►
From YouTube: CESM Climate/Land Ice/Earth System/Polar Climate/Paleoclimate Working Group Meetings Day 1 AM
Description
No description was provided for this meeting.
If this is YOUR meeting, an easy way to fix this is to add a description to your video, wherever mtngs.io found it (probably YouTube).
A
Evolving
and
then
one
with
everything
else,
evolving
so
that
everything
else
is
solar
variability
ozone,
volcanoes,
Landry,
Salon
cover
change.
So
this
allows
you
to
look
at
additivity
because
all
of
the
forcings
are
evolving
in
in
one
of
these
simulations,
motivated
by
the
work
that
was
done
for
the
description
paper,
we've
also
run
a
secondary
Ensemble,
which
was
run
in
the
same
way
as
csm1,
where
everything
is
evolving,
except
the
aerosol
forcing
and
the
aerosol
forcing
is
kept
fixed
at
1920s
value.
So
we
just
had
so
just
re
revised.
A
The
paper
had
minor
revision,
so
hopefully
the
description
paper
will
be
out
soon
and
so
now,
we'll
just
give
a
quick
run
through
of
the
kind
of
future
simulations
that
are
either
started
or
about
to
start
or
will
be
going
on
in
the
coming
year.
So
there's
a
couple
of
offshoots
from
the
single
forcing
large
ensembles
so
motivated
by
the
fact
that
we
do
see
a
pretty
big
method.
Dependence
on
the
aerosol
influence
on
global
mean
temperature
in
csm2,
which
you
can
see
here.
A
So
the
the
solid
is
where
only
the
aerosol
is
evolving
and
the
dashed
is
where
everything,
but
the
aerosols
are
evolving
and
the
Marine
is
csm2,
and
you
see
this
big
method-
dependence
in
csm2
that
wasn't
really
there
in
csm1,
where
you
compare
the
teal
lines,
but
we
have
pretty
big
differences
in
the
forcings
that
were
used
between
csm2
and
csm1.
So
we
are
going
to
run
some
complementary
ensembles
where
only
see
where
cement
5
aerosol
forcing
is
being
prescribed.
So
we
can
see.
Is
it?
A
Is
it
having
different
forcing
really
key
to
having
this
different
method
dependency,
although
there
are
arguments
to
be
made
for
why
the
models
are
also
different
and
contributing
to
that
another
one
is
motivated
by
again
some
of
the
results
with
the
single
forcing
large
Ensemble.
So
if
you
look
at
the
green
line
here,
this
is
the
everything
else
ensemble
and
in
the
early
21st
century,
in
the
late
20th
century,
you
can
see
that
it's
warmer
than
it
was
in
the
compared
to
1920
to
1940.
A
A
There
are
ongoing
simulations
with
the
Next
Generation
grid
of
Cam,
so
cam
is
moving
to
a
higher
top
80
kilometer
top
and
a
higher
vertical
resolution,
so
here
you're
seeing
the
grid
spacing.
So
this
is
the
difference
between
the
each
levels
as
a
function
of
height
red
is
the
new
grid,
so
we
go
to
a
500
meter
grid
spacing
and
we
go
to
a
much
higher
height,
and
one
of
the
motivations
for
going
to
this
new
grid
is
that
we
now
start
to
represent
the
Quasi
biennial
oscillation.
A
So
they're
going
to
be
many
changes
in
cam
7
compared
to
cam
6,
but
as
part
of
the
working
group
allocation.
What
we've
done
is
to
look
into
kind
of
the
intermediate
step
of
how
it
is
changing.
Only
the
vertical
resolution
impact
on
the
simulation.
The
next
generation
will
also
have
enhanced
resolution
in
the
boundary
layer.
A
We're
not
changing
that,
because
when
you
change
that
you
have
to
start
retuning,
so
we
have
a
100
year,
long
pre-industrial
control,
and
then
there
will
be
three
coupled
historical
into
SSP
370
simulations,
which
can
be
compared
with
the
large
Ensemble.
There
are
some
amip
simulations
and
Yaga
and
Nan
are
also
running
some
nudged
Qbo
simulations
for
a
part
of
the
the
Qbo
into
comparison
project,
so
these
data
will
hopefully
be
available
sometime
later
this
year.
A
So
this
is
a
grid
that
was
generated
by
Rob
who's
here
and
Adam,
and
it's
going
down
to
a
1,
8
degree
resolution
in
the
North
Atlantic
and
so
we're
running
a
1958
to
present
day
amip
simulation
with
this
grid,
with
the
motivation
of
seeing
how
it
is
having
that
kind
of
unprecedentedly
high
resolution
in
the
Atlantic.
For
this
long
of
a
period
impact
on
Ocean
atmosphere,
coupling
in
the
North
Atlantic
jet
stream,
variability.
A
This
is
a
project
that
is
being
led
by
Flavio
laner,
so
he
is
interested
in
you
know.
We
have
had
one
kind
of
realization
of
the
force
response,
plus
the
internal
variability
that
we've
looked
at
in
the
real
world,
but
there
are
many
other
ways
that
internal
variability
could
have
gone.
So
this
is
using
a
kind
of
synthetically
generated.
A
His
histories
of
sea
surface
temperatures
using
a
linear
inverse
model
to
sample
what
what
could
have
been
in
the
past
and
taking
kind
of
extreme
members
ones
that
give
you
more
of
a
warming
in
the
tropical
Pacific
and
ones
that
give
you
more
of
a
cooling
in
the
tropical
Pacific
to
see
what
the
impact
of
those
different
kind
of
outcomes
of
variability
would
be,
particularly
on
North
American
hydroclimate.
So
those
are
underway
and
we'll
probably
make
those
available
when
Flavio
has
written
his
paper
about
these
okay.
A
A
couple
of
other
projects
that
are
ongoing,
Brian
Medeiros
is
running
regionally,
refined,
Tropic
simulation
with
1
8
degree
in
the
tropics
to
look
at
the
impact
of
having
that
resolution
on
tropical
variability
and
will
soon
start
an
ssp-585
medium
Ensemble.
The
plan
here
is
to
have
15
members,
and
so
once
we
have
that
we'll
have
kind
of
a
reasonably
large
Ensemble
size
for
each
of
these
different
scenarios
and
I
guess
SSP
585.
A
B
We
have
two
sets:
one
is
historical
simulation
which
we
need
to
run:
10
top
member
from
1550
to
at
least
2021,
and
after
2014
we
will
use
ssp-585
housing.
After
that
control
simulation.
We
have
three
sets
of
pacemaker
simulation
is
specific
Atlantic
and
in
the
ocean
separately,
each
we
run
10
and
some
members
for
the
SSC.
B
The
assay
will
be
relaxed
to
Observation
for
the
full
field
between
10
degree,
North
and
10
degrees,
south
and
the
transition
zone
is
10
and
10
to
30,
South
and
10
30
knots,
and
this
will
be
run
under
cbcwt
working
group
allocation
data
said
that
the
simulation
is
because
Peacemaker
handclass
experiment
for
this
experiment.
It
will
first
be
run.
We
need
to
get
a
new
condition
by
relaxing
Global,
SSC
and
SSS
to
observations
when
we
run
the
period
in
1982
to
2021.
B
After
that,
we
will
run
a
set
of
Peacemaker
simulation
in
the
for
the
Global
and
then
its
specific
Atlantic
and
in
the
ocean.
Each
of
these
assimilation
will
start
and
we
have
four
actually
starting
bounds.
A
February
1st,
May,
1st,
August,
1st
and
November
1st.
Each
of
the
assimilation
will
run
for
12
months
and
the
February
1st
that
one
is
required
for
every
regroup
to
do
and
for
others
it's
way
better.
Do
it,
but
not
required
to
participate.
This
project.
C
So
so
one
of
the
things
that
Kay
found
when
I
ran
the
historical
run
for
this
mechanically
decoupled
configuration
was
that
when
we
don't
allow
external
forcing
to
drive
changes
in
the
ocean
circulation
that
the
model
doesn't
warm
as
much
and
so
naturally
one
of
our
questions
was
sort
of
what's
the
role
of
aerosols
versus
greenhouse
gases,
and
so
we're
going
to
run
a
10-member
Ensemble
that
just
a
greenhouse
gas
only
over
the
historical
period
and
then
the
other
kind
of
major
questions
that
we've
got
in
the
past,
particularly
at
the
pattern
effect
Workshop
was
they're
hard.
C
It's
hard
to
compare
the
the
historical
runs
to
all
the
climate.
Sensitivity
runs,
so
we
plan
to
do
a
one
percent,
CO2
simulation
and
a
four-time
CO2
simulation.
So
we
can
kind
of
apples
to
apples
compared
to
the
fully
coupled
version,
and
those
should
be
run
this
summer.
A
I
think
we're
a
little
behind
to
be.
We
have
some
time
at
the
very
end
of
the
day,
so
I
guess,
if
you
have
questions,
maybe
try
and
remember
them,
and
we
can
have
a
bit
of
a
discussion
about
any
of
these
things
or
any
other
things
you'd
like
to
see
the
working
group
do
so
I
guess
we'll
get
started
then,
with
the
talks
so
I'll
see
you
first.
A
So
our
first
speaker
is
Carlos
Martinez.
What
am
I
enter
full
screen?
E
Okay,
I'll
just
do
this,
because
it's
just
easier,
hi
everybody
online,
just
a
second
okay
I'll,
just
leave
that
there
so
the
Caribbean,
so
many
stakeholders
rely
on
rainfall
for
their
socioeconomic
needs
and
that's
why
climate
and
weather
Services
rely
on
information
regarding
rainfall,
especially
given
the
projected
changes
to
the
cycle,
as
well
as
the
hydrometeorological
disasters
that
have
affected
the
region
that
have
exacerbated
the
concurrent
socioeconomic
issues
in
the
region.
So
from
this,
I
was
really
intrigued
into
investigating
this
region.
E
I'm
also
born
in
Puerto
Rico
I
have
family
there
as
well,
so
there's
a
personal
antidote
and
when
I
looked
at
the
literature,
I
found
numerous
dynamical
mechanisms
that
many
studies
picked
one
or
two
to
that
influence.
The
rainfall
cycle,
and
so
I'll
break
this
down
bit
by
bit
to
show
from
my
study
Martinez
at
all
2019
what
these
mechanisms
are
when
I
developed,
more
comprehensive
understanding
of
the
mean
state
of
the
cycle,
so
we'll
start
with
the
large
scale
features
the
first
being
the
inter-tropical
Convergence
Zone
or
the
itcz.
E
E
So
you
can
see,
as
you
can
see,
there's
a
lot
of
features,
including
the
localized
forcings
like
sea
breezes
as
well,
and
I
was
able
to
find
that
the
two
main
features,
the
itcz
and
Nash
are
the
main
contributors
for
moisture
across
the
Basin,
and
then
you
have
some
modifiers
and
then
some
transient
activity
that
have
a
marginal
impact
in
the
region.
So
from
this
refined
understanding
with
more
information,
you
can
look
at
those
at
that
previous
paper
and
to
see
how
that
plays
throughout
the
seasons
of
the
rainfall
cycle.
E
I
wanted
to
look
into
modeling,
given
that
modeling
Studies
have
not
looked
at
or
explored
to
the
extent
that
I
found
in
the
observed
hydroclimate
the
those
governing
mechanisms
that
impact
the
Caribbean
region
and
look
at
it
on
a
sub-regional
lens,
as
well
as
a
seasonal
lens,
and
so
these
are
several
of
my
objectives.
First
is
to
look
at
what
Global
circulation
models
or
gcms
simulate
the
cycle
in
the
mechanisms
that
govern
it.
E
Do
models
perform
better,
for
example,
if
an
over
ocean
only
grid
spaces,
where
you
have
less
complexity,
such
as
land
and
those
land
interactions
that
you
have,
are
there
biases
specific
to
the
entire
Basin
or
specific
to
a
sub-region
of
the
Caribbean?
And
then
is
there
a
relationship
between,
for
example,
model
resolution
and
precipitation
estimates
or
the
configuration
of
a
model
such
as
when
you
prescribe
ssts
and
what
we
find
in
those
precip
values?
E
So
what
I
did
was
I
spent
a
lot
of
time
collecting
a
lot
of
data,
as
you
see
on
your
right,
are
all
of
the
models
that
I
collected,
including
the
members
that
were
used.
Their
atmospheric
model
I
also
have
their
ocean
model.
I
just
haven't
I,
do
I,
don't
have
it
described
here
in
the
the
atmospheric
model
used,
including
the
resolution,
the
first
top
right
section,
that
is
the
cesm.
E
E
Grid
in
the
ocean,
which
allows
me
to
assess
the
value
of
a
higher
resolution,
cesm
versus
the
one
degree
approximate
grids
from
the
other
runs
and
then
I
also
looked
at,
what's
known
as
the
cmip6
high-rismate
project,
which
allows
me
to
better
assess
having
a
model
with
just
differences
in
resolution
between
their
low
and
high
resolution
versions.
E
So
they're
the
exact
same
model,
just
differences
in
that
and
then
I
looked
at
several
observational
data
sets,
including
several
rain
gauge.
Data
sets
from
the
Caribbean
Institute
of
meteorology
and
hydrology,
and
several
satellite
or
Blended
data
sets
to
have
a
robust,
observational
Benchmark
to
compare
with
the
models.
E
Okay.
So
with
that,
the
first
part
that
I'm
presenting
is
looking
at
the
Caribbean
hydroclimate
between
observations
and
models
for
this
section
and
for
the
entirety
of
the
presentation,
I'll
be
showing
just
one
seasonal
component
of
the
rainfall
cycle,
known
as
the
early
rainy
season,
which
is
from
April
through
June.
That
is
the
growing
season.
E
And
what
we
see
here
is
a
box
whisker
plot,
showing
an
example
of
one
of
the
sub-regions
or
the
central
Caribbean,
which
is
over
Puerto
Rico
and
the
northern
part
of
the
Virgin
Islands
in
the
Eastern
Caribbean,
and
what
we
have
are
several
of
the
different
model
groups,
including
the
fully
coupled,
runs
and
then
the
aimip
versions
as
well.
E
What
we
see
is
that
many
of
the
fully
coupled
runs,
including
the
individual
members
shown
in
the
dots,
have
values
that
are
below
what
we
see
in
the
observational
spread,
whereas
in
the
aim
of
friends,
we
have
values
that
are
closer
to
what
we
see
with
the
observations
for,
at
least
in
this
instance,
precipitation
totals
during
the
early
rainy
season
when
I
examine
the
seam
of
six
runs
and
expand
into
the
different
model
groups.
That's
what's
shown
in
the
second
row.
E
What
we
see
is
this
difference
between
low
resolution
versus
high
resolution
versions
or
looking
between
the
blues
and
the
Reds,
and
then
also
a
difference
between
the
circles
and
squares
or
the
fully
couples
and
the
aimip
versions.
