over
this
five-year
period.
C
Where
this
x
is
the
point
of
that
time
series
I
showed
earlier
where
we
were
looking
at
the
marine
heat
wave
statistics,
and
so
what
you
see
is
that
okay,
so
you
you
see
el
nino,
obviously,
but
these
really
warm
blobs
of
marine
heat
waves,
don't
just
stay
in
one
place.
They
are
moving
around
they're,
passing
through
that
point
and
they're
leaving,
and
so,
if
you're,
just
looking
at
a
single
point
or
a
fixed
boxed
average
region,
you
might
be
missing
part
of
the
story.
C
Well,
it's
either
it's
due
to
their
forcing
so
it's
either
air
sea
heat
fluxes,
causing
that
warming
to
evolve
ocean
heat
transport
and
or
the
influence
of
remote
tele
connections,
and
so
now
I
want
to
turn
the
the
discussion
over
to
the
drivers
and
marine
heat
waves
to
understand
how
they
are
moving
around
the
ocean,
and
this
is
a
schematic
that
I
worked
with
neil
holbrook
from
the
university
of
tasmania
to
produce,
and
so
we
comb
through,
I
don't
know
hundreds
of
papers
to
kind
of
look
through
this
to
find
what
the
drivers
of
marine
heat
waves
are
and
then
map
them
onto
both
their
time
scale
and
their
spatial
scale.
C
And
so
this
is
what
we
found.
This
is
kind
of
a
broad
generalization
of
what
drives
marine
heat
waves,
everything
from
the
smaller
scale
process
level
to
the
large-scale
climate
drivers.
This
talk,
I'm
I'm
kind
of
avoiding
or
ignoring
the
long-term
climate
change.
Warming
signal,
that's
a
different
talk,
but
I
just
want
to
now
focus
in
on
some
of
these
different
scales
so
down
in
the
lower
left
corner.
These
are
local
processes
that
affects
the
mixed
layer.
C
Temperature
heat
budget
and
the
heat
budget
is
a
very
common
tool
that
we
use
to
look
at
the
drivers
of
marine
heat
waves.
So
let's
look
at
what
the
heat
budget
is.
I'm
sure
most
of
you
are
very
well
familiar
of
this,
but
for
those
who
aren't
you
have
the
temperature
within
the
next
layer
is
controlled
by
a
number
of
processes.
C
The
first
is
likely
horizontal,
advection,
so
heat
coming
in
or
out
of
this
mixed
layer
volume
you
have
horizontal
mixing,
whether
that's
mesoscale,
and
then
you
have
vertical
mixing
within
the
base
of
the
mix
layer
up
into
the
in
the
upper
layer
and
entrainment.
C
C
C
So
here's
a
study
done
by
eric
oliver
and
his
co-authors,
where
they
fix
a
box
in
the
tasman
sea
with
a
fixed
depth,
mixed
layer
during
a
marine
heat
wave,
and
so
what
they're
doing
is
a
a
integrated
heat
budget.
From
a
the
point
of
when
the
marine
heat
wave
starts
and
so
on
the
bottom
you
have
that
temperature
tendency
integrated
from
september
2015
through
april
2016,
and
the
total
temperature
tendency
is
showing
the
black
line.
C
Okay.
So
another
nice
study
by
dylan
maya
used
another
heat
budget
in
the
north
pacific
to
look
at
the
summertime
drivers
of
the
2019
marine
heat
wave,
and
so
what
he
found
was
over
here
at
the
end
of
the
time
series,
there's
a
really
big
jump
in
the
radiation
of
heat
flux
terms,
and
then
that
was
really
what
was
responsible
for
driving
the
warming
which
you
can
see
in
the
blue
line.
In
this
boxed
region,
an
advection
really
played
a
very
miniscule
real
role.
If
any
another
way
you
can
go
about.
C
So
here's
another
paper
by
key
chen
at
all,
looking
at
the
contributions
to
warming
along
the
mid-atlantic
bite
during
the
2012
marine
heat
wave,
and
so
what
he
found
here
was
that
again,
the
air
sea
heat
fluxes
contributed
the
greatest
to
that
the
warming
in
this
region,
and
it
really
mapped
on
spatially
to
where
you
had
high
positive
q,
fluxes
and
temperature
change,
where
the
advection
slightly
cooled
the
region
and
it
was
weaker
okay,
so
those
are
some
of
the
the
ways
that
we
can
look
at
these
local
processes
affecting
marine
heat
waves
at
a
particular
location.
