►
From YouTube: 2023 Chemistry Climate Model WWG- Day 2 AM Session
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
B
A
D
A
A
A
A
A
F
C
So
I'm
gonna
share
this
share
the
folders
and
what's
I'm
so
sorry,
what's
your
last
name,
one
more
time.
A
C
I
think
we're
gonna
have
seats
at
the
table,
yeah
I'm,
sending
it
right
over,
and
so
it's
going
to
be
under.
So
if
you're
looking
for.
D
D
A
F
D
F
E
E
G
D
A
H
G
D
D
I
A
I
B
Julio
is
stuck
in
with
the
traffic
so
stay
still
and
and
and
share
the
morning
station.
B
So,
first
of
all
two
important
reminders
for
everybody
online,
and
here
in
this
room
we
are
being
recorded.
So
please
keep
that
in
mind
and
then
a
reminder
to
everybody
attending
this
meeting
in
person
at
11.
There
will
be
a
cgd
seminar,
so
after
this
morning
session
we
will
have
to
move
to
that
room.
So
after
these
two
reminders,
let's
get
started
with
our
morning
session,
our
first
speaker
will
be
Dave
Randolph.
Who
will
give
us
some
update
of
Earthworks.
E
E
So
most
of
you
know
by
now
that
Earthworks
is
a
five-year
project
funded
by
the
computer
science
part
of
the
SF
size
and
the
goal
is
to
starting
from
cbsm,
develop
a
global
storm,
resolving
coupled
model
so
using
a
single
mirror,
uniform
grid
for
all
the
components,
so
the
coupler
has
less
to
do.
Of
course,
it's
still
there,
so
the
components
are
chemistics
physics
with
the
impasse.
Non-Hydris
got
a
local
core
develop.ng.
E
E
So
before
we
will
discuss
the
importance
of
using
the
components
of
these
scenes
that
are
including
a
specialist
index
and
we
are
working
to
observe
compatibility
with
the
CBS
and
quote
base.
We
don't
want
to
break
anything,
and
we
also
want
to
maintain
all
of
the
features
of
what
mq,
including
Regional,
refinement
capability
and
that's
for
both
the
atmosphere
of
the
ocean.
E
So
our
goal
is
our
Target
grade
is
3.75
kilometer
uniform
or
almost
you
know,.
E
But
up
to
now,
we've
been
using
NSM
machines
which
I'm
talking
about
so
initial
objectives.
Point
the
ocean
and
atmosphere.
Sorry,
the
the
ocean
and
sea
house
models
into
cbsm.
That
was
actually
a
pretty
good
job
that
was
done
in
CSU
by
Donald,
dazzling
and
finished
more
than
a
year
ago.
So
then
assemble
all
the
pieces
into
the
independent
configuration
which
I've
already
got
in
mind
and
perform
a
fully
coupled
simulations
initially,
which
is
120
kilometers
and
then
along
the
way
test.
E
The
physics
and
the
convection
committee
Bridge
mixing
and
that's
what
bill
will
be
talking
about
I'm,
not
going
to
say
anything
about
today
and
then
according
to
GPU,
Sportsman
and
I'll,
say
something
about
that
later.
So
we've
done
both
amip
simulations
and
holy
couple
simulations
with
the
120,
60
or
30
kilometer
prints.
We've
done
aquapuna
simulations
with
all
of
the
grids
very
short
in
3.75,
but
just
a
test
and
we've
done
a
simulation.
E
Corporation
refinement
and
today
I'll
show
you
some
numbers
from
a
test
just
in
the
last
10
days
or
so
3.75
of
the
planet
on
a
NSF
missionary,
Texas.
E
So
I
talked
about
this
briefly
at
HW
and
AMS,
so
most
of
your
property,
but
we
have
to
neutralize
the
ocean
model
for
a
couple
of
reservoirs.
We
have
to
initialize
the
sea
ice
and
in
our
first
attempts
what
we
did
was
prescribe
the
sea
ice
in
a
very
simple-minded
way
shown
here.
So
the
yellow
is
the
ice.
E
And
so
in
the
northern
hemisphere
there
was
two
and
a
half
meter
thick
ice
everywhere
in
the
sea
ice
was
present.
This
was
the
January
first
national
Commission
and
in
the
southern
hemisphere
two
and
a
half
meter
thick
sea
ice
from
67
to
the
Pope
in
the
northern
hemisphere
is
seven
years
more
physical.
E
E
You
can
see
my
cursor
starts
to
lie
and
it
tends
to
increase
from
time
and
of
course,
there's
some
attendant
pooling
for
those
of
that
we
have
something
like
a
one
and
a
half
watt
per
square
meter
in
top
of
atmosphere,
radiative
imbalance,
so
that
was
no
good
and
okay
I,
don't
see
but
make
sure
all
later
and
Bill
Khan
gave
us
some
advice
on
using
ocean
model
in
comparison
project
initialization
protocol,
which
is
quite
a
mouthful,
as
documented
in
the
paper
by
Griffeys
at
all
I
listed
here,
and
it
involves
running
the
ocean
and
sea
ice
with
data
atmosphere
for
62
years
behind
the
times.
E
310
years
of
simulation
and
during
that
restoring
the
surface
solidity
shows
more
details
here.
That
I
won't
talk
about
so
we're
currently
doing
this
and
I
expect
to
be
able
to
report
results
in
a
month
or
two
we're
doing
this
on
the
120
kilometer
grip
course
to
start
with,
and
then
Our
intention
is
to
initialize
the
ocean
at
high
resolution
by
interpolate.
E
As
a
graduate
student
at
CSU
and
he's
in
the
early
stages
of
a
project
to
look
at
Tropical,
Cyclones
and
Earthworks,
eventually
looking
at
something
to
do
with
the
climate
intervention,
but
we're
not
there
yet.
So
this
is
a
single
figure
showing
I.
Don't
know
if
you
see
the
little
red
and
green
dots
that
represent
tropical
Cyclones
on
a
stormy
day
in
1994
in
English,
Andrew,
essentially
reformulated
the
couple
of
cyclone
detection
algorithm
of
shallow
and
held
at
all.
E
This
rewrote
it
using
cyclone
and
has
used
that
to
produce
some
statistics
so
on
the
3400
reports
in
80
to
902cs
in
Europe
might
be
slightly
hydrated.
E
They
had
live
walls
and
they
move
in
more
or
less
realistic
way
and
the
right
places
The
Wind
Springs,
that's
measured
by
minimum
surface
pressure.
Sorry,
the
intensity,
that's
measured
by
minimum
surface
pressure
are
fairly
strong
on
the
30
kilometer
grid,
like
up
to
at
four
or
so,
but
in
terms
of
surface
wind
speed
they're
only
about
an
attack
too.
So
there's
something
we're
not
quite
right
there.
That
probably
involves
property
in
some
way
and
I
can't
tell
you
any
more
details.
E
What
we're
looking
into
this
it
may
have
something
to
do
with
club,
which
we'll
be
doing
about
later
and
me
one
slide
on
GPU
porting.
So
many
of
you
probably
know
that
the
NFL
atmosphere
was
ported
to
gpus
by
IBM
and
a
car
and
a
collaboration
several
years
ago,
and
the
MPS
ocean
was
supported
computers
by
the
television
and
right
now
we're
not
worrying
about
the
CIS,
we're
letting
that
go
CPU,
but
the
physics
reporting.
E
E
So,
as
I
mentioned
a
few
minutes
ago,
we've
recently
run
very
short,
distribute
time
steps
like
opening
simulation,
qtc6,
Honda
3.7.4
grid
on
1024
nodes
of
Frontera
shown
in
the
picture
here,
which
is
that
the
Texas
Advanced
Computing
Center
is
an
intercept
machine
and
we
give
access
to
that
as
well
as
some
of
your
weight
machines.
Frontera
runs.
This
particular
configuration
about
almost
two
times.
E
E
There
are
some
issues
that
we
found
at
3.75
and
they're
circled
in
red
here,
so
the
speed
of
the
Cutler
is
slower
than
we
expect
this
not
doing
much
more,
of
course,
but
the
memory
usage
is
not
scaling
as
expected,
but
a
big
thing
is
here:
initialization.
E
We've
looked
at
the
scaling
on
course
of
Reds.
This
is
perhaps
slower
than
you
might
expect
on
the
30
kilometer
grip,
but
you
know
this:
we
can
live
with.
E
3
100
seconds
is
just
not
okay
and
Rich.
Love's
comment
was
that
it
appears
that
the
initialization
is
scaling
as
if
it
was
serial
so
I,
don't
know
if
that's
the
case,
but
this
needs
to
be
investigated,
and
so
what
happens
with
us
every
time
they
go
to
higher
resolution,
we
find
even
problems.
This
is
something
that's
showing
up
at
3.75.
E
But
relating
to
our
goals,
which
I
already
mentioned,.
E
And
based
on
the
results
we're
getting
from
tac,
we
can
estimate
that
this
may
be
achievable
on
the
gpus
that
we
expect
to
be
available
because
machines
by
20045
so
we'll
see,
but
Bitcoin
does
not
appear
to
be
out
of
state.
