►
From YouTube: Ocean Model Working Group - 2022 CESM Workshop Day 2
Description
The 27th Annual CESM Workshop will be a virtual event. Specifically, the Workshop will begin with a full-day schedule on 13 June 2022 with presentations on the state of the CESM; by the award recipients; and two presentations from our invited speakers in the morning, followed by order 15-minute highlight and progress presentations from each of the CESM Working Groups (WG) in the afternoon.
- Welcome and logistics: Frank Bryan
- The Andes affect ENSO statistics: Weixuan (Rosa) Xu
- High-order multirate explicit time-stepping schemes for the baroclinic-barotropic split dynamics in primitive equations: Lili Ju
- Regional MOM6/CESM configuration of the Caribbean Sea: Giovanni Seijo
- Status of fully-coupled simulations with MOM6: Gustavo Marques
- Eddy CPT outcomes and plans: Scott Bachman
- Intro to discussion of mesoscale and submesoscale parameterization in CESM-MOM6: Ian Grooms
- Discussion
- Applying implicit energetics ideas to simulate full-water-column ocean mixing: Robert Hallberg
- Intro to discussion of diapycnal mixing parameterization in CESM-MOM6: Ali Mashayek
- Discussion
- Open Discussion of OMWG Plans and Priorities: Frank Bryan / Ian Grooms
To learn more:
https://www.cesm.ucar.edu/events/workshops/2022/
A
A
A
A
A
A
D
A
D
A
E
F
A
E
A
D
C
D
D
D
The
new
encar
machine
will
be
arriving
supposedly
in
january
now,
they've
been
having
trouble.
I
guess
procuring
enough
chips
to
build
it,
so
we'll
see
what
happens,
and
that
will
be
a
system
with
a
substantial
fraction
of
the
nodes
being
gpu
based
so
exactly
how
that
gets
allocated,
I'm
not
totally
sure
yet,
but
based
on
our
prior
allocations,
I'm
expected
that
our
guidance
will
be
for
requesting
somewhere
in
the
range
of
30
to
40
million
core
hours,
and
we
typically
split
that
about
50
50
between
what's
called
production,
which
is
typically
science
problems.
D
And
then
I
thought
we
could
also
touch
on
whether
there
might
be
some
opportunities,
some
additional
opportunities
for
collaboration
across
the
mom
six
community.
As
you
know,
as
we're,
you
know,
at
least
from
our
perspective
as
we're
pulling
together
a
new
system,
whether
there
might
be
some
opportunities
for
joint
experimentation
in
a
controlled,
multi-model
kind
of
framework
for
gftl,
emc
and
car,
doing
some
kind
of
experimentation
together
to
understand
particular
aspects
of
sensitivities
or
motion
processes
together.
So
those.
A
D
A
E
So
I'm
curious,
I
know
that
there's
a
lot
of
work
going
on,
especially
with
the
secret
version
of
cse.
Are
there
any
particular
difficulties
with
the
ice
ocean
coupling?
I
I'm
not
seeing
any
disgust
on
the
agenda,
but-
and
it
may
not
even
be
something
that
people
fully
understand
yet,
but
is
it?
Is
that
going
smoothly,
or
is
that
something
where
there
are
interesting
challenges
that
would
be
worth
discussing
in
the
few
minutes?
We
have
here.
D
Deal
for
us,
we
are,
you
know
there
were
there's
a
few
little
bays
here
and
there
that
things
have
gotten
clogged
up,
because
the
ice
model
has
no.
You
know
zero
velocities
at
a
channel
that
we
thought
was
open
and
isn't
really
so
so
we
need
to
do
a
few
little
tweaks
to
our
landmass,
but
other
than
that
kind
of
thing.
We're
not
I'm
not
seeing
anything
that
alarms
me
dave.
Do
you
have
any
from
your
preliminary
looks?
Are
you
seeing
anything.
H
Well
again,
I
I
think
most
of
our
focus
has
been
on
the
salt,
fresh
water
and
heat.
We
have
not
had
to
worry
about
dynamics
and
momentum
to
this
point,
but
I
think
that
that
is
going
to
be
an
issue
coming
in
the
next
year
for
sure
so,
but
and
and
I
know
this
is
your
favorite
conversation
bob,
but
we
we
are
still
levitating
this
the
sea
ice.
E
H
H
A
D
A
D
A
D
D
A
D
D
D
D
D
We
will
I've
got
it
on
auto
here.
I
have
a
few
science
talks
this
morning.
First,
in
the
first
part
of
the
meeting,
then
we'll
go
into
some
of
the
recent
work
here
at
ncar
with
mom
6
development,
both
regional
models
and
global
models,
and
then
we're
going
to
try
something
a
little
bit
different
for
our
working
group.
We're
sort
of
solicited
talks
on
process
representation
in
mom
6,
first
on
mesoscale,
some
as
the
scale
processes
and
scott
bachmann
and
ian
grooms
will.
