►
Description
The 26th Annual CESM Workshop will be a virtual workshop with a modified schedule on its already scheduled date. Specifically, the virtual Workshop will begin with a full-day schedule on 14 June 2021 with presentations on the state of the CESM; by the award recipients; and three invited speakers in the morning, followed by order 15-minute highlight and progress presentations from each of the CESM Working Groups (WG) in the afternoon.
On 15-17 June 2021, working groups and cross working groups have half-day sessions, some with presentations and some that are discussion only.
E
A
Yeah
yeah:
well,
it
looks
like
I'm
the
last
post
there
well
todd's,
just
reposted
it,
but
yeah.
It
should
be.
B
H
A
G
A
G
A
I
was
just
I
was
just
teasing
because
it
was
like
sea
ice
prediction.
Predictability
is
leading
the
way
so
oh
well,
marika's
voted
so.
A
A
I
J
G
E
M
C
F
Yeah,
my
my
team,
belgium,
is
playing
in
an
hour
and
a
half.
So
oh
I'm
getting
nervous.
I
C
Now
in
the
netherlands
it
will
be
tonight,
but
for
me
it
will
be
yeah
tomorrow,
yeah,
due
to
the
time
difference.
Why.
I
C
Yeah,
it
was
a
bit
yeah
yeah.
It's.
K
F
Okay,
everyone,
I
think,
it's
time
to
get
started
there.
We
go
recording
in
progress,
so
this
is
being
recorded.
Welcome
everyone.
This
is
the
polar
climate
and
landis
working
group
session.
F
My
name
is
john
denards,
I'm
from
the
university
of
colorado
boulder
and
I'm
co-chair
of
the
landis
working
group
together
with
bill
lipscombe,
and
then
we
have
co-chairs
marika,
holland
and
hansi
singh,
and
also
our
wonderful
liaisons,
aldous,
duvivier,
dave,
bailey
and
gunter
leggy,
so
in
name
of
all
of
them.
Welcome
ver
welcome
to
all
of
you
good
morning,
good
afternoon,
good
night,
depending
on
where
you
are.
F
F
B
I've
just
posted
the
full
pdf
of
the
code
of
conduct
in
the
chat.
F
There
we
go.
Thank
you
hansi.
So
next
up
is
a
really
important
item
for
this
morning.
We're
trying
to,
even
though,
if
even
though
this
is
a
virtual
meeting,
we're
we're
trying
to
make
it
interactive
so,
as
a
last
part
of
our
session
from
11
30
to
noon
mountain
time,
we're
going
to
have
breakout
sessions
and
you
can
choose
your
favorite
topic
for
the
breakout
session.
By
going
to
the
discussion
topic
topics,
google
form
that
is
posted
several
times
now
in
the
chat.
F
So
all
of
you
should
be
able
to
click
on
the
link
and
choose
your
favorite
topic.
So
there
are
options
already
there.
You
can
also
choose
your
own
topic
of
choice.
So
add
a
new
one,
and
apparently
you
can.
You
can
vote
multiple
times
to
you
know,
keep
it
interactive
and
going
so.
This
form
closes
at
10
am
at
the
end
of
the
break,
which
is
also
our
poster
session.
So
you
have
time
to
vote
on
your
topic
and
we'll
see
after
10
a.m.
F
We'll
see
what
topics
are
most
popular
and
choose
from
those
and
then
form
the
great
breakout
room.
So
we'll
we'll
come
back
to
this,
we'll
come
back
to
this
later
in
the
in
the
morning
and
then
we'll
also
have
google
docs
to
share
notes
to
compile
notes
and
to
share
them
with
other
groups
at
the
end
of
the
of
the
session,
to
really
compile
our
all
our
ideas
and
share
them
with
you
after
the
meeting
okay.
So
this
is
schedule
at
least
the
first
part
of
the
schedule.
F
So
we
have
four
blocks,
the
first
of
which
starts
in
two
minutes
shared
by
the
lipscomb,
and
then
we
have
the
poster
session
and
the
break
half
an
hour
from
9
30
to
10,
chaired
by
hansi
singh.
F
Then
we
have
two
more
presentations,
so
oral
presentation
blocks
one
starts
at
10
and
then
the
last
one
starts
at
11
and
then
the
last
block
is
the
one
way
to
break
out
sessions
and
then,
after
the
breakout
sessions
at
noon,
we
also
have
a
separate
lunch,
which
is
very
informal
for
the
polar
climate
and
the
land
eyes,
because
we
thought
you
know
being
together
is
great,
but
also
having
smaller
separate
groups
would
also
also
be
very
nice.
F
So
I
think
that's
it
in
terms
of
the
schedule.
What
I
do
want
to
give
away
to
the
speakers
is
that,
unless
this
is
something
we've
communicated
before
that
you
have
eight
minutes,
you
will
be
reminded
that
you
have
one
minute
left
after
seven
minutes.
So
please
make
sure
you
you
keep
your
time
to
eight
minutes,
and
then
we
have
two
minutes
for
questions
and
answers,
and
also
for
speaker
switching
if
you
have
your
slides
on
your
screen,
ready
and
you're
comfortable
sharing
your
screen.
F
F
If
you
have
questions
for
the
speakers,
please
raise
your
hand,
which
is
a
zoom
features.
I'm
sure
all
of
you
know
how
to
do
this
now
or
type
in
the
chat
and
we'll
keep
as
conveners
we'll
keep.
You
know
we'll
keep
track
of
questions
and
ask
them
afterwards.
If
there's
no
time
in
the
two
minutes,
we
can
also
try
to
reserve
some
time
at
the
end
of
each
block,
depending
on
what
how
the
time
goes.
F
Okay,
so
hanzi,
is
there
anything
else
you
wanted
to
add.
B
Thank
you
and
yes,
there's
one
final
thing,
and
this
is
regarding
computing.
So
many
of
you
have
reservations
or
have
have
chatted
with
us
about
running
on
the
polar
climate
working
group
computing
allocation,
and
so
we
would
love
to
see
the
cpu
hours
burning
and
the
machine
humming.
So
if
you
are
scheduled
to
run-
and
you
have
not
yet
started
your
run-
please
either
talk
to
us
about.
You
know
any
delays
that
are
going
on
or
get
your
run
started.
B
F
Okay,
thank
you.
Fancy
I'll
give
the
virtual
to
bill
now
have
fun
everyone,
and
if
you
have
any
questions
for
us
logistical,
please
put
them
in
the
chat
as
well
and
happy
to
help.
Thank
you
enjoy.
L
Thank
you,
yan,
and
I
will
go
ahead
and
introduce
our
first
speaker,
renee
weingart,
who
will
be
speaking
about
understanding
and
improving
the
representation
of
mountain
glaciers
in
cesm
you'd
like
to
go
ahead
and
share
your
screen.
Renee.
C
L
I
I
Here,
let
me
just
open
it
up.
Okay,
if
you
wouldn't
mind
sharing
time
and
what's
what's
the
name
of
it,
because
they
all
have
different
looking
names.
What
am
I
looking
for.
C
Okay,
it's
called
ally,
wg
session
workshop
and
then
zero
six
2021.
I
L
Thank
you,
todd
and
renee.
I
will,
as
jan
mentioned
I'll,
give
you
a
a
shout
out
when
you're
seven
minutes
in.
C
C
First
of
all,
I
would
like
to
thank
everybody
who
contributed
to
this
project
so
far
next
slide,
please,
I
will
give
a
yeah
first,
I
will
start
with
a
short
outline.
I
will
start
with
scientific
goals,
then
I
will
give
a
short
introduction
on
the
high
mountain
asia
grid,
and
then
I
will
go
into
more
detail
on
the
representation
of
mountain
glaciers
in
csm,
mainly
focusing
on
glacier
surface
mass
balance,
variable
elevation
profiles
and
updating
the
clam
glacier
dataset
next
slide.
C
Please,
okay,
the
main
goal
of
this
project
is
to
find
out
what
the
capability
of
an
esm
with
a
regionally
refined
grade
of
seven
kilometer
will
be
in
the
simulation
of
the
surface
mass
balance
of
glaciers
in
high
mountain
asia.
C
Main
reason
for
this
is
because
most
es
stamps
and
gcms
are
not
able
to
capture
glaciers,
and
that
is
why
it
will
be
interesting
to
find
out
whether
we
could
do
it
with
a
variable
resolution
model.
C
C
Do
you
hear
me
yeah?
Okay,
thank
you
oh
and
there,
but
something
with
the
figure
but
never
mind.
I
will
start
yeah
introduction
on
the
high
mountain
asia
grid,
so
we
started
more
than
a
year
ago
with
developing
a
grid
for
high
mountain
asia
with
a
116
degree,
refinement
over
the
high
mountain
asia,
mountain
ranges
and
the
tibetan
plateau
included.
C
We
in
order
to
able
to
be
able
to
assimilate
surface
mass
balance.
We
had
to
define
a
new
glacier
region
which
is
yeah
covering
the
high
mountain
asian
domain,
and
we
used
the
same
settings
for
greenland
as
well
as
for
antarctica.
But
the
difference
is
that
we
have
a
maximum
snow
depth
of
one
meter.
That's
actually
how
we
started
and
further
we
use.
We
decided
to
use
a
36
lfh
class
scheme
with
200
meter.
200
meter
elevation
bins
each
next
slide,
please.
C
The
main
outcome
of
the
first
simulations
that
were
complete
last
year
was
that
the
surface
mass
balance
we
simulated
that
it
delivered
quite
a
negative
surface
mass
balance.
Unfortunately
too
negative,
especially
over
the
southeastern
part
of
highmountain
asia,
and
there
was,
however,
also
some
positive
side.
C
Yeah
thing,
and
that
is
that
over
the
especially
the
southern
parts
of
the
high
mountain
asia,
as
you
can
see
in
the
inlet
that
there
were
some
regions
where
positive
surface
mass
balance
was
simulated
and
glaciers
can
in
fact
grow,
but
still,
however,
we
still
want
to
find
out
how
to
improve
the
negative
surface
mass
balance
next
slide.
Please.
C
So
because
of
that,
we
did
several
experiments
and
sensitivity,
runs,
etc,
and
based
on
that,
we
were
able
to
now
to
do
several
tunings.
The
first
is
that
we
increased
a
maximum
snow
depth
from
one
meter
to
five
meter
in
order
to
allow
for
more
refreezing
capacity,
we
increased
the
bear
ice
on
needle
to
to
have
less
ice
melt
and
we
widened
the
range
now
re-partitioning,
which
is
in
close
agreement
with
observations
over
high
mountain
asia.
C
Further,
we
used
adam's
variable
resolution,
tunings
on
cloud
cover
and
sea
ice
to
correct
underestimation
cloud
cover
over
high
mountain
asia
and
yeah.
That
goes
a
bit
too
fast
yeah
there
and
we
turned
off
the
long
wave
radiation
downscaling.
The
most
important
thing
is
that
we
did
also
a
new
spin
up
more
spin,
spin
up
approach
and
so
far
we
were
able
to
improve
the
signal.
C
The
the
surface
mass
balance
significantly,
as
you
can
see
in
the
lower
row
figures,
and
also
the
number
of
grid
cells,
for
example,
where
positive
surface
mass
balance
is
simulated,
is
now
quadrupled
next
slide,
please,
in
order
to
understand
more
especially
over
what
was
happening
in
the
southeast.
C
We
yeah,
we
went
more
into
the
model,
also
looking
on
grid
cells
and
to
see
also
the
the
elevation,
the
variable
elevation
relations
and,
as
you
can
see,
what
we
found
out
is
that
for
the
surface
temperature
that
we
found
some
some
temperature
inversion
above
around
four
or
five
thousand
meter
altitude.
C
Also,
we
found
that,
for
example,
for
the
lower
figure,
where
you
can
see
the
surface
mass
balance,
ice
flex
figure
that
there's
a
very
negative
surface
mass
balance.
I
choose
that
grid
cell
for
your
information,
but
this
is
mainly
probably
because
of
the
glaciers
that
have
ended
up
lower
than
actually
where
they're
supposed
to
be.
I
will
go
into
that
later
next
slide.
Please.
C
So,
in
order
just
to
start
with
the
the
temperature
inversion,
we
want
to
find
out
whether
it
is
a
real
thing
or
whether
it's
some
kind
of
modal
issue.
But
actually
it's
not.
We
found
out
that
what
what
is
happening
is
that
when,
when
you
go
to
higher
elevations
that
the
yeah
that
at
some
point,
snowpack
starts
to
develop
due
to
better
conditions
for
the
snowpack,
and
this
increases
the
surface
albedo,
which
results
in
a
decreased
net
short
wave
radiation
and
a
faster
decline
in
ground
heat
with
altitude.
C
The
second
thing
is
that
the
snowpack
functions
as
an
insulator
for
the
underlying
soil
and
therefore
causes
a
soil
temperature
inversion,
which
you
can
see
in
the
in
the
lower
grass,
but
also
in
the
table,
where
you
can
see
that
if
you
move
from
lfh
in
class
16
to
17,
where
there's
a
change
in
snowpack
that
there's
a
one
kelvin
increase
in
soil
temperature,
which
is
yeah
equivalent
to
the
surface
temperature,
you
saw
in
the
four
in
the
former
slide
next
slide.
Please.
C
C
So
what
we
also
found
out
was
that,
especially
over
the
southeastern
part
of
high
mountain
asia,
which
shows
in
the
in
the
upper
left
figure,
is
that
there's
yeah
some
discrepancy
and
that
the
elevation
difference
between
glacier
and
grid
is
very
high
in
some
regions.
The
glaciers
end
up
2000
meter
lower
than
the
grid
cell
amine
elevation,
and
I
know
for
this
region.
C
The
reason
for
this
discrepancy
is
that
there
are
some
inconsistencies:
inaccuracies
in
the
glacier
outlines
of
random
glacier
inventory
version,
one
version
two
relative
to
randolph
glacier
version
six,
but
also
there
are
some
likely
uncertainties
in
the
globe.
Topography.
C
And
further
the
glaciers
that
are
there
in
the
in
the
in
the
former
data
sets
or
in
the
data
set
that
is
usually
used.
There
are
only
61
elevation,
bins
of
100
meter
each
where
the
last
one
covers
6
000
meter
till
10
000
meter
altitude.
But
that
means
that
we
are
missing
those
that
those
glaciers
that
are
there
between
six
thousand
and
seven
thousand
meter
altitude
and
that
we
also
miss
some
part
of
the
accumulation
that
can
yeah
eventually.
C
Yeah
I'm
almost
finished.
Potentially
it
can
increase
the
surface
mass
balance,
so
we
wanted
to
know
to
it.