C
Now,
let's
think
about
how
climate
modes
large
scale,
you
know
what
atmospheric
blocking
or
ocean
waves
influence
marine
heat
waves
and
one
study
that
really
comes
to
mind.
Is
this
paper
by
delorenzo
mantua,
looking
at
the
to
see
if
there
is
a
dynamical
linkage
between
this
multi-year
persistence
of
anomalies
in
northeast
pacific?
So
this
is
their
their
climate
hypothesis
for
the
generation
and
evolution
of
north
east
pacific
anomalies
during
the
winter
2014
and
2015
on
the
right
here
and
then
2014
2013
2014.
C
Excuse
me
on
the
right,
2014
and
2015
on
the
left,
and
so
what
they
found
was
that
there
were
teleconnection
dynamics
at
play
between
the
extra
tropics
that
connected
to
the
tropics
and
then
back
into
the
extratropics
that
contributed
to
the
multi-year
persistence
of
these
anomalies.
And
so,
during
the
winter
of
2013
and
2014.
C
Those
temperature
anomalies
were
generated
by
a
persistent
ridge
of
high
pressure
over
the
north
atlantic
and
that
was
influenced
or
attributed
to
the
north
pacific
oscillation,
which
is
the
leading
pattern
of
atmospheric
variability
that
that
p
npo
response
is
one
precursor
for
el
nino
by
kind
of
tickling,
the
meridianal
modes
through
the
wind,
evaporation,
sst
feedback,
and
so
that
can
produce
a
positive
response
to
el
nino,
where
it
would
develop.
C
Those
rossview
waves
had
an
effect
on
the
aleutian
low,
so
dilution
low
was
deeper
and
more
southeast
and
and
that
mechanism
can
then
ignite
a
positive
pdo
like
pattern.
C
Now,
one
of
the
interesting
things
that
they
wanted
to
to
know
is
how
are
these
patterns
associated
with
these
marine
heat
waves
going
to
change
in
the
future,
and
so
what
they
did
was
they
looked
within
the
cesm
large
ensemble
for
at
these
marine
at
these
sst
anomalies,
and
then
they
computed
the
eofs
of
the
sea
surface
temperature
anomalies
in
the
north
pacific
during
january
february
march,
and
what
they
found
was
that
the
first
two
modes
refers
to
empirical
orthogonal
functions
resembled
a
lot
of
resembled
the
patterns
they
saw
in
the
marine
heatwave
footprint,
and
so
using
the
pcs
of
these
aos.
C
They
were
able
to
see
how
these
patterns
will
change.
The
amplitudes
will
change
into
the
future
okay.
So
this
is
one
way
to
kind
of
diagnose
the
evolution
of
the
marine
heat
wave
anomalies
and
to
look
at
these
different
patterns.
C
Looking
at
the
different
eofs
or
just
looking
at
the
maps
of
the
forcing
my
my
project
really
wanted
to
to
kind
of
take
a
more
unusual
approach
to
look
at
the
evolution
and
so
for
the
remainder
of
the
talk,
I'm
going
to
introduce
a
tool
that
I
call
osh
track
to
essentially
identify
marine
heatwave
objects.
I
call
them
and
follow
them
throughout
the
ocean.
C
Okay,
so
some
existing
techniques
that
have
been
used
for
object,
tracking
and
identification,
one
is
called
the
watershed
method,
and
so
this
method,
here's
an
example
in
david
john's
paper,
looking
at
hill
forecasting,
so
it
identifies
a
local
maximum
and
I
think
it's
like
total
wall,
total
water
column
grapple
and
from
that
maximum.
It
grows
the
objects
outward
until
they
reach
some
type
of
area
or
intensity
criteria.
C
Another
tool
that
exists
out
there
that
I
actually
used
early
on
and
what
helped
motivate
my
work
was
called
tempest
extremes,
and
this
is
a
a
tool
that
is
used
to
track
tropical
cyclones
and
it
has
several
different
kernels.