D
Sort
of
lead
that
part
of
the
discussion
and
we'll
leave
some
time
to
address
those
kind
of
issues,
and
you
know
how
we
can
advance
that
science
through
csm
and
how
that
how
we
should
be
pursuing
our
development
path
in
those
directions
and
then,
similarly
for
diaphysnal
mixing
processes
and
bob
hallberg
and
ali
marshayek
have
agreed
to
read
that
part
of
the
discussion
and
then
we've.
We
have
some
time
towards
the
end
there
for
some
open
discussion
about
the
working
group
plans
and
priorities
over
the
next
year
or
so
so.
D
That's
the
plan
for
today.
Just
the
logistics
here
usual
zoom
etiquette.
Please
stay
muted
and
turn
your
camera
off
while
others
are
presenting.
You
have
a
question
during
the
talk.
Please
put
it
in
the
chat
and
we'll
address
it
and
then,
after
the
talk
is
completed,
feel
free
to
turn
your
camera
back
on
during
the
discussion
and
we'll
try
to
use
the
raise
hand,
function
to
take
turns
and
take
questions
in
turn.
D
F
I
Okay,
I'll
get
started
good
morning.
Everyone
welcome
to
the
first
talk
and
for
this
talk,
I'm
going
to
share
with
you
my
research
about
how
the
andes
affect
the
ends
of
statistics,
and
I
am
a
graduate
student
at
brown
university
and
I
finished
this
project
with
my
advisors,
professor
german
lee
and
professor
bela
fox
kemper,
and
we
collaborated
with
dr
yan
plankton
and
dr
michael
mcfadden
from
noaa
pmel.
I
So,
firstly,
just
to
remind
you
what
is
enso,
it
is
the
inter-annual
variations
happening
in
the
eastern
coast
of
the
tropical
pacific
during
el
nino.
There
is
abnormal
warm
in
the
eastern
coast
and
then
during
la
nina.
There
is
abnormal
cold
here,
but
the
impact
from
enso
does
not
restrict
in
this
local
area.
I
I
I
I
C
C
A
B
B
So
today
I'm
going
to
talk
about
some
numerical
methods
for
the
ocean
models
and,
of
course
I
thank
the
organizer
for
giving
me
the
chance
to
present
some
of
my
recent
works.
So
the
title
here
is
here:
we
call
the
high
order
market
rate
explicit
time
stating
scheme
for
the
biokinetic
biotropic
split
dynamics
in
the
primitive
equation.
So
this
is
a
joint
work
with
my
post
doc
right
and
my
colleague
chu
wong
and
the
professor
max
genderberg
from
florida
state
and
the
fear
jones
from
los
angeles.
B
B
What
is
the
dynamic
splitting
and
the
spatial
degradations
then
we're
talking
about
the
high
order,
split,
split
time,
stepping
scheme
based
on
the
framework
of
strong
stability,
preserving
the
quota
method
and
what
we
can
do
to
use
margin
rate
in
the
slp
rk
framework
to
achieve
high
order
the?
Finally,
we
present
some
numerical
experiments
to
demonstrate
the
performance
of
our
schemes,
so
this
is
a
primitive
equation.
Okay,
usually
it's
the
incompressed
bosnic
approximation
under
the
hydrostatic
balance.
Okay,
so
you
have
the
sickness
equation.
That's
from
the
volume
conservation
you
have
momentum
equation.
B
Also
you
have
some
tracer
equations,
for
example
the
potential
temperature
and
the
salinity.
Also
you
have
the
hydrostatic
condition
the
equation
of
state,
so
these
are
some
notations,
so
we
present
here
here,
the?
U
is
mean
the
horizontal
velocity.
W
is
the
word
velocity
h,
it's
a
flow
thickness.
In
fact,
you
have
to
understand
in
a
special
way,
okay
and
then
to
consider
the
stratification
effect.
So
we
partition
the
fluid
into
air
layers.
B
Then
some
variables
you
have
to
put
in
each
layer,
for
example,
the
horizontal
velocity
we
a
uk,
the
layer,
thickness
hk
and
the
trace
variables
5k
you
have
put
in
the
middle
of
each
layer,
but
the
word
velocity
put
in
the
top
and
the
interfacial
height
you
defined
from
the
bottom,
the
capital
h,
the
total
column
height
at
rest.
Then
you
add
back
each
so
the
top
seed
surface
level
will
be
the
zero
okay.
So
that's
the
basic
idea:
what
is
the
marked
layer
shallow
water
equation
here?
B
Is
you
just
apply
some
central
different
skin
along
the
z
directions?
So
here
I
highlight
in
red:
that's
all
the
you
discretize
the
odd
operators
in
z,
direction.
Okay
in
horizontal
direction
still
is
a
continuous
okay.
So
you
get
the
marked
layer,
that's
defined
for
each
layer.
L
layers
and
k
equals
one
to
l.
Okay,
then,
what
is
the
dynamic
splitting?
So
the
bark
ninja
biotopic,
in
fact,
because
in
each
layer
your
horizontal
velocity
called
the
uk
okay,
then
the
faster
mode
is
a
two-dimensional
biotropic
mode.
B
It's
with
a
flow
thickness
weighted
average
of
the
velocity
and,
of
course,
the
park
mini.
It's
just
e
for
each
layer.