Let
me
skip
ahead
here,
because
I
have
another
slide
and
I
know
where
it
is
no
I.
Don't
so
I
deleted
it!
Well,
sorry
about
that.
It's
a
slide
that
I
got
from
Bill
Shamrock.
That
shows
memory
scale,
I.
H
E
So
when
we
think
about
scaling,
we
often
think
in
terms
of
you
know,
wall,
clock
performance
speed,
but
it's
also
possible
to
think
about
memory
scaling
and
that's
what
the
slide
deals
with.
So
the
idea
is
that,
as
we
let's
say
for
a
given
grid,
we
increase
the
number
of
nodes
to
try
to
make
the
thing
go
faster,
that's
strong
scaling
or
we
increase
the
resolution,
make
the
grid
bigger
but
simultaneously
add
nodes
in
the
same
proportion.
E
So
this
is
a
bit
complicated,
but
down
here
you
see
results
obtained
with
mpas,
and
this
is
the
m
cubed
version
of
that
pass.
The
version
that
is
not
running
inside
cpsm
and
the
actual
parameters
of
the
simulations
here
are
different,
so
the
space
between
these
lines-
and
these
is
not
being
full.
Okay,
that's
not
the
point
of
his
life,
but
the
first
comment
to
say
is
the
change
of
memory
per
thread
on
the
vertical
access
with
columns
per
threaded
notice.
E
The
numbers
increased
to
the
left
here
is
is,
as
expected,
you're
seeing
results
on
both
the
30
kilometer
gray
and
blue
and
the
60
kilometer
grid
black
and
looks
pretty
good
when
we
go
to
the
version
of
the
model
that
runs
and
this
cycle
planet
again
running
inside
the
cpsm
infrastructure.
We
get
these
curves
and
you
can
see
there
are
various
resolutions.
These
English
ones
that
don't
have
a
line
drawn
through
them
are
15
kilometer
grid
the
green
here,
all
right,
the
30
kilometer
grid
and
the
and
so
on.
E
So
there
are
sort
of
two
points.
The
first
one
is
that
something
strange
is
happening
over
here
on
the
right
side
when
we're
going
to
very
large
and
normal
remote
notes.
E
Essentially,
the
memory
usage
increases
in
a
way:
that's
not
reasonable
or
effective,
but
we
plan
to
be
using
on
the
other.
You
know
2
000
columns
per
ship
in
the
actual
applications,
so
we'd
be
over
here
somewhere,
but
over
here
we
see
this
separation,
so
the
blue
and
the
red
curves
are
pretty
much
on
top
of
each
other
here,
but
then
the
winter
is
using
more
memory
and
the
pink
dots
are
using
even
more
and
we're
only
at
15
kilometers
and
we
still
have
two
more
doublings
to
go.
E
So
this
is
another
issue
and
we
don't
know
the
crossover
at
this
point.
But
again
it
does
not
happen
again.
E
And
I
forgot
to
say
that
the
initialization
timing
that
I
mentioned
the
excessive
initialization
I
also
does
not
happen
in
the
abuse,
so
these
issues
appear
to
have
something
to
do
with
the
way
CBS,
and
this
is
constructed
and
we're
looking
into
that,
we'll
find
a
problem,
we'll
fix
the
problem
and
that
will
benefit
everybody
who's.
Trying
to
remember
so,
we'll
let
you
know,
but
that's
all
I
can
report
right
now.
E
Okay,
so
back
to
the
summary,
we
can
get
on
track
to
achieve
our
goal.
But
you
know
over
the
first
five
years.
E
E
G
E
We
just
got
the
results
10
days
ago,
so
we
don't
know
yet,
but
that's
one
of
the
things
if
I
could
say
it's
not
the
Peter's
thing
from
a
Frontera
results
once
you
get
to
once
you
get
to
integrating
the
memory
usage.
Is
this
flat
and
that's
right.
E
B
E
So
so
what
we
are
doing
to
test
the
physics
that
connector
scale
measures,
essentially
three
kilometer
cell
spacings-
is
that
we're
using
the
variable
resolution
capability
event
pass
with
a
60
between
kilometer,
the
three
kilometer
region.
Is
it's
not
very
big,
but
it's
big
enough
to
hold
convector
systems,
tropical
Cyclones
Etc
and
what
we've
done
is.
E
First,
one's
a
Squall
line,
here's
the
best
you
can
see
depicted
on
the
right
and
that
inner
circle
is
essentially
cell,
spacing
less
than
four
kilometers.
So
we've
got
a
fair
amount
of
the
central
U.S
and
what
we're
doing
is
just
forecast
mode.
So
we're
running
a
couple
days,
initialized
from
the
GFS.
This
is
with
the
58
level
configuration
which
is
kind
of
the
Workhorse
one
projected
soon
for
cam
in
that
next
release,
where
the
monotop
around
40
kilometers,
so
so.
E
So
this
is
the
thing
to
look
for
I
realize
this
is
kind
of
small
on
the
screen
on
your
left
is
the
observed
reflectivity,
and
this
is,
after
a
day
of
simulation
time,
the
model
spins
up
very
quickly,
so
it
spun
up
in
terms
of
convection
on
the
right
here
is
a
simulation
with
empaths
using
morph
physics,
the
Standalone
version,
and
what
you
see
here
is
this
is
a
classic
Squall
line.
E
You
can
see
the
bright
red
on
the
Easter
side,
that
line
that's
the
Leading
Edge,
that
propagates
from
west
to
east
and
behind
it
the
yellow
reflectivity
is
the
stratiform
region
and
you
can
see
the
same
kind
of
signature
in
the
empath
stimulation
of
War
physics,
and
this
is
kind
of
standard.
Typically,
three
kilometers.
These
models
will
kind
of
overdo
the
convection
a
little
bit
and
you
won't
see
some
aspects
of
the
stratiform
region,
but
but
you
get
most
of
it.
E
So
this
is
a
reasonable
simulation
and
I
would
say
this
is
poreless
state
of
the
art
or
or
for
models
here
for
non-hyprostatic
models
and
it's
kind
of
been-
and
this
is
relatively
strongly
forced,
there's
a
nice
cold
front
coming
through
a
Vero
Clinic
wave,
but
the
convective
system
is
out
in
the
warmth
sector
that
way,
so,
while
the
front
kicks
off
convection
that
system
propagates
out
ahead
and
it's
kind
of
very
standard
Squall
line.
So
this
next
figure
is
what
happens
when
you
just
run.
E
You
can't
search
physics
and
passing
Camp
same
exact
setup,
but
just
running
Camp
six,
although
we
are
running
NT3,
so
that
means
we're
included
a
gravel
hail
category.
If
you
want
to
think
of
it.
That
way
and
as
you
can
see
well
for
one,
the
Strathmore
region
is
is
way
too
large.
It's
too
wide
and
also
the
convection
is
not
concentrated
at
the
front
of
the
line
it's
at
the
center
and
what's
going
on
is,
is
it's
staying
tied
to
the
cold
front?
E
We're
not
seeing
a
propagation
of
the
system
in
here,
so
so
it
just
doesn't
look
very
good.
So,
okay,
that
was
first
out
of
the
box,
and
one
thing
we
tried
was
one
of
the
main
differences
between
the
microphysics
or
the
microphysics
and
macrophysics
in
cam
6.
Compared
to
most
cloud
model.
Implementations
is
the
fact
that
the
saturation
adjustment
is
done
inside
of
Club,
using
the
two
moments
using
the
two
PDFs,
and
so
what
we
thought
we
would
try
is
to
collapse.
E
Those
PDFs
to
Delta
functions,
to
try
to
mimic
how
it's
done
in
the
regular
Cloud
models.
We
did
that
and
what
you
get
is
the
figure
on
the
far
right,
which
shows
a
much
nicer,
Squad
line,
but
still
we're
not
seeing
that
convection
a
thin
line
of
convection
right
at
right
on
the
left
side
on
the
right
side
of
it
at
the
Eastern
Edge
and
we're
not
seeing
a
good
stratiform
region,
so
we've
gotten
part
of
the
way
there,
but
not
all
the
way.
E
So
that's
one
one
change
we've
gotten
this
given
us
ways,
at
least
in
terms
of
the
test.
This
is
a
depiction
of
a
cross-section,
a
west
to
east
cross,
section
of
the
potential
temperature
over
the
first
six
kilometers
of
the
the
atmosphere.
And
if
you
look
at
the
the
cold
colors
down
low,
the
the
blues
and
the
greens,
that's
the
gust
front
on
the
far
left
is
the
impacts
or
physics.
So
what
you
see
is
you
see
a
a
deep
coal
pool
with
a
very
sharp
Leading
Edge?
E
It
extends
up
about
a
kilometer
and
that's
where
the
the
flow
over
that
gust
front
is
is
triggering
new
convection
and
you
can
see
in
the
middle
one.
The
standard
configuration
with
Cam
6
physics
we're
not
seeing
that
deep
coal
pool
at
the
Leading
Edge
of
where
that
gust
front
is.