It
uses
to
detect
points
of
interest
in
the
data,
so,
whether
it's
extrema
or
it's
thresholding,
it
finds
closed
contours
around
those
points,
and
then
it
has
a
way
of
stitching
it
together
to
track
it
in
time.
C
One
thing
to
note
about
any
of
these
methods
is
that
there
are
often
parameters
involved
when
you
are
setting
thresholds
or
different
radiuses
of
of
the
object.
You're
looking
for
and
those
parameters
can
change
the
composition
of
your
object
set,
and
so
it's
important
not
just
to
take
one
set
of
parameters
and
run
with
it.
But
to
do
an
ensemble
approach
where
you
can
then
look
at
the
sensitivities.
C
C
Okay,
the
second
step
is
object,
detection.
So
from
those
candidate
points
we
determine
if
there's
a
closed
contour
that
exists
around
a
cluster
of
candidate
points,
and
for
that
we
use
morphological
operations
which
I'll
I'll
go
into
more
detail
using
morphological
operations.
We
can
then
remove
those
small
or
sorry
that's
not
after
we
do
the
more
orthogonal
operations.
We
then
remove.
C
C
Okay,
so
now
I'm
going
to
walk
through
the
visual
description
of
those
jack.
That
was
just
you
know,
words
before
so
we're
using.
For
this
example,
sea
surface
temperature
from
the
oh
data
set
again,
it's
monthly
average
monthly
from
september
1981
through
present,
and
the
first
step
is
to
extract
those
features
like
I
said.
So.
C
We
want
to
essentially
dissolve
away
any
of
the
points
that
don't
meet
our
criteria
for
a
marine
heat
wave
and
so
that's
set
at
the
90th
percentile
and
then
we're
going
to
just
take
that
map
of
the
candidate
points
and
feed
it
into
our
object
detection,
and
so
just
to
remind
you,
we've
removed
to
get
the
anomalies
on
the
bottom
map.
We
removed
the
mean
seasonal
cycle
and
trend
computed
over
1982
to
2020,
and
then
we
standardize
those
anomalies
by
the
local
standard
deviation
to
account
for
differences
in
variants
in
the
ocean.
C
So
after
we
have
our
features
that
we're
interested
in
working
with,
we
do
object,
detection
and
just
as
an
example,
I'm
going
to
pull
out
this
feature
in
the
northeast
pacific
since
we're
kind
of
talking
about
the
blob
okay.
So
this
is
kind
of
what
our
input
looks
like
to
ostrack
it's.
This
pixelated
image
with
you
know
points
of
interest
that
we
want
to
track.
C
So
more
morphological
operations
is
a
series
of
kind
of
image.
Manipulation
where
you
are,
you
know
dilating
or
eroding
the
image
to
shape
it
in
a
way
that
you
want,
and
so
closing
is
a
series
of
dilation
and
erosion,
and
so
you
can
see
now
we
turn
this
image
that
we
start
with
into
a
binary.
So
all
of
the
red
pixels
become
one
all
the
white
becomes
zero,
and
so
morphological
operation
only
works
on
binary
images.
C
We
thought
most
represented
what
they
should
look
like
from
the
original
image,
okay,
so
morphological
operations.
I
thought
I
just
walked
through
an
example
if
people
are
not
familiar
with
this,
the
first
thing-
that's
really
important
to
remember-
is
that
we're
working
with
binary
images
so
as
an
example
up
here
on
the
top,
this
kind
of
funky
shape
here
we're
just
going
to
say
that
is
our
binary
image.
So
you
can
imagine
everything
inside
of
that
shape
is
a
one.
C
Everything
outside
is
a
zero,
and
then
we
also
have
to
define
s,
which
is
a
structuring
element,
and
this
element
can
be
any
type
of
shape
or
size.
You
want,
for
our
purposes
we're
using
a
radial
looking
structuring
element
where
you
can
change
the
the
radius
of
it
to
suit
your
to
kind
of
tune
it
if
you
will
and
what
this
structuring
element.
Its
purpose
is
to
be
be
used
as
a
mask
to
pass
over
your
image
to
erode
and
dilate
your
binary
image,
and
so
for
erosion.
This
mask
is
split.
C
It
slides
across
the
image
at
the
origin
around
across
your
across
your
image.