We
thought
okay,
it's
time
to
to
update
the
clam
glacier
data
set
and
we
did
it
based
on
the
random
glacier
inventory
version
6,
that
machine
version,
3
version
2
for
the
greenland
and
antarctic
ice
sheets,
and
we
used
the
30
arctac
gm
tet,
which
is
all
more
yeah
consistent
with
what
is
already
used.
Unfortunately,
I
was
not
able
to
complete
it
entirely.
C
It
is
still
going
on,
but
I
can
show
you
these
figures.
The
upper
figure
shows
the
glacier
fraction
for
all
the
glaciers
from
the
random
glacier
infantry
and
the
lower
figure
shows
the
glacier
fraction
for
the
bed
machine
of
greenland
and
antarctic
ice
sheets.
That
are
there
from
the
right
from
the
bat
machine,
and
actually
that
is
about
it.
That
was
my
presentation
like
yeah.
The
next
slide
is
thank
you,
so
yeah,
okay,.
L
Thank
you
very
much.
Renee
dave
lawrence
has
a
comment
in
the
chat
dave.
Would
you
would
you
like
to
just
summarize
what
you
put
there.
I
I
don't
think
it
was
included
in
the
original
downscaling
that
you've
been
that
I
think
you
developed
built,
and
the
second
thing
is
that
this
with
the
hill
slope
model,
we
are
accounting
for
slope
and
aspect
and
again
probably
pretty
critical
for
for
glaciers,
most
likely,
mostly
on
north,
facing
slopes
right.
So
I
think
we
could
have
a
conversation.
C
L
L
If
you
could
stop
sharing
and
then
our
next
speaker,
david
bonan,
can
start
sharing.
L
C
L
All
right
right
and
david
dave:
could
you
go
ahead
and
tell
us
about
transient
equilibrium,
responses
of
the
amok
to
global
warming
and
couple
of
climate
models.
N
Yeah
thanks
thanks
bill.
This
is
work
that
I've
been
doing
with
my
advisor
andy
thompson
at
caltech,
as
well
as
emily
newsom
who's,
a
postdoc
at
the
university
of
oxford,
shantung,
sun,
a
postdoc
at
caltech
and
maria
rogenstein
who's
at
csu,
and
so
really.
N
There's
this
abrupt
a
partial
recovery
of
the
amok
and
then
there's
this
kind
of
grow,
slow,
gradual
re-strengthening
of
the
overturning
circulation
and
then,
in
contrast,
cnrm
kind
of
weakens
on
that
centennial
time
scale
as
well,
but
then
really
stays
prolonged
weakening,
has
a
prolonged
weakening
or
even
a
collapse
of
the
overturning.
N
And
so
just
this
plot
below
actually
just
shows
the
hog
muller
plot
of
the
depth
of
the
depth
of
the
stream
function
in
csm,
and
this
is
in
the
north
atlantic
and
again
you
see
that
weakening
in
conjunction
with
the
shoaling
of
the
overturning
and
then
there's
this
gradual,
deepening
and
re-strengthening
the
returning
and
csm
and
then,
in
contrast,
cnrm,
barely
recovers
and
begins
to
show
more
of
a
recovery
towards
the
end
of
its
simulation.
And
so
really.
N
What
I
want
to
try
to
highlight
today
is:
can
we
identify
and
understand
processes
that
influence
the
transient
and
equilibrium
responses
of
the
amok
to
warming
and
coupled
climate
models,
and
so
to
do
this?
I'm
going
to
kind
of
actually
borrow
some
some
language
or
from
more
theoretical
literature
from
nicaraguan
and
dallas
in
2012
and
multi-yanson's
work
with
these
idealized
toy
models
of
the
oceans
over
terrain
circulation,
and
so
what
they
say
is
that
you
can
relate
the
stream
function
to
the
density
gradient
in
the
north
atlantic
and
throughout
the
basin.
N
By
assuming
that
there's
no
flow
at
the
base
of
the
ocean
and
there's
no
flow
at
the
surface,
and
so
from
this
you
can
back
out
a
stream
function
and
I've
kind
of
just
schematized
this
here
on
the
top
panel,
which
these
are
the
isominals
and
csm
which
remain
relatively
flat
throughout
most
of
the
basin,
and
then
they
outcrop
towards
the
north
atlantic.
And
so
here's
just
that
density,
gradient
flipped.
N
I'm
sorry
for
the
sign
here
and
I
can
back
out
a
stream
function,
just
assuming
the
boundary
conditions
are
zero
at
the
surface
and
the
and
the
bottom,
which
is
denoted
here
in
purple,
and
you
can
see
that
compared
to
the
stream
function
that
you
get
from
csm
itself.
N
It
matches
both
the
the
depth
and
the
structure
and
the
magnitude
of
the
amoc
as
well,
and
so
we
think
that
this
is
might
be
a
good
way
to
kind
of
diagnose.
N
What's
mechanically
going
on
in
these
two
different
models,
and
just
to
convince
you
more
that
this
works
really
well,
here's
that
same
time,
series
and
block
of
csm's
overturning
strength
in
the
north
atlantic
and
then
purple
is
just
the
scaling
as
if
I
diagnose
the
density
profiles
throughout
the
entire
basin
and
in
the
north
atlantic
and
reconstruct
that
overturning,
and
you
can
see
that
it
captures
both
the
weakening
that
partial
recovery
and
that
gradual
strengthening,
although
it
tends
to
under
predict
the
weakening
and
so
then
we've
done
this
for
all
the
models
that
we
we
did
in
the
study
and
I've
just
calculated
linear
trends
through
different
150
year
periods
in
these
models.
N
What's
going
on
in
these
models,
I'm
just
going
to
focus
on
three
different
phases:
the
weakening
phase,
which
is
centered
at
that
100-year
time,
scale,
that
partial
recovery
phase,
which
is
a
couple
centuries
after
the
weakening,
and
then
this
re-strengthening
phase,
which
is
over
the
the
next
thousand
years.
And
so
this
top
panel
is
just
showing
the
density
gradient
change
in
these
two
models.
N
And
then
these
middle
panels
and
the
bottom
panel
are
showing
the
density
change
in
the
north
and
the
density
changes
in
the
basin
partitioned
into
density
changes
due
to
temperature
and
salinity.
And
so
what
you
can
see
is
that
they
tend
to
show
more
weakening
than
the
north,
largely
due
to
a
warming
of
that
water
column,
with
some
kind
of
input
from
fresh
water
as
well
and
then
the
basin.
N
It
just
tends
to
warm
more
with
this
kind
of
formation
of
some
salinity
anomaly
that
that
begins
to
show,
and
so
then
the
models
agree
on
what
they're
doing
on
this
weakening
phase.
But
once
you
go
to
the
recovery
phase,
you
see
kind
of
different
behavior,
you
see
in
csm.
N
The
stream
function
begins
to
recover
slightly
and
same
with
that
density
gradient,
and
this
is
somewhat
driven
actually
by
positive
salinity
anomaly
that
forms
throughout
the
subtropical
atlantic
and
then
it's
kind
of
infected
towards
the
north,
the
north
atlantic,
but
cnrm
tends
to
show
more
of
a
weakening
of
the
stream
function
and
a
weakening
of
that
density
gradient,
which
is
largely
due
to
the
fact
that
there's
just
this
high
latitude
freshwater
anomaly
that
begins
to
form
in
the
north
atlantic,
and
so
then.
N
N
The
strengthening
phase
on
this
time
scale
is
the
fact
that
the
basin
warms
more
relative
than
the
north,
and
so
then,
this
density
change
is
actually
a
bit
greater
than
the
density
change
throughout
the
north
atlantic
and
that's
driven
by
this
low
latitude
warming
signature
right
here
and
then
there's
this
kind
of
slight
roll
for
the
fact
that
this
limited
anomaly
in
the
north
atlantic's
positive
here
and
it's
it's
fresh
here
in
the
cnrm
model
and
so
to
kind
of
schematize.
N
What's
going
on
here.
This
is
the
hob
muller
plot
of
salinity,
just
say
between
the
1000
and
2000
meters
in
csm
and
plotted
over.
That
is
actually
our
annual
arctic
sea
ice
area,
and
what
you
can
see
is
that
the
salinity
anomaly
forms
when
the
amoc
is
weakened
and
then
slowly
throughout
the
simulation
it
gets
infected
up
into
the
high
latitudes
and
it
begins
to
erode
that
freshwater
anomaly
due
to
sea
ice
melt
or
an
intensified
hydrologic
cycle
and
then
in
cnrm.
N
N
And
so
just
as
a
summary
of
this
there's
a
schematic
of
what's
going
on
here.
These
two
models
are
very
similar
on
the
weekend
time
scale.
N
So
this
is
a
strong
amok
recovery
and
we
game
off
recovery
through
that
first
100
years
they
both
weaken
largely
due
to
a
high
latitude
temperature
increase,
but
in
csm
that
large
pot
that
large
partial
recovery
is
really
driven
by
the
fact
that
there's
this
limited
anomaly
that
is
affected
into
the
high
latitudes
and
the
fact
that
the
low
latitudes
warm,
which
increases
that
density
gradient
and
cnrm.
N
You
see
the
same
signature
of
low
latitude
warming,
but
you
see
this
freshwater
signal
in
the
high
latitudes
that
prevents
that
salinity
anomaly
to
reinvigorate
that
deep
convection
and
then
on
this
like
longer
time
scale.
We've
noticed
that
these
models
are
very
similar
in
which
the
low
latitudes
begin
to
keep
warming
and
that
really
drives
a
stronger,
overturning,
that's
even
stronger
than
the
initial
strength,
and
so
just
to
kind
of
provide
some
more
mechanisms.
The
long
run
mip
has
been
a
myth
of
opportunity,
so
there's
a
lot
of
output
that
really
isn't
available.
N
So
it's
been
hard
to
kind
of
diagnose
specific
mechanism
what's
going
on
here,
so
in
the
most
recent
times
I've
been
trying
to
develop
a
conceptual
model
of
the
ocean's
overturning
circulation,
and
I
would
say
this
is
kind
of
the
canonical
view
that
everybody
have
has
at
the
ocean,
overturning
in
which
there's
buoyancy
loss
in
the
high
latitude
north
atlantic
that
water
sinks,
it
flows
adiabatically
back
towards
the
southern
ocean,
where
it
up
wells
along
isopignils.
N
It's
transformed
and
it
goes
back
towards
the
higher
latitudes,
and
then
you
also
have
this
lower
cell
here,
and
so
what
I've
been
doing?
Is
I've
been
saying
that
okay,
we
can
keep
track
of
temperature
and
salinity
evolution
equations
in
each
basin,
so
the
atlantic
north
atlantic
and
the
southern
ocean,
which
are
going
to
be
a
function
of
some
surface,
forcing
here
some
dipignal.
N
Thanks
yeah
and
some
abduction
terms-
and
we
can
relate
these
two
I'll
wrap
this
up
very
quickly
and
actually
just
build
a
more
conceptual
model
of
what's
going
on
here,
and
so
I
actually
forced
this
model
with
a
specific
warming
scenario,
where
it
was
four
degrees
of
warming
in
the
north
atlantic,
two
degrees
of
warming
in
the
basin.
N
I
can
talk
about
this
more
later
if
people
are
interested,
but
you
actually
do
see
the
signature
of
that's
limiting
emily
form
in
the
basin
and
how
it
gets
affected
towards
the
higher
latitudes
going
forward
and
so
yeah.
This
is
work.
That's
been
written
up
into
a
paper
for
journal
of
climate
and
I'm
currently
trying
to
write
this
up
the
conceptual
model
for
another
paper.
So
thank
you.
L
Thank
you
dave.
Do
we
have
any
quick
questions.
I
So
this
is
a
peter
jen.
Can
I
make
a
comment
yeah?
There
are
examples
of
a
couple
models
where
the
a
mark
has
recovered
after
several
thousand
years.
I
So
I
think
that
if
that
crnm
integration
went
further,
it
would
have
recovered
and
there's
some
from
the
nabi
and
stauffer
and
another,
and
I
talked
about
them
in
a
paper,
I
wrote
called
a
commentary
on
the
a
mark,
stability
and
climate
models.
If
you're
interested.
H
N
Yeah,
it's
been
hard,
the
freshwater
fluxes
in
the
output.
We
have
aren't
specifically
partitioned
into
sea
ice,
runoff
or
like
a
hydrologic
cycle,
but
that's
the
motivation
for
kind
of
this
conceptual
model
is
to
maybe
provide
a
little
bit
more
mechanisms
of
that,
but
I
do
know
that
crm
loses
its
sea
ice
more
quickly
than
csm,
but
it
does
also
have
a
lot
more
runoff.
I
believe
the
hydrologic
cycles
are
very
similar
between
the
two
models.
N
L
Okay,
thank
you
again,
dave
we'll
move
on
to
our
next
speaker,
lily
wren,
who
will
be
speaking
about
source
attribution
of
arctic,
black
carbon
sulfate,
aerosols
and
arctic
surface
warming
during
1980
to
2018.
lily.
Are
you
able
to
share
your
screen
great
yeah.
A
O
O
The
active
has
warmed
the
rabbit
since
the
1980s,
with
a
1.5
k
increase
in
the
service
air
temperature,
which
is
about
three
times
faster
than
the
global
average.
A
number
of
studies
have
exempted
possible
mechanisms
that
cause
the
rapid,
active
economy
of
observations
and
monitoring.
Studies
suggest
that
variations
in
aerosols
are
a
good
explanation
for
the
faster
active
economy.
O
The
subject
concentration
shows
a
considerable
decreasing
change
from
1980
1980
to
2000,
which
then
slows
down
after
2000
years.
The
decrease
in
savage
during
this
time
period
results
from
the
reduction
in
emissions
from
europe
and
russia.
It
contributes
to
nineteen
percent
of
the
decline
of
the
of
the
arctic
surface
and
savage
concentrations.
O
We
also
studied
the
changes
in
those
contributions
to
the
annual
my
vertical
profile
of
savage
and
black
carbon
concentrations
over
the
arctic
between
2014
and
2018
and
1980
and
1984
below
six
kilometers.
Due
to
the
effective
emission
reductions.
The
contributions
from
both
europe
and
russia
to
the
arctic
surface
was
decreased
between
10
and
15.
Kilometers.
Contributions
from
south
asia
and
east
the
east
asia
increased
after
the
upper
troposphere,
which
is
consistent
with
the
increase
in
emissions
of
these
regions,
leading
to
a
combined
increase
in
surface
concentration.
O
O
Previous
studies
have
suggested
that
the
arctic
climate
responds
not
only
to
local
arctic,
so
arctic,
forcing
but
also
through
outside
forces
due
to
the
marine
energy
transport
change,
to
estimate
the
relative
rows
of
regional
aerosol
chains.
In
advocate
in
affecting
the
arctic
warming,
we
looked
into
the
temporal
variations
of
anaerobic
radiator,
forcing
of
surface
and
black
carbon
in
different
national
bands
during
1980
to
2018..