When
we
make
that
change
to
the
saturation
adjustment.
E
We
can
see
that
the
cold
pool
looks
a
little
better,
but
we're
not
seeing
quite
that
depth
and
strength
that
we
expect
to
see
so
seeing
that
we've
actually
gone
into
the
mg3
microphysics
and
made
the
change.
Try
to
increase
the
evaporative
cooling
in
the
boundary
respectively,
and
you
can
see
we're
getting
a
somewhat
deeper
cold
pulled.
But
you
can
see
it's
not
quite
as
strong
and
as
deep
as
we
would
expect
from
the
other
simulations
of
what
we
know
from
observed
qualities.
E
We
expect
to
see
so
essentially
we're
moving
forward
with
some
tuning
we're
getting
better,
but
we're
not
quite
where
we'd
like
to
be
we're,
certainly
not
state
of
the
art.
Yet,
but
at
least
there's
a
suggestion
that
perhaps
increased
tuning
and
changes
might
get
us
closer
to
where
we
want
to
be
with
the
camp
6
physics,
which
is
our
goal
Okay.
So
tropical
Cyclones,
we
have
simulations
of
tropical
Cyclone
Maria.
E
This
was
back
in
2017
a
lot
of
damage
in
Puerto
Rico,
for
example,
when
this
made
landfall
there-
and
this
is
the
mesh
we
used
to.
You-
can
see
the
three
kilometer
region
down
in
the
Caribbean
there.
It's
a
very
strong
storm
during
this
period
cap
four
cat
five,
it's
been
a
very
intense
cyclone
and
essentially
the
simulation.
We're
we're
going
to
see
from
the
models
was
done
when
it
was
always
over
the
sea.
So
we're
not
dealing
with
I
think
land
issues
here.
E
Okay,
so
on
your
left
is
the
Standalone
impasse
with
the
war
physics.
This
is
top
of
the
atmosphere
olr,
and
this
is
essentially
a
brightness
temperature
observed
brightness
temperature
from
from
satellite
they're,
not
the
same
color
scales.
What
I
want
you
to
notice
is
that
the
scale
of
the
tropical
Cyclone
the
eye,
the
bandage
structure,
you
can
see
them
in
both
and
they're
above
the
right
scale,
and
this
is
after
two
days
of
simulation,
so
which
is
more
than
enough
to
spin
up
Cyclones
in
the
model.
Absolutely
so
so.
E
This
is
a
reasonable
depiction
and
this
is
kind
of
what
we
expect
to
see
so
on.
The
next
slide
is
what
we
get
with
Camp
six
physics,
then
pass
same
mesh
same
setup,
same
initialization,
just
running
Camp,
six
physics
with
mg3
the
same
depiction
of
the
olr,
so
we
have
a
change
of
color
scales
or
anything,
and
what
you
see
is
that
we
get
a
much
larger
system.
We
don't
see
the
bandage
structure
very
well
at
all,
and
if
you
look
at,
for
example,
reflectivity
and
other
measures
updrafts
the
band
is
structure.
E
Just
does
not
look
good
compared
to
the
observation.
This
is
I,
think
the
experience
with
Cam
6
physics
on
mesoscale
grids,
for
example-
maybe
30
kilometer
cell,
spacing
that
you
get
these
big
Donuts
I've
heard
it
referred
to
them.
The
recent
account
they're
easy
to
find,
but
when
you
look
at
the
structures,
they're
not
what
we
want
to
see.
So
we
we
made
some
changes,
and
this
is
the
first
change
using
the
collapsed
PDF
and
the
saturation
adjustment
in
clubs
and
right
away.
E
You
can
see,
while
it's
still
too
big
but
we're
starting
to
see
the
band
destruction.
That's
a
good
sign!
That's
a
good
sign
and
we've
made
a
few
other
changes,
including
we've
changed
the
the
the
cooling
and
the
downdrafts
evaporation
plus
we've
taken
some
of
the
changes
that
have
come
from
the
process
team.
E
That's
been
looking
at
that
that
column's
been
involved
in
a
number
of
other
people,
including
some
folks
at
mpug
and
now
we're
seeing
a
much
better
bandage
structure
here,
so
we're
going
in
the
right
direction,
but
still
it's
too
big,
although
I'm
not
sure,
given
such
a
strong
Cyclone
that
this
is
necessarily
indicative
when
we
make
those
changes
of
what
we
should
expect
on
average.
E
So
we're
going
to
be
doing
a
lot
more
cases,
but
what
we're
trying
to
do
is
kind
of
get
to
the
first
order
result
that
we
can
maybe
sort
of
live
with.
So
that's
where
we
are-
and
this
is
the
the
summary
slide.
Essentially,
we've
made
some
changes
to
the
camp
such
visits
in
terms
of
tuning
or
actually
going
into
changing
code
to
collapse.
The
PDFs,
but
essentially
I,
think
we
expected
to
make
changes
to
the
microphysics
because
they
haven't
been
run
at
convective,
permitting
scales
in
models
very
much.
E
So
we
knew
we'd
have
just
some
tuning
there,
but
I
think.
The
big
question
is:
how
do
we
address
the
PDFs
and
Club
and
there's
a
group
of
us
that
are
talking
about
that
now
and
we're
going
to
see
look
at
some
options
in
terms
of
how
we
do
that.
The
last
thing
I
want
to
mention
is
especially
on
the
variable
resolution
scheme,
the
Deep
convection
scheme
cem.
It
tends
to
scale
its
precipitation
with
time,
step,
we'd
like
it
to
scale
it
with
the
mesh
size,
for
example,
in
a
60
kilometer
region
of
the
mesh.
E
When
we're
running
the
variable
resolution,
with
a
three
kilometer
time
step,
we
would
like
the
60
kilometer
region
of
the
mesh
to
have
the
convective
parameterization
a
deeper
Vector
parameterization
produced
what
it
would
using
that
60
kilometer
time
step.
If
you
had
a
universe-
and
we
don't
have
that
yet
so
we're
going
to
be
addressing
that
too.
So
that's
what
I
have
to
say
and
I'll
be
happy
to
entertain
two
questions
here:
foreign.
E
We
we
didn't
look
there
in
some
sense
in
a
lower
resolution,
I'm,
assuming
that
that
club
is
fairly
well
tuned
for
the
half
degree
to
degree
measures.
So
I
would
like
to
be
able
to
leave
it
like
that
to
leave
it
to
behave
like
that
in
that
in
those
regions.
D
See
what
the
presentation
one
point
it's
supposed
to
scale
with
the
light.
There
is
a
length
scale
in
there
and
it's
actually
supposed
to
scale
with
a
length
scale,
so
it
should
adjust,
but
it
doesn't
seem
to
be
effective
so,
ideally
like
he
was
talking
about
with
the
Deep
convection.
There
is
a
length
scale
term
in
there
that
should
change
the
PDF
as
the
mesh
size
changes.
D
B
E
E
D
E
Kilometers
doesn't
provide
any
new,
it
doesn't
change
the
structure
of
the
Cyclones
it
it
does,
but
it
does.
D
E
C
I
saw
like
in
the
lower
left
corner,
there
was
also
a
very
big
difference.
If
you
looked
away.
E
In
kind
of
the
trail,
a
region,
so
so
that
depends
on
where
the
cold
front
is
it's
okay,
so
so
in
the
impasse,
with
more
physics
from
the
cold
pool,
the
the
cold
front
that
progressed
just
a
little
bit
further
east
than
what
you
see
in
cam
runs:
okay,
I,
don't
think!
That's
a
problem
necessarily
with
Cam
physics,
okay,.
H
I
E
B
So
you
think
you're
built
next
Adam
Harrington
will
talk
about
covering
a
variable
resolution.
Atmosphere
to
the
part
two
things.
D
D
D
D
A
E
So
I'm
going
to
be
talking
about
a
couple
in
a
variable
resolution:
atmosphere
940s
unless
you've
been
done
by
9
45
the
latest.
This
is
a
joint
CSM
and
CNA
activity
on
the
CEO
Sunny
Side.
This
is
a
long-standing
cross-working
group
with
the
Landis
working
group
and
then
I
also
want
to
acknowledge
Zeke
Yin,
who
is
a
PhD
student
at
CU,
formerly.
D
G
E
Okay,
so
this
is
the
Arctic
grid.
I
write
about
it
in
this
James
paper.
It's
quarter
degree
resolution
over
the
broader
article,
100
elsewhere,
we'll
be
running.
The
CSM
2.2
code
base
go
over
sort
of
challenges
such
as
tuning
go
over
the
pi
control
climate,
I'll,
sort
of
a
one
percent,
CO2
experiment
and
the
book
is
going
to
be
on
the
Greenland
ice
sheet
and
so
system
is
on
and
we'll
be
focusing
on
its
response.
E
Here
is
you
want
to
compare
it
to
the
one
degree
work
course
and
see
what
resolution
does
for
us
if
anything
at
all,
so
the
challenges,
there's
there's
two
problems
that
we
keep
bumping
into
and
using
conventional
physics
packages
on
variable
resolution
grids
and
inadequate
scale.