You
subtract
that
from
the
original
velocity
uk,
so
you
can
regard
it
as
some
kind
of
mathematician,
regardless
kind
of
a
probation
okay.
So
here
so
for
this,
you
know
for
the
first
mode
is
only
one
layer
or
just
2d,
okay
for
the
slow
mode.
It's
a
3d
because
you
have
k
layers
along
the
z
directions
for
the
kinetic
mode.
B
The
important
thing
is
to
remember:
it
wait
use
the
flow
thickness
to
weight
it
then
the
weighted
average
is
zero.
So
then
you
get
the
speed
system.
You
can
solve
the
topic.
Brackeninic
mode,
the
equation:
12a,
okay,
here
look
at
this
g
bar
okay
and
biotropic
mode.
12B,
here
the?
U
bar
right,
you
you
bar
is
the
topic
velocity,
but
here
you
have
to
introduce
the
data
data
is
the
first
system:
height,
perturbation,
okay,
also
the
flow
thickness
equation.
The
three
systems
here
was
a
g
bar.
B
B
B
Digitalizations
you
get
this
dual
mh,
then
you
do
the
finite
warning
degradation.
In
fact,
the
twisk
is
a
secret
stuttering.
Okay,
you
do
the
discrete
divergence
gradient
and
the
curve
operator
by
final
orientations.
Of
course
also
there's
some
reconstruction
operator.
You
need
it
okay
to
get
this
tree
scheme.
So
that's
not
our
focus.
Our
focus
is
after
subjective
gradation.
You
get
a
specially
discrete
systems.
Now
that's
called
a
system
of
odes,
okay.
So
now
the
time
how
do
the
time
integration
so
suppose
we
have
a
partial
v,
partial
d
equals
f.
We
fuse
the
flux.
B
Okay,
then
remember
this
famous
responsibility,
predominantly
it's
a
convex
combination
of
forward
oilers.
Why
we
use
this
one
hybrid
explanator,
so
forward
oiler
is
from
first
one
from
tn
to
the
tm
plus
one.
What
is
sap
rk2?
The
two
stage
is
two
forward
euler.
Then
do
average
that
mean
from
tn
to
tm,
plus
one
ten
plus
one
to
kn,
plus
two
then
put
the
average
result
at
tn
and
tm
plus
two?
So
also,
you
have
sap
h3.
It's
a
convex
combination
of
three
forward.
Euler
time
statements.
B
Okay,
so
then
you
get
here
is
you
saw
the
parking
mode?
You
use
the
ford,
euler,
okay,
suppose
from
t0
to
t1
the
t
there
is
current
time.
Then
you
can.
Its
basic
idea
is
what
how
to
compute
the
g
bar
cheaper,
remember,
make
a
weighted
average
parallel
kinetic
mode
to
be
zero.
So
I
first
forward
the
braking
system.
Without
the
g
bar
term,
then
I
compute
what
kind
of
g
bar
to
make
the
weighted
average
of
the
predictive
predicted
braking
mode.
B
That's
from
the
idea
of
hitting
2005.
So
here
the
g
bar
just
comes
as
a
first
order.
Approximation
of
the
true
forcing
at
current
time
states
then,
because
this
forward
euler,
so
I
can
do
a
combination,
a
comma
command
for
lr,
based
on
the
ssp
frame
cprq
framework
to
get
high
order
schemes
here
also,
is
a
word
mixing
one
mention
here:
what
mixing
is
the
the
rate
term
in
fact,
is
vertical
diffusion.
Okay
people
say
okay,
sometimes,
for
numerical
stability
want
to
separate
this
one
okay,
so
you
saw
the
other
term
explicitly.
B
Then
do
the
vertical
mixing?
What
is
missing,
because
it's
only
in
vertical
direction,
so
they
are
in
the
total
parallel
for
all
the
vertical
directions.
So
that
means
you
can
use
the
triagonal
solver
to
get
a
quick
solution.
The
image
scheme,
in
fact,
is
a
imax
skin.
In
fact,
also
it's
just
first
order
imax
schemes,
so
the
the
trouble
is
sometimes
it
does
not
fit
into
the
ssp
arcade
framework.
B
Okay,
so
then,
okay,
that
means
we
use
data
t
to
forward
the
slow
mode,
the
clinical
mode
for
the
faster
mode.
At
each
stage.
You
have
to
use
the
substating
here.
If
the
delta
t
is
large
time
step
size,
then
directly
over
m
m
is
the
substating.
Okay,
so
use
small
term
time
step
side
for
each
step,
and
then
we
can
get
this
sap
rk
2
a
splitting
speed
of
multi-rate
schemes.
The
idea
is
what
okay,
firstly
do:
the
pre-processing,
you
do
the
decomposition?
Okay,
you
have
velocity
at
each
time.
B
You
know
current
velocity,
you
do
the
decomposition
get
the?
U
bar
n
and
the?
U
tilde
and
uber
is
the
first
mode.
Utility
is
the
slow
mode,
but
mine
is
2d
one
3d.
Okay,
then
what
do
you
do
face?
The
you
use
rt
to
advance
the
splitting
system.