O
You
can
see
that
within
the
arctic.
The
magnitude
of
savage
radiative,
forcing
due
to
aerosol
radiation
in
actions
decreased
from
minus
0.21
water
per
square
meter
in
the
in
the
1980
to
1984,
to
minus
0.1
water
per
square
in
2014
to
2018,
indicating
a
warming
effect
in
the
arctic
from
the
local
surface
change
over
the
middle
latitudes.
O
The
service
radiator
force
relative,
forcing
due
to
aerosol
radiation
interactions,
decreases
from
minus
0.87
to
minus
53
watt
of
square
meter.
O
Then
we
then
we
apply
arctic
climate
sensitivity,
vectors
to
calculate
the
recent
arctic
surface,
temperature
change
related
to
the
variations
in
surface
and
the
black
carbon
radiator
proteins
over
the
different
national
bands
in
the
right
figure,
as
mentioned
above
due
to
the
decrease
in
savage
and
the
at
the
middle
latitudes.
A
spin
center
meridianal
heat
transport
cost.
O
Another
the
combined
total,
the
combined
total
effects
of
savage
and
black
carbon
produce
the
arctic
service
warning
of
pointers.
L
Thank
you
very
much
lily.
Do
I
don't
see
any
questions
yet
in
the
chat?
Do
we
have
any
questions
for
lily?
I
don't
have
a
question.
I
just
have.
C
A
quick
comment
here,
if
you're
not
speaking,
can
please
mute
yourself.
I
A
Yeah
sure
so
I
I
it's
interesting
that
you
were
saying
about
the
partition
of
aerosols
from
terrestrial
sources
and
local
sources
in
arctic.
So
did
you
sort
of
categorize,
which
ones
are
more
important?
The
local
sources
I
mean-
and
I
assume
you're
talking
about
ships
when
you're
talking
about
local.
A
O
Sorry,
my
dishwasher
is
not
aware
of
yes,
the
emissions
from
south
and
east
asia
lead
to
the
positive
change
in
active
surface
and
black
and
black
carbon,
but,
as
you
said,
the
the
emissions
from
the
arctic
dominated
to
the
black
carbon
concentrations
in
the
lower
atmosphere.
O
It
is
up
to
the
different
different
kilometers
and
atmosphere.
L
Thanks,
thank
you
very
much
lily
you
shower.
Are
you
online.
L
Okay,
I
have
your
talk
here,
I'm
going
to
go
to
full
screen.
Okay
and
just
let
me
know
when
you'd
like
me
to
advance.
J
J
So
this
is
arctic
amplification.
I
think
everybody
one.
I
think
everyone
knows
everyone
knows
it.
Arctic
worms,
land,
the
rest
of
the
globe;
okay.
Next,
some
previous
study
trying
to
study
the
drivers
of
arctic
amplification
and
next
place,
and
most
of
them
using,
for
example,
doubling
co2
or
at
the
most
full-time
co2
experiment.
However,
in
the
in
the
future,
the
co2
concentration
could
be
even
higher
than
four
times
co2
relative
to
pre-industrial
concentration,
for,
for
example,
on
the
right
panels
in
ssp,
5,
8.5,
co2,
projected
scenario
in
the
future.
J
You
can
see
that
at
the
end
of
21st
century
the
co2
level
is
about
four
times
d2
relative
to
industrial
level.
However,
during
2200
and
2300,
the
co2
concentration
could
be
as
high
as
8
times
co2.
So
in
this
study
we
further
want
to
explore
what
is
the
arctic
amplification
change
in
different
labels
of
co2
concentration?
Next,
please
so.
J
Oh,
it's
okay
yeah,
so
the
scientific
questions
do
we
want
to
answer
here
is
how
the
arctic
amplification
in
response
to
different?
Oh
I'm,
sorry
yeah.
So
the
arctic
amplification
also
shows
some
seasonal
characteristic
because
it
is
dominated
by
different
mechanisms
during
different
seasons.
So,
if
we
just
look
at
the
annual
mean
arctic
implication
may
be
somewhat
misleading.
J
J
So
the
scientific
questions
we
want
to
answer
here
is
the
how
arctic
amplification,
in
response
to
different
level
of
co2
and
the
second,
we
want
to
addre
seasonal
cycle
change
in
response
to
different
co2
concentration.
I
J
Yeah
sure
so
yeah
I
can
keep
speaking
yeah,
so
the
methodology
we
are
using
is
that
we
are
using
ivami
type
in
mitarovsky.
The
recent
published
gro
paper
in,
in
which
he
run
abrupt
co2
simulation
with
different
level
of
co2
from
one
time
co2
all
the
way
to
a
times.
Co2,
okay
and
the
this
rung
is
conducted
for
155
150
years,
and
we
also
look
into
corresponding
level
slab
ocean
model
simulation
with
60-year
simulation
and
the
co2
simul.
J
L
Okay,
I
pressed
resume
share.
J
J
Yeah,
okay,
thanks
yeah,
so
yeah!
So
here
just
what
what
I
say:
yeah!
So
the
methodology
we
use
fully
couple
simulation
and
stabilization
model
with
different
co2
con
con
concentration
force
next,
please
so.
Here's
the
arctic
climate
response
in
in
response
to
different
co2
processing.
So
on
the
panel
a
we
see
surface
air
temperature
response
for
fully
coupled
model
and
the
slabs
model,
and
as
expected
they
are
the
temperature
response
increases
as
a
function
of
co2.
Next,
so
you
may
see,
there's
a
like
there's
an
abrupt
drop,
abrupt
drop
in
the
for
the
full-time
co2.
J
And
correspondingly,
the
ci
cs
loss
increases
as
a
function
of
co2
next
so
and
correspondingly,
we
also
see
that
the
turbulent
heat
fluxes
from
the
ocean
to
the
atmosphere
also
increases
so
this.
So
this
is
possible
that
these
three
components
are
highly
coupled
with
it
with
it,
with
each
other.
It's
more
artiforming
and
more
sea
ice
laws
open
more
waters,
allowing
more
heat
from
the
ocean
to
the
atmosphere.
J
Next,
so
after
more
heat
into
the
atmosphere,
you
also
thicken
the
arctic
tropospheric
air
column.
Okay.
Next
so
now
we
focus
on
the
our
annual
mean
arctic
amplification,
which
is
defined
by
the
by
the
temperature
arctic
temperature
change
divided
by
global
temperature
change.
So
in
panel
a
you
can
see
the
evolution
in
each
co2,
abrupt
co2
experiment.
So,
roughly
speaking
about
70
to
80
years,
this
arctic
amplification,
which
quarter
equilibrium
state
next
and
you
can
see
in
the
in
the
bottom
left
panel.
J
So
now
we
move
on
to
our
focus
to
the
arctic.
Amplification
of
the
signal
cycle
in
response
to
co2
processing
so
on
the
right
on
the
left
panel
is
the
arctic
temperature
response
in
this
last
30
year
mean
seasonal
cycle,
and
I
colored
in
blue
it's
two
times
co2
and
the
colored
in
red
or
a
times
co2.
So,
as
you
can
see,
the
peak
of
the
arctic
warming
shifted
from
november
to
december
into
january
from
two
times
co2
to
eight
times
co2.
J
J
Next
and
you
may
wondering
our
experiment,
it
might
be
somewhat
idealized,
but
what
could
happen
if
we
prescribe
more
realistic,
for
example,
cm6
scenario,
greenhouse
gas,
forcing
so
on
the
top
two
panels,
I'm
showing
the
cso
one
large
example
exactly
the
same
model
we
use
in
the
product,
co2
experiment,
and
in
panel
a
you
can
see.
You
can
also
see
this
shift
as
time
as
time
advanced
toward
the
end
of
the
century,
the
corresponding
on
panel
b.
J
You
can
see
the
peak
of
the
arctic
amplification
also
shifted
from
november
to
december,
and
the
next-
and
this
is
not
just
a
model
dependent
feature
because
also
also
see
in
gfc
or
cm3
large
ensemble.
These
ships
all
also
also
clearly
shows,
and
we
also
look
into
40
76,
multimodal
means,
although
the
shift
is
not
as
clear
as
single
model,
but
we
still
see
its
seasonal
shift
tendency
next,
please.
J
So,
just
briefly
summarize
so,
article
warming
response
seems
like
closely
coupled
to
cs
laws
and
the
surface
heat
fluxes
with
as
a
function
of
increasing
co2
bursting
and
the
strength
of
arctic.
Amplification
actually
decreases
as
a
function
of
co2
and
the
peak
phase
of
arctic
amplification
tends
to
shift
from
november
to
december
or
january,
depending
on
how
strong
the
co2
forcing
it
is.
J
We
are
not
only
seeing
our
abrupt
skill
to
experiment,
but
also
see
in
the
in
a
more
realistic
semi,
semi,
semi,
co2
or
greenhouse
gas,
forcing
scenario,
and
these
results
have
important
implications
for
rte
ecosystem
and
a
commercial
ship.
Okay,
I'll
stop
here
and
I'm
ready
to
take
any
questions.
Thank
you.
H
If
you
have
an
idea
which
of
the
feedbacks
is
responsible
for
the
slowdown
of
the
increasing
or
well
the
lack
of
how
do
you
say
this,
the
weakening
of
the
response
with
increasing
co2?
Do
you,
I
have
a
suspicion,
but
I'm
curious
what
you
think.
J
So
is
this
probably
you
know,
is
the
heat
exchange
that
dominate
this
seasonal
shift,
but
I
also
look
into
the
radiative,
the
short
wave
and
the
long
wave
radiation.
J
Although
I
just
look
at
the
net
surface
component
but
seems
like
this
short
wave
and
the
long
wave
component
does
not
show
the
seasonal
shift,
so
I'm
so
I
so
I
I
need
more
analysis
to
confirm,
but
I
guess
probably,
the
local,
air
or
ocean
atmosphere,
heat
exchange,
dominate
these
processes
and
produce
the
lake.
L
P
Thanks,
can
you
see
that
yep.
P
Okay
and
still
see
it
presentation
mode
here:
okay,
yeah
thanks.
I
know
we're
running
a
little
behind
so
I'll.
Try
to
make
this
quick,
and
this
is
a
little
bit
different
than
the
other
talks
here.
P
I
know
this
is
a
model
modeling
workshop,
I'm
a
bit
of
an
interloper
here
with
observations,
but
I
just
wanted
to
present-
and
I
want
to
thank
csm
and
yan
and
the
organizers
for
for
allowing
me
the
opportunity
to
present
on
a
new
version
of
our
climate
data
record
that
I
think
is
useful
for
modeling
modeling
applications,
and
so
I
just
kind
of
quickly
present
go
through
and
present.
This
folks
may
be
aware
of
this
already.
P
This
is
the
sea
ice
concentration,
climate
data
record,
it's
officially
a
noaa
product,
noaa
nsidc
product
funded,
the
specific
development
of
the
cdr
is
funded
through
the
noaa
ncei,
but
it's
based
on
nasa
and
the
nasa
dac
products.
So
it's
kind
of
a
joint
noaa
nasa
effort.
Originally
we
developed
this
in
2011
and
it
was
limited
as
far
as
the
cdr
to
1987
to
present
to
this
ssmi
and-
and
that
should
say
ssmis
part
of
the
record.
P
We
did
include
the
nasa
data
sets
that
included
simmer
going
back
to
1979
or
late
1978.,
but
these
didn't
have
the
full
transparency
and
the
price
processing
reproducibility
that
the
cdr
requires.
And
it's
it's
a
combination
of
the
nasa
team
and
bootstrap
algorithm.
I
won't
go
into
the
details
of
that
and
there's
qa
and
qc
uncertainty
fields
and
the
new
version
I
hope
to
be
out
by
now,
but
we
keep
running
into
little
glitches
and
kind
of
delivery
things,
but
it
should
be
out
by
next
week.
P
So
just
an
example.
Yeah.
This
is
your
typical
passive
microwave,
it's
25,
kilometer,
daily
fields
and
monthly
fields
for
the
arctic
and
the
antarctic,
and
then
this
just
as
an
example
of
the
the
standard
deviation
field
and
that
that's
something
that
we
provide
based
on
the
the
standard
deviation
of
the
nasa
team
and
the
bootstrap
concentrations
in
a
spatial
like
a
three
by
three
grid
cell
area,
that
that
gives
an
indication
of
the
relative
uncertainty.
P
It's
not
an
absolute
uncertainty,
you
know
not
a
quantitative
uncertainty,
but
it's
it
would
be
potentially
useful
for
weighting
the
observations
versus
set,
for
example,
models,
and
you
can
see
the
the
highest
uncertainties
tend
to
be
near
the
ice
edge.
This
is
in
march
in
the
belt
season.
You
also
see
high
uncertainties
within
the
within
the
ice
pack
or
higher
uncertainties
because
of
the
the
mel
ponding
and
surface
melt,
the
major
enhancements
I
kind
of
alluded
to
earlier.
P
P
that
had
not
been
included
before,
because
it
had
never
been
processed
at
nsidc
and
hadn't
been
processed
really
by
anyone
since
the
the
early
1990s,
and
that
data
has
a
lot
of
issues
with
it.
It
takes
a
lot
of
finessing
to
work
with
that
data,
because
it's
it's
early
sensor.
The
technology
was
not
as
good.
There
were
a
lot
of
sensor
issues
with
it,
but
we've
managed
to
put
that
together
and
put
it
into
the
cdr
without
any
manual
qc
goddard
when
they
created
their
their
record.
P
They
included
simmer,
but
they
basically
did
a
lot
of
manual
cue,
seeing
going
through
image
day
by
day
image
by
image
and
manually,
taking
out
or
correcting
data,
which
was
not
tracked,
so
you
don't
have
any
provenance
on
that.
We
still
have
some
bad
days
that
we
did
remove
because
the
data
was
just
too
corrupted
to
use.
N
P
We
also
we
have
valid
ice
maps
which
are
used
for
qceing
the
data
to
remove
ice
where
there
isn't
any
ice
possible
and
we
took
those
from
monthly
to
daily,
which
makes
for
a
little
bit
smoother
and
a
more
finer
qc
there.
One
of
the
other
big
things
is
we
added
spatial
and
temporal
interpolation.
P
This
had
been
done
with
the
nasa
goddard
products
as
well
for
there's
for
small
scattered,
missing
cells.
We
interpolated
at
the
brightness
temperature
level,
which
then
feeds
into
the
algorithms
for
larger
areas
where
there's
missing,
swaths
or
or
missing
days.