Awareness
is
obviously
a
big
one
and
largely
unsolved,
and
this
is
just
an
example
of
how
bad
it
can
get
where
this
is
an
opera
planet
where
the
igcc.
D
D
E
And
within
the
refined
reason
you
get
these
much
larger,
resolved
updrafts
that
facilitated
Divergence
a
lot
and
then
downstairs
outside
of
the
refined
region.
You
get
conversion,
so
you
have
a
pretty
circulation
that
if
you
move
the
patch
of
Border
use
grinder
the
circulation
mechanism.
So
it's
not
something
physical,
but
we're
refining
the
Arctic
there's
not
a
lot
of
diagnetic
forcing
heating.
So
we
really
don't
have
to
worry
about
that.
E
The
other
issue
is
this
very
large
sensitivity
of
the
climate
to
the
physics,
Time
stuff
and
I
mean
I.
Guess
you
could
just
run
at
the
one
degree
time
step
and
you
wouldn't
have
to
retune,
but
I.
Don't
think
that's
a
very
smart
decision,
because
I
mean
you
have
massive
termitation
errors
that
you're
introducing.
If
you
use
two
course,
physics,
Time
stuff
and
that's
just
Illustrated
with
a
simple
moist
bubble,
experiment:
multiple
experiments,
each
dot
is
the.
E
The
excess
he's
gone
from
one
degree
to
quarter
degree,
and
these
are
the
vertical
velocities
associated
with
those
bubbles.
And
what
you
want
to
see
is
this
one
over
DX
type
scaling
which
goes
with
the
pressure
gradient.
But
if
you
hold
the
physics
time
Step
at
one
that
one
degree
physics
time
step
so
a
half
hour
as
you
go
to
quarter
degree,
these
vertical
velocities
get
more
and
more
truncated.
E
What
you
want
to
be
doing
is
giving
more
resolved
connection,
but
you're
trying
to
be
able
to
do
it.
So
I
don't
like
to
do
that,
so
we're
gonna,
we're
gonna,
go
ahead
and
use
a
small
time
step
and
reach
you
in
the
model.
E
Basically
just
to
give
you
an
idea
of
this
this.
What
I
do
is
I
just
take
a
one
degree
model
and
I
tune
that,
with
a
small
physics,
time
step
that
that's
good
enough,
it
seems
to
be.
It
took
me
about
250
years
of
simulations,
so
about
one
and
a
half
two
million
core
hours
to
tune
this
thing,
which
I
don't
think
is
that
unreasonable?
E
So
let
me
go
into
the
actual
simulation,
so
I
branched
off
of
this
bg7
control,
which
you
look
at
the
global
ocean
temperatures
of
what
I'm
branching
off
of
this
is
just
the
one
leg
of
the
control
and
it's
it's
a
slightly
warmer.
It's
warming
there's
a
drift
in
the
pi
control
and
it's
also
a
drift
in
the
pi
control
in
the
official
scene
of
six
deck.
E
E
You
can
see
the
arctic's
a
bit
colder,
but
it
also
reduces
region,
so
it
was
too
hot
yeah.
E
So
the
reason
it's
cooler
is
there's
a
two
or
three
Watts
meter,
squared
reduction
in
solar
radiation.
E
That's
not
shown
but
I'm
showing
here
that
cloud,
forcing
changes
due
to
my
tuning
and
as
you
can
see,
I
I
thickened
up
the
strike,
cumulus
decks
off
of
the
Coastal
weapon
zones
in
the
southern
hemisphere
in
particular,
and
also
a
dicons
region,
and
we
look
at
the
SSD
change
due
to
the
tuning.
You
can
see
it
cools.
The
southern
hemisphere.
E
Yeah
obviously
see
a
big
Cooling
in
the
northern
hemisphere
over
the
polar
cap,
and
that's
because
one
of
my
tuning
knobs
was
the
Albedo
of
snow
over
sea
ice,
and
so
you
do
have
a
colder
Arctic
and
the
sea
ice
thickens
a
bit.
E
You
see
this
warm
anomaly
over
the
antibiotic
action
center
and
if
you
look
at
the
time
series
of
Anger
Management,
you
can
see
that
in
my
arctic
configuration
it's
quite
a
bit
more
intense,
but
there's.
E
So
if
we
take
a
look
closer
look
at
the
amok,
it
is
more
intense.
This
is
what
it
looks
like
relative
to
the
control
of
branching.
Off
of
so
it's
a
good
25,
more
intense,
overturning
stream
function
and
I
didn't
have
to
search
very
hard
for
a
culprit.
I
realized
it's
not
coming
out
too
well
in
this
in
this
plot
or
in
this
slide.
But
this
is
supposed
to
be
the
Grid
overlaid
on
the
sensible
and
latent
heat
level
compared
to
the
control
and
well
I.
E
Guess
we
can't
see
it,
but
there
is
a
25
watts,
computer
squared
reduction,
increase
in
the
maintenance
of
Speedy
clutches
over
the
impact,
overturning
centers,
and
so
it's
likely-
and
you
see
it
in
the
wind
stress
as
well,
and
so
it
seems
that
the
higher
resolution
atmosphere
is
has
storms
with
greater
variability.
The
surface
winds
that
are
mapping
onto
the
surface
fluxes
and
pulling
heat
out
of
the
ocean
and
facilitating
greater
deep
water
formation.
E
Try
to
prove
it
wrong,
a
little
harder
but
anyways
moving
on
out
of
the
ocean
and
talking
about
Greenland
Greenland's
a
bit
of
a
challenge,
because
the
precipitation
rates
are
very
sensitive
to
reservation
in
Greenland.
And
basically
what
happens?
Is
it's
in
this
cartoon
schematic
over
resolution?
E
In
contrast,
in
the
quarter
degree
and
eighth
degree
8th
degree
models,
the
green
one
actually
to
sufficiently
state
that
it
rings
out
the
existence
at
the
coastlines.
Some
of
the
rain
involves
into
the
ocean.
Some
falls
on
Greenland.
The
net
result
is
that
your
priesthood
biases,
which
are
shown
here
on
the
bottom
and
one
to
two
degree
models,
are
much
larger
than
at
one
quarter
to
A3
degree.
Models
shown
here
in
the
Arctic
growth
is
that
called
the
quarter
degree
point
now.
E
So
here's
what
the
ice
volume
of
the
green
and
ice
sheet
looks
like
in
the
bg7
control
is
in
blue,
and
so,
if
I
look
at
the
tricks
and
I
just
restart
this
run
again
I'm
just
running
the
Workhorse.
Wanted
me
we're
not
talking
about
the
Arctic
if
I
lift
the
tricks,
you
get
this
this
purple
line,
and
you
actually
just
now
starting
to
grow
and
to
become
more
faithful
to
the
actual
surface
mass
balance
it's
receiving.
E
It's
still
spinning
up
when
I
discovered
this
I
decided
I
wanted
to
redo
the
one
percent
CO2
experiment,
with
a
with
a
more
accurate
ice
sheet
for
a
better
Wonder
wheat
one-to-one
comparison.
E
E
Let
me
just
give
you
one
slide
on
the
sort
of
broad
response
of
the
one
percent
CO2
configuration
in
the
Arctic
grid
here,
I'm,
comparing
it
to
this
much
War
paper,
which
is
the
one
percent
CO2
experiment
that
has
the
tax
in
the
green
and
ice
sheet,
but
we
can
still
look
at
Global
metrics
like
this
AMAC.
You
can
see
it's
diving
down
immediately
in
both
the
Arctic.
E
And
I
guess
I
should
say
this
is
what
the
one
person
this
is.
What
the
person
looks
like
this
is
one
percent
CO2
until
it
hits
4X
CO2
fixed.
E
One
first
result
is
we
looked
at
the
transient
climate
response
and
the
solid
line
here
is
the
Arctic
where
the
dotted
line
is
the
Workhorse
and
it
systematically
says
to
be
lowered
by
a
few
tenths
of
a
couple
tenths
of
the
Kelvin,
and
so
it
doesn't
like
there
is
a
reduction
in
the
declining
sensitivity,
at
least
in
the
short-term
response
of
which
you
know,
I
could
imagine
100
ounce
novels,
and
maybe
this.
E
Lastly,
I
just
want
to
go
over
the
Greenland
ice
sheet
response
in
the
BG
Arctic,
because
we
did
carry
out
the
Arctic
configuration
in
this
one
percent,
CO2
and
well.
The
agreement
I
sheet
is
is
retreating
rapidly.
The
ablation
area
is
about
10
in
the
pi
control
and
it
goes
up
to
50
percent
at
the
end
of
280
years
of
the
stimulation.
E
This
is
the
surface
mass
balance,
the
ablation
Zone
and
the
pi
control
is
just
this
little
red
zone
and
then,
by
year,
280
or
so
50
of
the
ice
sheet
is
and
manipulations
out.
So
this
thing's
a
donor.
This
is,
and
it's
just
not
a
sustainable
refuel
of
inflation
areas,
the
accumulation
ratio,
it
also
accelerates,
because
the
more
the
inflation
zones
expand.