So
if
you,
if
you
plop
this
structuring
element
on
top
of
your
image
and
you
you
pass
it
along
wherever
there's
a
one
in
your
structuring
element
and
a
one
in
your
image,
the
pixel
at
the
origin,
where
it's
overlapping
between
the
structuring
element
in
your
image
would
then
become
a
zero,
and
so
essentially
this
and
this
this
continues
for
the
entire
image.
C
C
So
let's
see
this
on
our
data,
so
here
is
that
image
that
we
have
after
we've
extracted
our
features
before
we
do
anything
we're
going
to
turn
this
into
a
binary
right,
so
one's
our
features,
zeros
or
the
background,
so
closing
is
a
dilation
followed
by
erosion.
So
you
can
see
now.
These
features
are
more
connected,
they're
they're,
much
more
pronounced
than
in
our
original
image,
and
then
erosion
shrinks
those
that
those
objects
back
down
to
size.
C
After
closing,
we
take
that
image
on
the
right
and
we
pass
it
into
opening,
so
we
erode
it
even
further
and
then
and
then
dilate
it,
and
so
now
this
final
dilation
two
is
the
mask
for
our
original.
C
C
Okay,
so
after
we
have
all
of
those
closed,
contour
objects
in
our
data
set,
we
can
go
through
and
calculate
the
area
of
each
one
and
and
then
what
we
do
is
we
create
a
distribution.
C
We
look
for
the
75th
percentile
of
that
distribution,
and
that's
what's
shown
here
as
this
vertical
dashed
orange
line,
and
we
keep
all
of
the
objects
with
an
area
that's
greater
than
that
75th
percentile,
and
so
you
can
see
you're
really
retaining
most
of
your
objects,
but
what
you're
doing
is
you're,
eliminating
these
really
small,
noisy
ones
that
you
know
we're
just
not
that
interested
in
looking
at
these
might
correspond
to
more
eddy-like
features,
and
this
choice
is
completely
up
to
you,
and
so,
as
of
now,
there's
really
two
choices
that
you
have
to
make
is:
one
is
the
size
of
that
structuring
element
the
shape
even
and
the
other
one
is
what
threshold
you
choose
for
your
size
criteria,
so
once
we
we
filter
based
on
the
size
that
now
we're
left
with
these
static
images
in
our
data
set.
C
C
Essentially
their
ids
will
be
joined,
and
so
we
define
this
connectivity
element,
it's
kind
of
like
the
structuring
element,
but
now
we're
in
3d
to
be
able
to
track
these
marine
marine
heatwave
objects
in
our
data
set.
One
of
the
unique
things
about
this
is
that
it
remembers
when
marine
heat
wave
objects
will
merge
or
when
they
split,
and
so
it
keeps
track
of
these
events
as
they
break
off
or
come
together.
C
Okay,
so
let's
look
at
an
example
of
of
a
marine
heat
wave,
so
this
is
in
the
gulf
of
maine.
In
2012
there
was
a
marine
heat
wave
and
we
ran
ostrack
on
the
data
and
found
what
sequence,
what
what
object
most
resembled,
what
we
thought
would
be
the
2012
marine
heat
wave
and
so
playing
this
movie
four.
You
can
kind
of
see
this
blob
kind
of
evolving
in
space,
so
it's
really
staying
put
in
the
northwest
atlantic.
It's
not
no
branching
off
and
going
to
the
tropics
or
south
the
south
atlantic.
C
So
it's
a
really
localized
event.
We
know
that
it
was
driven
by
the
atmosphere
in
this
region
and
the
the
plot
on
the
bottom
right
is
just
showing
you
the
overall
footprint
of
this
event
in
terms
of
its
cuen
of
intensity.
So
it
gives
you
an
idea
of
where,
where
it
was
greatest,
the
impacts
were
greatest.
C
We
can
also
look
at
the
centroid
so
for
each
object
contained
in
a
marine
heat
wave,
so
each
snapshot
in
time.
We
can
compute
the
centroid
of
that
marine
heat
wave
to
form
more
of
a
trajectory,
and
so
you
can
see
that
this
marine
heat
wave
really
didn't
go
too
far.
It
was
in
the
same
area,
it
moved
kind
of
towards
land
and
then
off
the
coast
there
a
little
bit
before
coming
back,
and
so
it
was
a
pretty
localized
event.