Where
again,
we
see
a
similar
result
where
we
see
the
aim
at
friends
showing
values
closer
to
the
observations
than
we
find
in
the
fully
coupled
runs
then.
Lastly,
these
two
ping
dots
which
I
didn't
share
earlier,
though
that's
the
high
resolution
cesm,
which
both
it's
fully
and
aim
at
versions,
show
values
that
are
similar
to
The.
E
Observations
show
so
from
this
I
looked
at
a
time
series,
because
I
was
really
curious
to
see
how
the
fully
coupled
and
amip
continue
how
these
fully
coupled
and
aim
of
versions.
Their
Ensemble
means
look
in
this
example
for
land
only
precipitation
totals
in
the
central
Caribbean,
and
this
is
why
I
want
to
show
the
early
rainy
season,
because
I
find
such
interesting
results
in
this
season
versus
some
of
the
other
seasons.
For
example,
in
this
red
box.
E
It's
it's
very
clear
that
the
fully
coupled
runs
show
for
many
models
in
their
Ensemble
means.
They
don't
even
show
an
early
rainy
season,
such
as
the
cesm
runs,
for
example,
versus
the
high
resolution
run
and
a
difference
between
the
low
and
the
high
resolution
versions
of
cmip6,
noting
their
Standard
TV
or
standard
errors.
In
the
shading
and
the
black
dots,
the
black,
the
the
black
dotted
lines,
showing
the
observations
when
we
look
at
the
aim
of
version.
However,
despite
their
course
resolutions,
we
still
we
see
a
a
much.
E
It
appears
to
be
a
much
better
look
of
the
early
rainy
season
in
the
structure
and
the
seasonality
of
the
cycle,
and
so
I
found
that
to
be
really
interesting
when
we
are
prescribing
those
ssts
in
the
models
they
do
perform
better
overall,
and
so,
lastly,
in
this
section,
I
wanted
to
show
Scatter
Plots
of
the
rainfall
cycle
to
better
examine
the
individual
models
and
what
their
values
show
between
land
only
precipitation
totals
and
the
y-axis
and
ocean
only
precipitation
totals
in
the
x-axis.
E
In
the
top
row,
we
have
the
central
Caribbean
between
the
fully
coupled
and
aim
of
versions,
and
then
this
is
another
sub-region,
the
Western
Caribbean,
which
is
over
Central
America,
and
what
we
see
is
a
very
clear
shift
between
the
fully
coupled
and
aim
it
versions
in
comparison
to
what
the
observations
are
showing
and
we
see
this
consistently
across
the
sub-regions.
When
I
am
looking
at
this,
we
do
see
this.
E
Similarly
set
for
the
ocean,
where,
again,
we
don't
have
as
much
complexity
as
the
land
and
for
all
of
the
seasonal
components
of
the
cycle,
so
when
we
shift
to
the
Dynamics,
so
what
I
did
was
I
took
correlation
coefficients
between
each
of
the
model's
value
of
a
variety
of
dynamical
mechanisms,
in
this
case
sea
level,
pressure,
zonal
winds
and
Mariano
low-level
winds
and
I
compared
that
with
precipitation
totals
in
the
central
Caribbean.
E
In
this
case,
what
we
do
see
are
these
intriguing
inter-model
differences
in
the
mean
state
where
we
have,
for
example,
in
the
fully
coupled,
runs
a
large
low
circulation
over
the
Caribbean
Basin,
accompanied
by
a
circulation
pattern
that
would
imply
a
weakening
of
the
North
Atlantic
subtropical
high,
as
you
would
have
anomalous
westerlies,
where
the
easterlies
are
and
anomalous
southerlys,
which
would
weaken
the
easterlies
or
shift
the
East
release
to
a
southeasterly
component,
thus
producing
more
moisture,
moisture
being
fluxed
into
the
Caribbean
Basin
in
those
precipitation.
E
Interestingly,
in
the
aimip
versions,
when
you're
even
prescribing
the
ssts,
we
still
see
a
similar
setup,
which
highlights
these
Intermodal
differences,
even
in
the
SSC
prescribed,
runs
as
well,
and
so,
when
I
took
a
closer
look.
For
example,
I
took
a
spatial
average
over
the
Caribbean,
Basin
and
I.
Looked
at
what
each
model's
ensemble
mean
value
was
for
sea
level,
pressure
versus
their
Central
Caribbean
precipitation
totals
for
the
month
of
May.
You
do
see
this
very
clear
relationship
where
models
that
have
less
precipitation
than
what
the
observations
are
showing.
E
You
also
have
higher
pressure
over
that
spatial
domain
and
in
the
aimip
versions.
Although
the
range
is
much
smaller,
we
still
see
this
inter
model
difference
and
to
an
extent
a
difference
between
the
Reds
and
blues
or
the
low
and
the
high
resolutions,
and
then.
Lastly,
when
I
looked
at
other
Intermodal
differences.
Thinking
that
you
know
there
could
be
something
here
with
Nash
I
calculated
the
center
of
the
North
Atlantics
of
tropical
high
in
those
fully
coupled
and
aim
of
runs
and
I'm.
E
Just
sharing
here
with
the
fully
coupled
runs,
are
showing,
which
is
that
most
models
have
the
center
of
the
high
further
west
or
closer
to
the
Caribbean
basin,
which
highlights
and
may
explain
the
enhanced
subsidence
and
thus
have
a
connection
with
the
dry
early
rainy
season
bias
that
we
see
in
those
fully
coupled
models.
And,
finally,
when
looking
at
June
ssts
In
Those
runs,
we
do
see
a
very
clear
relationship
in
that
most
models
have
cooler,
ssts
denoting
also
less
precipitation
in
the
region
and
again,
these
findings
are
very
similar
across
the
other
sub-regions.
E
In
my
analysis,
and
so
lastly,
this
is
the
total
moisture
budget
analysis
that
I
computed
looking
at
aim
of
minus
fully
coupled
runs,
and
what
we
see
is
that
the
aimip
runs,
which
you
know
again.
They
highlight
and
show
closer
values
to
the
observational
spread
in
both
the
Dynamics
and
precipitation.
E
What
we
see
is
more
moisture
being
flexed
into
the
Caribbean
Basin,
which
again
you
have
anomalous
southeasterlies
versus
what
we
see
in
the
fully
coupled
runs
and
what
appears
to
be
a
low
cyclonic
circulation,
that's
very
similar
to
what
I
found
in
those
Intermodal
differences.
E
When
I'm,
at
least
in
this
case,
comparing
the
aim
of
versus
fully
coupled
runs
when
we
look
at,
for
example,
the
high
resolution
cesm,
we
actually
see
the
values,
those
anomalies
quite
muted
and
that's
again
reflecting
how
the
aimip
and
the
fully
coupled
configuration
of
the
high
resolution
CSM
are
quite
similar
to
one
another
and
so
that
the
fully
coupled
run
in
HRC
esm
does
a
fairly
well
job
and
similarly,
to
some
extent
with
the
cmip6
runs.
E
And
so
in
conclusion,
I
found
most
fully
coupled
runs
underestimate
early
rainy
season,
land
and
ocean
precept
totals
across
the
region.
I
do
also
see
those
fully
coupled
runs
also
showing
a
stronger
and
Westward
shift
of
the
of
the
Nash,
which
may
cause
those
enhanced
easterlies
and
Divergence
over
the
Basin.
E
In
their
low
resolution,
counterparts
and
I
didn't
share
the
other
Seasons,
the
midsummer
drought
and
late
rainy
Seasons,
but
I
do
find
similar
findings
as
well,
and
that's
it
happy
to
answer
any
questions.
Thank
you.
A
D
F
E
That
in
here,
but
I
do
have
those
and
then
here
you
kind
of
see
the
relationship
with
the
SSD.
Sorry,
which
are
those
Contours
which
I
didn't
mention.
But
you
do
find
a
an
influence
in
the
Eastern
Pacific
as
well.
E
G
So
we
we
looked
at
the
RS
map,
Ensemble
for
Global,
precipitation
and
Extremes
in
the
battle
adult
paper.
I
think,
and
we
didn't
find
your
the
same
thing
as
you
here,
but
the
high
resolution
was
necessarily
bear
in
the
low
resolution,
so
so
I'm
interested
to
yeah
to
understand
why
in
this
particular
region,
you
would
find
this
Improvement,
while
on
other
regions,
tropical
regions
in
particular,
we
could
not
even
I
mean,
of
course,
there
were
more
precipitation
in
the
HR
versus
the
LR.
E
Cases
we
do
see
this.
For
example,
the
high
resolution
cesm
versus
what
we
see
in
the
cesm
lens
runs
I
for
specifically
in
the
Caribbean,
because
you're
dealing
with
Islands
in
particular
you're,
going
to
have
most
likely
in
what
I
do
see
in
many
cases,
is
that
these
high
resolution
versions
are
able
to
see
those
localized
features
more,
perhaps,
and
thus
you're
able
to
characterize
those
precipitation
totals
better
than
what
you
would
find
in
a
more
course
model.
E
But
that's
not
always
the
case
and
I
also
show
here
that
there
are
other
factors
at
play,
such
as
the
large-scale
Dynamics,
where
you
do
see,
even
in
high
resolution
models.
They
do
not
show
the
the
large-scale
Dynamics
correctly,
which
does
have
an
overarching
role
in
the
Hydra
climate,
given
what
I
see
in
the
mean
Flex
in
convergence,
where
you
do
see
Nash
and
the
itcz
as
the
dominating
modes
for
mean
State
rainfall.
So
it's
a
combination.
E
E
I
would
go
back
to
look
at
my
slides,
but
I.
Don't
have
them
with
you,
but
yeah.
The
range
would
be
about
one
degree.
I
would
say
the
thank
you.
The
high
resolution
cesm
actually
fared
the
best
out
of
the
other
model
groups
where
you
had
SST
in
the
fully
coupled
about
a
0.25
versus
what
we
see
in
the
other
ones,
but
I
do
agree.
You
know
the
cmip3
looking
at
Martin
and
Schumacher,
which
looked
at
cmip3,
you
find
a
very
similar
result
with
the
amip's
improving.
So
it
is
consistent.
H
Okay,
thumbs
up
up,
there
wait
someone
said
no
yeah,
it's
fine,
okay,
now,
okay,
okay,
yeah,
so
I'm
going
to
tell
you
about
some
new
variable
resolution,
simulations
with
the
with
Cam
SE
cam6,
where
and
show
some
science
results
and
show
that
resolving
weather
fronts
seems
to
increase
the
large-scale
circulation
response
to
Gulfstream
SST
anomalies.
H
So
compared
to
the
simulation
that
Isla
mentioned
with
high
resolution
ssts
and
a
name
app
simulation
I
can
be
looking
at
some
SST
anomaly,
experiments
that
are
fairly
idealized,
and
so
this
is
work
in
collaboration
with
with
Adam
Harrington
and
Isla,
and
also
David
Batista
at
the
University
of
Washington.
So
some
motivation
for
this
is
the
apparent
underestimation
of
multi-decatal
variability
in
atmospheric
circulation
in
mid-latitudes
in
coupled
models
compared
to
observations.
H
H
These
on
a
wind
there's
multiple
variability
in
the
zonal
winds,
that's
not
found
in
the
models.
So
there's
been
this
long-standing
idea
that
mid-latitude
ssts
don't
matter
very
much
for
the
atmospheric
circulation.
There's
nice
review
paper
by
Joachim
Kushner
in
2002,
basically
is
one
of
the
main
papers
about
this.
But
then
there's
been
some
more
recent
work,
suggesting
that
this
could
be
an
artifact
of
not
fully
resolving
the
relevant
processes.
If
you
go
to
higher
resolution,
you
might
get
a
larger
response
of
atmospheric
circulations
to
mid-latitude
SST
anomalies.
H
So
then
this
multiplicative
variability
of
SST
then
gives
you
more
multiple
variability
of
the
atmosphere
and
so
just
a
a
paper
lending
some
support
to
this.
Here's
looking
at
SST
anomalies
in
the
curricio
extension
region
and
the
response
at
two
different
resolutions,
a
quarter
degree
you
get
this
response
and
upper
troposphere.
At
one
degree
you
don't
and
then
there's
been
some
more
Regional
modeling
studies,
looking
at
even
higher
resolution,
so
here
looking
at
12,
kilometer
or
about
eighth
of
degree
resolution
showing
that
you
get
a
much
stronger
Ascent.
H
If,
if
you
have
a
sharp
SSD
gradient,
but
that
you
can
only
get
the
strong
Ascent
if
you
have
a
sufficiently
high
enough
resolution.
So
if
you
have
eighth
of
a
degree,
but
if
you
have
40
kilometers,
which
is
closer
to
quarter
degree,
then
you're
not
fully
getting
this
response
to
the
S
of
the
atmospheric
Ascent
to
the
SSD
gradient.
H
So
but
then
you
can't
from
these
Regional
model
simulations
you
can't
really
think
about
how
this
influences
a
large-scale
circulation.
On
the
other
hand,
most
of
the
existing
high
resolution,
climate
modeling
efforts,
so
climate
modeling
being
decade
or
longer,
simulations
really
tops
out
at
about
a
quarter
degree.
Atmospheric
resolution.
So
this
includes
the
high
resolution
model
inner
comparison
project,
as
well
as
the
ihas
effort
with
casm.
Both
of
these
have
only
quarter
degree
atmosphere,
even
though
they
often
go
to
higher
resolution
in
the
ocean.
H
So
what
we're
doing
is
we're
taking
advantage
of
the
variable
resolution,
capabilities
in
in
cam,
SE
to
and
and
developing
a
new
grid,
with
with
Adam
Harrington
doing
a
lot
of
the
heavy
lifting
on
this
effort
to
Design
This
nicely
grid,
where
we
have
14
kilometer
or
about
eighth
of
a
degree
resolution
in
the
North
Atlantic.
But
you
can
still
resolve
the
rest
of
the
globe
at
the
lower
110,
kilometer
or
one
degree
resolution.
H
And
so
then
the
simulations
I'm
going
to
show
are
basically
a
reference
simulation
with
climatological
seasonally,
varying
ssts
and
important
to
note
here.
I'm
I'm
still
keeping
the
ssts
at
one
degree
resolution,
so
I'm
only
changing
the
resolution
of
the
atmosphere
and
seeing
what
influence
that
has
then
the
two
experiments
I'm
I'm
running
are
shown
in
the
bottom
here:
there's
an
assist
they're,
both
SST
anomalies
in
The,
Gulfstream
region.
H
One
is
an
SSD
gradient
anomaly,
with
warm
in
the
South
and
cold
in
the
north,
and
one
is
just
a
square
patch
of
warm
ssds
raising
the
temperature
by
two
degrees
Celsius
throughout
the
whole
region.