You
more
bear
as
you
expose,
which
is
darker
than
snow,
and
so
you
see
these
changes
in
Albedo
that
are
accelerating.
E
You
can
also
see-
and
this
is
what
the
ice
sheet
thickness
looks-
there's
there's
places
where
you're
losing
a
kilometer
and
just
a
vise,
and
then
lastly,
we
can
look
at
Outlook
later
detail,
actually
because
it's
a
four
kilometer
I
shoot
up,
and
so
here's
the
example
of
writer,
Glacier
Northwestern,
the
surface
mass
balance
at
different
succession
years
or
more
negative
and
negative,
and
so
it's
thinning
and
thinning
the
ice
sheet,
things
out
with
Glacier,
less
tragic
stress
to
drive,
Glacier
flow,
and
so
the
velocities
of
the
flow
velocity,
slow
down
same
thing
happening
with
health
regulation
and
Southeast
Greenland.
E
But
it's
not
retreating.
Just
an
idea
of
the
detail
can
get,
and
so
that's
that's
basically
it
you
know
like
there's
lots
to
talk
about,
but
it
took
a
lot
of
work.
To
put
this
together.
Do
is
this
something
that
people
want
to
be
able
to
reproduce
my
results,
which
is
this
a
configuration
that
maybe
we
should
think
about
supporting
in
CSM
going
forward?
E
It's
not
very
forward-looking,
because
I'm
using
pop
two
and
we're
moving
to
mod
six
and
so
later
this
year,
I
may
be
doing
a
similar
thing.
A
couple
in
this
dual
polar
grid
to
mount
six.
D
So
Adam:
do
you
think
you're
I
think
your
coupled
tunings
for
any
120
VR
would
be
useful
because
you
did
it
at
low
resolution?
Would
that
be
useful
even
for
doing
like
aim?
Simulations
for
VR
runs
with
others
most
of
the
planet
at
Boca,
small
ones,
for
anyone
20
times
that
so
you're
asking
whether
it's
worth
tuning
in
a
map.
Amf
configuration
well
now
I'm
asking
if
you're
a
couple
tunings
might
work
in
amip
or
if
it's
not
worth
tuning.
E
So
the
clouds
don't
change
quite
as
much
as
in
the
B
concept,
but
the
general
rule
of
thumb
is
you
increase
the
piece
you
reduce
the
physics
time
step,
your
clouds
go
away,
and
so
an
aim
if
you
do
see
a
reduction
in
clouds
and
so
the
tuning
would
be
easier.
E
H
He's
curious
about
with
that.
You
know
that
ocean
response
and
the
to
the
increased
winds
if
you
had
more
unscreened
variability,
so
that
must
be
a
general
problem
that
bridge
works
is
going
to
encounter
anyone
who's
gone
to
high
resolution
that
you
or
anyone
know
like
to
start
with
the
strategy
is
to
sort
of
deal
with
that.
As
you
move.
E
To
high
resolution
to
well
it's
a
we're
realizing
it
as
it
moves
down
the
pipeline
that
this
might
be
an
issue.
Is
it
an
issue?
Maybe
we
like
the
yoga
training,
the
oceans
regulation,
buyer
with
the
20
degree,
atmosphere
forcing
I,
don't
know,
I
had
a
best
friend,
but
just
in
general,
the
variable
resolution
experiments
that
are
projects
that
I'm
involved
with
just
starting
to
find
sensitivity
of
synoptic
scale,
storms
to
resolution
going
from
one
degree
to
quarter
degree.
E
C
World
I
was
going
to
say
a
quick
I
think
comment
on
that
is
answers
yeah.
Actually,
what
we
generally
find
is
that
synoptic
from
a
synoptic
perspective.
D
A
lot
of
changes
are
in
the
structure
of
the
storms,
the
gradients
and
the
invariability,
which
I
think
is
one
thing
you
could
be
seeing.
C
D
E
Boundary
and
until
I
see
something
I
think
it's
fine,
but
maybe
we
should
be
comparing
store
trap
locations
in
the
one
degree.
D
E
I
B
K
So
high
resolution
modeling
at
convection
route,
resolving
scales
has
merged.
E
K
So
analysis
to
the
gray
Zone
case,
study
approach,
We
Run,
The,
coincident,
CSM
simulations
on
a
hierarchy
of
horizontal
and
vertical
resolutions
and.
E
We
compare
the
results
with
contemporaneous
output
from
core,
so
from
work.
We
have
three
one-year
simulations
that
were
run
with
a
four
kilometer
resolution
with
a
domain
over
South
America
and
using
your
fine
breach
online.
With
that
work
domain,
a
hierarchy
of
reasonable
grades
were
created
from
100
down
to
a
six
kilometer
horizontal
resolution,
with
each
of
those
having
a
global
base
resolution
of
100
kilometers
and
for
the
horizontal
grid.
There
are
two
variable.
E
Are
aligned
with
the
warmth
levels
so
for
consent
for
consistency
with
the
warp
model
runs
the
boundary
conditions
outside
and
along
the
perimeter
are
opposed
by
strong
nudging
to
era5.
E
So
we
focus
our
attention
here
on
the
the
environmental
side.
Note
as
it
represents
a
dominant
countries
from
the
overall
variability.
The
errors
and
the
animal
cycle
provide
a
measure
of
how
well
fundamental
processes
are
being
represented,
and
this
mode
exists
at
the
intersection
between
the
chaotic
Behavior,
the
weather
regime
and
the
longer
large-scale
variability.
E
E
E
Going
to
show
are
going
to
be
at
the
six
kilometer
25
kilometer
and
50
kilometer
resolutions,
as
the
other
behavior
of
the
electric
resolutions
can
be
occurred
from
these
for
the
vertical.
K
Resolution
I'm
just
going
to
show
the
32
level
results,
the
70
level
results
and.
E
Basically,
the
only
difference
is
that
the
biases
in
in
70
layer
runs
are
larger.
K
So
this
this
figure
here
goes
along
this
top
panel
on
top.
K
E
E
K
All
the
scaled,
the
more
precipitation
that
concurrency
assembly
at
the
six.
E
Kilometers
scale
for
work,
the
precipitation
has
isolated
in
the.
E
K
Along
the
ampy's
bridge
at
all
resolutions,
and
it's
just
accompanied
by
a
strong,
persistent
precipitation
ordered
over
the
Andes
Bridge.
K
So
if
you
look
at
the
the
the
three
profiles
of
water
vapor
across
the
cross-section.
K
Their
CSM
profiles
compared
to
work,
there's
mountains,
water
vapor
over.
E
At
six
kilometers,
the
CSM
results
better,
maintaining
sharp
gradient
but
they're
still
leakage
that
does
occur.
That's
not
stronger.
K
K
And
then,
to
finish
it
up,
I
want
to
get
a
chronic
state
of
assessment
of
how
modifications
improve
predictive
scale
and
then
once
the
modification
demonstrates
an
improvement
in
scale,
dependent
bias
and
potential
skill
and.
E
E
E
E
Increases
these
are
the
fundamental
differences
that
are
generally
uniformed
across
all
scales.
So
here
we're
going
to
focus
on
just
from
here
on
how
just
the
orographic,
precipitation
and.
K
From
an
analysis
of
the
results
of
discrepancy
in
the
governing
equations
abilities,.
E
Represent
tangential
motions
along
the
surface
boundary
from
that
you
know,
develop
a
a
corrective
term
and
presumably
come
up
with
a
justification
for
a
current
good
prioritization,
but
there
were
also
diagnostic
tests
that
showed
and
suggested
that
the
problem
might
originate
in
the
spectral
element.
Dicor
so,
and
so
we
really
want
to
avoid
is
a
situation
where
we
develop.
E
The
government
equations
to
the
hospital
scrapments
in
this
started
with
replacing
the
hydrostatic
balance
with
a
hyperdynamic
balance
constraint
here,
which
we
place
the
assumption.
K
K
Dry
equations,
so
in
addition
to
that
there
were
some
moisture
reduction
terms
added
to
the
thermodynamic
equation.
E
The
gradient
of
pressure
had
substantial
numerical
errors.
Something
had
to
be
done
about
that
and
for
damping
hyper
viscosities,
perhaps
not
appropriate,
for
low
order,
truncated
column
limit
functions
rather
than
acting
to.
K
So
basically,
the
the
current
implementation
of
the
hyperviscosity.
E
Dampening
violates
every
single
one
of
the
conservation
principles.
E
With
an
alternate
spectral
band
being,
which
allows
for
Selective
removal
of
the
grid
scale
variability
and,
more
importantly,
to
ensure
that
the
form
of
the
applied
damping
does
not
violate
the
conservation
principles
that
the
government
patients
are
based
on
and
one
last
mathematical.
Inconsistency
which
is
related
to
the
wet
dry
problem
is
that
if
the
relations
in
two
blue
boxes
are
true,
then
the
relation
in
the
red
box
is
also
true,
which
means
that
anywhere
that
the
hybrid
vertical
coordinate
is
used
on
dry
surface
pressure
is
a
source
of
error
in
the
model.