So
basically
you
see
the
13a.
You
advance
the
clinical
mode
before
the
euler.
Okay,
the
same
time
you
get
a
g
bar
g
bar
and
zero.
Okay,
that's
the
forcing
a
first
order,
plasmid
forces.
B
It
gets
predicted
by
clinical
velocity
at
the
tn
plus
one,
but
to
get
how
the
unit
go
on
right,
but
to
go
in
the
bar
clinical
mode.
You
have
to
use
assistance
from
the
flow
thickness
and
the
biotopic
model,
so
you
have
to
do
certain
b,
certain
c
certain
d
and
e.
What
I
mean
you
compute
the
paratropic
mode
you
substituting
okay,
then
combine
together
data
predicted
velocity
at
tm
plus
one.
Then
you
can
update
the
flow
thickness
and
the
system
higher
perturbation.
The
first
stage
you're
done.
B
Okay,
then
you
do
the
second
stage
of
parking
it,
because
now
the
parking
indica
you
can
do
the
forward
euler
from
tm,
plus
one
to
tm
plus
two.
So
you
get
you
get
a
prediction
at
a
tm
plus
two.
Also,
you
get
a
g
bar
and
one
that's
another
fourth
thing
for
the
second
step.
Okay,
so
this
is
then
the
broccoli
is
done
because
you
get
a
prediction
at
ten
plus
one
based
on
the
sap
arcade
framework,
you
can
do
average,
get
the
two
update
at
tm
plus
one.
B
So
you
get
14b,
then
what
do
we
do
for
biotropic
mode
and
the
flow
signals
so
prototypic
mode?
You
say:
can
I
go
ahead
update
it?
Then
you
don't
get
high
order,
because
each
step
is
for
the
biotropic
model.
It's
substituting.
It
does
not
fit
to
the
ssprk
framework
to
get
high
order.
You
have
to
use
the
14a
substating
from
tn
to
tm
plus
one,
but
what
is
the
forcing
the
forcing
now
it
becomes
the
average
of
g
bar
and
zero
and
g
by
n
one.
Otherwise,
you
lost
the
high
order
after
this
one
down.
B
Then
you
recover
the
update.
I
get
the
updated
velocity
at
the
tm
plus
1.
Then
you
update
the
h
the
flow
thickness
and
the
c
surface
height
a
perturbation
state.
That's
the
idea,
similar
things,
so
you
you
can
do
the
sap
arcade
three
scheme.
You
can
do
the
update
here
right
for
each
one,
so
similar
idea,
you
get
each
first
pre-processing
with
current
velocity
to
the
splitting
to
update
the
parking
mode.
B
Okay
stage,
one
station
two
stage:
three:
okay,
you
use
substating
for
the
biotopic
velocity,
but
the
large
time
state
directive
for
a
bar
clinic,
but
in
order
to
get
high
order
by
clinic
you
can
compute
at
each
stage
you
get
a
prediction:
g
bar
and
zero
g
bar
and
one
and
the
g
bar
and
one
half
okay,
so
other
the
topic
only
for
systems.
No,
it's
not
to
update.
So
after
that
is
done.
Then
you
use
the
true
forcing
combination
the
interpolation.
Here
I
write
down
the
red
part.
That's
each
stage
you
get!
B
What
is
the
true
forces?
Then
you
update,
then
you
start
to
correct
the
biotopic
mode.
Otherwise
you
lose
the
accuracy.
Then
you
update
the
flow
thickness
and
the
the
system
hydration.
So
we
think
okay!
Is
that
enough?
The
idea
I
get
spr
k2
x3,
so
I
hope,
because
we
saw
the
first
mode
always
use
the
small
time
step.
The
slow
mode
is
a
large
time
state,
but
unfortunately,
remember
here,
the
system
had
absorbation
can
be
determined
from
what
from
the
paratropic
subsystem
soldered
by
probably
substitute.
B
Also,
you
can
get
from
the
update
of
flow
thickness,
so
in
the
pda
levels
is
the
same,
but
in
the
this
time
fully
discrete
level
it's
over
determined
for
our
for
this
algorithm
right
now,
it's
always
determined
from
update
of
the
flow
thickness,
but
you
are
substituting
for
the
biotopic
solver.
Are
they
consistent?
If
they
are
not
consistent,
then
your
motor
saw
are
still
tangled
together,
you
don't
have
the
stability,
so
you
want
to
keep
the
accuracy.
You
want
the
stability.
You
have
to
carefully
be
careful
here.
B
The
idea
is
here:
we
we
thought
one
year
is
reconciled
the
system
with
hyper
perturbation.
So
the
idea
is
for
sap
rk2.
How
do
you
can
reconcile
the
system?
Perturbation
is
okay.
There
are
two
substations.
There
are
two
updates
for
the
flow
thickness.
You
just
check
for
each
one.
What
is
the
flux?
Okay,
the
flask
for
the
back
biotopic
substating,
the
flux
for
flow
signaling,
update?
Okay,
it
takes
difference
on
the
difference
of
their
flux.