I'm
running
each
of
these
simulations
in
the
this
new
variable
resolution
grid,
as
well
as
a
reference
Grid
at
about
one
degree
resolution
and
just
to
give
an
idea
of
the
cost
of
this
it.
H
This
does
take
about
35
times
the
core
hours
of
a
one
degree
simulation
to
run
this
variable
resolution
grid,
but
it's
still
about
a
tenth
of
the
cost
of.
If
you
were
to
run
this
resolution
globally,
foreign-
and
this
allows
me
to
run
for
after
some
spin
up
for
15
years
for
these
high
resolution
simulations
so
get
some
good
statistics
and
here's
the
main
result.
So
this
is
showing
the
djf
sea
level
pressure
response
to
these
two
different
SST
forcings
and
in
the
high
resolution.
H
You
get
this
these
large
responses
that
are
significant
compared
to
the
internal
variations.
These
are
quite
large
responses
of
up
to
or
even
greater
than,
four
hectopascals
in
some
regions.
This
is
really
big.
Nao
like
anomaly
in
these
simulations.
We
are
extending
these
out
a
little
bit
further
to
see
if
this
holds
up
to
longer
simulations,
but
all
indications
so
far
are
that
this
is
a
very
significant
anomaly.
H
In
the
high
resolution
case
at
low
resolution,
you
get
a
similar
response
and
to
the
warm
SST
anomaly
case,
but
weaker
amplitude
and
then
you're
kind
of
completely
missing
the
response
in
the
to
the
SST
gradient
case
and
in
order
to
get
the
statistical
significance
here,
I've
had
to
run
these
simulations
out
further,
which
is
of
course
possible
because
of
the
low
lower
resolution.
H
Despite
these
big
differences
in
the
atmospheric,
large-scale
atmospheric
circulation
response,
there's
not
a
very
big
difference
in
another
quantity
that
I
thought
would
be
relevant,
which
is
the
local
precipitation
over
the
SST
anomalies.
If
you
compare
the
the
two
different
resolutions
in
these
two
different
experiments,
they
look
pretty
similar
in
the
different
resolutions:
you're,
not
getting
some
big
enhancement
of
the
of
the
precipitation
in
in
the
forcing
region.
It
looks
pretty
similar
between
the
two
different
resolutions,
yet
somehow
you're
getting
a
much
deeper
temperature
response
into
the
free
tarposphere.
H
So
this
is
showing
averaged
over
the
forcing
longitudes
the
latitude
versus
height
versus
pressure,
cross,
section
of
atmospheric
temperature
potential
temperature
and
then
the
V
and
Omega
wins.
So
you
can
see
in
high
resolution
cases
you
get
this
warm
anomaly
extending
up
into
the
free
troposphere
in
both
cases
so
kind
of
similar
in
both
both
of
these
high
resolution
cases.
You
get
that
similar
kind
of
response,
but
weaker
in
magnitude.
H
In
this
one
degree
warm
anomaly
case
consistent
with
its
then
it's
similar
but
weaker,
large-scale
circulation
response
and
then,
in
the
one
degree
gradient
anomaly
case,
you're
getting
the
temperature
anomalies
instead
going
more
meridianally
and
not
really
reaching
up
as
far
into
the
free
troposphere.
H
So
I
think
this
is
explaining.
This
is
key
to
explaining
the
differences
in
large-scale
circulation
response.
How
does
this
temperature
normally
get
up?
There
kind
of
hinted
at
this
already
with
the
precipitation
anomaly
plots,
but
it's
not
it's
not
really
a
story
of
differences
in
diabetic
heating.
If
you
look
at
the
diabetic
heating,
including
latent
Heating
and
and
also
radiation,
it's
not
hugely
different
between
the
different
cases
in
the
free
troposphere,
especially
comparing
top
and
bottom
so
I.
Don't
think!
H
That's
the
difference,
but
the
other
thing
that's
relevant
for
temperature
anomalies
in
the
free
troposphere
is
also
what
the
transient
Eddies
are
doing,
and
so,
if
you,
if
you
look
at
the
energy
budget
and
find
that
the
total
transient
Eddy
heating
as
a
residual
in
the
high
resolution
cases
you're
getting
that
transient
Eddies
are
acting
to
heat
the
upper
terposphere.
This
is
the
the
yellow
orange
Contour
is
here
versus
in
the
lower
resolution
cases,
and
in
particular
this
any
30
gradient
anomaly
case.
H
So
what
I
think
is
going
on
is
basically
the
transient
of
these
are
moving
heat
from
near
the
surface
up
into
the
free
troposphere
in
the
high
resolution
case
versus
you're,
more
energizing,
the
storm
track
with
this
forcing
and
moving
heat
meridianally
in
the
low
resolution
cases
and
one
other
interesting
quantity
I
wanted
to
share
that
I
think
is
related
to
this.
Is
that
you're
getting
a
huge
increase
in
the
the
vertical
velocity
variance
in
these
high
resolution
cases?
H
This
is
kind
of
an
expected
result
of
going
to
higher
resolution
that
you
get
more
Vertical
Velocity
variants.
A
lot
of
this
is
happening
at
very
small
scales,
where
it
doesn't
necessarily
always
influence
the
the
heat
fluxes,
but
basically
you
do
have
a
lot
more
Ascent
going
on
within
your
synoptic
storm
systems,
and
some
of
this
is
acting
to
modify
the
vertical
Heat
fluxes
by
by
the
synoptic
storm
systems
and
latitudes.
H
So
this
is
I
should
have
said.
This
is
the
climatology
and
Contours
and
then
the
response
to
SST
anomalies
and
shading
and
both
are
increasing
by
about
an
order
of
magnitude.
For,
as
you
go
from
one
degree
to
to
the
high
resolution.
H
It's
checking
time
two
minutes.
Okay,
so
should
speed
up
a
little
bit.
This
was
all
nice
and
simple
so
far,
but
if
you
add
in
a
quarter
degree
case
which
I
wanted
to
see,
do
you
really
need
this
eighth
of
a
degree
to
get
this
response?
Things
get
a
little
bit
messy
you're,
still
not
getting
the
gradient
anomaly
response,
but
then,
in
the
warm
anomaly
case,
even
though
you
had
about
the
right
pattern,
just
a
weaker
amplitude
in
the
one
degree
case
now
at
quarter
degree
you're
getting
something
completely
opposite.
H
But
if
you
look
at
the
same
quantity,
what
what
are
the
transients
doing
it's
it's
kind
of
emphasizing
the
same
story.
Basically,
if
transientalities
heat
the
free
troposphere,
you
get
this
large
atmospheric
circulation
response.
If
they're
acting
to
instead
move
heat
pull
word.
H
This
has
been
engaging
more
of
a
storm
track
like
response
and
not
getting
this
big
influence
on
the
the
mean
winds,
so
ultimately,
I
think
it
comes
down
to
where,
in
the
Cyclones,
the
ascent
is
occurring
in,
in
the
eighth
of
degree,
case
you're
getting
a
scent
in
the
cold
front
which
can
get
you
anomalies
efficiently.
H
Up
into
the
free
troposphere,
in
the
quarter
degree
case,
you're,
emphasizing
more
ascent
and
rainfall
anomalies
in
the
warm
sector
of
cyclones,
where
this
can
move
the
heat
more
forward
and
not
upwards
and
I
have
full
animated
versions
of
these.
That
didn't
make
it
into
the
the
presentation
computer
here
today,
but
if
you
want
to
see
them
come
come
find
me
and
I
can
show
you
the
animated
versions
of
these.
Oh,
never
mind
it
got
switched
it
tested
before
and
it
didn't
work.
H
But
basically
the
eighth
degree
is
starting
to
look
like
more
what
you're
used
to
seeing
from
a
weather
map
and
actually
resolving
all
these
fronts
and
the
vertical
velocities
associated
with
them.
And
then
once
you've
seen
this.
The
one
degree
simulation
really
looks
quite
like
a
blob
in
comparison,
and
it's
kind
of
remarkable
that
this
does
do
as
well
as
it
does
considering
we're
not
resolving.
All
of
these
frontal
features
in
a
one
degree
model.
H
So,
in
conclusion,
if
you
go
to
14
kilometer
resolution,
you
get
a
much
larger,
no
like
response
to
warm
Gulf,
Stream
SST
anomalies
that
is
weaker,
absent
or
of
opposite
sign
and
lower
resolution
simulations,
and
this
results
from
a
large
increase
in
resolved
Ascent
within
mid-latitude
Cyclones
that
acts
to
bring
heat
up
into
the
free,
troposphere
and
I
think
the
opposite
response
at
the
intermediate
resolution
is
coming
from
preferring
warm
sector
Ascent
versus
cold
sector
Ascent
within
the
Cyclones
kind
of
out
of
time.
H
Here
just
wanted
to
say
that
this
really
has
a
potential
big
implications
for
making
Decatur
predictions,
because
we
can
predict
SST
anomalies
on
multi-year
time
scales
and
if
they
have
a
bigger
influence
on
the
atmospheric
circulation.
This
is
then
very
important
for
predicting
climate
over
the
surrounding
land
regions,
and
this
atmospheric
response
could
also
influence
the
further
evolution
of
SST
anomalies.
I.
Think
there's
some
open
questions
related
to
what
part
of
the
SST
pattern
matters.
F
H
H
Yes,
we,
there
are
definitely
remote
influences
of
this
response
outside
of
the
North
Atlantic,
but
it's
starting
to
get
a
little
bit
harder
to
get
the
statistical
significance
from
these
15-year
simulations,
so
I'm,
hesitant
to
say
anything
about
those
until
the
simulations
are
extended
a
few
more
years.
I
Oh
yeah,
good
morning,
everyone
yeah,
my
name,
is
Hui
Ding
and
my
talk
is
about
summer
relationship
between
the
North
Pacific,
high
and
upwinding
winds
along
the
west
coast
of
North,
America
and
I.
Did
this
work
with
Michael
Alexander
from
nowhere
and
I
mean
fun
team
from
Columbia
University?
I
So
here
is
a
schematic
for
the
California
Current
system
along
the
North
American
West
Coast
in
summer
on
the
left.
It's
a
North
Pacifica
ocean
and
in
some
in
some
you
know,
specifically,
is
dominated
by
subtropical
high
and
we
often
call
this
the
North
Pacific
high
and
near
the
coast.
Equator,
water,
winds,
preview
and
these
winds
move
water,
offshore
by
ecuma
transport
and,
of
course,
coaster
of
winning
and
coastlab.
One
is
important
because
it's
the
universe,
nutrient
Rich
water
from
the
deep
ocean
to
maintain
a
productive,
Marine,
ecosystem
and
yeah.
I
I
Actually,
several
studies
already
examined
the
impact
of
the
North
Pacific
High
species.
The
surface
high
pressure
cell.
I
Along
the
west
coast,
but
this
yeah
but
results
from
this
paper
Shield
shows
the
impact
of
the
high
pressure
crisis,
only
a
small
impact
on
costing
up
when
in
summer
the
impact
mainly
occurs
in
spring
in
the
winter,
so
yeah.
So
this
motivates
us
to
look
at
the
other
aspect
of
the
North
Pacific
High
and
you
know
we
future
changes
in
cost
of
winning
has
been
extensively
started.
I
So
this
figure
is
a
summer
means
state
in
the
eastern
North
Pacifica
and
the
shedding
in
each
panel
is
a
500
hectic.
Pascal
pressure,
Vertical
Velocity
or
you
are
current
sharing.
So
okay,
what
is
the
mouth.
I
I
disappears
again.
Thank
you.
I
I
So,
where
am
I
okay,
yeah
shading
in
these
two
panels
is
500
Pascal
pressure,
Vertical
Velocity
and
the
factors
are
surface
winds
at
1000,
hectic,
Pascal
and
near
surface
winds
is
left
panel
is
eif-5
and
right
panel
is
the
large
Ensemble
and
we
can
see
a
near
surface
Equity,
liquid
Wood
Winds
in
near
the
West
Coast
near
California
and
yeah.
I
These
winds
are
avoid
in
favorable
since
the
point
to
the
equator
in
the
Northern
Hemisphere
and-
and
we
also
notice
this-
this
equator
World
winds
occur
in
regions
only
with
strong
subsistence
above
actually.
This
is
consistent
with
the
fertility
balance
in
the
atmosphere.
I
According
to
previous
studies-
and
we
can
see
this
relationship
also
along
the
west
coast
of
North
America,
so
this
figure
shows
a
cross
section
along
the
west
coast
of
North
America,
and
the
x-axis
in
both
panels
shows
latitude
of
costing
great
sales
and
y-axis
using
height
and
shading
in
both
panels
shows
Meridian
wind
from
each
level
only
and
we
can
see
near
the
surface.
Equator
Woodwinds
occur
only
in
the
in
the
southern
part
of
the
Coast,
with
a
strong
subsidence
above,
and
this
relationship
is
also
noted.
I
So
since
we
have
these
two
indices,
we
cover
theme
with
Meridian,
surface
Meridian,
wind
in
the
East
and
North
Pacifica
yeah.
This
in
this
panel
is
the
correlation
between
the
sea
level
pressure
index
with
Meridian
wind,
and
we
can
see
the
surface
high
pressure
seal
is
the
only
weakness
correlation
with
wind
in
the
coastal
areas.
I
The
subsidence
has
a
strong
impact
on
Meridian
Wing,
surface
Meridian,
wind
along
the
coast,
and
the
correlations
about
minus
0.6
and
minus
0.7.
So
this
is
Meridian
wind
only
and
we
can
also
concede
both
toys
and
Hawaii,
and
then
we
can
calculate
upwinding
favorable
along
showing
stress
for
each
grade.
Cr,
which
is
this
right
panel,
so
on
y-axis,
shows
the
latitudes
of
each
coasting
great
cell
within
correlate.
The
operating
favorable
launch,
showing
stress
with
the
twin
disease
and
again
we
can
see,
is
x-axis
shows
the
correlation
values.
I
Again,
we
can
see
the
surface
high
pressure.
Co
is
only
weekly
correlation
with
the
avoiding
favorable
launch
of
Industries,
which
is
Peru
line,
and
the
correlation
is
about
0.3
only
or
need
above,
but
on
the
other
hand,
the
subsidence
as
a
strong
control
and
account
for
about
50
of
the
entire
availability
in
this
latitude
between.
I
I
So
this
is
in
the
eye
availability
and
in
the
following
insta
slides.
We
show
some.
We
show
some
work
for
future
change
in
the
North
Pacific
high
and
up
winning
wins,
so
this
is
stormed
by
the
ensemble
mean
epic
difference
and
so
the
last
30
years
minus
30
years
at
the
beginning,
so
on
this
figure
shows
the
Epic
difference
in
sea
level,
pressure
on
the
left
and
the
surface
pressure
on
the
right
on
the
left,
the
dash
and
the
sun.