P
We
do
an
interpolation
of
a
temporal
interpolation
from
up
to
plus
or
minus
five
days,
and
so
that
gives
us
much
more
complete
fields
and
it
also
allowed
us
to
to
basically
to
create
daily
simmer
fields,
simmer
only
operated
every
other
day
for
the
so
for
the
1978
to
1987
period,
even
in
the
goddard
fields,
there's
only
daily
data
every
other
day,
but
we
used
the
temporal
interpolation
and
interpolated
to
the
to
those
missing
days.
P
So
there's
daily
fields
for
virtually
the
entire
record,
except
for
a
few
missing
periods
where
there's
longer
data
gaps.
The
well-known
one
is
december,
1987
to
january
1988.
july
1984.
I
also
had
some
corrupt
data
and
a
few
other
days
here
and
there,
and
then
we
also
filled
the
the
pole
hole
in
the
in
the
arctic
with
a
pretty
basic
interpolation.
That's
something
we
might
look
at
more
in
the
future,
but
it
gives
you
a
complete
field.
P
You
don't
have
a
a
missing
whole
area
to
deal
with
and
then
just
a
comparison
of
the
extents
here.
Comparing
you
know
the
the
cdr
with
the
goddard,
the
two
algorithm
products,
the
nasa
team
and
the
bootstrap
the
cdr
is-
is
most
closely
related.
It's
a
combination
of
both,
but
it
matches
more
closely
with
the
bootstrap.
P
So
the
nasa
team
has
larger
differences,
but
it
you
know,
there's
some
offsets
in
the
the
way
that
we
process
versus
the
way
that
goddard
process,
but
overall
they're
they're,
pretty
small,
there's,
not
a
lot
of
bias.
There
there's
a
little
bit
in
the
simmer
period
in
the
southern
hemisphere,
particularly
that's
still
a
difficult
thing.
P
To
look
to
to
handle,
but
again
we
we
have
data
that
we
didn't
have
before
and
overall
it's
it's,
I
think,
pretty
consistent
with
the
the
goddard
product,
so
I
think
it's
a
suitable
long-term
data
set.
This
is
september
of
the
arctic.
The
trends
are
are
almost
the
same,
very
small
differences.
Some
other
months
have
a
little
bit
bigger
differences
but
they're
pretty
consistent
between
the
cdr
and
the
bootstrap,
as
well
as
the
nasa
team.
P
I
left
that
off
here
for
clarity,
so
just
to
conclude,
this
version
four,
which
again
is
going
to
be
out,
should
be
out
next
week.
It's
a
near-complete,
long-term,
consistent
sea
ice
concentration
record.
I
think
it's
suitable
for
model
validation,
comparison
and
simulation
and
there's
the
the
website
for
those
you
can
go
and
register
and
see.
You'll
get
a
notification
when
the
when
the
update
is
actually
made
and
get
regular
updates
and
there's
the
new
real-time
version.
P
That's
updated
daily,
that's
ongoing
as
well
and
again
supported
through
noaa
the
nci
cdr
program,
and
then
nasa
has
this
through
the
dac.
So
I'll
stop
there
and
hopefully
finish
up
on
time.
L
P
P
Appreciate
the
opportunity
to
present
it
to
this
group
as
well
so
and
folks
can
feel
free
to
email
me
for
further
questions
or
information
about
the
data
set.
L
Okay,
jan.
F
P
Yeah,
it's
a
good
question.
We
chose
the
the
poll
hopeful.
It's
been
responsive
some
to
users,
basically
where
they,
they
basically
wanted
data
everywhere,
and
so
we
provide
that
now
it's
a
it's
a
just,
a
basic
simple
and
it's
the
high
arctic.
So
there's
not
a
lot
of
variability
and
we
don't
know
exactly
what
the
variability
is.
So
it's
just
an
average
of
the
surrounding
days
that
is-
and
I
want
to.
I
should
note
that
all
the
interpolation,
all
the
filling
and
everything
is
all
flagged.
P
So
if,
if
you
don't
want
to
use
that
data,
it's
very
easy
to
denote
that
flag
as
as
a
pole,
hole
and
and
take
that
out.
So
it's
it's
there
to
be
flexible
for
users
that
that
want
that.
L
All
right
tell
we'll
go
ahead
and
go
to
the
break.
Now
we
will
reconvene
for
the
next
session
of
talks
on
time
at
10.,
so
our
break
will
be
a
little
bit
shorter
hansey.
Would
you
like
to
tell
folks
about
how
the
break
will
go.
B
B
B
Kevin
rader
will
tell
us
about
the
cam6
dart
ensemble
re-analysis,
as
well
as
data
sets
for
machine
learning
that
come
from
this,
and
then
elizabeth
hunky
will
tell
us
about
the
size,
consortium
and
advancing
sea
ice
modeling.
So
these
three
posters
will
be
available
in
breakout
rooms
and
we
invite
you
to
go
to
these
breakout
rooms,
check
out
the
posters
while
you're
taking
a
break
and
and
then
reconvene
here
in
about
20
minutes.
It
looks
like
so
that's
it.
A
I
B
B
I
I
I
think
it
would
be
great
if
people
can
go
from
room
to
room
so
that
they
can
choose
which
one
to
go
to.
Let.
G
A
Yeah
sure
I'm
happy
to
present
in
the
main
room
with
that,
and
then
I
guess
you
know,
I'm
not
sure
what
time
we
were
gonna
close
it,
but
we're
gonna
close
it
soon.
So.
G
G
I
I
Dave
or
anyone
are
you
seeing
the
breakout
rooms?
Are
you
able
to
enter.
A
I'm
sharing
my
screen.
Okay,
I
did
just
get
an
invite
to
the
one
breakout
room.
I
Let's
see,
but
if
you
go
back
out
to
the
breakout
room
menu,
do
you
see
the
list
of
rooms
and
there's
a
join
button
next
to
them?
No,
only
the
one
that
we've
been
assigned
to
okay,
I'm
going
to
have
to
close
the
rooms
and
reopen
them.
I
D
D
No,
no,
you
don't
need
to
recreate
them.
I
just
I
just
will
check
the
box
to
allow
them
to
choose
a
room.
L
Ryan,
I
just
got
an
invitation
to
join
one
particular
room.
I
don't
see
how
I
would
choose
based
on
what's
on
my
screen,
so.
D
If
you
hit
cancel
on
that
and
then
open
up
the
breakout
rooms
tab
at
the
bottom,
you
should
be
able
to
hover
over
the
number
next
to
the
breakout
room.
Does
that.
L
L
A
D
D
Hey
dave
lsd
wants
to
do
a
quick
screenshot
test
before
we
get
started
for
the
next
one.
That's
fine.
A
Okay,
all
right,
I
I
can
stop
sharing
for.
D
D
A
D
D
G
G
No,
I
think
it's
fine,
I'm
just
trying
to
figure
out
how
we
should
split
up
the
the
breakout
rooms
right.
I
A
Right
yeah,
I
think
I
think
those
two
we
could
just
add
into
sort
of
one
of
the
other
discussions.
G
Maybe
we
should
like
diagnostics,
maybe
could
be
something
relevant
to
all
of
the
different
breakout
groups.
G
A
Yeah
yeah
yeah.
I
think
that
you
know
you
could
maybe
just
sort
of
if
there's
time
people
could
throw
that
into
the
future
projections
of
antarctic
change.
People
could
maybe
so.
Q
Or
the
variable
high
resolution
I
was
going
to
ask:
I
want
to
test
with
ryan
screen
sharing,
since
that
was
kind
of
weird
the
last
time.
Do
you
guys
mind
if
I
do
that?
Go
ahead,
yeah.
I
G
A
Well,
I
think
we
again
are
we
going
to
just
let
people
sort
of
go
between
the
rooms.
I
A
Okay,
I'm
gonna
go
ahead
and
close
the
responses.
I
think
we're
stuck
at
32.
A
A
I'm
sending
you
the
link
to
the
responses,
so
let
me
know
if
you
see
that
and
then
so,
I'm
guessing
based
on
the
responses.
We
should
do
all
six
of
the
main
topics
and
then
we
can
sort
of
combine
extra
two.
So.
I
I'm
sorry
crit
create
all
six
and
then
another
one
with
two.
A
I
A
A
I
A
Okay,
als,
what
which
of
the
two
would
you
most
prefer
the
stakeholder
integration?
Historical
data
sets.
G
Q
I
I
mean,
if
you
combine
them,
that's
fine.
If
I
had
to
choose
I'd,
probably
know
more
about
know
more
about
his
the
stakeholder
integration.
I've
thought
more
about
that
recently
than
okay,
I'm
not
sure
exactly
what
historical
data
means
like
hansi
said
yesterday.
Is
this
whaling
logs
or
because
I
know
nothing
about
those.
A
G
So
we'll
get
started
with
the
talks
in
just
a
minute,
so
we're
in
the
process
of
setting
up
these
breakout
rooms
thanks
everybody
for
for
answering
the
survey
and
giving
your
input
it
looked
like.
There
was
interest
in
all
of
these
topics,
so
we'll
probably
set
up
a
breakout
room
for
each
topic
or
we
might
combine
the
historical
data
sets
indigenous
knowledge
and
stakeholder
integration
and
communication
into
one
room.
G
There
were
a
couple
of
additional
topics
that
people
put
in.
One
was
on
diagnostics,
and
so
because
there
was
only
one
response
requesting
that
we
just
suggest
that
diagnostics
be
part
of
the
conversation
in
all
of
the
breakout
rooms
and
then
another
was
on
greenland,
and
I
know
that
a
lot
of
the
variable
and
high
resolution
modeling
work
with
like
refined
regional
pam,
is
focused
on
on
greenland
simulations.
G
So
I
want
to
make
sure
we
stay
on
time
because
we
do
have
a
tight
schedule,
so
we're
going
to
get
started
here
with
our
second
set
of
talks,
as
in
the
earlier
session.
Every
speaker
has
a
10-minute
slot,
but
we
want
you
to
do
an
8-minute
talk,
and
so
I
will
cut
you
off
at
seven
minutes
or
tell
you
to
hurry.
G
Hur
tell
you
to
finish
up
in
about
seven
minutes,
and
so
without
any
further
ado,
and
for
those
of
you
who
don't
know,
I'm
marika
holland,
I'm
one
of
the
coaches
of
the
polar
climate
working
group.
So
we
are
going
to
go
ahead
and
get
started
with
molly
ranga.
I'm
talking
about
data
assimilation,
applications
for
urtic
sea
ice,
so
molly.
If
you're
able
to
share
your
screen
and
go
ahead,
yeah.
R
Let
me
can
everybody
see,
hopefully
not
the
presented
version.
It
looks
great,
okay,
wonderful,
great.
Well,
then,
I
will
get
started
hello.
Everyone
thank
you
for
having
me.
I'm
really
excited
to
to
be
here.
I'm
a
second
year
student
at
the
university
of
washington,
and
I'm
excited
to
talk
to
you
about
kind
of
the
beginning
of
our
team's
efforts
to
improve
21st
century
estimates
of
arctic
sea
thickness
variability
and
change,
in
particular
through
a
data
assimilation,
application
I'd
like
to
to
start
by
acknowledging
the
rest
of
the
team.
R
That's
helping
me
do
this,
my
advisor
cecilia
bits
and
doctors
alec,
petty
and
andrew
roberts
at
nasa
and
los
alamos
national
lab
respectively.
I'd
also
like
to
acknowledge
the
votes
at
encar,
who
have
been
really
helpful
in
getting
this
project
off
the
ground,
as
well
as
support
from
the
university
of
washington
and
from
nasa
so
to
contextualize
what
I'll
present
today.
Our
interest
in
this
problem
is
really
motivated
by
a
desire
to
understand
recent
and
future
sea
ice
change.
R
Some
of
this
motivation
can
be
explained
by
this
figure,
which
shows
three
commonly
sea
ice
thickness
products,
icesat
in
royal,
blue
cryosat-2
and
teal,
both
with
an
uncertainty
bound
and
then
biomass,
which
is
a
reconstructed
product
in
the
dash
black
line,
while
piones
does
a
decent
job
of
capturing
the
trend
of
cs
thickness
over
time
and
is
continuous
over
many
decades
several
decades,
its
variability
doesn't
line
up
with
the
satellite
products,
and
it's
well
known
to
underestimate
thick
ice
and
overestimate
the
nice.
R
So
we
believe
that
we
can
address
kind
of
our
two
wish
list
items
and
our
forecasting
challenge
and
our
reconstruction
problem
by
improving
that
by
assimilating
sea
ice
thickness
data
into
a
fully
dynamic
sea
ice
model.
So
our
long-term
goal
is
to
produce
a
new
sea
ice
thickness
record
from
the
early
2000s
to
the
present
day.
R
The
work
that
I'm
going
to
present
today
is
a
far
cry
from
that
goal,
but
it
builds
the
foundation
for
eventually
reaching
it
so
to
build
a
framework
and
an
intuition
for
how
data
assimilation
works
with
sea
ice.
We've
been
testing
its
efficacy
in
with
observing
system
simulation,
experiments
or
aussies.
R
R
To
then
observe
that
state,
by
creating
synthetic
observations
from
the
truth
from
our
randomly
selected
truth
and
then
to
localize
our
assimilation,
forecast
forward
and
evaluate
the
impact
we
sample
the
state
using
a
30
member
ensemble
of
size
5.,
the
model
is
configured
using
a
slab
ocean
and
a
jri
55
reanalysis
atmosphere.
R
We
then
randomly
select
one
of
those
ensemble
members
to
be
our
our
truth
and
then,
and
the
figure
that
I'm
showing
for
the
ensemble
here
on
the
right,
the
ensemble
mean
is
shown
in
red
and
our
randomly
selected
truth
member
is
shown
in
blue,
and
so,
as
you
can
see,
our
truth
is
systemically
lower
than
the
ensemble
mean
in
these
experiments,
and
so
we
infer
that
the
ensemble
in
this
perfect
world
is
biased
high.
R
R
We
observe
the
state
by
using
the
truth,
integration
to
mimic
real-world
satellite
measurements,
so
this
mimicry
is
crude
in
our
experiments,
but
it
captures
the
daily
domain
of
the
icesat-2
satellite,
which
is
the
newest
cs,
thickness,
observing
system
the
available
ice
set
to
a
long
track.
Data
for
april
15
and
2019
are
shown
on
the
left
from
open
altimetry,
while
the
pseudo
tracks
that
I
generated
by
sampling
along
the
model
grid
from
the
truth
member
are
shown
on
the
right.
R
We
ran
a
number
of
experiments,
assimilating
different
members,
different
numbers
of
these
pseudotracks,
so
the
darker
colors
on
the
figure
just
represent
the
first
assimilation.
While
the
lighter
colors
represent
additional
tracks
that
were
added
in
each
experiment,
we
assimilate
a
single
day
of
the
pseudotrack
observations
into
the
ensemble,
using
a
localization
of
700
kilometers,
so
limiting
the
effect
of
that
assimilation
spatially
and
we
chose
700
because
it
was
shown
to
be
the
length
scale
of
the
cs,
thickness
anomaly.