C
The
next
one
I
want
to
show
is
the
blob
and
then
and
the
pacific
ocean
and
you'll
see
that's
a
much
broader
scale
event
and
what's
really
fascinating,
is
that
this
marine
heat
wave
as
it
evolves?
You
know
right
now.
You
can
see
that
kind
of
more
typical,
canonical
pattern.
You
see
this
big
el
nino
developed,
but
what's
really
unique,
is
that
we
were
able
to
trace
that
warmth
all
the
way
back
to
the
western
south
pacific.
C
You
can
see
it
kind
of
continues
off
into
the
south
atlantic,
but
all
of
these
blobs,
as
the
movie
is
playing
and
moving
around
they're
all
connected.
Somehow,
it's
not
to
say
that
the
dynamics
are
connected
in
any
way,
but
that
this
pattern
of
a
warming
progressed
in
this
such
map
and
a
matter
like
this
okay,
so
here
in
the
bottom
right,
is
the
queue
of
intensity
for
this
event,
and
so
you
can
see
that
it
really
had
most
of
the
it
lingered
most
in
the
northeast
pacific
in
the
el
nino
region.
C
But
you
can
see
kind
of
the
how
far
and
wide
it
had
an
effect
here
by
these
lighter
colors.
C
And
then
I
mean,
if
you
look
at
the
centroid,
it's
all
over
the
place,
but
you
can
trace
it
back
through
the
development
of
the
el
nino
kind
of
like
that
dilorenzo
paper,
and
so
it's
all
makes
sense
in
terms
of
dynamics.
But
this
is
showing
you
kind
of
the
pathway
it
took
and
its
evolution,
and
so
we
calculate
this
event
to
be
a
53
month.
Event,
which
is
much
you
know
on
par,
but
a
little
longer
than
what
others
described.
C
So
here
this
is
the
maximum
intensity
for
each
marine
heat
wave
identified,
and
so
you
can
see
that
it's
a
bit
positively
skewed
with
some
intense
events,
but
most
they
kind
of
kind
of
congregate
on
the
three
degrees
c.
C
And
then,
if
you
plot
each
event
as
a
on
a
scatter
plot,
you
can
see
comparison
between
the
gulf
of
maine
and
the
blob
event
and
the
northeast
pacific.
So
the
blob
was
just
a
lot
more
intense
than
the
gulf
of
maine
event,
which
was
really
regional.
C
We
could
do
the
same
for
duration,
and
so
the
duration
is
as
a
very
heavy
tale,
and
so
a
lot
of
events
are
more
in
the
you
know
two
to
four
month
range,
but
there
are
some
that,
like
the
blob
that
went
on
for
a
very
long
time
that
was
connected
through
the
tropics,
and
so
that's
shown
again
here
on
the
bottom
panel,
the
difference
between
the
gulf
of
maine
and
the
blob.
C
So
the
gulf
of
maine
was
a
much
shorter
event,
the
blob
really
persisted
for
much
longer
and
then
finally,
we
can
calculate
the
area.
So
this
is
the
total
area
over
the
lifetime
of
the
event,
and
so
again
it's
also
heavy
tailed
with
a
lot
of
you
know
smaller
sized
marine
heat
waves,
but
there
are
very
few
really
massive
marine
heat
waves
like
we
saw
the
blobs.
The
blob
was
quite
unusual
in
its
duration
and
area
and
even
intensity.
C
Okay,
so
I
just
want
to
pop
this
slide
on
here,
because
I
talked
a
little
bit
about
choice
of
parameters,
and
so
the
two
that
we're
looking
at
for
osh
track
is
the
size
of
that
structuring
element.
So
that's
the
smoothing
radius
on
the
y-axis
and
then
that
minimum
size,
percentile
and
that's
shown
on
the
x-axis.
C
So
what
this
is
showing
is
that
when
you
have
larger
structuring
elements
and
a
larger
percentile
you're,
just
your
number
of
marine
heat
waves
is
is
less
essentially,
you
have
less
marine
heat
waves
and
we
think
that's,
that's
probably
because
you
have
less
marine
heat
waves,
but
they're
also
really
big
marine
heat
waves,
because
that
structuring
element
is
connecting
more
of
the
the
features
that
we
extracted
from
it.