K
E
K
E
One
through
three
or
I've
been
implemented
installed,
but
the
hyper
viscosity
adapting
has
only
been
replaced
in
the
energy
and
mass
conservation
creations.
The.
E
K
So
for
this
test,
I
kind
of
found
the
middle
middle
middle
value
that
balances
these.
E
Are
in
the
middle
column
and
on
the
left
are
the
corresponding
dwarf
upscale
morph
simulation
results
and
on
the
right
are
the
original
reference
simulations.
E
So
we
see
here
that
for
temperature,
the
bias
and
the
Amazon
basin
and
the
line
the
Andes
is
reduced
and
for
water
vapor
we
see
the
reduced
amplitude
and
the
water
vapor
surges,
and
particularly
away
from
the
river.
E
For
precipitation,
we
see
that
the
dry
Halo
is
mostly
eliminated
and
the
precipitation
here
is
more
localized
into
stronger
migrating
cells
and
there's
a
corresponding
decrease
in
the
precipitation
over
the
Andy
three
age.
Now.
E
Maintain
the
sharp
gradient
water
vapor
along
model
levels,
okay,.
K
Yes,
and
for
the
last
result,
if
you
look
at
the
monthly
immune
precipitation.
E
With
the
alternate
spectral
damping
for
Tracer
equation,
we
see
in
the.
K
E
But
I
need
to
finish
the
implementation
of
the
and
testing
of
the
Tracer
damping.
Then
move
on
to
replacing
the.
K
Damping
of
momentum
equation
and
then
once
then
it'll
be
to
the
point
where
they
don't
want
to
run
through
a
the
sequence
of
three
hierarchies
to
evaluate
its
impact
on
the
on
the
scale,
dependent
bias
and
based
on
these
results.
It
seems
like
we.
E
Should
get
some
some
good,
measurable
Improvement
and
then
we
want
to
run
that
through
the
predictive,
the
S2s
framework,
to
get
a
measurement
of
how
the
predictive
scale
has
been
affected
by
the
changes
and
we'll
complete
that
and
repeat
the
process.
Next
time
you
know
before
you've
seen
about
the.
K
Two
leader
wallpaper
models,
so
in
closing
I
just
think
with
a
question
for
mwg,
and
that
is:
should
we
establish
a
standard
set
of
regional
hierarchy,
hierarchical
grids.
K
E
K
B
B
B
J
J
Okay
thanks
so
everyone,
my
name
is
Xi'an
Jiang
from
UCLA
JPL,
Joint
Institute.
So
this
work
is
a
close
collaboration
between
our
UCLA
JPL
team,
including
myself,
along
Duo
and
uncut
team,
the
Patrick
Adam,
Julio
and
leech.
So
this
study,
mainly
that
will
analyze
the
regional
refund
at
the
cam
6
simulations
that
the
Patrick
just
introduced
so
wait,
but
we
mainly
focus
on
the
downside
of
the
Eastern
Amazon
region,
and
this
study
is
supported
by
the
NSF
CPT
project.
J
So
as
we
know
that
Amazon
River
Basin
plays
a
crucial
role
for
the
global
hydrological
and
carbon
Cycles.
Therefore,
the
accurate
representation
of
key
climbed
processes
of
this
living
will
be
crucial
for
regional
and
even
Global
Climate
projections.
So
in
this
study
we
mainly
concern
model
representation
of
precipitation
over
over
this
Legion.
J
J
Mode
we're
seeing
Lily
it's
a.
J
J
B
Don't
know
yeah
the
third
side
now.
J
So
it's
better,
like
I,
will
try
to
make
it
larger,
yeah,
yeah,
so
yeah.
So
so
this
slides
shows
the
the
our
current
model.
That
shows
the
serious
buyers
in
simulating
the
synonym
precipitation
over
the
Amazon
region,
so
here
with
many
focus
on
the
body
of
spring
season,
from
March
to
May
and
may
end
so
that
this
panel
shows
the
Mam's
in
the
mean
the
mean
precipitation
the
based
on
the
GPM
observation.
J
So
in
this
study
we
mainly
focus
on
the
eastern
part
of
the
Amazon
region,
so
that
near
the
coastal
engaging
so
the
market
by
this
green
box.
So
actually
we
can
see
that
this
observed
the
rainfall
maximum.
During
the
mam
of
this,
the
Eastern
Amazon
region
is
largely
significantly
underestimating
Cloud
Model
simulations,
including
the
cam
6
so
shown
here.
J
This
is
the
mam
synonym
precipitation
simulated
by
the
one
degree
chem6
so,
which
shows
only
a
patent
correlation
less
than
0.7,
and
we
can
also
see
the
dry
bias
over
this
East
Amazon
region
with
the
Southwest
shifter
for
lean
belt
and
the
more
serious
model
bias
is
found
in
this
club
and
map
version
of
Cam
6
so
which
is
under
development.
Actually,
the
with
the
S2
just
introduce
the
new
tomorrow's
talked
about
you
all
that
this
model
actually
can
generally
capture
a
well
capture
the
the
climatology
of
global
precipitation
in
the
cloud.
J
But
in
this
case
we
can
see
the
Amazon
East
Amazon.
The
mam
rainfall
is
almost
completely
missed
in
this
model
and
this
drive
virus
in
itself.
The
reading
is
also
a
very
typical
the
problem
in
synthesis
model.
You
can
see
this
example
means
based
on
more
than
20
models
and
also
the
individual
model.
J
Then,
in
order
to
understand
the
model,
biasing
simulating
the
signal
mean
precipitation
over
Eastern
Amazon.
This
slide
shows
the
climatological
the
dyno
evolution
of
precipitation
just
based
on
the
GPM
observation
and
the
two
chemistics
model
simulations.
So
we
can
see
that
in
the
based
in
the
observations,
the
synonym
precipitation
of
this
green
box
region
over
Eastern
Amazon
is
largely
characterized
by
the
the
significant
the
annual
evolution
of
lean
belt,
which
first
appears
over
the
coastal
region
in
the
early
afternoon,
then
propagate
Westward
into
the
Inland
throughout
the
nighttime.
J
However,
you
can
see
in
the
cam
6
models
in
that
this
case
is
one
degree
simulation
you
can
see.
This
Dino
evolution
of
ring
belt
is
the
largely
the
list,
in
the
one
degree,
poorly
simulated.
In
one
degree,
the
cam
6
simulation
and
almost
completely
gone
missing.
This
club
MF
version
of
Cam
6.,
so
this
result
suggested
that
the
model
virus
in
simulating
the
seasonal
mean
mam
mean
the
the
presentation
of
this
is
the
Amazon
region
could
be
closely
associated
with
the
model
by
seeing
down
the
cycle.
J
So
this
is
the
main
motivation
of
this
study
we
want
so
in
the
following.
We
mainly
focus
on
two
points,
so
the
first
question
we
want
to
address
is:
what's
the
key
process
is
regulating
down
the
cycle
lymph
of
the
Easter
Amazon
region,
and
then
we
want
to
explore
the
the
can
representation
of
this
down
cycle
over
the
Eastern
Amazon
be
improved
in
high
resolution,
cam
simulations
as
the
Practical
introduced
at
the
regionally
fund,
this
simulations.
J
So
for
the
first
question,
actually
we
mainly
focus
on
based
on
the
observation.
So
the
specific
question
we
want
to
introduce
the
address
key
is
what
controls
the
westward
application
of
the
animal
looking
for
events
over
the
Easter
Amazon
region
that
which
you
can
you
can
see
the
very
clearly
by
this.
The
homologator.
So
here
is
the
this.
Is
the
time
repeated
for
during
two
days
this
longitude
that
this
coastal
region
so
the
for?
J
For
this
analysis,
we
mainly
borrow
this
precipitation
buoyancy
framework,
which
is
this
recently
proposed
by
the
fires
Ahmed
and
that
they
were
needing
at
all
paper.
So
the
main
idea
is
that,
based
on
their
close
relationship,
then
the
lower
troposphere
buoyancy
can
be
used
as
a
proxy
for
the
collection.
J
So
this
equation
is
for
the
lower
drop
sphere.
The
simplified
version
of
the
low
top
sphere
buoyancy,
which
contains
two
terms
one,
is
that
the
corresponding
to
undiluted
Cape
another
is
the
moisture
deficit.
Then
the
lower
panel
shows
the
homolog
to
these
two
terms,
so
we
can
see
the,
although
cable
can
may
explain
some
the
convection
onset,
the
near
the
coastal
Coastal
leading
the
of
the
this,
the
dyno
convection,
but
we
can
see
the
systematic,
what's
good
propagation
of
the
presentation
is
mainly
dominated
by
the
this
westward
migration
of
lower
troposphere,
the
moisture
deficit.
J
So
since
this
Master
University
involved
with
the
difference,
the
Sita
Sita
e
and
the
saturated
Theta
e,
then
the
further
analysis
shows
that
this
monster
deficit
is
largely
dominant
by
the
westward
propagation
of
the
lower
troposphere
moisture.
And
meanwhile
it
can
be.