R
And
then
we
assimilated
that
on
april
15th
and
then
also
on
october
15th
of
2019
attempting
to
capture
the
impact
of
the
assimilation.
In
times
of
both
ice
loss
and
ice
growth.
R
R
The
results
for
the
spring
forecast
are
shown
here
so
to
just
kind
of
warrant
you
to
the
figure
on
the
top
row,
I'm
showing
you
the
root
mean
squared
error
on
day
45
on
the
bottom
row
at
day.
75.
The
first
column
is
a
case
with
no
assimilation.
R
The
middle
column
is
if
we
assimilate
just
a
single
track
and
then
the
third
column
is
if
we
assimilate
the
full
15
tracks
across
our
spatial
domain,
and
you
can
see
that
as
we
move
from
the
left
to
the
right
on
this
figure
as
we
simulate
more
data
in
a
single
day,
our
mean
squared
error
drops,
notably,
which
is
good.
R
We
like
that
and
then
the
other
interesting
thing
to
kind
of
pull
away
from
this
figure
is,
if
you
look
at
this
one
track
assimilation
case,
there's
a
notable
reduction
in
root
mean
squared
error
at
day
45,
but
that
effect
is
really
lost
by
day
75.
R
So
there's
an
advantage
to
incorporating
more
data,
because
your
reduction
root
mean
squared
error
will
be
persistent
through
longer
times,
which
we
can
look
at
a
little
more
specifically,
I'm
just
looking
at
the
the
15
track,
assimilations
for
a
longer
integration,
I'm
showing
you
the
time
series
of
that
integration
on
the
top
right,
where
the
green
is
the
the
ensemble
mean
unadjusted,
the
red
is
the
truth.
R
You
see,
we
reduce
down
to
less
than
0.2
meters
root,
mean
squared
error,
and
while
we
start
to
get
some
of
that
bias
back
as
we
move
forward
in
the
integration,
the
the
250
days
after
the
assimilation
is
still
a
drastic
improvement
over
the
unassimilated
case,
so
to
kind
of
pop
up
yeah.
Very
briefly,
thank
you.
Marika.
R
Our
main
result
is
that
in
a
perfect
world
assimilating
along
track
arctic
sea
ice
thickness,
estimates
persistently
reduces
model
bias
for
approximately
three
months,
and
it
still
improves
on
unassimilated
cases,
for
up
to
eight
months
could
potentially
be
longer
a
stop.
The
integration
at
the
end
of
2019
and
then
next
steps
we're
looking
at
doing
a
sequential
day
by
day,
assimilations
and
experimenting
with
realistic
errors
by
testing
some
direct
assimilation
of
the
cs,
thickness
distribution
and
then
eventually,
we'll
also
transition
from
these
perfect
model
experiments
to
real
world
data.
R
So
with
that,
thank
you.
I
appreciate
the
opportunity
to
speak.
I'd
be
happy
to
take
any
questions
and
I'll
just
put
the
summary
slide
back
up.
E
I
was
wondering
if
there's
a
computing
cost
with
with
the
increasing
number
of.
R
There
is
a
cost
to
to
running
dart
with
csm
and
with
sis,
given
that
the
adjustment
happens
in
the
restart
files
for
the
model,
so
there's
a
lot
of
stopping
the
model
adjusting
it
restarting
it,
and
so
there
there's
there
is
a
high
computational
cost
to
doing
this
kind
of
online
assimilation
and
a
lot
of
a
lot
of
files
that
get
output
and
need
to
be
managed.
So
there's
also
a
big
storage
management
aspect
as
well.
G
G
M
Rika,
can
you
see
my
screen?
Okay,
yep.
M
Yeah
hi
everyone,
I'm
abby
smith
and,
as
I
said
today,
I'm
going
to
be
discussing
model
satellite
comparisons
of
sea
ice
melt,
onset
with
the
satellite
simulator
and
I
am
at
the
research
applications
lab
at
ncar
now,
but
most
of
this
work
was
done
as
a
final
part
of
my
phd
thesis
that
I
recently
completed
so
happy
to
be
back
and
getting
another
chance
to
talk
about
that
and
see
the
polar
climate
working
group
again.
So
thanks
for
having
me.
M
However,
observational
male
onset
products
are
derived
from
satellite
observations
of
brightness
temperatures,
which
are
not
currently
simulated
by
climate
models,
and
so
that
means
that
model
definitions
of
male
onset
tend
to
be
based
on
other
model
variables
that
are
sensible
but
vary
such
as
snowmelt
or
surface
temperature.
M
So
what
I'm
showing
here
is
an
example
from
the
cesm1
large
ensemble
and
the
red
line
are
the
observed,
melt,
onset
dates
for
each
year
and
the
blue
line
down
here
is
one
ensemble
member
of
mel
onset
using
a
thermodynamic
ice
volume,
tendency
definition
as
well
as
the
ensemble
spread,
and
then
we
have
surface
temperature
and
snowmelt
definitions,
as
well
as
an
additional
ensemble
spread
of
surface
temperature,
no
onset,
and
what
we
see
is
that
that
volume
tendency
definition
that
is
impacted
by
ice
melt
is
dominated
by
bottom
melt
and
tends
to
occur
earlier
than
melt
at
the
surface.
M
And
that's
this
is
a
pan
arctic
mean,
but
that's
particularly
true
where
warm
water
is
entering
the
arctic
from
further
south,
causing
those
melt.
Onsite
dates
to
be
particularly
early,
but
the
surface-based
definitions,
such
as
snowmelt
and
surface
temperature,
tend
to
fall
closer
to
the
observations,
but
slightly
earlier
than
observed,
melt
onset.
So
what
we
can
see
from
this
is
that
no
one
model
definition
is
exactly
capturing.
M
Brightness
temperatures
in
the
model
that
are
observed
by
satellites
and
we're
doing
that
by
using
the
arc
30
satellite
simulator
that
was
recently
developed
by
clara
burgard
and
dirk
knotts
at
mpi,
and
we
have
adapted
that
to
run
with
the
ces
m2
jra
55
ocean
ice
heimkast,
so
really
trying
to
get
at
a
close
as
possible
representation
of
what
the
satellites
would
be
seeing
by
using
that
re-analysis
atmosphere.
M
And
so
what
this
enables
us
to
do
is
actually
create
simulated
brightness
temperatures
that
we
can.
There
then
compare
to
observed
brightness
temperatures.
It
also
gives
us
the
opportunity
to
create
a
new
metric
and
we're
going
to
call
this.
The
simul,
the
earliest
snowmelt
estimation
and
the
great
thing
about
this
new
metric
is
that
it's
a
slightly
simpler
version
of
the
next
big
step
into
processing
those
brightness
temperatures.
M
And
while
we
have
these
new
metrics,
we
can
also
compare
them
to
the
most
current
method
of
comparison,
which
is
to
use
satellite
derived
sea
ice
melt,
onset
products
that
are
brightness
temperatures
that
have
been
processed
through
different
algorithms
and
various
algorithms
exists
and
comparing
those
to
what
I
mentioned
before,
which
is
not
onset,
dates
derived
from
other
model
variables.
M
What
I'm
showing
here
is
that
new
metric
that
I
have
introduced
that
we
can
apply
to
both
the
simulated
and
observed
brightness
temperatures
in
the
same
way
to
produce
earliest
snowmelt
estimation
dates,
and
that
is
shown
for
two
different
satellite
products,
the
dmsp
as
well
as
amster
and
then
on.
The
right
is
the
earliest
snowmelt
estimation
dates
from
this
esm2
jra
55.
M
If
we
remember
those
definitions
of
melt
onset
that
I
mentioned
before,
we
can
apply
them
to
the
same
data.
What
I'm
showing
here
is
just
an
example
from
2003,
and
we
see
a
definition
based
on
snowmelt
and
one
based
on
surface
temperature
and
the
comparison
that
we
would
have
done
before
to
those
algorithm-derived
male
onset
dates
from
the
satellite
observations.
M
G
M
That's
great
thanks.
I
think,
where
I
left
off
was
probably
just
making
the
point
that
what's
great
about
these
more
direct
comparisons,
is
that
you
can
actually
feel
more
confident
that
we
can
diagnose
a
model
bias.
So
we
haven't
changed
the
direction
of
the
bias
or
made
it
go
away.
By
introducing
this
new
metric,
we
can
just
more
confidently
categorize
it
as
a
model
bias,
since
we've
removed
that
uncertainty
due
to
definition,
differences.
M
One
thing
I'll
mention
before
wrapping
up
quickly-
and
this
is
just
an
example
from
one
grid
cell
of
the
evolution
of
brightness
temperature
from
january
to
through
june,
and
we
can
see
that
the
simulated
brightness
temperatures
in
red
tend
to
closely
follow
the
observed
brightness
temperatures
over
this
time
period,
and
so
that
gives
us
confidence
that
what
we're
representing
in
the
model
is
what
we
actually
are
observing
on
the
ground.
M
Using
these
satellite
observations,
and
since
we
have
this
similarity,
what
we
can
also
do
is
to
leverage
other
model
variables
to
inform
what
these
changes
in
brightness
temperature
actually
mean.
So
there
is
uncertainty
in
these
retrieval
algorithms
that
process
brightness
temperatures
into
mel,
as
shown
by
a
blis
at
all
paper
in
2017,
and
so
we
have
an
idea
of
what
these
brightness
temperatures
represent,
but
that's
difficult
to
verify.
M
Approximately
75
of
it
will
melt
by
the
time
that
satellite
algorithm
picks
up
and
it
detects
continuous
melt
onset.
So
this
is,
of
course,
only
just
using
one
model,
but
is
a
really
interesting
example
of
how
we
can
also
use
these
more
direct
comparisons,
not
only
to
evaluate
climate
models
but
to
inform
our
process
understanding
of
what
observations
actually
mean.
So
I
think,
I'm
probably
over
time
already
so
I'll.
Just
leave
my
conclusions
up
here
but
happy
to
answer
any
questions.
Thank
you.
G
M
Have
really
adapted
the
input
data
from
the
cesm2
to
fit
within
the
current
existing
framework
of
the
simulator,
which
is
publicly
available
on
github,
and
I
can
put
the
link
to
that
in
the
chat.
If
anyone
is
interested
since
I've
kind
of
changed
gears
a
little
bit,
I
probably
will
not
be
carrying
forward
the
simulator
development
and
into
the
into
the
actual
size
code,
but
I
think
there's
a
lot
of
opportunity
for
that
and
it
would
be
a
really
cool
advancement.
M
Yeah,
and
so
I
think,
right
now-
we've
done
I've
only
shown
one
year
today.
We've
done
this
for
three
years,
but
there
are,
you
know
we
do
have
ideas
and
plans
to
expand
it
to
kind
of
a
multi-decatal
time
series,
and
that
would
be
something
that
we
would
definitely
provide
on
an
either
data
center
or
somewhere
else
to
be
used
by
other
people.
I
M
I'm
not
I'm
not
sure,
I'm
exactly
understanding
the
question,
but
yes,
the
dissertation
is
published
in
the
full
version
of
the
the
paper
draft
is
in
that
document
and
we're
hoping
to
submit
to
the
cries
here
pretty
soon
too.
I
did
have
maps
of
the
earliest
amount
dates
here.
If
that's
what
you
were
looking
for,
but
there's
also,
you
know
in
my
past
work
to
other
other
representations
of
the
melt,
onset
dates
derived
from
algorithms
as
well,
and
other
people
have
looked
at
that
too.
G
So
in
the
interest
of
time,
why
don't
you
guys
take
this
offline
and
make
sure
that
you
guys,
you
know,
get
the
correct
pdf
to
the
correct
person,
but
we
need
to
move
on
to
the
next
talk,
so
thanks
abby.
That
was
great.
So
so
now
we're
gonna
move
on
to
chris
wyburn
powell
and
chris.
Hopefully
you
can
share
your
screen
and
he's
gonna
be
talking
about
realism
of
simulated
internal
variabilities.
Yes,
okay
thanks
chris.
E
E
So
the
figure
on
the
bottom
right
here
we
have
the
csm
large
ensemble
in
blue
and
observations
in
red
and
because
we
only
have
one
realization
of
reality.
It
is
hard
to
use
the
same
methodology
to
compare
observations
and
models
with
the
large
ensemble.
We
can
easily
say
that
the
internal
variability
can
be
represented
by
the
spread
between
these
ensemble
members,
so
this
is
where
resampling
comes
in.
E
E
So
the
resampling
technique
we
use
uses
the
idea
that
if
we
have
a
time
series,
the
linear
trend
from
that
time,
series
can
be
represented
by
the
forced
response
and
internal
variability
is
represented
by
anomalies
from
that
linear
trend.
So
the
figure
in
the
bottom
left
shows
the
september
sea
ice
area
from
1979
to
2020
the
gray
line
showing
the
linear
trend
and
then
on
the
right.
We
have
the
anomalies
from
that
linear
trend
representing
internal
variability,
and
this
is
following
mechanimetal
2017,
where
they
looked
at
north
american
surface
temperatures.
E
E
Once
we've
done
this
free
sampling
1000
times,
we
can
call
the
output
from
this
a
synthetic
ensemble,
and
this
basically
can
be
shown
in
the
figure.
The
bottom
right
here,
where
there
are
a
thousand
different
evolutions
of
sea
ice,
which
we
determine
could
have
been
possible
if
internal
variability
were
different,
and
we
note
here
it
still
has
the
same
trend
for
all
of
these
evolutions.
E
Although
the
variability
year
to
year
is
different
and
we
can
do
the
same
thing
with
resampling
for
sea
ice
concentration,
and
so
here
we're
showing
the
anomalies
from
the
trend
of
each
of
these
grid
cells
for
years
96
to
2001.,
just
as
an
example
and
where
red
shows
anomalies
above
trend
and
blue
below
trend.
E
E
Then,
when
we
create
a
synthetic
ensemble
of
1003
samplings,
we
can
also
take
the
standard
deviation
across
those
1000
free
samplings.
So
the
distribution
has
a
resampling
for
the
each
individual
member
in
the
middle.
Here
we
show
member
3
from
csm1
and
then,
when
we
sample
the
observations,
we
also
have
the
standard
deviation
of
that
distribution,
and
so
these
are
the
three
metrics
we'll
be
looking
at
in
the
results.
E
So
the
figure
on
the
right
here
shows
the
ratio
of
that
so
values
below
one
shows
that
the
large
ensemble
spread
is
larger
than
that
which
we
have
recreated
with
our
1000
free
samplings,
and
so
what's
important.
To
note
here
is
that
approximately
65
to
90
percent
of
the
standard
deviation
of
the
large
ensemble
is
reproduced
when
we
do
our
resampling
technique.