C
Duration.
Also,
it
increases
as
you
increase
your
structuring
element,
but
it
actually
isn't
as
linear,
because
when
you
increase
your
percentile,
you
just
don't
have
as
large
of
a
population
size
to
calculate
the
duration.
So
you
see
that
it
kind
of
peaks
more
at
a
smaller
area
threshold.
C
And
so
we
we
did
this
to
look
at
the
sensitivity
of
the
parameters
and
it
actually
helped
us
kind
of
land
on
the
ones
that
we
used
to
look
at
the
marine
heat
waves
and
we
also
conduct
this
study
using
different
thresholds
for
marine
heat
waves
and
different
resolutions
of
data
set
okay.
So
what
new
insights
does
ostrack
reveal
about
marine
heat
waves?
C
Well
from
what
I've
shown
you
there's
really
two
different
types
of
marine
heat
wave
patterns,
so
there's
these
localized
events
that
kind
of
grow
and
decay
in
the
same
general
location.
So
that's
that's
most
typical
of
the
gulf
of
maine
marine
heat
wave
and
then
there's
these
really
large
dynamically
linked
marine
heat
waves
that
can
move
connected
through
the
tropics
and
that
was
the
blob
was
an
example
of
that
type
of
marine
heat
wave.
C
So
these
globally
tropical
connected
marine
heat
waves,
they
tend
to
last
longer
and
and
then
I
also
showed
how
these
marine
heat
waves
can
split
and
merge
and
come
back
together.
Okay,
so
why
should
you
use
ostrack?
Well,
I
think
that
we've
overcome
some
limitations
of
characterizing
the
complex
facial
connectivity
and
temporal
behavior
of
marine
heat
waves.
By
coming
up
with
this
tracker,
we're
not
just
now
relying
on
a
single
point
or
an
average
analysis
over
some
domain,
we're
kind
of
letting
the
anomalous.
C
Self
move
around
this
potentially
provides
catalog
for
observing
humans
that
others
use,
and
you
can
apply
oats
track
to
other
phenomenon.
It's
not
just
limited
to
extremes
in
temperature,
but
you
could
look
at
oxygen
minimum
zones,
which
is
work
that
that
I
might
be
doing
next
and
then
looking
at
harmful
algal
blooms
or
even
trying
to
track
the
temperature
anomalies
related
to
el
nino
around
the
pacific-
and
this
is
my
last
slide
here
before
questions,
and
I
just
wanted
to
pop
this
up
here
in
case.
C
Anyone
is
interested
in
using
osh
track
or
wants
to
get
involved.
That
would
be
great
if
you
try
those
track.
If
you
found
something
that
wasn't
working
and
submitted
an
issue
on
the
github
page
or
even
improved,
it
would
be
even
better
or
you
can
email
me
directly
and
I
can
help
set
you
up,
there's
going
to
be
another
talk
related
to
ostrack
at
scipy
in
july,
and
that
will
be
more
demo
technically
faced.
So,
if
you're
interested
in
learning
more,
I
encourage
you
to
go
to
sci-fi.
A
Thanks
so
much
hilary
for
a
really
nice
talk,
thanks
for
describing
your
work
on
nose
track,
so
we
have
about
10
minutes
for
questions,
which
is
great.
So
as
always,
if
you
have
questions
for
hillary,
you
can
type
them
in
the
chat
and
I'll
read
them
out
loud
or
you
can
use
the
raise
hand,
feature
and
zoom
and
ask
your
question
out
loud.
A
I
guess
I'll
start
with
a
with
a
question
I
had,
which
you
you
sort
of
answered
at
the
end,
which
is
using
it
for
other
phenomena,
and
so
I
was
just
wondering
if
you
had
even
some
initial
thoughts
on
how
the
process
might
be
different
when
you're
looking
at
things
like
algal
blooms
or
other
other
types
of
extreme
events.
C
Yeah,
so
it
really
depends
on
the
data
set.
You
have,
I
think,
so
what
the
resolution
is
both
spatially
and
temporally.
C
What
I
think,
having
some
a
priori
knowledge
of
the
phenomenon
you're
trying
to
characterize,
is
very
helpful.