B
Then
you
based
on
the
weight
at
the
full
thickness
of
each
layer,
you
design
so-called
transport
velocity
adjustment
and
to
make
up
the
difference.
So
that's
the
idea
here
then:
okay,
no
matter
what
you
use
the
full
thickness
layer
thickness
or
use
the
biotopic
saw
you
get
same
results
for
the
sisterhead
preparation
data,
okay,
same
thing
for
the
third
order,
skin,
so
idea
here
you
can.
Of
course
there
are
three
stages
you
have
to
compare.
Then
you
update
our
original
algorithm.
Then
we,
our
numerator,
shows
this
then
become
very
stable.
B
Okay,
so
we
use
the
academic
channel
test
case.
Here
is
a
large
eight,
a
d
simulation
okay,
so
we
use
20
vertical
layers
and
the
horizontal
mesh
is
10
kilometer
resolution
activity
image
and
for
this
example,
the
work
mixing
is
no
work.
Missing,
has
no
effect
in
this
test.
In
fact,
we
numerically
also
verified,
and
we
use
because
there's
no
exact
solution
to
verify
the
convergence
order.
So
we
spent
a
very
small
time
step
as
a
benchmark
solution.
We
find
this
system
header.
B
Reconciliation
is
very
useful
for
stability
and
accuracy
when
the
sub-saving
is
large.
Also
we
find,
based
on
this
framework,
sap
arcs3
se
still
does
not
recover
the
fourth
third
accuracy.
Okay,
what's
the
motivation
for
our
work
is
here
because
we
use
the
current
split-speed
approach
in
the
m-pass
ocean?
We
do
the
benchmark
test
for
time
accuracy.
B
We
find
in
fact
it's
at
most
first
order:
accuracy,
okay,
it's
stable,
but
it
does
not
follow
some
mathematic
principle
for
designing
high
order
scheme,
in
fact,
does
not
give
you
high
order
accuracy
in
time.
Okay
and
for
our
this
work,
you
can
see
here.
No
vertical
mixing
is
applied
to
this
balcony
channel
test
case
the
left
hand,
side,
no
ssh
system
head
reconciliation
and
the
right
one
is
scissor
head
probation.
You
can
see
with
the
civil
reconciliation
is
where
it's
much
simpler
than
the
other
one.
B
Also
we
find
for
long
time
simulation,
it's
become
more
stable.
Okay,
this
one
is
snp
rk2
and
you
can.
We
can
get
perfect
second
order
schemes.
Of
course,
one
m
is
very
large
because
the
inconsistence
at
the
error
term,
for
the
first
mode
flow
mode
is,
is
highly
imbalanced,
so
it'll
affect
the
overall
time
accuracy
a
little
bit
okay,
but
it's
not
that
big.
You
can
see
you
get
a
perfect
convergence.
Okay,
just
get
high
order,
accuracy
for
sap.
I
guess
three.
You
look
at
here
the
left
hand
side
without
ss
ssh
reconciliation.
B
We
only
get
a
second
order
and
with
resolution
reconciliation
we
get
a
still
not
a
little
bit
better
than
a
second
order,
but
still
not
the
third
order.
In
fact,
we,
if
you
go
back
to
the
numerical
scheme,
you
see
there's
still
some
violation
of
the
ss
ssp
rk
framework.
That's
why
the
truncation
arrow
in
our
paper,
you
can
see
our
paper,
the
truncation
error
analysis,
so
it
cannot
give
you
certain
order,
but
at
least
second
order.
So
why
what's
the
benefit?
I
said
I
guess
three.
B
It
has
better
stability
and
the
better
accuracy.
Okay,
so
we
did
some
simulations,
that's
the
we
use
the
m
equals
16,
that's
meaning
the
time
step
for
the
topic
is
15
seconds
for
the
brightening
of
20
240
seconds,
okay
and
we
compared
so.
We
also
coupled
with
the
tracer
solver
in
the
impulse
ocean.
Now
we
can
see
that
the
how
the
skin
does
play
an
important
role,
the
the
comparisons
of
a
current
and
past
splitting
speed
scheme.
The
absolute
difference
keeps
increasing
along
the
time
with
even
different,
drag
coefficients
for
the
bottom
okay.
B
So
we
want
to
see
you
have
to
the
high
order,
then
also
to
the
global
ocean
test,
and
this
one,
of
course,
is
not
a
laplacian,
it's
a
bioharmonic
operator
for
stabilizations
and
we
observe
almost
the
similar
behaviors
okay
for
this
one.
We
we
know
sh,
essentially
because
it's
important,
so
we
always
apply
it
without
vertical
mixing
and
with
vertical
mixing
so
for
the
global
ocean
seems
to
have
some
difference
and
we
show
second
order
you
can
get,
but
after
applying
the
vertical
mixing
like,
I
analyzed
before.
B
In
fact
it's
the
first
order,
mx
scheme.
It
does
not
fit
the
ssp
framework.
Sometimes,
if
it's
play
is
very
influential
like
here,
we
reduce
to
the
first
order,
if
you
add
vertical
mixing,
maybe
it's
more
stable,
but
the
accuracy
will
drop
okay
for
sap
request
three,
it's
same
things:
okay,
without
vertical
mixing,
it's
high
order,
but
what
can
mixing
a
job
to
first
order?