E
So
it's
important
to
note
here
that
the
ratio
that
we
get
as
a
result
is
highly
dependent
on
the
realization
which
we
use
and
so
in
the
rest
of
the
results,
especially
spatially,
we'll
be
looking
at
the
average
of
the
medium
from
these.
All
of
the
realizations
from
the
large
ensembles.
E
Okay,
yeah
the
I'll
go
through
this
quickly.
So
in
conclusion,
from
the
results
we
have
the
spring,
we
have
much
larger
internal
variability
by
this
metric
than
the
observations.
Although
in
september,
when
we
look
at
the
pan,
arctic
models
and
observations
agree
fairly.
Well,
we
can
see
the
ratio
here
is
close
to
1..
E
E
So
in
conclusion,
here
we
have
shown
that
resampling
captures
a
majority
of
the
large
ensemble
internal
variability,
allowing
more
direct
comparisons
of
observations,
although
this
mainly
focuses
on
inter-annual
variability,
and
there
is
less
low
frequency.
Variability
captured
with
this
technique
in
general
models
simulate
larger
internal
variability
than
observations
when
compared
with
resampling
and
in
summer
internal
variability
was
modeled
relatively
well,
with
the
largest
overestimation
in
march,
due
to
the
sea
ice
edge
and
around
the
coastal
regions.
Thank
you
for
listening
and
I'll.
Take
any
questions.
I
Yes,
nice
talk.
How
did
you
come
up
with
the
number?
A
thousand
for
that
resampling
and
do
the
results
depend
on
that
number.
E
Yeah,
thank
you
for
the
question
so
that
the
number
of
resamplings
shouldn't
influence
the
the
distribution
of
data
we
get
out.
The
resampling
technique
should
reproduce.
If
we
have
a
standard
deviation
in
our
data,
then
the
sampling
should
also
produce
a
sorry.
E
A
normal
distribution
in
our
data,
a
thousand
is
just
seems
to
be
a
typically
used
number,
which
is
more
than
sufficient
to
accurately
reproduce
the
same
distribution
of
data,
and
so
that
was
what
was
used
in
the
atar
2017
paper
and
that
seems
to
be
consistent
across
previous
research.
A
thousand
is
kind
of
a
conservative
number
to
make
sure
that
that
is
a
robust
way
of
measuring
it.
G
Thanks
chris,
so
if
you
have
any
other
questions
for
chris,
please
put
them
in
the
chat
and
we'll
move
on
to
our
next
talk,
which
is
from
alice
to
vivier.
Talking
about
impacts
of
sea
ice
mushy
thermodynamics
in
anarchic
simulations.
Q
And
it
looks
great
alice.
Okay,
great
thanks.
Let
me
just
move
all
the
pictures,
so
I
can
see
the
time
okay,
so
the
purpose
of
this
experiment,
or
that
what
I'm
going
to
talk
to
you
about
today
is
we
wanted
to
further
explore
impacts
of
sea
ice
thermodynamics
in
the
coupled
system.
If
people
were
at
the
polar
climate
working
group
and
meeting
in
february,
I
talked
about
this
some,
so
this
is
a
bit
of
a
conclusion
from
that.
Q
Okay,
all
right,
so
I
think
chris
and
I
were
looking
up
the
same
graphics
about
apples
and
oranges.
But
what
just
to
give
you
a
little
summary
of
what
we
found
here?
Hansey's
paper
on
analyzing,
the
cesm2
experiment,
found
that
while
the
sea
ice
state
was
not
super
different
in
the
antarctic,
we
found
that
there
were
big
differences
in
the
frazzle
ice
production.
So
hopefully
you
can
see
my
mouse
here,
but
you
can
see
there's
like
a
ring
of
fire
around
the
coast
of
antarctica,
but
that's
not
actually
volcanism.
Q
That's
frazzle
ice
production
or
production
of
ice
in
open
ocean
waters,
and
that
was
not
present
in
cesm1,
and
so
one
of
the
big
questions
is:
why
does
this
exist
and
one
of
the
the
immediate
culprits
is
that
cesm2
uses
this
new
sea
ice
thermodynamics
that
we
call
the
mushy
layer
of
thermodynamics,
developed
by
adrian
turner,
where
there's
prognostic,
sea
ice,
salinity
and
ice
is
a
combination
of
like
mushy
solid
ice
and
it's
a
mushy
combination
of
solid
ice
and
brine.
Q
In
contrast,
cesm1
uses
the
bits
and
lipscomb
99
thermodynamics,
where
the
salinity
is
not
prognostics
prognostic,
but
there
are
a
ton
of
other
differences
between
these
two
models,
so
we
wanted
to
be
able
to
compare
more
apples
to
apples
again
looking
up
similar
graphics.
So
what
we
did
is
we
ran
two
cesm2
experiments
where
the
only
thing
we
changed
was
the
sea
ice
thermodynamics.
So
one
of
these
uses
the
bits
and
lipscomb
99,
and
one
of
these
uses
mushy
they're,
both
pre-industrial
experiments,
they're
100
years
daily
and
monthly
data
are
available.
Q
Q
So
what
I
want
to
point
out
is
that,
where
you
see
the
biggest
differences
in
frazzle
ice
production
are
right
here
along
the
coast
and
this
area.
This
coastal
area
accounts
for
about
56
of
the
total
thermodynamic
sea
ice
growth
in
the
winter
in
the
antarctic.
So
this
is
one
of
the
reasons
I'm
focusing
on
what's
happening
on
the
coast.
Q
When
we
look
at
the
coastal
ice
mass
budget
throughout
the
throughout
the
year,
we're
looking
at
this
figure,
I
want
you
just
to
focus
on
the
net
budget,
the
net
thermodynamic
terms
and
then
the
dynamic
terms.
So
those
are
the
colors
here.
I
just
want
to
point
out
that
there
are
significant
increases
in
the
thermodynamic
growth,
so
this
gray
line
right
here
in
all
months.
Q
The
little
diamonds
indicate
that
the
difference
is
significant,
so
throughout
the
whole
melt
or
throughout
the
whole
growth
season,
there's
an
increase
in
thermodynamic
growth
in
the
mushy
experiment.
Again,
I'm
showing
differences
here
and
but
that
is
somewhat
compensated
by
an
increase
in
dynamic
loss.
So
that's
what
we're
seeing
right
here
and
so
on
the
whole.
The
net
budget
is
not
very
different.
It's
a
little
different
at
the
beginning
of
the
melt
season
or
at
the
beginning
of
the
growth
season.
Sorry,
and
then
it
becomes
not
statistically
significant.
Q
Differences,
so
what
we're
really
interested
in
is
what
this
does
to
the
ocean
and
the
atmosphere
around
antarctica.
So.
Q
Here
is
a
cross
section
averaged
across
this
admins
and
bellinghaus
and
c
sector.
That's
the
star
here,
and
what
we
see
is
that
when
we
look
vertically
so
the
the
y-axis
here
is
depth
through
the
ocean,
the
the
ocean
water
has
become
more
saline
in
the
mushy
experiment.
Q
Again
these
are
differences,
mushy
minus
bits
and
lipscomb
99,
and
so
we
find
that
originating
at
the
surface
where
the
sea
ice
is
forming,
the
ocean
is
becoming
denser,
and
this
is
true
in
all
the
sectors,
so
I'm
just
showing
the
the
cross
section
here
for
the
admins
and
bellingham
c,
but
you
can
look
when
we
look
when
we
just
look
at
100
meter
depth
that
there
are
increases
in
density
almost
everywhere.
Q
So
that's
the
red
colors,
and
when
we
look
at
the
the
changes
in
salinity,
we
see
that
there
are
increases
in
salinity
pretty
much
everywhere
in
all
the
sectors
as
well,
and
they
very
the
the
spatial
pattern
is
very
similar
to
what
we
see
for
the
density
itself.
Q
This
is
a
reflection
of
the
increase
in
thermodynamic
growth
in
these
coastal
regions
in
the
mushy
experiments.
So
in
cesm2,
when
ice
is
formed,
fresh
water
is
basically
removed
from
the
ocean,
and
so
the
ocean
will
become
more
saline
and
that
the
changes
in
temperature,
which
is
what
I'm
showing
here
in
the
far
right
column,
are
not
significantly
contributing
to
these
changes
in
density
and
they're,
not
consistent,
necessarily
with
depth
or
bisector.
Q
So
why
this
is
relevant
is
that
it
impacts
the
overturning
of
the
the
impacts,
the
meridional
overturning
circulation.
So
what
you
can
see
here
is
just
the
annual
mean
of
the
moc
for
the
mushy
experiment,
so
we
have
the
upwelling
kind
of
water
here
and
then
down
welling
here
and
when
we
look
at
the
difference
we
can
see.
There
are
significant
decreases
here
in
the
mushy
experiment
of
the
overturning
at
lower
densities.
This
is
density
space
here
and
then
we
have
increases
at
higher
densities.
Q
And
so
what
that
looks
like
when
we
look
at
a
time
series
over
the
whole
hundred
year,
experiments
is
that
this
dark
line
is
the
10-year
running
mean,
but
the
antarctic
bottom
water
production
is
like
slightly
higher,
but
the
difference
is
significant
through
pretty
much
the
entire
hundred
years,
with
the
mushy
experiment
compared
to
the
bits
and
lip
guns.
So
basically,
the
volumetric
flow
rate
has
increased
for
these
higher
densities.
The
higher
densities
exist
because
we're
making
more
ice
and
getting
saltier
denser
water.
Q
I
told
you
that
I
would
talk
a
little
bit
about
what's
happening
in
the
atmosphere.
The
short
answer
is
much
less.
When
we
look
at
differences
in
the
sea
level,
pressure
they're
insignificant,
pretty
much
everywhere,
so
there's
not
really
any
significant
circulation
change.
There
are
small
differences
in
the
precipitation,
but
they're
not
consistent
by
sector,
and
they
don't
seem
to
be
driving
any
of
the
changes
in
snow
ice
formation.
That's
related
to
the
thermodynamics
as
well
the
mushy
thermodynamics.
Q
We
do
find
that
there's
a
slight
decrease
in
the
cloudiness
around
antarctica,
but
it's
very
small,
and
even
though
it's
significant,
it's
not
a
huge
difference.
So
I
know
I'm
almost
done
with
my
time
here.
Marika,
you
don't
have
to
remind
me,
but
where
this
is
going
now
is
what
I'm
working
on
at
this.
The
newest
work
that
I've
been
doing
is
using
machine
learning
or
self-organizing
maps,
with
some
of
the
cesm2
output
to
try
to
understand
the
processes
driving
these
polenias
or
open
ocean
areas
around
the
coast.
Q
I'm
looking
in
more
regional
areas
and
I'm
trying
to
understand
if
what's
changing,
that,
what's
driving
the
plenius
changes
over
time
and
if
that
might
change
into
the
future,
so
I'm
just
showing
this
is
very
much
in
progress,
but
a
song,
that's
trained
with
sea
ice
thickness
in
the
raw
sea
and
you
can
see
just
the
coastal
sea
ice
thickness.
There
are
some
times
where
it
finds
there's
you
know
very
little
very
thin
ice
around
the
rossi
and
we
know
that
a
polenia
occurs
here.
Q
D
Q
So
I
would
say
that
the
mushy
layer
is
more
more
consistent
with
how
the
sea
ice
itself
evolves
in
time.
So
I
think
that
I
would
say
that
maybe
that
one
is
more
realistic.
I
haven't
done
a
close
comparison.
You
know
I've
tried
to
find
observations
of
ice
density
and
ice
temperature,
like
you
know,
throughout
the
ice
in
antarctica
and
those
don't
really
exist.
So
it's
hard
to
to
say,
like
you
know,
it's
getting
it
spot
on
or
something
like
that,
but
I
do
think
the
prognostic,
salinity,
evaluation
or
evolution.
Q
Sorry
is
is
more
realistic
with
what
actually
happens
in
sea
ice.
D
Q
A
There's
a
caveat
too
that
we're
not
doing
true
salt
flux,
coupling
we're
still
assuming
for
psu,
and
so
that's
one
of
the
things
I'm
working
on
fixing.
G
H
Know
it's
a
technical
question
too
good
neat
talk,
alice,
so
good
to
see
this
work
done
and
I'm
curious
if
you
thought
about
trying
to
adjust
the
fixed
salinity
profile
in
the
bits
and
lipscomb
model
to
something
more
like
a
saline,
fresh
ice
distribution
or
profile,
because
currently
it's
very
fresh
at
the
surface.
Q
Yeah
I
have
thought
about
that
and
like
how
how
you
and
bill,
I
guess,
decided
on
the
profile
you
use.
I
have
not
done
experiments
where
I
did
adjust
that,
because
part
of
the
purpose
of
these
is
of
this
analysis
is
just
to
understand.
Q
You
know
that
consists
that
evaluation
more
consistent,
I
think
you
know
one
of
the
things
we
saw
was
there
were
there
were
differences
in
the
overturning
the
antarctic
bottom
water
formation
in
cesm2,
and
so
that
was
another
one
of
the
the
motivations
is
to
understand
like
is
this
being
caused
by
the
thermodynamics
or
other
things
going
on,
and
and
so
then,
if
we
wanted
to
see
how
it
was
different
from
csm1,
I
wanted
to
keep
the
same
profile
that
you
guys
had
used,
but
that
is
something
we
could
do:
try
to
make
it
more
similar
to
mushy,
because
this
the
ice
profiles
are
very
different.
Q
G
H
H
This
is
really
largely
the
work
of
june
jai
and
myself
at
university
of
washington.
So
I'm
going
to
acknowledge
her
best
contribution
to
this
work
and
what
we're
interested
in
is.
H
Can
we
actually
avoid
a
model
like
a
dynamical
model
and
instead
use
a
machine
learned
model
of
sea
ice
motion
to
make
a
superior
forecast,
and
by
that
I
mean
kind
of
what
the
next
time
step
forecast
of
sea
ice
motion
is
so
why
we
might
hope
to
do
this
is
that
you
know
the
sea
ice
motion
feels
very
complex,
and
there
are
lots
of
you
know
sophisticated
dynamics
that
go
into
the
dynamical
model,
computation
of
things
that
could
cause
it
to
kind
of
lock
up
at
the
ice
edge
and
exhibit
you
know,
complex,
dynamical
behaviors
and
it's
you
know
questionable
whether
we
actually
know
the
right
physics.
H
I
think
we
obviously
have
the
right
momentum
equation
for
f
equals
m
a,
but
it's
this
interaction
of
ice
with
itself.
That
is
uncertain,
and
it's
really
largely
in
models
based
on
kind
of
rules
of
what
we
think
are
the
material
relationships
of
sea
ice,
specifically
the
stress
to
the
strain
rate
or
possibly
other
scalers,
that
dictate
this
ice
material
property.