So
for
me,
with
marine
heat
waves,
I
knew
I
wanted
to
characterize
these
really
the
big
events,
not
I
don't
want
to
look
at
eddies
or
like
smaller
marine
heat
waves
and
so
having
an
idea
of
what
you're
looking
for,
is
helpful
and
that
will
help
inform
your
decisions
of
the
parameters
to
choose.
C
So
the
things
like
the
structuring
element
and
your
size
criteria,
and
so
I
think,
there's
just
a
lot
of
playing
around
you-
can
do
with
those
criteria
to
adjust
it
to
kind
of
you
tune
it
to
see.
If
what
you're
getting
out
looks
to
what
you're
the
phenomena
you're
trying
to
identify
and
track,
but
I
could
see
using
it
eventually,
it
doesn't
necessarily
work
on
satellite
data
currently,
but
there's
many
ways
for
it
to
expand
to
that
level
as
well.
A
Thanks
yeah,
I
thought
those
sensitivity.
Plots
were
also
really
interesting.
Where
you
talked
about
the
different
parameter.
Choices
looks
like
we
have
a
question
from
bill.
Large
go
ahead.
B
Yeah
thanks
very
much
yeah
there's
a
lot
of
applications
for
this
sort
of
stuff
and
and
you
you
had
many
instances
to
use
thresholds
to
make
decisions
and
you
use
90
kind
of
things.
Is
there
any
literature
or
a
lot
of
use
where
it
it's
actually
different
criteria
to
get
out
of
a
characterization
that
will
go
in
so
it's
might
be
hard
to
get
in,
but
it's
easier
to
stay.
B
You
know
what
I
mean.
Well,
I
don't
so.
What
do
you
mean?
Well,
like
you
might
say,
I
need
a
90th
percentile
to
be
characterized,
but
if
I'm
already
characterized,
I
might
have
to
get
maybe
I'll
I'll
be
okay
with
80
percent,
it's
easier
to
get
in
the
it's
hard
to
get
in,
but
it's
easier
to
stay
you,
don't
you
don't
running
across
that
idea
at
all?
I
I
don't
know,
I'm
just
that's
my
question.
B
C
I'm
not
sure
I
understand
completely
crossing
up
yeah,
I
mean
yes.
C
Yeah,
I
I
actually
heard
the
beginning
part
of
the
question
and
I
I
think
it
was
somewhere
like
once.
You've
set
your
threshold
for
the
marine
heat
wave
say
at
90
percentile.
It
was
something
related
to
what
you
could
get
out
of
it.
B
Everything
you
did,
it
had
to
have
that
same
threshold
to
stay,
but
there
might
be
applications
where
you
say
well,
once
I'm
in
a
state,
it's
it's
it's
hard
to
get
into
the
state,
but
it's
easier
to
stay.
That
was
the
way
and
I'm
just
wondering
if
you
run
across
that
idea
or
or
used
it
or
thought
about
it,.
C
It
could
be
pretty
valuable
yeah,
so
easy
is
it
for
you
to
stay
above
that
threshold
for
that
marine
heat
wave.
I'd
have
to
think
about
it,
a
little
more
yeah
and
also
too
like
with
with
the
morphological
operations,
because
it's
adding
and
removing
different
pixels
around
the
perimeter
of
the
objects
it's
it
could
potentially
be
adding
pixels
that
didn't
meet
your
criteria
to
begin
with,.
B
D
Yeah
hi
hillary
thanks
a
lot
for
this
introducing
this
tool
and
we
had
been
working
also
on
similar
kind
of
tool
for,
but
for
different
application,
like
a
drought,
identification,
for
example,
to
track
the
space-time
evolution
of
the
drought
and
the
similar
sort
of
object
and
similar
sort
of
issues
arises,
and
you
clearly
identified
or
said,
like.
Okay,
the
choice
of
the
parameter
of
splitting
and
merging
also
makes
big
differences
removal
and
we
should
be
very
cautious
about
its
use
in
the
in
any
kind
of
trend.
D
Analysis,
for
example,
how
many
number
of
events
that
could
increase
in
future?
That
would
very
much
depends
on
what
parameters
that
we
fix
it
in
this
removal.
One
one
thing
I
was
just
wondering
about
once
you
have
the
identification
of
the
clusters,
then
how
are
you
splitting
and
joining
this
cluster?