This
effect
can
be
Amplified
by
a
cooling
Factor
during
that
time,
when
the
the
convection
propagates
into
the
Inland
due
to
reduce
the
saturated
Theta
e.
So
the
next
question
is
we
want
to
answer
what
leads
to
the
this
last
word:
application
of
the
the
lower
chops
were
moisture.
J
Then
we
analyze
the
this
is
most
budget
at
the
11
pm
at
this
time.
The
connection
is
located
over
this
longitude,
so
this
is
a
vertical
versus
the
longitude
around
the
equator
see.
This
is
directly
output
from
the
EC
model,
so
we
can
see
the
corresponding
to
the
the
strong
rest
of
the
propagation
of
the
moisture.
J
So
we
can
see
the
the
Western
into
the
west
and
joining
to
the
east
of
the
convection
that
in
the
lower
and
the
middle
troposphere,
then
if
we
average
the
use
the
worst
tendency
vertically
in
the
lower
drop
sphere,
we
can
see.
This
is
the
total
muscular
tendency.
It's
largely
due
to
a
two
canceling
terms.
One
is
the
most
you
think
due
to
connection
and
the
moisture
Source
by
the
Dynamics.
J
Then,
if
we
decompose
this
Dynamics
into
the
vertical
Direction
and
the
horizontal
action,
we
found
that
what
you're
the
vectoring
almost
completely
canceled
by
the
moisture,
the
reduction
by
correction
and
this
total
muscular
tendency
at
the
the,
namely
the
moistening
to
the
west
and
the
joining
to
the
east-
is
mainly
contributed
by
the
horizontal
direction
of
moisture.
So
as
we
know
that
during
the
the
mam
season,
the
Eastern
region
is
characterized
by
a
easily
the
wind,
which
is
the
the
so-called
the
Amazon
low-level
jet.
J
So
this
results
suggested
that
the
westward
application
low-level
moisture
is
largely
contributed
by
the
horizontal
moisture
Direction
by
this
easterly,
the
low
level
jet.
So
this
will
be.
This
funding
will
be
used
to
examine
the
model
results.
So
next
I
will
move
to
the
model
simulation
of
the
in
the
regionally
fund,
the
chemistic
simulations.
So
we
analyzed
the
six
five
simulations
with
the
horizontal
resolution,
ranging
from
one
degree
to
about
the
607
degree
Patrick
just
introduced.
J
So
this
slides
shows
that
also
the
climatological
the
down
cycle
done
evolution
of
lean
for
the
based
on
the
three-year
simulation
during
the
mam
season.
So
we
can
see
that
again
the
the
the
observe
that
this
evolution
of
a
particularly
establishment
of
this
the
link
belt
over
the
coastal
region
in
the
early
afternoon
is
largely
missed
in
the
lower
solution
simulations
and
thus
started
to
emerge.
J
We
can
see
started
to
emerge
the
in
the
higher
resolution,
for
example,
that
this
is
the
quarter
degree,
and
we
can
see
that
this
propagation
of
lean
belt
can
be
reasonably
well
resolved,
particularly
in
the
highest
resolution,
the
six
kilometer
so
which
is
more
clear
Ed
by
this
next.
This
is
how
communal
diagram
again,
we
can
see
the
westward
propagation
of
this
diagonal
convection.
The
in
the
observations
is
largely
missing
in
the
completely
missed
in
the
in
the
lower
solution
simulations
and
started
to
pick
up
in
the
higher
resolution,
particularly
in
the
six
kilometer
simulation.
J
We
can
see
the
reasonably
well
resolve
this
Westward
propagating
the
the
connection,
the
the
this
with
onset
near
the
coastal
region.
This
coastal
line
the
dash
line.
Then
we
want
to
understand
how
the
increased
resolution
can
improve
this
stimulation
of
Donald.
What's
for
the
propagation,
so
this
shows
the
again
that
this
is
total
interpretation
in
the
five
simulations.
Then
we
decompose
the
total
into
the
convective
and
the
large
scale
connection
we
can
see.
The
Improvement
of
this
Wastewater
location
is
mainly
through
the
the
in
the
improve
increased,
enhance
this
large-scale
precipitation.
J
Well,
Collective
presentation
tends
to
be
decreased
with
increased
horizontal
solution,
so
this
results
largely
consistent
with
previous
studies.
Regarding
this
highly
solution.
Modeling,
then,
the
question
is
based
on
our
previous
start,
the
observational,
the
results,
the
westward
population
of
convection,
can
be
largely
the
contributed
by
the
westward
population
of
low-level
moisture.
So
this
motivated
us
to
further
check
the
diagonal
evolution
of
the
low-level
moisture
in
this
five
stimulations.
Actually,
it's
a
little
surprisingly.
We
can
see
that
the
regardless
they
are
horizontal
resolution.
J
All
these
five
stimulations
can
capture
the
this
Westward
propagation
of
low
level,
moisture
reasonable
reason
reasonably
well,
particularly,
for
example,
this
half
degrees
relation.
The
this
shows
the
smooth,
the
westward
population
of
the
low
level
moisture,
which
is
comparable
to
the
to
the
very
the
highest
simulation
that
the
solution,
the
six
kilometer
simulation,
so
this
result
suggested
that
the
difference
in
the
convective
behavior
in
this
different
simulations
could
be
due
to
their
different
convective
response
to
lower
troposphere,
moisture,
forcing
or
buoyancy
forcing
so.
J
This
is
supported
by
that
this
last
slide,
I
have
which
shows
the
relationship
between
the
precipitation
versus
the
lower
chops
where
lower
charge
sphere
buoyancy.
So
this
is
the
total
precipitation
and
buoyancy
the
code.
Color
means
the
represents
the
lower
resolution
and
the
warm
the
higher
the
solution.
So
we
can
see
the
waste
increase
of
the
horizontal
resolution,
corresponding
to
the
same
the
same
value
of
the
lower
troposphere
buoyancy.
We
can
see
the
presentation
significantly
increased
in
high
resolution
simulations,
so
this
enhanced
sensitivity
of
the
precipitation
in
responding
to
the
lower
chop
sphere.
J
Buoyancy
is
mainly
through
the
large-scale
precipitation,
as
well
previously
mentioned,
so
this
results
the
again
the
the
suggest,
the
need
of
the
skill
awareness
when
parliamentalization
accumulus
so
I
think
this
is
the
main
results.
I
have
so
very
brief.
Summary
we
found
the
dyno
cycle.
A
B
B
D
E
I'm
going
to
continue
on
the
theme
of
South
America
I'm,
using
the
variable
grid
that
Patrick
developed
and
talked
about
to
try
to
learn
something
like
orographic
flow
around
the
endings,
the
usual
backgrounds
anyway.
E
This
is
what
I'll
be
talking
about:
I'm
going
to
show
mudging,
Tendencies
and
then
machine
learning
model
to
try
to
learn
about
the
Tendencies
and
then
some
questions
Etc,
so
orographic
drag
may
not
need
any
introduction,
but
you
know
just
this
is
a
nice
kind
of
cartoon
that
shows
all
the
different
sorts
of
things
that
happen
around
topography.
You
know
mountains
are
barriers,
so
the
air
has
to
figure
out
what
to
do
go
around
go
over
Etc
or
just
get
blocked
anyway.
E
So
these
are
the
things
that
high
resolution
might
do
better
than
low
resolution,
and
the
idea
here
is
to
use
use
Patrick's
grid
here
at
the
highest
resolution
that
Patrick
ran
seven
kilometers
or
six
kilometers
somewhere
in
there
I'm
using
the
32
level
configuration
I.
As
you
mentioned,
these
are
very
preliminary
results.
I'm.
You
know
I'm
not
expecting
this
to
turn
into
a
parameterization
time
soon,
but
anyway.
E
So
what
I'm
going
to
look
at
are
Imagine
Tendencies
when
we
coarsen
results
from
the
seven
kilometer
32
level,
South
America
great,
with
no
convection,
not
that
it
really
matters
for
what
I'm
doing
so.
This
is
this
is
what
I'm
using-
and
you
know
you
all-
might
Express
skepticism
enough,
hydrostatic,
non-hyperstatic
I,
don't
think
for
orographic
flows,
especially
at
low
levels.
E
First
I'm
aware
of
you
know
the
fact:
that's
being
used
by
Chris
brotherton's
machine
learning,
modeling
group,
so
basically,
nudging
is
bigger
relaxation
to
a
desired
solution
with
some
time
scale
in
merch
to
find
you
take
a
fine
model,
a
high
resolution
model,
and
course
in
some
way,
and
then
you
nudge
the
course
model
to
that
for
the
coarsened
fine
model.
E
So,
as
I
said,
you
know,
Chris
Brown's
group
is
looking
at
machine
learning
combined
with
much
defined,
and
one
issue
is
that
much
defined
around
topography
can
be
tricky.
So
the
first
issue
we
have
to
contend
with
is
how
to
how
to
you
know
how
to
portion
things
when
they're
big
mountains
in
the
middle
of
your
domain.
E
So
this
tries
to
illustrate
the
problem.