H
So,
let's
just
see
if
we
can
avoid
all
that
and
go
right
ahead
and
calculate
the
ice
motion
field
with
a
statistical
model,
and
I
think
this
might
be
a
good
example
of
a
test
problem
for
this
kind
of
work,
because
there's
such
a
wide
range
of
sea
ice
concentration
seasonally.
So
we
really
have
this
wide
range
of
data
from
which
to
train
our
model
and
the
basic
method.
H
We're
going
to
use
is
to
train
our
statistical
models
with
the
sea
ice
concentration
at
one
time,
step
time,
step
0
and
the
ice
motion
at
that
time.
Step
as
well,
for
both,
u
and
v
components
of
the
ice
velocity
and
then
we're
going
to
include
the
winds
of
the
current
time
that
we're
interested
in
forecasting.
H
So
we
assume
that
we
know
the
wind,
but
we
don't
know
the
ice
motion
and
concentration
also,
so
that's
our
basic
model
and
the
time
step
I'm
going
to
propose
is
one
day
and
that's
because
the
ice
velocity
data
that
we
have
is
available
only
once
daily,
but
we
first
started
this
whole
exercise
with
using
ice
and
ice
motion
output
from
cesm2,
as
well
as
the
concentration
and
thickness.
H
H
So
we
tried
various
models
built
on
these
predictors
and
the
method
I'm
showing
results
for
here
is
a
convolutional
neural
network
to
just
show
you
what
kinds
of
statistics-
and
these
are
our
skill,
statistics
of
the
correlation
or
anomaly
correlation
coefficient
and
one
minus
a
normalized
root-
mean
square
error
and
you
can
see
actually
don't
vary,
a
ton
which
is
good
news,
and
that
really
means
hey.
We
can
probably
just
chuck
the
thickness,
which
is
great,
because
obviously
we
don't
observe
the
thickness
very
well.
H
And
basically
the
lesson
here
is
that
ice
motion
or
winds
ice
motion
from
the
previous
step
or
winds
of
the
current
step
are
good,
predictors
themselves,
which
is
a
very
fortunate
situation
to
be
here
and
the
metric
or
the
the
time
step
here
was
one
hour,
but
if
we
shrunk
it
down
to
one
day,
it
actually
increased
the
skill.
So
that's
nice
too,
so
just
to
show
you
a
whole
range
of
models
to
begin
with
compared
to
the
truth
here,
which
is
a
satellite
based
observation
from
a
past.
H
Sorry,
the
polar
pathfinder
primarily
compared
to
the
convolutional
neural
net,
the
dynamical
model
size
at
a
one
degree
resolution.
Another
machine,
learned,
model,
multi-layer,
perceptron,
a
random
forest
machine,
learn
model
and
then
just
a
simple
linear
regression,
and
obviously
all
of
them
look
pretty
good,
but
they
do
differ
in
some
qualities,
such
as
statistically
I'll.
Show
you
in
a
minute,
but
for
this
just
one
random
day,
you
can
kind
of
get
the
sense.
H
That
slice
looks
like
it's
too
strong
and
the
pattern
looks
good
and
then
there
aren't
forest
tends
to
be
a
little
bit
kind
of
erratic
linear
regression
is
a
little
too
smooth.
So
let's
just
go
and
now
look
at
some
actual
comprehensive
statistics
for
all
of
one
year
of
test
training
data.
So
this
is
for
year
2015.
H
I
should
say
we
train
the
model
on
something
like
1990
to
2014,
so
something
like
25
years
of
training
data
and
what
we
found
for
this
correlation
statistic.
H
It
is
like
an
anomaly
correlation
coefficient
where
x
is
the
truth
and
y
is
the
model
and
you're
looking
at
basically
the
correlation
between
the
two
in
a
pattern
and
then
summing
over
all
locations
in
space,
and
we
did
this
for
u
and
v
separately,
just
sort
of
made
one
large
vector
for
the
index
I
here
and
what
we
see
is
they're,
all
pretty
good
but
size,
maybe
for
this
group,
is
of
interest
it's
kind
of
in
the
middle
of
the
pack.
H
H
I
should
say
we
were
forcing
the
size
here
with
the
same
reanalysis
product
that
we
used
for
the
wins
in
our
models.
That
are
statistic
at
least
the
other
scale
score
I
used
here
is
more
like
the
normalized
root
mean
square
error
that
I
have
previously.
I
apologize
for
changing
my
name
of
this
variable
or
this
metric.
H
It's
a
ratio
of
standard
deviations
and
that's
quite
a
bit
harder
metric
to
to
look
successful
at
especially
if
the
magnitudes
are
slightly
different
and
that's
where
science
looks
poor
and
that's
because,
according
to
this
version
of
truth,
that
we've
used,
it
overestimates
the
magnitude
size
does
compared
to
to
the
truth.
So
I
think
probably
we
ought
to
redo
this
with
buoy
data,
which
would
be
an
independent
source
of
observations
from
the
training
data
set.
Obviously,
we
separated
in
time
our
testing
and
training
data,
but
it's
still
the
same
basic
training.
H
H
So
now
I'll
just
say
well,
what's
next
is
to
try
to
do
this
comparison
to
a
high
resolution.
Size
version
to
like
I
said,
evaluate
the
methods
against
we.
Data
alone,
evaluate
things
like
divergence
and
shear.
We'd,
really
love
to
implement
this
machine
learning
method
within
size.
Instead
of
so,
it's
basically
strip
out
the
dynamical
model
and
input
a
machine
learning
model,
because
it
should
speed
up
the
code
significantly
and
maybe
increase
the
accuracy.
H
But
I've
been
really
eager
to
try
to
learn.
Actually
the
stress
strain
relation
itself.
Instead
of
doing
the
ice
motion
directly.
So
with
that
I'll,
just
it
and
take
questions
while
leaving
up
my
conclusions.
D
Yeah,
just
totally
a
quick
question:
have
you
I
mean,
given
that
there's
you
considered
like
five
methods,
if
you
consider
looking
at
the
skill
as
a
multi-method
average
or
ensemble,
rather
than
just
solely
focusing
on
machine
learning,
because
if
you
combine
them,
it
looked
like
their
end.
Result
might
be
a
lot
better.
H
I
hadn't
thought
of
that.
It's
a
cool
idea
because
my
motivation
obviously
was
to
actually
eliminate
the
dynamical
model
inside
in
a
climate
model.
It
wasn't
really
on
my
radar,
but
I
think
that's
a
cool
idea.
H
G
H
Interesting
point
and
yeah
the
marginal
improvement
isn't
that
great,
so
maybe
it's
worth
just
going
with
linear.
It
certainly
would
be
simple,
though:
it's
not
that
hard
actually
to
implement
swanky
neural
network
models,
either
it's
just
more
black
boxy
yeah.
G
So
so
bob
so
we
are
gonna
eat
into
our
break
in
just
oh.
We
don't
have
a
break
anymore,
so
I'll,
let
bob
go
and
then
we're
going
to
take
a
one
minute
break
and
come
back
for
the
next
session.
L
H
Great
question-
and
you
know
I
started
by
bragging
about
how
this
might
be
the
perfect
data,
because
a
problem,
because
there's
such
a
wide
range
of
cs
coverage
over
the
seasonal
cycle,
so
you
do
really
have
this
broad
amount
of
large
amount
of
range
of
data
to
train
on.
Another
point,
though,
is
that
the
ice
velocity
hasn't
changed
very
much
over
the
observational
record.
H
There
has
been
some
claim
that
it's
sped
up
a
little
bit.
I
think
that's.
We
don't
really
see
that
evidence
very
strongly.
It's
certainly
not
large
compared
to
the
noise.
So
I
appreciate
your
question
and
I
don't
have
an
answer
other
than
to
say
this
might
be
a
particular
problem
that
is
less
of
an
issue.
G
Thanks,
okay,
so
I
think
there
may
have
been
some
more
questions.
So
if
you
have
other
questions
for
cc,
if
you
could
put
them
in
the
chat
and
because
we
ate
into
our
break
really
effectively,
we
were
supposed
to
have
10
minutes.
We
don't
have
any
time
at
all.
Why
don't
we
come
back
at
1105?
Is
that
okay,
john?
Is
the
next
convener?
F
Absolutely
because
we
have,
we
have
one
presenter
who
canceled
at
the
last
moment,
so
we
have
a
little
bit
of
room,
so
let's
take
a
five
minute
break
and
start
at
1105
with
a
lot
of.
Thank
you
thanks.
G
Yeah,
no,
it's
it's
very
interesting.
I
mean
it's
yeah.
Do
you
how
how
updated
is
the
polar
pathfinder
data
I
mean?
I
guess
like
if
you
were
doing
this
in
practice
and
trying
to
do
a
forecast
for
the
next
day?
Is
that
problematic
that
that,
like
the
ice
motion
data,
it's
not
available
at
you
know
re
at
near
real
time.
H
H
G
It
would
be
interesting
to
see
how
quickly
it
goes
wrong.
I
mean
because
you're,
it's
all
based
on
knowing
the
information
today
to
predict
the
information
tomorrow.
So
if
you
were
doing
that
sequentially
in
a
climate
model
like
do
the
errors
grow
to
the
point
where
you
know
a
month
out,
your
forecast
is
crummy.
G
O
F
Okay,
let's
move
to
our
last
block
of
our
session
with
two
more
speakers
and
then
we
have
our
awesome
breakout
sessions
following
those.
So
we
have
two
more
speakers
focused
on
land
eyes
and
the
first
speaker
is
laura
muntuff
laura
you,
I
think
now
you
can
go
ahead
and
share
your
screen
and
unmute
yourself
there.
You
are.
S
Yes,
so
let
me
just.
I
S
F
It
should
be
somewhere
in
your
view,
let's
see
what
it
looks
like
for
me
view
and
then
full
screen
mode.
F
S
S
S
So
in
that
sense
it's
a
nicely
comparable
pace
of
of
that
and
and
for
the
remainder
of
the
simulation,
which
is
another
210
years.
We
maintain
the
four-time
co2
levels
and
here
I'll
focus
on
two
periods
of
analysis,
which
is
around
the
quadrupling
period
here
in
the
blue
shade
and
at
the
end
of
the
simulation,
the
20-year
average
and
the
results
are
the
the
ic
is
losing
mass.
Of
course.
S
S
S
So
for
the
results,
I
will
pick
two
two
parts
to
talk
about
some
more
one
is
the
cloud
radiative
forcing
and
the
other
one
is
the
ground
heat
flux,
so
the
cloud
radiative
forcing
we
have
defined
so
shortwave
clouds,
I
know
to
reflect
radiation
defined
as
the
the
short
wave
net
radiation
at
the
top
of
the
atmosphere:
minus
the
short
wave
net
radiation
in
clear
sky
conditions.
S
So
this
is
usually
a
negative
quantity
and
cooling
and
long
wave
radiation.
We
define
as
the
outgoing
long
wave
radiation
in
clear
sky
conditions
at
the
top
of
the
atmosphere
minus
the
outgoing
long
wave,
normal
outlet
algorithm
at
the
top
of
the
atmosphere.
Usually
this
is
a
positive
quantity,
as
it's
remitting,
a
long
wave
radiation.
S
S
The
top
panels
here,
spatial
panels
they
show
the
first
one-
is
the
pre-industrial
average.
The
middle
one
is
the
around
stabilization,
so
between
so
around
year
140
and
the
last
one
is
at
the
end
of
the
simulation.
We
see
that
the
most
clouds
cooling
in
the
short
wave
is
in
the
ablation
zone.
S
The
red
line
is
the
equilibrium
line
altitude
and,
as
that
shifts
off
up
more
cooling
is
also
realized
there
from
from
clouds
the
long-wave
cloud
radiative
forcing
is
changing
very
little
and
not
significant
at
all,
and
it's
interesting
to
see
that,
if
I
add
them
up
together
to
find
the
total
cloud
radiation
forcing
its
positive
quantity.
So
in
the
pre-industrial
time,
clouds
are
warming.
The
ingredients
I
see
slightly
with
a
few
watts
per
square
meters.
S
These
are,
is
the
bottom
row
three
maps
and
the
brown
contour
is
the
zero
contour
and
as
the
as
the
climate
is
warming,
we
see
that
it's
switching
signs
and
at
the
end
of
the
simulation,
there's
net
cooling
on
the
ice
sheet,
except
for
this
small
area
here
in
the
highly
elevated
area
in
the
east,
and
the
other
thing
I
wanted
to
talk
with
you
about
is
the
ground
heat
flux,
so
the
ground
heat
flux
we
have
defined
as
on
the
interface
between
the
atmosphere
and
the
lens.
S
So
this
is
also
including
the
snow,
so
it
would
be
an
energy
flux
of
heat
transfer
can
be
into
the
snowpack
or
out
it's
a
generally
small
component
of
the
surface
energy
balance
compared
to,
of
course,
the
radiator
fluxes
and
also
the
turbulent
fluxes.
But
it's
interesting
to
see
what's
happening
in
the
in
this
multi-decade
or
in
centennial
time
scale
in
the
pre-industrial
time.
We
see
that
it's
a
positive
so
again
it's
a
positive
defined
as
contributing
to
the
surface
energy
balance
and
negative.
It's
withdrawing
energy
from
the
instruments.
S
It's
positive
here
around
the
south
in
the
ablation
zone
and
around
the
equilibrium
line
altitude
and
as
the
climate
warms
and
the
equilibrium
line
altitude.
Also,
the
area
where
the
ground
heat
flux
is
positive
is
increasing
and
we
spend
some
time
on
thinking
what
this
means,
the
heat
conduction
towards
the
surface
from
out
of
the
snowpack,
and
we
reckon
that
it's
releasing
energy,
that's
that
got
into
the
snowpack
earlier
from
latent
heat
releases
of
refreezing
melt
energy,
I'm
sorry
of
refreezing
melt
water.
S
So
in
essence
it's
I'm
preparing
the
snowpack
for
for
a
full
melt.
S
Another
interesting
one,
one
minute.
Yes,
almost
done,
another
interesting
feature
is
if,
if
I
cross
compare
this
map
so
for
the
ground
heat
slots
to
the
refreezing
maps,
we
find
there's
also
re-freezing
in
the
north,
but
not
so
much
a
positive
ground
heat
flux
there
and
we
reckon
that
the
north
snowpack
has
sufficient
cold
content
to
buffer
such
heat
releases
of
the
melt
and
therefore
it's
a
it's
a
much
more
stable
there.
S
So
that's
what
I
was
that
was
for
my
point,
so
the
clouds
have
initially
a
warming
effect
on
the
ice
sheet.
As
the
climate,
warms
clouds
become
to
have
a
cooling
effect,
which
is
due
to
more
with
cloud
reflection
in
the
short
wave
radiation
and
the
ground
heat
flux.