I
It
Contours
shows
a
surface
high
pressure
cell
and
we
can
see
sea
level
pressure
increases
only
in
the
northern
and
southern
part
of
the
high
pressure
seal,
but
not
much
in
the
middle
of
the
cell.
So
in
the
end
the
Epic,
the
future
change
in
in
Sil
C
Level
pressure
index
on
the
bottom
right.
It's
pretty
small
and
it's
not
significant.
I
And
yeah
interesting,
we
can
see
here
we
see
in
a
large
lump
yeah
a
lot
of
increases
level
pressure
in
continent,
and
this
is
also
confirmed
by
surface
pressure,
so
yeah.
I
On
the
other
hand,
we
we
do
see
a
significant
kind
of
future
change
in
in
500,
hectic
Pascal,
Vertical,
Velocity
and
surface
wind
stress
in
the
bottom.
Two
panels
and
in
the
upper
two
panels
is
a
summer
means
state
from
the
two
on
different
from
the
earlier
in
the
late
future
Apex
in
the
right
panel
minus
the
left
panel
gives
the
effect
difference
in
500,
hectic,
Pascal,
vertical
pressure,
Vertical
Velocity
and
the
surface
wind
stress,
and
we
can
see
the
subsidence
in
this
area
decreases,
but
we
see.
I
On
the
other
hand,
we
see
an
anonymous
subsidence
in
the
North
and
according
on
on
the
right,
we
see
a
Polaroid
when
the
stress
surface,
wind
stress
and
the
equator
water
stress
in
the
north-
and
we
can
see
this
may
be
better
along
the
west
coast
of
the
North
America.
So
this
is
a
upper
panel
is
a
cross
section
along
the
west
coast
west
coast.
It's
its
effect
chain
is
epic
difference,
and
the
bottom
panel
is
epic
difference
in
upwinding,
favorable
launching
stress
hygiene
and
adults
indicate
95
significance.
I
Now
we
can
see
on
this
yeah
boruto
this
Brutus.
This
means
appointing
favorable.
I
Okay
and
yeah
and
above
we
can
see
Anonymous
ascending
motion
near
California
coast
and
the
descending
Motion
in
the
North.
Okay,
this
one's
the
last
one.
So
this
one
is
the
intermember
Scatter
Plots
between
500
hectic
Pascal
pressure,
Vertical,
Velocity
and
yeah,
so
it's
x-axis
and
upwinding
favorable,
launch
of
Industries
for
y-axis
in
the
south
and
north,
and
we
can
see
here
in
the
South.
The
correlation
is
very
high.
I
So
this
suggests
future
change
in
the
subsidence
modulate
about
50
percent
of
future
change
in
upcoming
favorable
launcher
when
the
stress
in
the
South
actions
California
coast-
and
this
relationship
also
exists
in
the
North
and
the
correlations
even
higher.
I
So
yeah
so
okay
I'll
be
sure
here
so
so
yeah.
The
the
message
is
on
the
subsidence,
has
a
strong
or
account
for
a
large
fraction
of
value
into
annual
availability
and
modulate
a
future
change
in
upgrading
favorable
launch
of
Industries
along
the
west
coast
of
North
America.
Thank
you
very
much.
A
And
I
guess
I'm
wondering.
Can
you
then
look
at
you
know
these
the
large-scale
metrics
that
you've
looked
at
are?
Is
there
consistency
across
models
and.
I
Yeah
I
think
it's
pretty
your
boss
in
Simi
fire,
but
it's
not
it's
not
clear
yet
about
cmf6.
A
Yeah,
okay,
let's
yeah
just
leave
it
there.
So
I
guess
are
you?
Do
we
have
to
stop
sharing
or
should
I
be
doing
that
here.
A
To
share
here
yeah,
we
we
need
to
stop
sharing
our
one
good,
just
a
second,
no
problem.
A
A
I'll
jump
in
at
12
minutes
to
give
you
a
warning,
so
yeah
feel
free.
Oh
no,
I!
I!
Guess
you
can't
share
yet
because
I
can
yeah
in
the
meeting.
We're
still.
A
You
able
to
share
now
sorry
now.
A
Hold
on
a
second,
let
me
just
check:
is
this
his
problem
or
our
problem?
That's
right.
L
If
you
could,
please
close
Firefox
and
reopen
it
like
it's
asking.
D
A
You
have
Chrome,
can
you
log
off
and
log
back
in
on
Chrome.
J
To
quit,
Firefox
one
second
and
then
I'm.
A
M
N
J
Okay,
I
cannot
hear
anyone
in
the
audience,
so
it's
like
everything
is
silent.
Here.
Go
on
the
the
first
light.
J
This
would
be
the
okay
good
if
you
have
to
ask
questions,
just
write
it,
because
I
cannot
hear
anything
so
hello.
Everyone,
sorry
for
these
little
issues,
I
will
present
today
a
new
methodology
to
quantify
probability.
These
changes
in
probability
distributions
directly
from
time
series
data
and
I
will
do
so
by
focusing
on
an
application
and
sea
level
rise
as
an
application.
Next
slide.
J
Next,
okay
I
will
start
from
introduction
and
motivation,
focusing
first
of
all
on
global
Regional
syllable
rise
and
then
I
will
make
the
case.
That
is
useful
when
trying
to
understand
climate
variability
and
change
to
actually
quantify
changes
in
the
full
probability.
Distributions
I
will
then
propose
a
methodology
to
quantify
sources
of
changes
in
probability,
distribution
directly
from
Time
series
and
I
will
show
some
results,
focusing
on
sea
level
rise
or
an
application
across
observation,
in
my
case,
type
gh's
data
and
jfdl
mode,
and
then
I
will
go
through
conclusion.
J
So
it
is
well
known
nowadays,
the
global
means
level
has
been
rising.
On
the
left
panel,
you
see
Frederick
a
figure
from
the
paper
of
Frederick's
Italian
2020,
where
we
see
an
increasing
Global
means
level
from
1900
to
2020
and
I
would
like
you
to
focus
on
the
blue
curve,
showing
changes
in
The
observed,
reconstructed,
sea
level,
and
it's
clearly
going
up,
and
there
is
in
the
last
you
know
decades
and
acceleration
and
is
going
up
with
an
average
rate
of
this
slope
of
1.35
millimeter
period.
J
On
the
other
hand,
if
we
look
at
Regional
sea
level
and
here
from
a
symmetry
from
1993
to
2014,
we
see
that
there
is
a
lot
of
heterogeneity
in
this
Trends
and
this
tutorial
genetic
come
mainly
from
local
variability,
local
climate
probability.
So
from
changes.
Local
changes
in
winds,
ocean
patterns,
ocean
warming
patterns
and
model
variability
next
and
when
we
usually
quantify
Trends,
are
changing
from
C
from
sea
level.
Time
series
in
general
Fountain
series.
The
first
thing
we
do
is
to
quantify
linear
regressions
on
trends
of
the
of
the
time
series.
J
This
allows
us
to
quantify
changes
in
the
conditional
mean
of
the
distribution
in
the
case
of
sea
level,
conditional
mean
of
sea
level
in
condition
to
type.
We
do
so,
therefore,
By
ignoring
higher
order
changes
in
the
distributions.
However,
next,
to
better
understand
climate
change
and
climate
variability,
it
is
useful
to
actually
understand
the
full
probability
distribution
as
climate
change
may
impact
variance,
asymmetries
and
Tails
of
the
distribution
too
next.
J
So
what
I'm
going
to
show
you
in
this
talk
is
very
briefly,
is
a
way
to
actually
do
so
to
decompose
changes
in
distributions
in
a
interpretable
way
and
I
will
do
this
in
a
pedagogic
way.
Next
slide
by
focusing
first
of
all
of
some
synthetic
time
series
and
then
going
through
sea
level
rise,
and
you
know
tight
gauges
and
modes,
so
consider
a
Time
series
of
discount.
You
have
time
on
the
x-axis
and
a
variable
s,
and
we
generated
these
time
series
ourselves
next
slide.
J
If
you
compute
a
linear
regression,
you
would
find
that
there
is
a
trend
in
this
10
series
and
there
is
a
positive
slot,
so
this
trend
is
going
on.
This
means
that
the
conditional
mean
of
the
distribution-
it's
a
conditional
mean
of
the
variable
s
condition
to
time
is
going
up.
However,
as
you
focus
on
the
conditional
mean,
we
can
nowadays
also
com,
compute,
conditional
quantiles,
so
conditional
median,
a
conditional,
quantile,
0.05
0.95
next
slide,
and
we
can
do
this
using
methodology
called
quantile
regression.
J
More
than
that,
we
can
actually
focus
on
a
lot
of
different
quantiles,
not
just
this
string
next
slide,
and
here
click
next
three
times
and
here
what
you
have
are
changes
in
the
in
the
that's
it.
You
can
see
changes
in
the
in
all
the
quantiles
of
the
distribution
from
0.05
quantile
to
0.95
point
time
on
the
x-axis,
you
see
the
probability
level
of
each
quantile
and
the
y-axis
there's
lot
next
slide.
J
J
So
here
we
know
that
in
distance
it
is
probably
both
the
mean
and
the
variance
may
change
next
next
slide,
but
what
about
other
other
higher
order
moments?
If
we
want
to
think
about
skewness
quarters
of
fifth
moment
and
so
on,
instead
of
becoming
more
and
more
difficult
to
interpret
with
this
plot,
also
how
to
deal
with
n
times
Series,
where
n
may
be
ten
thousand
times
six,
we
cannot
anymore
look
at
this
plot
and
how
to
deal
with
statistical
significance.
Well,
you
know
it
may
involve
some
boostropic
or
reshuffling,
and
so
on.
J
Let's
light
and
a
strategy
that
was
first
proposed
as
far
as
I
know,
by
Karen
McKinnon
in
a
jgr
paper
in
2016
is
actually
to
reduce
the
dimensionality
of
the
of
all
these
on
tiles
and
focus
on
how
changes
in
moments
of
a
distribution
can
explain.
These
changes
in
quantiles
and
specifically
independent
changes
in
moments
of
a
distribution
next
slide.
J
So
here
we
show
inspired
by
the
the
methodology
proposed
by
Karen
McKinnon.
We
actually
tried
to
look
at
this
this
framework
and
analytical
kind
of
way
here
you
have
quantiles
and
that
I
call
them
QP,
quantiles,
skew
and
probability
level
P.
So
if
quantile
QP
equals
0.5,
it
means
that
you're
looking
at
quantile
0.5
or
the
mean
we
look
at
non-stationary
distribution.
J
You
can
find
this,
but
a
little
bit
later,
for
you
know
in
backup
slides,
but
you
can
find
a
relationship
between
next
Slide.
The.
J
But
what
it
means
is
these
polynomials
are
actually
orthogonal
in
the
in
the
probability
space
from
zero
to
one
in
this
polynomial
orthogra
to
each
other,
so
next
slide
so
quite
beautifully.
What
the
what
the
quantifies
binomial
means
is
how
quantize
of
a
distribution
change
when
you
shift
its
moment
one
at
a
time.
So
if
I
change
the
mean
of
the
distribution
without
changing
any
one
of
the
other
moments,
the
quantity
will
change
is
a
constant
and
this
constant
is
B1
P.
J
So
the
blue
curve,
if
you
change
just
the
variance
but
not
the
mean,
and
not
the
other,
not
the
other
moments,
then
the
changes
will
be
shown
by
the
orange
curve,
B2B
next
slide
and
so
on
and
so
on
foreign.
So
we
obtain
a
simple
formula
where
we
have
on
the
left
side:
the
computed
quantile
regressions.
So
these
are
the
slots
and
quantiles
I
showed
you
before
and
then
on
the
right
side.
J
You
have
this
polynomials
that
we
also
have,
and
the
only
unknown
is
this
chain
is
involvement,
so
you
can
fit
these
changes
in
Moments
by
a
simple
linear
regression
using
this
basis,
this
polynomial
basis
next
next
slide,
and
so
what
you
see
here
is
that
focal.
This
is
an
example,
and
we
have
many
of
these.
But
if
you
focus
on
the
top
panel,
so
Upon
A
and
B,
you
see
time
series
that
are
generated
by
a
drifting
gaussian
distribution.
J
J
This
is
a
consistently
generated
by
a
beta
distribution,
and
here
we
know
that
changes
come
only
from
variance
cuteness
and
vertices,
and
the
method
is
actually
able
to
find
these
changes.
Excellent,
so
we're
trying
to
apply
this
methodology
also
next
slide
to
sea
level
rise
as
a
first
application.
Here
we
focus
on
daily
sea
level
and
we
focus
on
the
period
1970
to
2017..
These
are
Tech
cages,
data
downloaded
from
sea
level,
Central
and
University
of
Hawaii.
Whenever
you
see
a
DOT
means
significant,
when
you
see
a
mix,
it
means
not
significant.
J
So
what
this
tells
you
is
that
changes
in
constancy
level,
at
least
for
the
tight
gauges
that
we
have
available
come
especially
from
Chinese
and
I
mean
so,
in
the
mean
all
bicycle.
Basically,
all
changes
in
the
mean
are
are
are
significant.
While
changes
in
higher
order
moments
appear
not
significant
under
you
know
our
statistical
significant
test,
so
this
means
that
you
can
explain
these
changes
in
in
sea
level.
Distributions
in
The
observed
world
as
a
simple
shift
of
the
probability
distribution
towards
in
general
higher
values.
J
J
We
look
at
changes
in
sea
level
for
time
series
that
have
at
least
more
than
80
years
of
data
focus
on
the
AC
level,
and
what
you
see
here
is
that
still
the
conclusion
still
stands,
the
main
changes
come
from
the
mean
and
not
from
virus
known
as
okay.
So
this
motivated
us
to
try
to
understand
how
changes
in
in
probability,
distribution
may
happen
because
of
in
model
outputs
or
with
you
know,
CO2
experiment,
I,.
J
Minute
more
next
slide
and
that
the
good
thing
about
using
a
model
is
that,
even
if
it's
biased
we
have,
we
can
understand,
we
can
decompose
the
level
in
different
in
different
ways.
O
J
Here
next
slide,
we
write
sea
level
in
them
as
a
summation
of
dynamic
sea
level,
which
is
a
summation
of
changes
in
water,
column,
mass
and
locator
facts,
and
this
you
cannot,
you
can
just
think
of
it
in
a
very
simple
way,
as
change
local
changes,
sea
level
that
comes
about
just
because
the
ocean
is
moving,
so
there
are
currents
and
so
sea
level
is
changing
locally
and
then
a
next
slide.
Another
term
that
is
the
inverse
barometer,
which
shows
us
how
changes
in
sea
level
can
be
driven
by
local
changes
in
sea
level.