To
tell
that?
This
is
a
one
event.
D
C
Yeah,
so
that's
a
question:
we
run
the
tracking
essentially
forward
and
backwards
along
the
data
set,
and
so
it
has
a
way
of
remembering
which
objects
have
merged
or
split
together
and
so
yeah,
quite
quite
simply,
you're
tracking
forward
and
you're
tracking
backwards
and
then
you're
matching
all
those
together.
C
C
Just
yeah
they
just
touch
on
the
you
know,
spatially
or
temporally
yeah
tempest
extremes.
Now
they
do
have
a
an
overlap
criteria
that
you
can
set.
It's
like
percent
overlap,
and
so
you
can
set
that
to
be
whatever
you'd
like.
D
A
A
C
C
C
Those
are
really
that's
a
good
question,
but
I
think
that's
where
your
frontal
boundaries
are,
and
so
those
marine
heat
waves,
any
kind
of
perturbation
in
that
position
of
that
current
can
create
a
really
intense
marine
heat
wave
but
they're,
very
energetic
regions
where
you
know
things
like
eddies
and
and
things
like
that
can
move
through
pretty
quickly,
and
so
the
duration
is
not
as
as
high
in
those
regions.
C
That's
my
understanding
of
it,
I'm
not
sure
like.
When
we
did
the
literature
review.
There
were
not
a
lot
of
examples
of
marine
heat
waves
in
the
western
boundary
that
was
actually
kind
of
a
gap
and
so
yeah.
I
think
it
would
be
interesting
for
someone
to
do
some
kind
of
analysis
looking
at
eddies
and
detected
marine
heat
waves
and
their
co-occurrence.
A
D
So
thanks
anyway,
if
I
understood
it
in
one
of
your
slides,
you
said
that
okay,
well,
you
identify
the
anomaly
with
respect
to
removing
the
seasonal
cycle,
and
then
you
remove
the
trend.
Did
I
got
it
right
that
you
removed
the
trend
from
the
before
calculating
the
90th
percent
and
what
right?
It's
a
really
basic
idea
trend.
Sorry,.
C
Yep,
so
we
removed
the
trends
with
the
seasonal
cycle,
essentially
so
you're
de-trending
the
data
and
then
you
compute
the
90th
percentile
on
that
de-trended
data.
So
we
wanted
to
avoid
the
situation
where
you
would
just
have
a
global
marine
heat
wave
because
of
the
trend
essentially,
and
this
tracking
it
wouldn't
quite
work.
C
So
so
it's
it's
basically.
D
C
Yeah,
I
think
that
you've
cut
out
a
little
bit
yeah,
so
it
we
are
not
looking
at
the
trend,
it's
more
of
the
the
variability
of
the
the
natural
variability
of
the
marine
heat
waves.
D
D
C
C
Yeah,
that's
true,
but
I
also
think,
even
when
we
remove
the
trend
that
the
blob
in
the
northeast
pacific
really
still
stood
out
to
be
quite
unusual,
and
so
it's
it's,
maybe
not
as
important
to
to
can
you're
still
able
to
capture
these
events.
A
Did
you
see
that
peter
you
have
your
hand
up?
Do
you
have
a
quick
question
before
we
before
we
conclude.
B
C
Yeah
I
so
australia
cannot,
but
I
am
that
one.
C
Those
those.
B
C
C
You
know
atmospheric
very
views
that
area
experience
rain
heat
wave.
It
doesn't
diagnose
any
of
the
drivers
quite
yet.
A
C
Sure
yep
so
ostrack
does
not
diagnose
drivers,
but
it
is
a
an
additional
data
set
that
could
be
used
as
a
mask
to
then
go
back
and
look
at
some
of
the
other
processes
going
on
in
those
regions
that
could
be
contributing
to
the
marine
heat
wave
in
that
location.
A
So
thanks
and
thanks
peter
for
the
question
I
had
a
similar
question,
so
I
think
we're
a
little
bit
over.
I
think
we'll
we'll
conclude,
but
thanks
again
hillary
for
a
really
nice
talk
and
introducing
us
to
your
work.
Next
week
we
will
hear
from
danielle
tuma
from
the
university
of
california,
santa
barbara,
so
hope
to
see
you
then.