It's
bringing
refreezing
energy
to
the
surface
of
prior
melt
water.
That
is
refreshing
in
the
snowpack
yeah.
That's
that's!
What
was
my
presentation
if
you
have
any
questions
I'll
be
happy
to
address
them.
F
Please
raise
your
hand
or
type
them
in
a
chat.
F
S
Carrying
young.
F
Yeah
you
can
you
can
keep
these
keep
these
slides
this
slide
on
for
for
a
little
while,
so
people
can
take
a
look
at
them
and
you
know
try
to
you
know,
remember
this,
so
I
have
a
quick
question
for
you
laura.
If
no
one
else
has
one.
So
you
mentioned
the
net
cooling
effect
by
clouds,
but
basically
everywhere,
especially
except
for
an
area
on
the
eastern
side
of
the
greenland
ice
sheet.
Did
I
understand
that
correctly?
S
Well,
I'm
not
really
sure-
and
I've
been
thinking
about
my
definition
of
the
shortwave
clouds
forcing
actually
a
bit
recently.
So
that's
also
why
I
was
I'm
presenting
this
that
I'm
thinking,
maybe
I'm,
including
the
effects
of
service,
albedo
change
and
attributing
them
to
the
cloud
change.
So
that
might
be
a
reason
and
of
course,
in
the
east,
that's
a
very
high
elevated.
S
F
Yeah,
absolutely
okay:
if
we
don't
have
any
more
questions
laura,
please
stop
sharing
your
screen
and
I
can
give
the
virtual
floor
to
ravindra
ravindra.
Please
start
sharing
if
you're
ready.
If
there
are
any
more
questions
for
laura,
please
type
them
in
the
chat
and
you
can
answer
them
right
now
and
ravindra.
Please
go
ahead.
This
looks
good
eight
minutes.
K
I
think
now
you
can
hear
me
right:
yeah,
perfect,
yeah,
the
the
screen
disappeared.
Sorry
all
right
good
morning,
everyone,
my
name,
is
ravindra
dudu,
I'm
an
associate
professor
at
civil
and
environmental
engineering
at
vanderbilt.
K
That's
a
recent
change,
and
for
the
past
few
years
I've
been
working
on
damage,
mechanics
formulations
for
calving
and
I've
always
been
presenting
work
that
is
outside
of
csm
and
so
I'll
guess
continue
that
tradition
even
today.
This
is
my
way
of
reaching
out
to
the
community
and
getting
feedback
and
trying
to
eventually
work
with
csm.
K
This
work
is
done
in
collaboration
with
alex
youth,
who
is
now
at
princeton
and
ben
smith,
who's
at
the
university
of
washington,
and
we
have
nsf
grant
and
a
new
nasa
grant
that
I
want
to
acknowledge
the
the
support
of
I
shall
fracture
an
iceberg.
Calving
accelerate
ice
sheet
mass
loss,
as
you
can
see
here,
the
larsen
bi
shelf
collapse
apparently
accelerate
the
ground
ice
flow
behind
the
ice
shelf
from
the
collapse
and
also
larson's
sea
ice
shelf.
Calving.
K
That
happened
in
2017
caravan.
A
big
iceberg
is
68
and
overall,
the
the
the
main
point
that
I
want
to
mention
is
that
I
shall
fracture
is
poorly
understood
and,
and
it's
poorly
represented
in
ice
sheet
models
right
and
and
the
and-
and
there
is
also
this
other
complication
of
atmospheric
warming
due
to
which
a
hydro
fracture
and
I
lose
it
in
a
loose
way-
can
be
a
factor
too.
K
So
so
go
ahead.
Yeah!
Sorry
about
that.
So
mainly
when
you
look
at
ice
sheet
models,
the
way
calving
mass
loss
or
the
calving
front
motion
due
to
calving
mass
loss
is
modeled
is
through
a
continuous
process
and
on
the
left
here,
I'm
I
have
a
paper
by
bonzio
and
and
involves
helen
sirus
and
matthew,
murligam
and
and
eric,
and
there
they're,
using
a
level
set
method
to
gradually
move
that
calving
front
right.
K
However,
on
the
other
end,
on
the
other
side,
I
want
to
focus
that
you
know
there
are
these
fractures,
which
are
really
discrete
features
on
the
ice
shells
and
when
of
course,
calving
happens,
you
have
discrete
events
where
icebergs
are
separating
and
that
can
contribute
to
a
huge
loss
in
the
ice
extent
right
or
the
ice
shelf
extent,
and
you
cannot
really
handle
that
as
a
continuous
process
and
then,
especially
if
the
melt
water
driven
fracture,
makes
calving
more
plausible
right
or
or
possible.
K
Rather
then
you,
you
really
have
to
have
a
method
to
handle
this,
so
that
has
been
our
motivation
and
I'll
just
play
a
video.
I
don't
know,
I
don't
think
this
is
playing
right,
it's
not
showing.
I
think
I
have
some
kind
of
a
bug,
because
I
wish
the
video
is
not
showing
sorry
about
that.
But
what
we
have
been
developing
is
a
generalized
interpolation
material
point
method
and
the
cs
community
has
started
using
these
sorts
of
method.
K
First,
in
fact,
elizabeth
and
cc
must
have
pioneered
some
of
these
works
and
we
have
for
the
first
time
adapted
that
to
glacier
I'm
sorry.
This
video
doesn't
seem
to
be
playing
right
now
for
me,
but
we
have
a
new
paper
that
just
got
accepted
a
few
weeks
ago,
where
we
described
how
this
method
can
be
used
for
shallow
ice
shelves,
and
we
have
shown
that
we
can
even
add
the
thickness
evolution
and
the
way
that
whole
process
worked
along
with
calving
now,
which
we
have
incorporated,
is
through
flow
work
as
follows.
K
So
we
first
saw
this
shallow
shelf
approximation,
which
is
a
depth
integrated
model.
Then,
from
that
we
get
the
velocity,
and
once
we
get
the
velocity,
we
can
get
the
deviatoric
stress
through
the
standard,
glen's
law
type
constitutive
model.
Then
we
assume
that
we
know
the
pressure
based
on
the
assumption
that
the
vertical
stress
is
hydrostatic
and
once
we
have
that,
then
we
implement.
K
If
you
look
at
the
bottom
right
hand
corner
we
implement
this
creep,
damage,
evolution,
law
and
in
that
we
can
even
parameterize
the
presence
of
water
and
and
the
hydro
fracture
parameterization.
If
you
may
call
that,
and
the
creep
damage
is
done
through
a
separate
subroutine
that
gives
in
a
quasi-3d
approach.
K
What
is
the
depth
to
which
the
crevasse
will
propagate
and
then
we
we
now
get
those
crevasse
propagation
profiles
which
is
shown
in
this
bottom
left
figure
and
from
that
we
depth
integrate
to
extract
the
damage,
which
then
goes
into
the
shallow
shelf
formulation.
So
that's,
and
then
we
do
this
using
a
certain
time
step.
The
time
step
is
typically
smaller
than
what
you
would
run
without
damage,
so
that
is
a
limitation
to
go
over
quickly.
What
the
equations
are,
on
the
left
hand
side
you
have
these
vertically
integrated
momentum
balance.
K
The
bottom
equation,
where
I
showed
the
deviatoric
and
pressure,
is
the
new
sort
of
addition
where
we
put
pw,
which
is
a
hydro,
fracture
or
parametrization
from
seawater
and
basal
pervasis,
and
then
we
evolved
damage
through
something
called
criterion
that
captures
brittle
and
ductile
fracture
in
a
very
empirical
way
and
finally,
the
last
equation
on
the
right
hand.
Bottom
is
the
depth
average
damage
tensor,
which
then
goes
to
effect.
K
The
viscosity
in
the
the
the
vertically
integrated
momentum
balance
equation
right,
and
we
described
this
whole
method
in
a
paper
in
james
that
just
is
tentatively
accepted.
So
we
have
two
papers
that
are
coming
out
in
james
as
part
one
and
part
two,
then
just
to
show
a
few
pictures
of
what
can
be
done.
We
we
did
the
miss
mip
plus
example.
We
got
calving
profiles
due
to
the
damage
that
look
like
a
curved
fracture
pad
that
fully
propagates,
that
sort
of
reminds
us
of
the
pine
island
glacier
fracture.
K
So
that's
one
thing:
we
also
applied
it
to
larson
ci
shell.
This
is
something
we
are
currently
working
on
to
tune
the
model
and
to
conduct
parametric
sensitivity
and
in
in
one
of
the
tuning
cases
we
are
able
to
reasonably
capture
the
the
calving
that
happened
after
2.2
years
with
the
so
the
time
doesn't
match
very
well,
but
we
can
match
the
fracture
path
and
which
is
evident
from
this
yellow
dash
line,
which
is
observations
and
the
red
one,
which
is
our
model
prediction.
K
And
finally,
we
also
developed
a
new
damage
law
that
can
capture
a
crevasse
propagation
in
a
consistent
way,
I'll
move
on
to
the
next
slide
in
the
interest
of
time.
But
here
hopefully,
this
movie
will
play
and
we
can
actually
simulate
this
sort
of
very
nice
cliff
failure
due
to
shear
and-
and
so
we
were
looking
at,
can
crevasses
progressively
fracture
at
the
terminus
of
a
of
a
grounded
glacier
or
a
marine
ice
cliff
and
eventually
cause.
K
You
know,
a
mass
loss
that
that
that
keeps
progressively
moving
inland
and
we
we
do
see
some
evidence
of
that
happening
and
particularly
the
sheer
stress
which
is
in
the
left
figure
in
this
in
the
subfigure
c,
seems
to
drive
this
kind
of
calving
or
fracture
process
near
the
terminus,
and-
and
so
that's
what
we
this
work
that
came
out
in
extreme
mechanics
letters
early
this
year
showed,
and
so
so
we
had
a
good
year,
three
papers
accepted
or
published.
K
K
We
have
to
work
on
that
and
then
anisotropic
damage
models
and
tuning
to
observations
is
is
something
our
future
work
is
is
tackling.
Thank
you.
F
There
is
no
not
much
time
for
questions,
so
I
would
suggest
everyone
who
has
questions
for
ravindra
to
put
them
in
a
chat.
In
the
meantime,
it's
almost
11
30.
anyway,
so
we
are
ready.
We
should
be
ready
for
the
breakout
rooms.
I
think
todd
prepare
them.
They.
I
F
Ready,
okay,
so
top!
Please
launch
them
and
you
can
choose
your
favorite
breakout
room
and
each
of
the
breakout
rooms
will
have
conveners.
F
So
please
pick
your
favorite
topic
and
we
will
be
in
the
breakout
rooms
for
probably
something
like
25
minutes
or
so
and
then
come
back
all
together.
Just
before
noon,
to
wrap
up
and
then
at
noon.
We
will
have
two
new
rega
breakout
rooms,
one
for
the
polar
climate,
one
for
the
land
ice
working
group
for
an
informal
lunch,
and
there
will
be
two
more
if
people
want
to.
You
know
discuss
something
very
specific
but
more
on
that
later.
F
So,
if
you're
ready
for
a
breakout
room,
please
pick
your
favorite
one.
There
we
go
and
we'll
see
all
of
you
back
in
25
minutes.
More
information
will
be
given
to
you
in
the
specific
breakout
room
enjoy.
I
I
I
F
I
I
G
Hi
everyone,
I
guess,
we're
we're
all
coming
back
to
the
main
room
now,
so
hopefully
the
breakout
groups
worked
well,
we
were
having
a
good
discussion
in
the
sea
ice
prediction
group
and,
of
course,
got
cut
off,
which
is,
I
guess,
what
happens
when
you're
in
breakout
groups
and
get
kicked
out
so
so
I
think
in
terms
of
wrapping
up,
you
know
we
have
all
the
talks
and
things
like
that
will
be
recorded
available
online.
So
if
you
missed
something,
please
go
back
and
take
a
look.
G
G
We
actually
are
gonna
move
into
sort
of
these
informal
lunches
and
I
think
we'll
have
breakout
rooms
set
up
for
that,
and
so,
if
you're
available
to
join
for
lunch,
it's
going
to
be
just
very
informal,
just
a
way
to
kind
of
chat
and
check
in,
and
so
please
join
us
for
lunch
if
you're
available-
and
you
know,
I
think
we
had
a
lot
of
really
great
science
presented-
it's
just
wonderful-
to
see
the
breadth
of
things.
G
People
are
doing
across
the
polar
climate
and
the
land
ice
working
groups
and
it's
just
really
exciting
stuff.
So
thank
you
to
those
of
you
who
presented.
I
don't
know
if
my
fellow
co-chairs
hansi
yon
bill
or
the
liaisons
grunter
and
dave
want
to
add
anything.
I
just
kind
of
took
took
over.
G
Yeah,
hopefully,
there's
a
lot
of
you
at
lunch
and
hopefully
we
see
you
in
person
at
winter
meetings
at
ncar,
so.
L
G
So
they're
they're
in
your
down
at
the
bottom
of
your
zoom
screen,
there's
breakout
rooms
and
if
you
click
on
that,
there's
at
the
very
bottom
you'll
see
everybody's
names
at
the
very
bottom.
There's
the
landice
working
group
lunch
table
and
the
polar
climate
working
group
lunch
table
you
can
on
the
right
of
the
panel.
You
can
click
on
that
to
join.
G
We
also
did
put
together
two
open
rooms,
partly
because
we
saw
that
in
the
chat
there
were
some
people
saying.
Oh,
we
should
discuss
this
later
or
oh
it'd
be
great
to
talk
more
about
this,
and
so
we
left
two
rooms
open.
So
if
there's
groups
of
people
that
want
to
just
go
to
one
of
those
open
rooms
and
chat
about
something
that
came
up
during
the
meeting,
you're
welcome
to
do
that.
So,
basically,
that's
how
you
join
one
of
these
lunches,
I'm
gonna,
take
a
five-minute
break
or
so,
and.
I
G
So
I'm
I'm
expecting
that
we'll
join
these
lunch
tables,
but
it
might
actually
be
five
ten
minutes
before
people
are
actually
there
and
please
grab
lunch
if
you
want
bring
it
back
but
yeah.
So
that's
how
you
join
and
if
things
come
up,
you
want
more
information
on
simulations
or,
as
hanzi
said,
you
need
to
use
your
computing
time
polar
climate
working
group
people,
please
contact
us
if
you
need
help
with
that,
because
we
don't
want
to
lose
these
computing
resources
and
we've
got
a
lot
of
great
things
that
people
proposed.
G
So
with
that,
I
guess
we'll
just
close
the
meeting
and
please
join
us
for
lunch.
I.
L
G
Okay,
well,
hopefully,
we
see
you
at
lunch
and
thanks
for
a
great
meeting,
it's
been
really
good
to
see
everybody's
work.