J
Pressure
an
example
is
strong.
Competitive
processes
would
give
a
strong
change
in
sea
level.
That
is
just
local
because
of
convection.
So
next
slide.
We
first
focus
on.
We
focus
on
Jeff
dlcm
for
data.
Here
is
0.25
degrees,
remapped
to
0.5
degrees.
J
We
focus
just
on
from
the
minus
60
to
60
inches
minus
60
South
to
60
North
latitude,
and
we
love
look,
look
it's
nearly
sea
level,
and
what
we
found
is
that,
if
you
focus
on
the
historical
run,
we
have
the
same
information
that
we
have
obtained
in
the
observations
in
this.
In
this
level,
the
composition
we
do
not
look,
also
the
global
means
11,
so
you
would
have
to
add
a
positive
term
to
all
these.
All
these
all
these
all
these
threats,
but.
J
J
However,
even
if,
in
the
historical
realm
we
see,
no
higher
order
moment
changes,
we
start
seeing
emergence
of
higher
order.
Changes
in
runs
such
as,
in
this
case
one
percent
CO2
a
year
from
pandasia
and
on
the
left
side
you
have
Dynamics
level
on
the
right
side,
Dynamics
level
plus
inverse
barometer,
and
you
see
you
start
seeing.
The
changes
in
virus
and
skewness
are
already
present
in
Dynamics
level,
and
changes
in
other
moments
are
always
Amplified
when
looking
at
the
inverse
barometer.
J
Can
you
go
on
next
slide
and
also
next
slide?
This
is
a
simple
example
of
changes
in
the
Mediterranean
Sea.
In
this
simulation,
without
looking
at
the
global
mean
changes,
so
you
will
have
to
take
that
time
series
and
put
it
up
of
a
trend,
but
it
would
not
change
the
probability
distribution
so
we'll
just
shift
it,
and
what
we
see
is
that
the
methodology
will
tell
you.
D
J
Slide
as
conclusions
I
will
not
over
it
because
I'm
already
over
time
and
I'm
sorry
about
the
technical
difficulties.
The
main
point
is
now.
K
J
A
So
we
since
Fabrizio,
can't
hear
us
anyway,
we
won't
take
questions
in
the
room,
but
if
any
of
you
online
have
questions,
you
can
go
ahead
and
put
them
in
the
chat
and
maybe
have
an
offline,
a
discussion
then
there.
A
A
Okay,
does
it,
it
looks
like
we're
still
sharing
our
screen.
Oh
yeah,.
A
She's
we
want
to
have
her
try
sharing
online.
A
L
N
L
N
Okay,
it
says
I
have
to
restart
Google
Chrome.
Okay
is.
N
Thanks
yeah
so
today,
I'm
going
to
talk
about
estimating
Trends
in
freshwater
fluxes
using
linear
response
Theory-
and
this
is
work-
that's
been
done
with
my
PhD
advisor
Lozano
next
slide
great.
So
to
start
off
with
some
background,
estimating
the
hydrological
cycle
over
the
historical
record
is
often
approached
using
salinity
changes
because
it's
hard
to
directly
measure
surface
freshwater
fluxes.
So
the
idea
is
that
the
amplification
of
the
evaporation,
minus
precipitation
pattern
imprints
on
Ocean
salinity
and
then
you
can
use
ocean
salinity
to
try
to
infer
what
those
freshwater
flux
changes
were.
N
So
to
look
at
the
salinity
changes
over
the
historical
record.
We
can
look
at
the
plots
on
the
right,
so
the
top
plot
is
the
mean
surface
salinity
over
the
period
1975
to
2019
and
then
the
bottom
plot
is
the
change
over
that
same
period.
So
it's
the
mean
of
the
last
five
years,
minus
the
mean
of
the
first
five
years
and
so,
broadly
speaking,
what
we
can
see
is
that
regions
that
are
salty
are
getting
saltier
in
regions
that
are
fresh
are
getting
fresher.
N
If
we
compare
those
two
plots-
and
this
to
some
extent,
is
an
imprint
of
that
kind
of
amplification
of
the
freshwater
flux
pattern,
but
the
the
complication
or
like
an
additional
Wrinkle
In
This,
is
that
the
change
in
local
surface
salinity
isn't
just
affected
by
surface
fluxes.
So
if
we
were
to
look
at
the
Tracer
equation
and
think
of
the
Tracer
here
as
salinity,
it's
affected
by
the
sources
minus
sinks
term-
that's
shown
at
the
end,
but
it's
also
affected
by
the
other
terms,
which
here
would
be
ocean
transport.
N
Next
slide,
please
great.
So
quite
a
few
studies
over
the
past
decade
have
worked
on
this
problem
of
trying
to
estimate
hydrological
cycle
amplification
from
salinity.
So
here
the
y-axis
is
amplification
per
degree
celsius
and
then
the
x-axis
is
various
studies.
So
they've
used
quite
a
few
different
methodologies,
so
I
won't
go
into
all
of
it,
but,
for
example,
drop
it
all
2012
and
zika
at
all,
both
use
surface
salinity
and
then
the
other
ones
that
are
shown
there
primarily
used
interior
salinity
next
slide.
N
Great,
so
the
the
goal
with
this
work
was
to
add
a
complementary
estimate
onto
the
previous
work
that
we're
showing
on
the
right,
and
we
aimed
to
do
a
couple
of
things
with
this
new
estimate.
So
the
first
thing
is:
we
want
to
focus
specifically
on
Surface
salinity,
because
observations
at
the
surface
are
a
little
bit
more
certain
than
in
the
interior,
and
then
the
other
thing
that
we
want
to
capture
is
that
the
impact
of
ocean
circulation
change
is
primarily
Regional
or
local.
N
So
when
people
talk
about
salinity
pattern
change,
often
what
they
talk
about
is
an
amplification,
so
the
entire
Global
pattern,
like
amplifying
in
its
existing
form.
N
But
if
we
look
at
the
map,
that's
shown
on
the
slide
there,
which
is
from
an
ocean
only
model,
that's
forced
with
just
heat
fluxes
and
so
we're
seeing
the
impact
of
salinity
basically
being
moved
around
by
heat
flux,
induced
ocean
transport
change,
and
so
what
we
can
see
is
that
here
there
are
regions
where
the
signal
is
very
strong,
so,
for
example,
in
the
Atlantic
there's,
there's
quite
a
large
salinity
change
just
from
that
ocean
transport
change.
N
But
it
doesn't
look
like
the
existing
pattern,
so
it
doesn't
look
like
a
scale
up
or
down
next
slide,
please
so
to
to
give
away
the
answer
from
the
very
beginning,
the
orange
dot
that's
been
added
to
the
plot
on
the
right
is
the
estimate
that
I'm
going
to
talk
about
for
the
rest
of
this
presentation,
so
we
ended
up
with
a
value
of
about
4.5
percent
plus
or
minus
error,
which
is
within
range
of
of
some
of
the
previous
studies,
but
is
definitely
below
the
clausius
clap
around
rate
of
seven
percent
and
adds
some
confidence
that
the
amplification
is
below
clausius
clap
around
next
slide.
N
N
So
on
the
left
and
light
gray
is
the
surface
salinity
distribution
and
then,
on
top
of
it,
we
fit
a
gaussian
mixture
model
with
six
mixtures
so
going
from
less
salty
up
to
the
saltiest,
which
would
be
number
six
and
then
on.
The
right
is
how
a
map
of
the
ocean
surface
would
then
be
clustered
based
on
this
mixture
model.
So
the
idea
is
for
each
point.
We
categorize
it
into
one
of
the
six,
depending
on
the
salinity
of
that
point.
N
So
the
next
ingredient
that
we
need
for
this
method
is
linear
response,
Theory,
which,
broadly
speaking,
finds
the
change
in
statistical
properties
of
a
dynamical
system
because
of
a
forcing
so
here
we're
considering
the
dynamical
system
to
be
the
climate
system
and
we're
going
to
apply
linear
response
theories
so
that
we
can
talk
about
the
change
in
salinity
and
temperature
in
each
of
those
gaussian
mixture
model
regions.
So
in
each
of
the
six
regions,
that's
shown
in
the
the
plot
on
the
bottom
right
there.
N
So
basically,
linear
response
Theory
allows
you
to
write
the
the
change
in
these
time
series
as
equal
to
the
convolution
between
the
response
of
a
step
function
and
the
strength
of
a
forcing
time
series
next,
so
yeah,
underneath
there
is
showing
that
in
equation
form.
So
here,
Delta
Y
is
the
change
in
salinity,
temp
in
temperature,
regionally
and
then
the
R
function.
There
is
the
response
to
a
step
function
and
then
the
f
is
the
strength
of
the
forcing
over
time.
N
Next,
please
so
I
won't
talk
about
this
too
much,
but
basically
to
be
able
to
apply
this
linear
response,
Theory
expression.
Of
course
we
need
those
responses
to
step
functions
and
we're
going
to
take
those
from
the
ocean
only
fast
map
project,
which
is
basically
an
inter-comparison
project
where
OSHA
models
are
forced
separately
with
three
different
flux:
perturbations
freshwater,
fluxes,
heat,
fluxes
and
wind
stress
change,
and
these
are
all
associated
with
a
doubling
of
CO2
and
they're,
applied
separately
to
those
ocean.
Only
models
next.
N
Okay,
so
the
final
setup
of
the
problem
is
basically
to
assume
that
these
Regional
time,
series
of
salinity
and
temperature
are
linear
combinations
of
the
response
to
heat
flux,
perturbations,
freshwater
flux,
perturbations
and
wind
stress
change.
So
that's
shown
in
the
boxes
on
the
slide
there
on
the
left.
Again,
we
have
the
the
time
series
of
salinity
and
temperature
change
and
then
we're
saying
that
can
be
broken
up
into
those
three
components.
N
Next,
so
to
each
of
the
individual
components,
we
apply
the
linear
response,
Theory
expression
that
I
talked
about
on
the
last
slide,
so
we're
we're
expressing
these
changes
in
salinity
and
temperature
in
terms
of
those
step
functions
and
the
magnitude
of
the
forcing
for
each
of
those
three.
And
so
what
really?
N
What
we're
going
to
focus
on
here
is
that,
like
using
this
expression,
we
can
solve
for
the
strength
of
the
four
settings
that,
like
are
attributed
to
the
time
series
and
so
we're
going
to
focus
on
the
freshwater
flux
forcing
strength,
but
as
we
solve
for
that
freshwater
flux,
forcing
strength
we're
kind
of
inherently
taking
into
account
the
salinity
and
temperature
change.
That's
happened
because
of
the
other
two
components
as
well:
next
slide:
okay,
So.
N
Eventually
we're
going
to
talk
about
applying
this
to
observations,
but
before
we
applied
it
to
observations,
we
tested
the
method
on
the
CSM
large
Ensemble
data
over
the
period
1975
to
2019.,
so
first
I'm
going
to
talk
about
applying
it
to
the
CSM
ensemble
mean
so
in
the
bottom.
Right,
for
example,
is
the
change
in
salinity
in
the
ensemble
mean,
and
there
would
be
kind
of
a
similar
plot
for
the
change
in
temperature.
So
this
is
in
each
of
the
regions
and
we're
showing
the
trend
in
time.
N
So
the
method
that
we
just
talked
about
makes
an
inference
from
this
from
like
these
time
series
and
gives
us
a
amplification
as
a
proportion
of
the
step
forcing
pattern.
So
it
the
the
kind
of
output
of
the
method
is
as
a
proportion
of
the
map
that's
shown
in
the
lower
left
here,
which
is
the
freshwater
flux
step
forcing
and
then
what
we're
looking
at
in
the
top
right
is
comparing
two
things.
N
So
the
first
is
just
applying
the
method
to
salinity
and
temperature,
and
then
we
can
compare
that
directly
against
model
fluxes
so
that
the
kind
of
top
row
there
is
from
the
linear
response
method
and
we
get
a
value
just
over
0.35
times
that
step
forcing
map.
And
then
the
bottom
is
from
model
fluxes,
and
we
can
see
that
the
the
value
that
we
got
from
linear
response
Theory
as
well
within
the
the
kind
of
error
range
of
the
true
value
for
model
fluxes.
N
So
truly
on
some
I
mean
we
find
that
we
can
recover
the
true
boxes
as
the
takeaway
here
next
and
then.
The
next
thing
we
wanted
to
test
on
is
individual,
Ensemble
members.
So
the
idea
here
is
that
observations
are,
to
some
extent
kind
of
equivalent
to
an
individual
Ensemble
member.
N
We
can't
take
a
true
ensemble
mean,
and
so
what
I'm
showing
on
the
bottom
right
is
kind
of
in
the
thick
lines
is
again
the
ensemble
mean
Trends
in
salinity,
but
around
region
six
and
the
the
lighter
yellow,
that's
hopefully
visible
I'm,
showing
the
different
members
that
are
making
up
that
Ensemble
Meme
and,
of
course,
each
of
each
of
those
trend.
Lines
has
individual
members
making.
G
N
N
So,
if
we
look
at
the
the
spread
of
those
yellow
lines,
we
see
that
there's
both
a
spread
of
force,
responses
and
there's
a
spread
of
the
amount
of
internal
variability
and
so
for
members
that
have
a
significantly
or
like
a
a
large
enough
signal
to
noise
ratio.
We
are
able
to
recover
the
true
response
and
for
members
that
are
dominated
by
internal
variability,
we
can't
see
the
force
response,
and
so
we
can't
recover
it.
N
N
Okay,
so
now
that
we've
tested
on
CSM
model
data,
we're
going
to
turn
to
observations,
so
here
we're
looking
at
salinity
and
temperature
data
from
Chang
at
all
2020.-
and
the
first
thing
to
note
here
is
that
this
data
kind
of
easily
meets
and
beats
the
significance
criteria
that
we
talked
about
on
the
previous
slide.
So
the
force
response
is
clearly
visible
beyond
the
internal
variability,
and
so
we
proceed
with
applying
the
method
as
we
did
before.
D
N
Yeah,
so
that
4.58
number
is
the
orange
dot
on
the
plot
on
the
right.
That
I
also
showed
at
the
beginning
and
yeah
so
I'm
just
going
to
conclude
here
so
ocean
transport
change
is
primarily
affecting
surface
salinity.
N
Regionally,
we
came
up
with
a
method
that
takes
this
effect
into
account
and
we
tested
it
on
cesm
model
data
and
we
found
that
we
were
able
to
find
the
true
amplification
in
this
model
data,
and
then
we
applied
it
to
observations
and
we
found
that
it
agrees
with
many
of
the
previous
estimates
of
hydrological
cycle.
Amplification.