Okay,
so
I
wanna
say
here
the
word
missing
is
stabilized,
but
it
will
destroy
your
accuracy.
Okay.
B
Also,
we
compare
with
current
twitter
forwards.
We
get
the
the
left.
One
is
the
potential
temperature
evolution
the
right
one.
Is
the
surface
salinity
difference?
Okay,
that
that's
the
difference,
not
the
true
value,
the
difference
with
the
empath
se.
I
wanna
show
the
how
all
the
accuracy
importance.
Also
we
did
the
performance
test,
so
we
use
different
substituting.
We
can
see
use
better
self-dating.
Of
course
you
get
almost
the
linear
speed
up
time
efficiencies.
B
Okay,
of
course,
here
the
3d
biotopic
saw
so
always
dominate
the
computational
cost,
just
in
the
query,
with
16
64
cross.
So
here
is
how
to
some
remaining
work:
how
to
recover
svr
gas
3
scheme
third
order.
Also
for
vertical
mixing.
Can
you
get
high
order
vertical
mixing?
Not
the
current
probably
used
in
ocean
modelling
also
how
to
do
the
try
high
order,
tracer
salt
coupled
to
the
dynamics
also,
can
we
do
because
the
speed
of
the
arrow
is
imbalanced
for
the
topic
of
bracket
mode?
B
C
Thank
you
lily,
so
we've
only
got
about
20
seconds.
Instead
of
having
questions,
I
think
I'm
just
going
to
try
and
get
the
next
talk
set
up.
But
if,
if
there
are
questions
for
lilly,
we
do
have
a
break
at
10
15
and
we
can
maybe
discuss
then
so
when
we
come
back
to
rosa,
do
you
think
you're
ready
to
go
yet.
I
Yes,
exactly
so,
just
to
remind
you,
my
research
is
about
the
end
so
and
then
I'll
jump
to
the
introduction
of
the
challenges
that
we
are
facing
in
simulating
the
tropical
pacific
climate,
and
this
figure
shows
the
csm
model's
bias,
impressive,
surface
temperature,
and
I
think
this
is
for
csm1,
so
the
error
it
is
having
that
is
outstanding
from
this
figure
is
that
the
precipitation
is
overestimated,
especially
in
the
central
south
pacific
region
and
for
sea
surface
temperature.
It
is
simulating
an
excessive
warm
pool
in
this
eastern
coast
region.
I
Aside
from
the
biases
in
the
background
climate,
there
are
also
biases
in
eso
variations
across
a
lot
of
climate
models.
So
the
errors
include
in
include
in
the
aspect
of
amplitude
and
frequency.
The
left
figure
shows
you
different
model
simulation
for
enzo
amplitude,
and
this
column
is
the
observation
data.
You
can
see
that
there
are
many
models
that
disagree
on
the
amplitude
of
enzo
variations,
and
the
right
figure
is
the
frequency
spectrum
of
the
answer
variation.
I
Therefore,
we
think
that
the
era
in
a
tropical
pacific
climate
simulation
might
be
originated
from
the
era
in
the
representation
of
the
andes.
So
this
figure
shows
you,
the
south
american
topography.
The
left
figure
is
the
high
resolution
satellite
data
and
the
right
figure
is
the
csm
model's
representation
of
the
topography.
With
the
resolution
of
about
2
degrees,
you
can
see
that
the-
and
this
is
a
very
tall
and
thin
mountain
range,
with
its
highest
elevation
of
above
five
kilometers,
but
with
the
smoothing
in
the
low
resolution
global
climate
models.
I
I
Therefore,
we
did
an
experiment
of
elevating
the
height
of
the
andes.
We
change
the
mountain
elevation
along
the
grid
cells
for
the
mountain
for
the
andes,
mountain
range
and
and
change
it
to
the
maximum
height.
According
to
the
high
resolution
satellite
data,
and
then
we
call
this
experiment
the
higher
end,
this
experiment
and
we
run
the
global
fully
coupled
model
together
with
this
default
elevation,
to
see
how
the
climate
simulations
are
changing.
We
use
the
csm
1.2.2,
fully
coupled
model
with
the
horizontal
resolution
of
2.5
degrees
times
1.9
degrees.
I
We
use
the
component
set
of
b2000,
which
means
the
co2
forcing
is
as
in
2000,
so
that
we
can
compare
our
model
result
with
the
observation
data,
because
this
is
a
fully
coupled
model.
We'll
expect
the
model
take
a
relatively
long
time
to
catch
up
with
a
relative
equilibrium
with
the
upper
ocean,
so
we
run
the
model
for
350
years
and
we
use
the
last
160
years
of
data
for
our
analysis.
I
Firstly,
I'm
going
to
show
you
the
result
for
the
background
climate.
This
is
the
sea
surface,
temperature
distribution
for
married
diurnal
and
zonal
directions.