So
over
here
is
the
seven
kilometer
topography
from
the
Sam
Wharf
grid.
Here
it's
been
conservatively
remapped
to
the
ne30
or
100
kilometer
grid,
and
here
is
what
we
actually
use
in
the
nd30
model
which,
since
all
die
boards
use
some
additional
smoothing
of
topography.
E
Just
kind
of
conservatively
coarse,
graining
topography
to
your
model
grade
is
not
really
actually
used
as
a
bottom
boundary
and
that's
part
of
the
problem.
This
bottom
here
is
the
difference
between
the
top
row
and
this
actual
topography
used
in
cam.
So
the
issue
is
that
everywhere
you
know,
you
see
red,
it
means
there
are.
You
know
big
Peaks
sticking
out
of
your
bottom
topography
and
that's
what
the
nudging
has
to
condemn
with
so
I'm
advocating
or
at
least
what
I'm
trying.
What
I
tried
in
this
study
is
the
simplest
possible
solution.
E
Well,
there
actually
is
a
simplified
there.
What
do
you
do
in
these
situations?
So
what
I'm
Ellis?
What
I'm
trying
to
illustrate
here
is
the
problem
that
is,
you
know
your
your.
Your
smooth,
smooth
topography
actually
is
below
the
just
coarse
grain
topography,
and
you
have
to
figure
out
what
to
do
in
these
points
where
the
coarse
grain
Topography
is
actually
above
your
smooth
topography.
So
in
the
you
know,
I
think
the
standard
nudging
application.
E
Is
you
actually
don't
do
anything
you
just
use
whatever
the
course
grading
gives
you
on
on
on
this
smooth
grid
I've
tried
a
different
option,
which
is
just
to
zero
out
the
Subterranean
sort
of
the
the
sub
training
wins
on
the
coarser
grid,
so
I'm
calling
those
the
not
Mo
option
in
the
block
option.
E
So
these
just
a
summary
of
the
steps
to
go
from
the
high
resolution
variable
variable
mesh
to
Global
nudging
on
100,
kilometer
and
E30
SE
grid.
So
let
me
just
probably
skip
this.
E
There's
the
the
basic
run
here
is
to
turn
off
the
turn
off
the
gravity,
with
scheme
totally
in
the
in
the
ne30
and
nudge,
then
to
the
nudge
to
the
Court
re-gridded,
fine,
so
I'm,
turning
off
the
gravity
wave
scheme
in
in
the
model
and
nudging
to
the
fine
model,
and
that's
supposed
to
tell
us
about
orographic
flow.
E
All
right,
so
these
These
are
the
mean
nudging
Tendencies.
When
you
turn
off
when
you
turn
off
the
orographic
drag
in
any
30.
and
the
the
three
rows
paparo
are
the
nudging
Tendencies
when
you
zero.
The
winds
in
these
Subterranean
proportion
points.
E
The
null
is
the
the
standard,
I
think
approach
in
matching
studies,
and
this
is
the
difference,
and
you
know
it's
not
not
surprising,
the
the
block
that
the
block
option
shows
that
there
is.
There
are
stronger,
Stronger
tendencies
in
nudging
when
you
know
when
you're,
when,
when
you
zero
the
wind
in
the
in
in
the
in
the
topographic
sort
of
in
the
projects.
E
Above
your
your
model
grid-
and
you
know
this
projection
of
topography
above
your
lowest
model
surface-
is
an
issue
even
when
you're
dealing
with
re-analysis,
because
re-analysis
Topography
is
not
the
same
as
the
topography
in
in
the
model.
E
So,
in
a
there
is
a
difference,
depending
on
on
how
you,
how
you,
how
you
prepare
data
for
nudging
in
this
sort
of
nudge,
Define,
much
defined
approach,
so
you
you
do
have
to
you
know:
I'm,
not
I'm,
not
claiming
at
this
point
that
one
approach
is
better
or
not,
but
you
do
have
to
realize.
You
have
to
accept
that.
E
There's
a
there
is
an
impact
on
how
you
nudge,
on
how
you
prepare
the
nudging
data
over
over
a
complicated
Terrain,
all
right
so
talking
with
Colin
a
few
months
back.
He
encouraged.
You
know
encouraged
me
to
try
scikit,
which
is
a
fairly
simple,
to
use
machine
learning
package.
So
this
just
shows
I'm
trying
to
illustrate
what
I'm
doing
so.
E
The
first
thing
is,
you
have
to
prepare
some
input,
data
from
the
machine
machine
learning
algorithm
and
this
this
this
red
box
here
shows
the
structure
of
of
the
input
data
that
I'm
putting
in
so
I'm
only
looking
at
the
11
lowest
model
levels
UV
the
winds
I
also,
you
know
because
it
is
a
part
of
all
gravity-way
schemes
I'm
putting
in
the
stratification
I
want
to
put
in
moisture.
Eventually,
but
I
haven't
done
it
yet
and
then
also
optionally.
E
There
are
rich
parameters
in
each
of
the
grid
boxes,
so
this
input
data
is
a
list
of
columns.
Basically
with
u
11
11
values
of
the
meteorological
profiles
plus
Ridge
parameters,
then
the
target
are
these
nudging
tendencies
again
I'm
using
11
the
lowest
11
values
from
the
run
in
which
there's
no
orographic
scheme
in
in
the
course
model,
and
then
this
is
all
thrown
into
a
random
Forest
regression.
E
So
this
is
the
basic
result
that
I
want
to
show,
because
this
side,
over
here
on
your
left,
is
a
scatter
diagram
of
the
predictive
Tendencies
from
the
machine
learning,
either
nudging
Tendencies
or
both-
u
and
v,
at
all
levels
versus
the
test
Tendencies.
So
this
is
what
the
model
actually
sort
of
this
is.
These
are
the
actual
nudging
Tendencies
from
the
model
on
this
side
is
trading
with
meteological
profiles,
plus
rigid
information
and
on
this
side
is
just
using
the
meteorological
profile.
So
you
know
there
is
there's.
E
Clearly,
just
meteorological
profiles
does
a
lot,
but
the
rigid
formation,
the
subgrid,
let's
just
call
it
subgrid
topography
information-
does
introduce
useful.
It
appears
to
introduce
things
that
the
machine
learning
can
use
to
make
a
better
prediction.
E
So
another
view
of
that
is
this:
where
this
this
just
shows
sort
of
doesn't
come
out
very
well,
there's
a
very
faint
cyan
line
there.
E
E
These
are
just
this
plot,
but
at
each
level
and
I'm
looking
for
the
the
slope
of
the
scatter
plot,
which
tells
you
you
know
how
much
bias
or
sort
of
you
know,
does
the
machine
learning
exaggerate
or
not
the
the
prediction
and
the
correlation
you
know:
does
it
fit
a
straight
line,
so
the
correlations
are
high,
pretty
pretty
much
everywhere.
E
You
know,
there's
this
I
think
meaningful
structure
that
you
know
the
machine
learning
model
is
doing
a
better
job,
predicting
things
at
lower
altitudes,
but
you
know,
there's
still,
you
know
they're
still,
at
least
in
the
correlation
some
skill
as
you
go
higher,
the
slope
is
also
best
predicted.
You
know
so
the
one-to-one
relationship.
E
Never
if
you
want
one
everywhere,
you
never
get
close
to
one
really
for
the
slope
like
if
you
do
get
closer
at
low
levels
and
also
there
is,
you
know
again,
obviously,
from
the
first
from
the
last
plot.
There's
been
some
information
imparted
by
the
ridge
information,
so
I'm
going
to
skip
this
I
think
there
is
I'll
just
say
briefly:
there
is
an
issue.
What
what?
What
do?
What
do
you
really
you
know
is:
are
the
learning
Tendencies
really
model
error
and
I'm?
E
Not
so
sure
that
that's
true,
but
for
now
we'll
pretend
to
you
know
well,
we'll
just
leave
it
at
that
for
now
so
anyway,
I'll
just
stop
here.
This
is
a
summary
I
think
you
know
details
and
how
you
pour
it
for
me
matter
in
much
defined.
I.
Think
I
didn't
talk
about
it,
but
I
think
we
need
to.
We
need
to
worry
about
what
model
error
really
is
and
how
to
how
to
figure
it
out.
E
I
do
think
information
for
sub
grip
topography.
If
you
wanted
to
replace
autographic
drag
with
machine
learning,
you
would
want
to
include
something
about
subgroup,
topography
and
then
you
know
first
in
the
future,
there
are
all
the
die
core
issues
that
Patrick
mentioned.
You
know
you
might
want
to
try
a
different
dive
board.
We
should
investigate
the
structure
of
the
machine
learning
model.
You
know
what
exactly
is
leading
to
this
drop
in
skill
with
vertical.
You
know
with
the
vertical.
I
E
You
know,
spread
the
training
around
a
little
bit
and
make
you
know
neighboring
grid
cells,
because
I
think
you
know
this
all
assumes
we
can
do
things
in
columns
which
may
not
be
true
and
finally,
I
think
you
know
we
need
to
understand
exactly
the
relationship
between
nudging
and
modulator.