N
The
caveat
here
is
that
error
bars
are
probably
a
lower
bound
because
we're
not
capturing
all
sources
of
uncertainty,
but
nonetheless,
I
think
the
takeaway
is
that
this
adds
some
confidence
that
the
rate
of
amplification
has
been
less
than
clausius
clap
around,
and
it's
an
additional
data
point
to
compare
models
against
cool
thanks.
A
P
N
Yeah,
so
it
wasn't
that
I
found
good.
So
basically,
a
subset
of
the
individual
Ensemble
members
had
significant
Trends
so
of
of
the
whole
set
about.
Seven
of
them
met
certain
significance
criteria
that
meant
that
we
could
recover
the
true
response
and
then
in
observations.
We
also
easily
met
the
significance
criteria,
so
I
think
the
takeaway
is
that
they're
closer
to
a
subset
of
of
the
CSM
Ensemble
members
in
terms
of
their
signal
to
noise
ratio.
N
K
Okay
thanks
a
lot
yeah
feel
free
to
carry
on
with
any
questions
in
the
chat,
I
guess
we'll
move.
Q
D
D
P
A
P
Can't
so
just
pretend
I
haven't
got
a
mic
on
cool,
so
we
end
up
with
a
massive
cold,
salty
water
at
the
continental
shelf
and
that's
denser
than
the
surrounding
water.
So
it
flows
down
over
the
continental
shelf
into
the
bottom
of
the
ocean
and
it
mixes
with
other
waters
along
the
way
and
what
we
end
up
with
is
Antarctic
bottom
water
flowing
into
the
bottom
of
the
ocean
and
drawing
in
other
waters
to
replace
it
and
that's
forming
the
lower
cell
of
global
Meridian
overturning
circulation.
P
Zooming
out
here
is
global,
meridianal
overturning
circulation.
We
have
Antarctica
in
the
center
and
branching
out
from
that.
The
three
main
ocean
basins,
an
Antarctic
bottom
water,
fills
the
bottom
of
all
three
of
those
ocean
basins.
So
anything
moving
heat
or
carbon
dioxide
around
in
the
bottom
of
the
ocean.
P
So
by
doing
lots
of
complicated
maths,
you
can
turn
this
distance
metric
between
the
two
satellites
into
an
estimate
of
mass
over
the
Earth
and
gravity
over
the
Earth.
Now
the
Earth
is
a
very
large
heavy
chunk
of
rock,
so
the
dominant
gravity
signal
of
the
earth
is
the
big
bit
of
Rock,
but
the
rock
doesn't
change
much
over
time.
So
if
you
subtract
the
time
mean
gravity
signal,
then
what
you're
left
with
is
changes
in
Water
Mass
over
the
Earth.
P
That's
a
really
simple
ocean
and
if
you
put
some
extra
mass
on
the
top
of
the
ocean,
that
directly
corresponds
to
a
higher
pressure
at
the
bottom.
So
these
Grace
gravity
measurements
are
measuring
ocean
bottom
pressure
and
if
we
move
to
a
slightly
more
realistic
ocean
where
I've
got
some
sort
of
surface
waters
and
then
an
Antarctic
bottom
water
layer
at
the
bottom,
the
grace
satellites
are
measuring
pressure
at
the
bottom
of
the
ocean,
down
both
sides
and
across
the
bottom,
specifically
they're.
P
So
in
the
southern
hemisphere,
if
you
have
a
positive
pressure
on
the
western
boundary
and
a
negative
pressure
on
the
eastern
boundary,
then
that
corresponds
to
a
velocity
into
the
screen,
so
very
simple
conceptual
model.
Theoretically,
the
grace
satellites
should
be
able
to
measure
and
tactive
bottom
water
velocities.
P
This
is
the
Australian
ocean
model,
access,
arm2
and
I'm,
using
it
at
0.1
degree
resolution,
because
at
that
finer
resolution
it
does
a
good
job
of
capturing
Antarctic,
bottom
Water,
Dynamics
and
flows
and
I'm,
estimating
Antarctic
bottom
water
across
a
transect.
So
one
example
here
is
in
the
Pacific
Ocean
and
then
I'm
looking
at
meridianal
velocities
in
different
density
classes.
P
At
the
bottom,
we
have
two
density
classes
going
northwards
and
so
I'm
calling
that
my
Antarctic
bottom
water.
That
I
would
like
to
be
able
to
measure
I'm
emulating
Grace
just
by
taking
the
ocean
bottom
pressure
anomaly,
which
is
what
it
should
be
measuring
and
then
averaging
it
to
the
really
coarse
spatial
resolution
of
the
grace
satellites
and
that's
my
emulated
Grace
product
and
then
I'm
linking
the
two
of
them
just
using
a
linear
sum.
P
So
I'm,
assuming
that
you
can
find
Antarctic
bottom
water
as
some
weight,
multiplied
by
one
particular
pressure
or
gray,
squid
point
plus
another
weight
times.
Another
grid
Point
another
weight
times,
another
grid
point
and
so
on
and
so
forth.
And
then
I
find
those
coefficients
or
weights
using
a
simple
linear
regression.
P
And
that's
what
I
get
so
before.
I
was
showing
pressure
anomalies
on
this
grid
and
I'm
now,
just
showing
the
coefficients
that
I
would
multiply
those
gray
squid
points
by
in
order
to
reconstruct
Antarctic
bottom
water
and
our
flag
that
we
have
a
patch
of
largely
positive
weights
on
the
western
boundary
and
a
patch
of
largely
negative
weights
on
the
eastern
boundary,
which
corresponds
back
to
that
theoretical
geostrophic
balance
link
that
we
have
positive
pressures
and
negative
pressures
on
on
different
sides
of
the
Basin
corresponding
to
Northwood
transport
of
Antarctic
bottom
water.
P
So
we've
got
fairly
High
skill
fairly
low
root
mean
square
error
that
suggests
that
the
Reconstruction,
so
the
red
dashed
line,
is
given
just
the
emulated
Grace
data.
What
do
I
think
Antarctic
bottom
water
would
be,
and
the
black
is
the
measured
and
actually
bottom
water
in
the
model
and
they
line
up
pretty
well.
P
And
then
we
add
that
three
centimeters
water
height
equivalent
of
uncertainty
and
we
get
slightly
lower
skill
but
still
fairly
decent
and
that
lines
up
surprisingly
well,
and
that's
because
I
spent
several
months
trying
to
optimize
it.
So
there's
so
many
different,
like
medic
constraints,
going
into
exactly
how
the
method
works
and
if
you
don't
include
the
noise,
it
doesn't
matter
and
then,
as
soon
as
you
include
the
noise
individual
metrics
make
a
lot
of
difference.
P
But
as
soon
as
you
start
considering
the
uncertainty
in
the
grace
satellite
observations,
then
it
starts
to
matter
that
you,
including
more
data
to
average
out
that
noise
and
I
guess
the
moral
here
is.
If
you're
trying
to
mimic
observations,
then
you
can't
just
mimic
them
in
the
perfect
world
in
the
model,
because
you
might
miss
some
opportunities
to
improve
your
reconstruction.
You
need
to
include
things
like
resolution
also
makes
a
massive
difference
and
the
noise.
The
uncertainty
in
the
data
anyway
I
verified
that
reconstruction
method
with
two
other
models.
P
So
the
top
is
the
same
ocean
model
access
again
at
quarter.
Degree
resolution
It
lines
up
fairly
well.
The
bottom
is
gfd
alarm
4,
which
is
using
mom
6.
Instead
of
mom
5
different
ocean
model
configuration,
it
has
slightly
larger
root,
mean
Square.
The
skill
like
signal
and
noise
is
awful,
because
the
bottom
one
was
run
with
a
repeat
your
atmosphere.
I
didn't
have
access
to
any
data
with
an
internal
event
varying
atmosphere.
P
Okay,
I
can't
change,
slides,
yeah
yeah
I
am
there
we
go
so
we
get
a
reconstruction
of
Antarctic
bottom
water
in
the
West.
Pacific
can
do
the
same
thing
for
the
whole
Pacific
Ocean,
and
so
that
is
the
first
estimate
of
Antarctic
bottom
water
into
annual
variability
in
the
Pacific
Ocean,
and
you
can
see
that
there
is
variability
beyond
the
magnitude
of
the
uncertainty.
P
The
error
bars
are
a
little
larger
than
I'd
like.
So
if
we
could
have
better
Grace
data,
then
that
would
make
things
a
lot
easier.
That's
where
most
of
the
error
is
coming
from,
there's
a
little
bit
from
the
model
differences
and
there
isn't
any
significant
Trend.
So
you
can
look
at
it
and
say:
oh,
it
might
be
decreasing
with
time,
but
but
absolutely
not
significant.
It
doesn't
really
answer
the
question
of
whether
declines
in
transport
are
leading
to
this
decline
in
volume
or
if
it's
just
changes
in
different
Water
Mass
properties.
P
Summing
up,
we
got
the
first
estimate
of
Antarctic
bottom
water
transport
variability
and
how
it
might
be
changing
in
response
to
climate
change,
but
we
don't
actually
get
any
answer
for
what
it
is
doing
in
response
to
climate
change
and
if
you're
trying
to
emulate
any
observations.
It's
important
to
include
the
uncertainty
that
course
resolution
and
the
actual
characteristics
of
those
observations,
rather
than
being
as
close
as
possible
in
perfect
model
world,
and
you
can
use
models
to
work
out
how
those
observations
might
be
used
and
I'm
happy
to
take
questions
now
or
by
email.
K
D
B
P
C
P
Theory
says
you
want
to
have
like
a
basin
with
walls
on
both
sides,
so
I
tried
splitting.
Basically,
every
time
the
topography
came
up
above
the
Antarctic
bottom
water
layer
and
then
back
down
again
and
so
I
get
the
way
specific
and
the
specific
and
then
I
also
tried
in
the
gap
between
Australia
and
New
Zealand.
But
the
grace
resolution
is
like
300
kilometers
and
you
get
two
or
three
tiles
in
the
Gap
and
then
you
actually
don't
get
any
useful
data
out
of
it.
P
A
Okay,
thanks
a
lot,
so
we
were
supposed
to
have
a
break
now,
but
we're
precisely
at
the
end
of
the
break
in
terms
of
time.
I
guess,
I'll
take
a
survey
and
I
feel
like
there's
two
things
we
could
do
either.
We
could
have
a
very
quick
five
minute
break.
A
We
have
three
talks
after
the
break
and
then
we'll
have
a
10
minute
break
before
the
seminar
assuming
there's
no
technical
issues,
or
we
could
just
power
on
through
and
then
have
a
15-minute
break
before
the
seminar
who
wants
option
one
having
a
five
minute
break
now,.
A
O
Hi,
can
you
hear
me
first.
A
A
O
Cool
yeah,
thanks
for
having
me
I'm,
sorry,
I
can't
be
there
in
person,
so
I
am
Danielle
I'm
a
post-doctor
and
car
in
CSU
and
I'll
be
presenting
some
really
preliminary
work.
It's
kind
of
experimental
and
I've
been
working
on
this
on
and
off
with
our
destiner
Dave
Lawrence
and
Jackie
Schumann
and
Carr,
so
my
focus
is
Extreme
fire
weather
and
the
motivation
is
pretty
simple.
We
have
been
seeing
observed,
increases
in
fires
and
burn
area
in
the
western
U.S
for
the
past
three
decades
or
so.
O
A
A
D
O
Okay:
okay,
let's
get
back
into
it,
so
so
yeah
we've
seen
recent
increases
in
area
burned
in
the
western
U.S
and
then,
if
we
look
towards
the
future,
there's
been
a
few
studies
showing
that
extreme
fire
weather
frequency
that
can
drive
really
large.
These
are
weather
conditions
that
can
drive
really
large
wildfires.
O
O
This
figure
from
a
paper
a
couple
of
years
ago
showed
that
the
signal
of
extreme
prior
weather
frequency
is
expected
to
emerge
above
the
natural
historic
variability
by
2080
in
many
parts
of
the
world.
So
the
goal
of
my
current
research
is
to
understand
how
forced
climate
variability
changes.
O
This
is
an
index
that
takes
daily
climate
impact,
maximum
temperature,
precipitation,
relative
humidity
and
wind
speed
calculate,
and
then
we
calculate
three
different
moisture
codes
that
represent
drought,
conditions
on
different
time,
scales,
oops,
sorry
daily
to
monthly
time
scales,
and
then
we
also
then
calculate
two
indices
and
then
finally
get
the
forest
fire
weather
index
and
this
index.
Basically,
we
have
a
daily
value,
and
that
tells
us
How
likely
a
fire
is
to
spread
and
how
large
that
fire
could
possibly
get.
O
So.
One
thing
that
we've
been
experienced
experimenting
with
is
figuring
out
how
to
calculate
an
extreme
threshold
based
on
this
fire
weather
index.
So
traditionally
or
in
recent
studies,
people
use
a
fixed
threshold,
so
this
is
based
on
the
historic
distribution.
So
in
our
case
it's
1980
to
2014
and
we
use
a
percentile
value.
So
in
our
case
the
99th
percentile
by
it
could
be
anywhere
from
the
90th
to
the
99.9th
percentile,
but
basically
I.
We
find
this
threshold.
O
O
We
allow
this
threshold
to
change
depending
on
the
time
that
we're
assessing
so,
for
example,
for
assessing
2045
to
2079.
We
would
now
use
this
blue
threshold
instead
so
and
you
can
see
that
it's
a
higher
threshold
because
of
changes
in
our
climate
and
lastly,
maybe
one
that
you
haven't
seen
before
is
using
this
hybrid
threshold.
So
this
hybrid
threshold
allows
the
mean
state
to
change.
O
So
it
allows
this
p50
to
change
over
time
using
the
moving
window
distribution,
but
it
does
not
allow
the
variability
to
change,
so
we
instead
impose
the
difference
between
the
99th
percentile
and
the
median
and
the
historic
period
on
top
of
the
median
of
the
moving
window
distribution.
So
you
can
see
that
this
is
slightly
higher
than
the
fixed
threshold,
but
lower
than
the
moving
threshold
for
a
future
climate
change
scenario.
O
O
The
hybrid
threshold
gives
us
only
Force
changes
in
the
means,
so
we've
allowed
the
mean
to
change,
but
not
the
variability
so
going
back
to
our
extreme
fire
weather
events,
so
we
also
wanted
to
kind
of
capture
extreme
fire
weather
events
more
holistically,
so
these
events
are
connected
in
space
and
time.
So
we
wanted
to
capture
that
as
well.
So
what
we
do
is
for
each
of
these
thresholds.
O
We
then
identify
grid
points
that
exceed
that
threshold
throughout
all
our
Ensemble
numbers
and
throughout
the
whole
analysis
period,
and
we
and
we
find
grid
points
that
are
connected
in
space
and
time.
So
this
is
an
example
of
one
of
these
events.