You
can
see
that
for
in
the
control
simulation,
it
is
simulating
a
higher
sea
surface
temperature
across
the
whole
tropical
pacific,
and
when
we
increase
the
height
of
the
andis,
we
not
only
lower
the
sea
surface
temperature
across
the
pacific,
but
we
also
increase
the
north-south
sea
surface
temperature
gradient,
which
is
more
consistent
with
the
observation
and
for
the
zonal
distribution.
It
is
similar
the
higher
end.
I
I
Secondly,
we
also
look
at
the
surface
wind
stress
curl,
which
is
consistent
with
the
which
can
be
influenced
by
the
temperature
and
can
influence
temperature
vice
versa.
So
this
is
the
zono
wind
stress
curl,
in
its
zonal
distribution.
One
thing
I
want
to
highlight
here
is
that
in
the
east
eastern
tropical
pacific,
the
wind
is
actually
getting
weaker,
because
this
is
the
zona
wind.
So
it's
the
negative
number
means
it
blows
towards
the
west
and
when
the
absolute
value
is
larger,
it
means
that
the
wind
is
stronger.
I
I
I
One
thing
you
can
see
here
is
that
for
the
higher
andex
experiment,
it
is
showing
a
smaller
vertical
temperature
gradient
and
it
is
more
consistent
with
the
observation.
It
means
that
when
we
change
the
height
of
the
andes,
the
upper
ocean
is
getting
less
stratified
here.
So
these
three
changes
are
what
we
observe
in
the
background
climate
and
since
we've
changed
some
properties.
In
the
background
climate,
we
are
expecting
some
changes
in
enzo
variations
as
well,
and
here's
what
I'm
going
to
show
you.
I
Firstly,
the
most
impressive
result
that
we
get
is
that
when
we
increase
the
height
of
the
andes,
the
amplitude
of
enzo
variation
is
getting
much
weaker
than
the
control
simulation.
The
left
figure
is
the
zonal
distribution
of
the
sea.
Surface
temperature
anomaly
standard
deviation.
It
can
represent
the
strength
of
the
enzyme
variations
here.
I
I
Second
aspect
that
we
are
looking
into
is
the
frequency
spectrum
you
can
see
here.
There
is
a
very
strong
peak
in
the
control
simulation
with
around
five
year
cycle,
which
means
that
in
the
control
run
the
end.
The
enso
is
happening
very
regularly
with
around
five
year
cycle,
which
is
not
true
in
the
observation
data,
because
it
does
not
see
a
very
strong
peak
in
this
frequency
and
when
we
increase
of
the
height
the
height
of
the
end,
is
the
frequency
spectrum
gets
more
similar
to
the
observation
again.
I
I
So
one
question
you
may
ask
is
why
does
these
changes
happen?
What
process
has
the
higher
end?
This
experiment
changed
to
quantify
the
process.
We
are
following
the
research
done
by
john
fedorov
in
2020,
we're
using
the
beard,
nice
stability
index
to
quantify
the
feedback,
strength
for
the
ano,
the
enzo
cycle,
so
the
bj
index
can
it
can
be
separated
into
five
terms:
the
thermal
damping
term,
which
quantifies
the
surface
latent
heat,
sensible
heat,
short
wave
and
long
wave
radiations.
I
The
second
is
the
mean
advection
damping
it.
It
is
consistent.
It
consisted
of
the
zonal
advection,
the
meridian
over
action
and
a
vertical
advection,
and
these
three
are
the
main
affections
of
damping
and
the
last
three
terms
are
the
positive
feedbacks.
They
are
the
zonal,
advection
feedback,
eggman
feedback
and
thermocline
feedbacks.
They
are
quantifying
the
anomalies,
sense
sensitivity
to
our
changes.
A
I
But
because
their
research
is
for
a
regional
climate
model.
They
cannot
simulate
the
complete
enzo
cycles,
though
so
they
did
not
look
into
this
aspect,
and
I
think
the
high
resolution
definitely
improve
the
simulations
there.
But
but
one
question
that
you
may
ask
is
how
high
the
resolution
is
enough
for
correctly
capturing
the
effective
effect
of
the
andes,
yeah.
A
G
I
I
One
similar
conclusion
is
that
they
both
increase
the
accuracy
of
the
simulation,
but
the
difference
between
the
slab
motion
and
fully
coupled
model
here
that
we
captured
is
that
the
range
that
the
slab
motion
model
actually
changes
is
larger
than
what
the
fully
coupled
model
actually
changes.
So
we
think
the
dynamic
ocean
circulation
is
slightly
offsetting.
Some
effects
from
the
andes.
A
A
I
A
I
still
have
a
quick
question.
Thanks
for
the
nice
talk,
so
I
was
wondering
if
you
looked
into
any
of
the
more
energetic
related
dynamics
or
diagnostics.
Sorry,
so,
for
example,
when
you
raise
the
andes,
you're
also
going
to
be
impacting
subtropical
clouds
and
where
the
itcz
is,
and
that
could
also
have
a
another
way
of
analyzing
what
could
be
going
on
how
enso
could
be
changing
through
cloud
diagnostics?
I
was
wondering
if
you've
taken
that
angle
yet
or
not,.