O
The
shading
is
the
fire
weather
index
on
a
given
day.
So
these
are
six
consecutive
days
and
the
circles
here
represent
grid
points
that
are
space,
spatial
temporarily
connected
to
each
other.
So
you
can
see
how
this
event
evolves
in
time
and
over
the
region,
and
so
for
each
threshold.
O
We
then
have
13
000
around
13
000,
extreme
fire
weather
events,
and
today
I'll
just
focus
on
California
that
we've
looked
at
on
a
lot
of
the
western
us,
and
each
of
these
events
has
an
area,
a
duration,
location,
and
then
you
can
get
all
sorts
of
variables
from
it
as
well.
I've
also
calculated
the
higher
weather
index,
the
maximum
index
and
the
average
index
as
well
so.
O
Now,
let's
see
how
these
characteristics
are
changing
over
time,
so
here
again,
I'm
focusing
on
California
for
the
summer
months
only
and
on
the
top
I
have
the
duration
of
events
and
the
number
of
days
and
on
the
bottom,
I
have
the
area
of
these
events,
so
the
total
coverage
over
their
lifetimes
and
and
each
I'm
showing
here
the
result
that
you
get
by
using
these
different
thresholds.
O
So
you
can
see
that
with
the
moving
thresholds,
you
have
virtually
no
change
in
your
the
duration
of
your
events
or
the
area
of
your
events,
so
this
by
allowing
the
thresh
the
extreme
threshold
to
to
change
in
time
and
follow
changes
in
the
mean
and
the
variability.
We
see
no
changes
in
the
these
events.
O
However,
when
we
use
the
fix
and
hybrid
thresholds,
we
see
that
events
are
on
average
10
days
longer
and
cover
around
20
times
the
area
they
would
have
compared
to
the
historic
period
and
I
yeah
and
I
also
forgot
to
mention
that
the
spread
here
so
the
mean
the
median
across
ensembles,
is
shown
in
the
thick
lines
and
the
the
25th
to
75th.
Quantiles
are
of
The.
Ensemble
are
shown
in
the
shading.
O
So
so
we
wanted
to
break
down
kind
of
we
wanted
to
isolate
the
changes
in
the
mean
and
the
variability.
So
to
do
that,
we
started
to
kind
of
play
around
with
these
different
results
from
these
different
thresholds.
So
first
we
looked
at
the
fixed
minus
moving
thresholds.
This
gives
us
the
force
changes
in
the
mean
and
gradability,
and
if
we,
if
we
take
the
moving
threshold
away
from
the
hybrid
threshold,
this
gives
us
the
first
changes
in
the
mean
and
when
we
take
the
fixed
minus
hybrid
results.
O
So
I've
taken
the
differences
here
for
both
the
duration
and
area
and
now,
let's
bring
our
little
cheat
sheet
back
of
what
all
this
means.
So
if
we
look
at
this
fixed
minus
moving
these
Force
changes
in
the
mean
and
variability,
we
find
that
that
has
the
largest
effect.
So
you
can
see
this
red
line
over
time
increases
and
this
effect
is
always
larger
than
the
two
other
effects,
and
this
is
because
it
represents
the
force
changes
in
both
the
mean
and
the
variability.
O
If
we
look
at
the
Hybrid
minus
moving
and
this
teal
color,
we
see
that
this
has
the
second
largest
effect,
and
so
this
shows
us
that
there
are
the
four
changes
in
the
mean
are
quite
large
as
well.
However,
if
we
looked
at,
if
we
look
at
the
force
changes
in
just
the
variability,
this
has
a
really
small
effect
over
time
and
this
effect
doesn't
really
change
throughout
the
simulation
period.
O
So
this
is
basically
very
preliminary
results
and
we're
trying
to
understand
continue
to
understand
these
different
effects
and
so
I'll
stop
here
and
take
any
questions.
A
M
O
Do
you
see
that
or
no
yeah
yeah,
we
can
see
it?
Okay,
cool!
Sorry,
I
didn't
hear
you
so
yeah.
So
it's
so
the
way
we
do
it
is
we
take
the
median
of
the
moving
away
the
distribution.
So,
for
example,
let's
see
our
moving
window
is
this
20
45
to
2079.
O
and
which
is
in
blue,
so
we
picked
the
median
of
that
and
then
we
impose,
we
simply
add
on
the
difference
between
the
median
of
the
historic
and
the
P99
of
the
historic.
So
this
is
kind
of
a
really
artificial
way
of
and
very
simple
way
of,
saying.
Okay,
the
tails,
the
distance
between
the
median
and
the
Tails
isn't
changing
over
time,
but
the,
but
the
median
is
changing.
O
So
that's
kind
of
our
the
way
we're
thinking
about
it,
but
I'd
be
really
happy
to
hear
if
you've
done
something
different
or
if
you
have
any
ideas
or
suggestions.
M
Thought
about
it
that
deeply
in
this
sort
of
context,
but.
O
A
D
A
O
So
that's
my
next
step:
we're
planning
to
recalculate
these
events
by
keeping
each
of
the
variables
constant
or
at
the
historic
levels.
So
yeah,
that's
my
next
step
and
I
was
hoping
to
have
results
for
that
for
this
presentation.
But
not
such
luck
so
but
yeah
I'll
I'll
share
those
with
you.
I
get.
Those
yourself
sounds.
A
Good
thanks
yeah,
if
you
could
stop
sharing.
O
R
R
R
So,
first
what
is
Amplified
warming
over
tropical
land?
Here
we
see
two
maps.
The
first
one
shows
the
summertime,
Ming
warming
relative
to
the
global
mean
and
the
second
one
shows
swarming
in
the
hot
hot
tail
relative
to
the
local
mean.
So
for
the
mean
warming,
we
see
landforms
more
than
ocean
and
dry
land
warms
more
than
moist
land
and
for
the
tail
warming.
R
We
see
that
across
almost
all
the
tropical
land,
the
extreme
hot
days
were
more
than
the
me
were
more
than
average
days
and
there
are
two
main
perspectives
on
understanding
this
phenomena.
The
first
one
is
based
on
atmospheric
Dynamics,
so
the
idea
is
taking
changes
in
more
static
energy
as
a
constraint
total
and
it
partitions
between
a
warming
component
and
a
moisten
component
So
based
on
a
weak
temperature
gradient
and
a
convective
quasi-equilibrium.
R
This
perspective
was
first
applied
to
to
understand
the
mainland
ocean
warming
contrast,
but
it
is
recently
applied
to
understand
the
variability
as
well,
and
the
second
perspective
is
from
the
surface,
so
it
focuses
on
the
partition
of
the
net
radiation
received
by
the
lens
surface
into
sensible
heat
versus
latent
heat,
and
so
the
argument
emphasized
that
when
soil
dries
or
plants,
physiology
changes,
latent
heat
decreases
and
surface
flux,
partitions
towards
sensible
heat
and
therefore,
warming
in
in
the
near
surface
is
Amplified.
R
And
so
we
aim
to
compare
the
two
perspectives
and
ask
how
they
connect
how
they
are
connected,
and
we
first
want
to
compare
a
full
spatial
temporal
distribution
of
the
key
variables
in
the
two
perspectives
and
before
we
do
that,
I
want
to
bring
a
note
that
land
is
very
complicated.
So
first
there
is
temporal
variability
and
so
when,
on
the
right
hand,
side
here,
you
see
the
this
is
one
example
grid
cell
from
the
southern
Africa
in
one
model.
R
So
this
is
this
shows
the
relationship
between
latent
heat
and
and
soil
moisture.
The
left
is
colored
by
press
patient
write
this
color
by
temperature
and
when
soil
is
dry,
when
it
is
on
the
dry
days
later
on,
he
is
low
and
then
it
increases
with
increase
of
soil,
moisture
and
then
at
some
point.
R
Latent
heat
loses
sensitivity
to
soil
moisture
and
then
it
decreases
during
the
rainy
days
and
so
the
the
stage
where
latent
heat
sensitively
increased
with
so
much
here
is
called
the
transitional
regime
and
in
this
regime,
trying
on
the
soil,
moisture
will
feed
on
will
feed
back
onto
the
temperature
and
warms
the
temperature
and
second
beyond
the
temporal
variability.
So
the
spatially
land
is
very
heterogeneous.
So
this
map
shows
the
person
how
the
clientological
a
rich
index.
R
We
have
places
like
the
desert,
which
have
which
has
a
very
high
aridity,
and
we
have
places
like
Amazon
or
like
central
Africa,
which
is
very
moist
and
the
latent
heat.
So
much
relationship
is
quite
different
across
different
regions
and
also
for
a
certain
geological
location.
Different
models
represent
the
relationship
differently,
so
the
a
newer
version
of
the
gfdl
model
shows
a
different
shape
and
csm2
shows
a
different
shape
and
also
other
models,
but
an
effective
way
to
examine
the
spatial.
R
Temporal
distribution
is
through
the
space
space
because
so,
if
we
just
not
carefully
just
simply
average
things
across
a
time
periods
or
across
regions
or
like
Aggregates
or
data
together,
it's
easier
to
lose
the
important
signals,
but
by
organizing
data
into
this
kind
of
face
space.
So
the
for
this
kind
of
face
space,
the
x-axis
is
formed
by
the
daily.
It's
almost
your
percentile,
so
along
the
axis,
we're
looking
at
a
local
dry
days
to
local
wet
days
and
y-axis
is
formed
by
the
percentile
of
the
climb
Plus
foreign
index.
R
So
along
the
y-axis,
we
are
looking
at
moist
region
to
dry
region
and
back
to
the
two
perspectives.
We
can
plot
the
changes
in
a
key
variable
in
the
space
space
and
along
the
black
curve.
We
are
looking
at
the
transitional
regime
where
latent
heat
is
sensitive
to
soil,
moisture
and
so
for
more
static
energy
increase.
R
We
see
that
so
for
the
most
conditions,
the
increase
of
MSE
is
uniform,
which
is
consistent
with
the
original
Theory
and
but
over
dry
regions
and
on
dry
days
the
increase
is
smaller
and
so
the
uniform
MSC
assumption
breaks
down
and
therefore
we
find
that
it's
hard
to
provide
an
accurate
prediction
for
temperature
increase
for
those
for
these
conditions
based
on
that
Theory.
R
So
this
includes
places
like
desert
and
in
fact,
and
the
reason
is
because,
in
those
conditions
the
MSC
is
not,
it
is
too
dry.
So
MSC
is
not
high
enough
to
reach
the
collective
threshold,
but
in
fact
those
conditions
are
the
conditions
that
warm
the
most.
R
R
This
strong
partition
between
the
sensible
and
latent
heat,
especially
in
the
transitional
regime
and
one
interesting
note,
is
that
we
might
also
expect
that
sensible
heat
increase
over
the
desert,
but
we
do
not
see
it
here
and
the
short
and
dirty
answer
is
that
surface
long,
wave,
upwelling
long
wave
radiation
is
partly
playing
a
role
for
sensible
heat
so
to
compare
the
two.
R
The
atmospheric
perspective
provides
a
general
principle
on
the
TQ
partition,
but
it
is
not
explicitly
coupled
to
the
surface
and
the
surface
perspective
is
process
based.
We
know
the
physical
process,
but
it
is
local
and
it's
not
coupled
to
the
broader
atmospheric
Dynamics,
but
through
comparing
the
full
spatial
temporal
distribution.
We
see
that
there
is
a
general
correspondence
of
the
two
perspectives,
especially
for
the
conditions
in
the
transition
regime,
and
we
do
note
that
there
is
a
discrepancy
in
terms
of
moistening.
R
So
in
those
conditions,
latent
heat
decreases,
so
evapor
transcription
decreases,
but
the
atmospheric
humidity
is
increasing.
Despite
that,
it
is,
it
is
increasing
less
than
other
conditions,
but
it
is
increasing.
So
this
is
because
the
atmosphere,
atmospheric
Q,
has
other
source
and
sinks
Beyond
evapor
transpiration,
such
as
the
large-scale
convergence
or
and
the
formation
of
condensation
of
into
condensates
and
a
note
for
relative
humidity.
So
relative
humidity
is
often
taken
as
an
indicator
for
atmospheric
dryness,
but
it
depends
on
both
q
and
T.
R
So
here
we
see
Redtube
Community
strongly
decreases
in
these
in
the
transition
regime,
in
correspondence
to
the
thermosphere
drying
and
the
latent
Heats
decrease.
But
the
decrease
of
the
moisture
humidity
here
is
is
like
strongly
affected
by
the
temperature
feedback,
and
next
we
ask
so
which
one
is
more
important
is
more
strongly
associated
with
a
strong
warming,
whether
it's
climatological
dryness
or
changes
in
dryness.
R
So,
in
the
atmospheric
perspective
in
their
moisture
constraint,
the
gamma,
this
ratio
that
that
makes
that
determines
that
tells
us
that
lead
moistens
less
is
determined
by
the
base
climate
state
only
so
we
do
not
need
land
to
get
drier
to
have
land
War
more,
but
in
the
surface
perspective
it
emphasizes
that
soil
dries
and
the
flux
partition
changes
and
to
investigate
this.
R
We
Define
a
partition,
Factor
PSI
that
represents
the
to
to
connect
the
temperature
component
to
the
total,
so
the
inverse
of
that
is
represent
the
ratio
of
the
temperature
component
to
the
total
for
both
perspectives
and
for
a
climate
change.
We
decompose
the
change
in
the
total
into
a
component
that
keeps
the
base
partition
and
a
component
that
has
a
changes
in
Partition
and
we
reorganize
the
equation
and
we
relate
the
we
can
relate.
R
The
magnitude
of
warming
to
one
is
changes
in
total,
which
is
the
analog
for
the
forcing
in
the
climate,
sensitivity,
studies
and
a
term.
That
is
a
sensitivity
which
decide
which
is
decided
by
the
base
kind
of
dryness
and
another
term
that
reflects
contribution
from
changes
in
Partition.
So
the
idea
is
given
a
certain
changes
in
total.
How
base
climb
dryness
and
changes
in
dryness
will
affect
the
magnitude
of
warming,
and
here
the
y-axis
is
the
warning
magnitude.
X-Axis
is
a
base
climate
sensitivity.
R
The
color
is
the
changes
in
Partition
and
each
dot
is
one
pixel
in
the
face
space
that
you've
seen
before.
So
we
see
a
general
relationship
between
the
warming,
magnitude
and
space
climate
sensitivity,
so
a
place
that
is
drier
in
the
base
climate
will
work
more
and
Beyond.
This
generally
quasi-linear
relationship
in
the
here
in
the
intermediate
conditions
between
wet
and
dry,
the
changes
in
Partition
further
enhances
or
dampens
the
warming
for
a
given
base
climate
sensitivity,
and
so
for
the
atmospheric
perspective.
R
Partitions
were
sensible
and
others
passions
were
latent
and
to
summarize
so
the
two
perspectives
show
General
correspondence
so
which
indicates
that
the
atmospheric
perspective
carries
a
strong
surface
information
and
there
and
there
is
discrepancy
in
musni,
which
results
from
other
source
and
sinks
in
the
atmosphere
of
humidity
besides
evap
transpiration,
and
we
show
that
the
base
climate
dryness
can
largely
expand
the
magnitude
of
warming
and
beyond
that,
during
the
intermediate
conditions
between
wet
and
dry,
the
changes
in
Partition
further
enhances
or
dampens
the
warming
and
after
this
I
will
work
with
Karen
Isla.
R
So
yes,
this
is
a
good
question.
It
is
calculated
across
considering
the
variability,
so
that's
the
it's
based
on
the
the
face
space,
so
the
so
the
base
five,
the
base
common
dryness
can
have
a
distribute
spatial
temporal
distribution.
So
it's
not
averaged
over
the
time.
R
H
Any
other
questions
also
I,
guess
just
a
bit
of
a
just
a
clarification,
but
this
side
that
you
defined
you
had
it
defined
in
different
ways.
For
the
two
different
perspectives.
Is
it
that
is
it
the
same
quantity
then
or
is
it
is
oh,
you
have
different
scripts
for
it.
Okay,
so
it
is
a
different
quantity,
but
it's
trying
to
quantify
the
role
of
the
partitioning
in
each
Theory.
A
Q
A
D
A
Q
So,
thank
you.
So
my
name
is
Haruki
I'm
a
postdoc
at
the
University
of
Victoria
in
Victoria,
Canada
and
today,
I'll
be
discussing
some
work
that
we
have
been
doing
with
our
collaborators
at
the
Palo
Alto
Research
Center
and
the
University
of
Washington,
where
we're
looking
at
using
an
AI
implementation
of
the
fluctuation
dissipation
theorem
in
order
to
project
climate
responses
to
radiation
anomalies.
Next
slide,
please
so.
Q
The
fluctuation
dissipation
theorem
is
a
theorem
that
comes
out
of
statistical
mechanics
and
but
simply
it
states
that
the
response
of
certain
dynamical
systems
can
be
estimated
using
the
statistics
of
the
internal
fluctuations
of
that
of
of
the
system
and
it's
being
hypothesized
that
the
climate
is
one
such
system
where
the
FDT
can
be
applied
so
that
there
have
been
several
attempts
throughout
the
years
to
do
this
with
varying
degrees
of
success.
Q
And
so,
for
example,
here
is
one
relatively
nice
example
by
gritson
and
brand
stator,
where
they
use
the
FDT
to
project
the
climate
response
in
the
community
climate
model,
zero,
two
heating
anomalies
in
the
atmosphere.
Q
Okay
next
slide,
please
so
one
common
formulation
of
the
FDT
for
climate
purposes
is
in
terms
of
a
linear
response
function.
So
in
these
cases
the
projected
response,
which
I'm
denoting
Delta
y
here
is
estimated
to
be
equal
to
the
product
of
the
linear
response
function
at
L
and
the
forcing
Vector
Delta
F,
where
L
is
estimated
from
The
covariance
Matrix
of
the
internal
variability
of
the
climate
system,
and
so
this
covariance
Matrix
is
calculated
at
a
range
of
different
time
lags
and
then
integrated
over.
Q
Those
time
lags
out
to
some
point
where
it's
assumed
that
the
Co
the
lag
covariance
essentially
converges
to
noise.
Q
So
next
slide
please
so
when
we're
applying
FDT,
there
are
several
assumptions
that
need
to
be
made
and
for
one
typically,
especially
for
the
climate,
we
need
to
do
a
pretty
substantial
Dimension
reductions
in
order
to
be
able
to
you
know
properly
calculate
the
covariance
matrices.
Another
assumption
made
is
that
the
statistics
of
the
variables
we're
interested
in
are
near
gaussian
and
also
that
the
responses
that
of
the
system
are
linear
and,
finally,
because
we
need
to
calculate
the
covariance
at
you
know
pretty
long
time
lags.
Q
We
also
need
a
large
reference
data
set
of
internal
variability
in
order
to
get
a
good
estimate
of
the
covariance
Matrix,
and
so
what
we
wanted
to
do
is
see
if
we
could
loosen
some
of
these
requirements
by
replacing
the
classic
linear
response
function
with
an
AI
response
operator
and
in
particular,
hoping
to
you,
know,
reduce
the
amount
of
Dimension
reduction.
We
need
to
do
and
loosen
the
assumptions
on
the
gaussian
statistics.
So
next
slide
please.
Q
So
the
AI
model
that
we
used,
or
rather
the
our
colleagues
at
Park
are
created,
is
a
spherical,
multi-layer,
perceptron
model
and
what
it
what
it
does,
is
it
emulates
the
so
it
it's
a
regression
model
that
maps
from
monthly
mean
radiation
anomalies.
So
here
we're
using
cloud
and
so
Cloud,
shortwave
and
long
wave
anomalies,
as
well
as
clear
sky
anomalies
at
the
surface
and
top
of
atmosphere,
and
then
it
Maps
those
anomalies
in
those
variables
to
anomalies
in
selected
output
variables.
Q
So
here
we've
chosen
a
precipitation,
surface
temperature
and
surface
pressure,
and
one
key
thing
we
do
is
that
we
regret
our
data
to
a
spherical
icosahedral
grid
and
this
is
to
avoid
dealing
with
grid
area
variations
which
might
complicate
the
implementation
of
the
AI
model.
Next
slide,
please.
Q
So
for
our
training
data
we're
using
the
CSM
to
large
Ensemble-
and
this
is
because
you
know
we
do
require-
the
AI
still
requires
a
large
amount
of
internal
variability
information
to
be
properly
trained,
and
so,
in
this
case
we're
using
data
from
the
csm2
smooth
biomass
burning,
historical
simulations.
So
that's
the
second
50
Ensemble
members
of
the
historical
simulations
and
in
order
to
pre-process
the
data
for
training.
Q
What
we
do
is
we
remove
the
ensemble
mean
at
each
grid
point
and
each
month
in
the
time
series
so
that
we
only
have
the
monthly
time
scale
or
month
to
month
internal
variability
at
a
given
at
each
of
the
grid
points
and
so
yeah.
One
key
thing
is
that
we
are
removing
the
seasonal
cycle
in
our
data
here
next
slide.
Q
Please
so
sorry
I'm
not
very
good
at
explaining
this
part
yet,
but
so
we've
essentially
developed
a
sort
of
ad
hoc
method
of
applying
FDT,
using
our
the
AI
models,
We've
trained
and
so
first
off.
What
we
do
is
we
train
the
air
range
or
essentially,
an
array
of
different
AI
models
at
the
different
time
lags,
and
then
each
of
these
AI
models,
as
I
said,
are
regression
models
and
essentially
what
they
do.
Q
Is
they
map
from
the
input
field,
X
of
I,
which
is
the
radiation
anomalies
to
some
yeah?
What
I'm
interpreting
as
the
average
of
possible
trajectories
of
the
output
variables
y
for
a
given
time
like
Tau,
and
so
in
order
to
project
the
climate
response?
We
first
run
the
model
many
times
on
different
sampled
input,
Fields
X
of
I,
and
then
we
run
them
again
with
this.
Q
Those
same
input
Fields
but
perturbed
with
some
radiation
anomaly,
Delta
F,
and
then
we
average
over
the
output,
the
difference
of
those
two
outputs
for
many
different
sampled
input
fields,
and
then
we
bring
in
the
our
FDT
idea
again
and
again
here
and
integrate
over
the
range
of
different
lag
models
that
we've
trained
we're
out
to,
in
our
case,
60
months,
which
is
maybe
too
short
for
some
of
the
processes
that
you
know
slower
ocean
processes.
But
this
is
sort
of
our
preliminary
upper
limit
that
we're
using
here
next
slide.
Please,
okay!
Q
So
as
a
test
case,
what
we
use
we're
using
Marine,
Cloud
brightening,
so
Marine
Cloud
branding,
is
a
proposed
solar
radiation.
Geoengineering
technology,
where
Sea
Salt
Air
cell
is
injected
into
Marine
boundary
layer
clouds
to
increase
the
Albedo
and
cool
the
surface,
and
so
what
makes
MCB
interesting
for
our
purposes?
Is
that
there's
a
wide
range
of
possible
MCB,
forcing
patterns?
Q
And
you
know
it's
not
really
practical-
to
assess
all
the
possible
different,
forcing
patterns
using
esms
and
so
we're
sort
of
hoping
that
this
sort
of
FDT
method
might
be
a
tool
that
can
be
used
to
perform
preliminary
assessments
of
different
intervention
patterns.
Next
slide,
please.
Q
So,
in
order
to
evaluate
our
model,
We've
taken
some
data
from
a
separate
project,
we're
working
on
where
we've
imposed
Marine
Cloud
brightening
in
csm2
in
this
case
by
setting
Cloud
droplet
number
concentrations
to
600
percent
rated
cubed
in
three
regions
with
that
have
a
high
concentration
of
clouds
that
are
sensitive
to
Cloud,
droplet
number
concentration,
perturbations,
and
so
you
can
see
them
here
in
the
in
panel
A
and
C,
which
are
cres,
which
is
the
cloud
shortwave
radiation
anomalies,
and
so
we're
increasing
cdnc
in
the
subtropical
Northeast,
Pacific,
Southeast,
Pacific
and
Southeast
Atlantic,
and
then
we
obtain
so
we
obtain
the
radiation
anomalies
that
we
impose
in
our
AI
model
from
fixed
SSD
simulations
in
csm2,
and
this
is
the
Delta
F
that
we
impose
in
our
AI
model.
Q
Okay,
so
here
I'm
showing
some
results
so
on
the
left,
I'm
showing
the
true
response
from
csm2
coupled
simulations,
the
response
of
the
couple
model
to
the
MCB
perturbations
and
then
on
the
right
I'm,
showing
the
projected
anomalies
that
we
get
out
of
the
AI
model.
You
know
integrated
over
the
different
time
lags,
and
we
see
that
yeah.
Q
So
the
AI
model
is
actually
doing
a
pretty
good
job
in
terms
of
projecting
the
pattern
of
the
climate
response
to
the
Marine
Cloud
brightening
interventions,
but
I,
you
know
importantly,
it
is
overestimating
the
magnitude
of
the
response,
but
nevertheless
I
think
there's
quite
a
few.
Nice
features
that
it
seems
like
the
AI
model
is.
G
Q
Know
properly
capturing,
for
example,
if
you
look
at
you
know,
C
and
D,
where
we're
showing
the
precipitation
anomalies,
you
can
see
that
the
AI
model
is
correctly
projecting
that
MCB
will
increase
rainfall
over
West,
Africa,
south
Asia
and
Australia,
but
it's
reducing
precipitation
over
the
Amazon
and
Southern
North
America.
And
you
know
these
sort
of
patterns
are
being
properly
captured
and
right.
So
next
slide,
please
so
I
guess.
Q
One
thing
to
note
is
that
in
the
previous
slide,
I'm
showing
quite
a
large
perturbation,
but
if
we
switch
to
smaller,
more
Regional
perturbations,
we
start
seeing.
Q
You
know
some
of
the
weaknesses
of
the
AI
method,
so
here
I'm
showing
the
temperature
anomaly
due
to
perturbations
in
each
of
the
three
regions
that
I
mentioned
before
the
Northeast
Pacific
Southeast,
Pacific
and
Southeast
Atlantic,
and
here
we're
seeing
larger
discrepancies
between
csm2
and
the
AI,
where
we're
getting
in
particular
for
the
Northeast
Pacific.
The
AI
model
is
projecting
a
stronger
linear-like
response,
and
so
it
yeah.
Q
It
appears
that
the
AI
model
is
perhaps
overweighting
and
so
in
when
it's
training,
but
there
are
indications
that
you
know
it's
learning
non-enco
signals
so,
for
example,
in
panels
enf
I'm,
showing
the
Southeast
Atlantic
response
and
we're
getting
this
Southeast
Atlantic,
Cooling
and
Amazon,
and
tropical
Pacific
warming
pattern
that
it.
We
also
see
in
the
CSM
simulations
next
slide.
Please,
okay,
yeah!
So
just
to
conclude
so.
Q
We've
developed
an
ad
hoc
method
for
implementing
FDT
using
an
AI
model,
and
we
do
find
that
the
large
data
pool
provided
by
the
csm2
ledge
Ensemble
is
crucial
for
our
AI
model.
Training
and
you
know,
applying
FDT
with
this
AI
method
is
able
to
skillfully
project
the
climate
response
pattern
to
Marine
Cloud
brightening,
but
it
pretty
substantially
overestimates
the
magnitude
of
the
response.
Q
So
you
know,
there's
obviously
more
work
to
be
done
in
order
to
improve
the
model,
but
you
know
I
think
there
is
a
good
indication
that
this
sort
of
method
could
be
a
useful
tool
for
generating
a
first
look
estimates
of
climate
responses
to
forcing,
for
example,
for
scenario,
development,
whether
it
is
for
geoengineering
or
for
you
know,
other
sorts
of
you
know
short-term
climate
forces
scenarios.
K
F
Q
Q
The
method
we're
using
is
sort
of
a
I
I
think
we
do
need
to
go
back
into
the
FDT,
Theory
and
and
try
to
you
know
if
you
recall
that
linear
response
operator
that
there's
the
well
first
is
the
covariance
Matrix
of
the
different
lags,
but
there's
also
the
inverse
of
the
auto
covariance,
Matrix
and
so
I
think
trying
to
figure
out
how
that
translates
into
the
AI
would
I
think
that
might
be.
A
Okay,
well
thank
well
thanks
a
lot
to
all
the
speakers
and
I
apologize
for
all
the
technical
difficulties.
Hopefully
things
will
go
a
bit
smoother
this
afternoon.
I
guess
there
are
a
couple
of
things.
I
want
to
say
before
we
take
a
quick
break.
One
is
that
we'll
have
a
happy
hour
at
4
30,
so
Elizabeth
sent
a
sign
up
sheet.
If
you
could
sign
up,
then
we'd
have
some
idea
of
the
numbers
so
that
we
can
yeah.
A
Okay,
so
for
those
online,
if
you
didn't
hear
that,
apparently
this
sign
up
sheet
is
already
expired,
so
we'll
try
and
get
that
fixed
and
we'll
have
a
rough
estimate
so
feel
free
to
join
us
at
4
30
at
Under,
the
Sun.
If
you're
around
the
other
thing
was
yeah
for
online
speakers,
you
need
to
go
to
YouTube
now
for
the
seminar
at
11
and
then
also
any
virtual
speakers
I.
Think
it's
probably
a
good
idea.