I
Yes,
actually,
this
is
the
second
step
of
experiment
that
we
are
doing
in
a
fully
coupled
model
and
before
this
one
we
run
a
similar
simulation
in
the
slab
ocean
model,
which
we
focus
more
on
the
change
in
the
atmosphere,
circulation
and
cloud
feedback.
And
what
we've
seen
is
that,
because
the
sea
surface
temperature
is
decreasing
in
the
southeast,
pacific,
cold
town
and
this
decrease
of
temperature
is
having
different
reasons
from
off
coast
and
like
near
coast
region,
so
near
the
coastal
region
of
south
america.
I
We
see
a
lot
more
low-level
clouds,
so
it's
blocking
more
solar
radiation.
So
this
is
the
reason
the
sea
surface,
temperature
is
decreasing
in
the
coastal
region
and
for
the
off-sea
region,
because
we
modulated
the
atmosphere
circulation.
So
the
southeasterly
wind
is
getting
stronger
and
induces
more
evaporation,
so
the
central
pacific
region
is
actually
cooled
by
evaporative
cooling
yeah.
C
All
right,
thank
you
again,
rosa
we're
gonna
go
on
next
to,
I
think
giovanni
yep
giovanni,
and
we
are
a
couple
minutes
off.
I've
allowed
us
to
get
a
couple
minutes
off
because
of
the
technical
difficulties,
but
I'm
going
to
give
everyone's
going
to
get
there
20
minutes
so
giovanni.
If
you
want
to
share
screen
yep.
G
G
G
Almost
ready
to
forge
him
from
j.r.a
ronoff
is
from
the
world.
Ocean
atlas
died
from
topics.
Poseidon
claudio
field
is
from
seaweeds
and
we're
using
65
c
star
vertical
levels,
and
this
configuration
serves
as
a
way
to
first
learn
how
to
configure
m6
and
system
regionally
properly,
make
sure
that
it's
working
correctly
and
for
us
to
be
able
to
establish
a
good
framework
and
workflow
for
this
configuration
in
order
to
move
forward
with
a
an
actual
working
configuration
that
we
can
use
in
the
future.
G
And
so
we
actually
just
finished
running
a
15-year
presentation
just
simulation
just
last
week,
and
this
is
what
you're
seeing
here.
It's
color
control
the
surface
salinity,
the
black
line.
Contours
our
sea
surface
height
we've
ran
for
15
years
from
2000
to
2000
at
the
end
of
2014
and
at
least
qualitatively,
when
we
haven't
seen
anything
we're
going
on,
we
can
see
the
amazon
river
plume
and
that
kind
of
highlights
the
retroflexion
very
nicely.
G
We
can
see
the
goal
stream
and
some
eddy
shedding,
and
so
again,
as
I
mentioned,
this
run
was
just
finished
last
friday,
so
we
haven't
done
any
much
of
in
the
quantitative
way,
but
at
least
qualitatively
it
gives
us
a
good
idea
of
what
the
model's
doing-
and
we
don't
see
anything
odd
going
on.
So
it
reassures
that
our
efforts
are
going
in
the
right
direction.
G
One
thing
that
I
noticed
that's
very
good
that
we're
capturing
because
again
we're
interested
in
salinity
is
that
minimum
in
2009
right
here
that
the
model
captures
which
is
associated
with
the
2008
to
2009
la
nina,
which
resulted
in
large
negative
salinity
anomalies
in
the
whole
caribbean.
So
the
model
captures
that
very
nicely,
which
is
promising
and
temperature
seems
to
be
doing
fairly
well
much
better
than
salinity.
I
would
say
at
least
again
qualitatively
speaking,.
G
Moving
forward,
we
have
a
few
other
benchmarks
just
for
peace
of
mind,
and
this
is
just
to
mention
a
few
comparing
the
model
to
a
full
graded
output
like
a
glorified
analysis
or
some
other
product
that
may
provide
better
information
than
the
argo
greater
product,
because
we
know
the
caribbean
there's
not
that
many
argo
floats.
So
the
observations
are
a
little
bit
limited
on
that
side.
G
So
we
know
that
we
can
run
month,
6
and
csm
in
a
regional
configuration.
So
what
we
want
to
do
moving
forward.
We
want
to
create
a
1
12
of
a
degree
grid
same
extent
as
the
quarter
of
degree,
and
then
we're
going
to
change
our
initial
conditions
or
forcing
a
runoff
to
the
csm
large
ensemble,
particularly
the
more
members
which
have
very
detailed
outputs
fields.
G
These
are
10
members
of
the
large
ensemble
and
we're
going
to
be
using
that
to
run
the
model
and
project
into
the
future,
so
that
we
can
assess
any
salinity,
viability,
changes
in
the
future
for
the
caribbean
sea
and
then
yeah.
I
I
need
to
highlight
james
thinking,
sand
and
liquid
cursor
at
rutgers.
They've
been
super
helpful,
with
helping
with
the
whole
workflow
and
setting
up
the
model
and,
of
course,
the
isp
right
visitors
program,
and,
as
I
mentioned,
this
was
on
the
much
shorter
side
of
the
dogs
today.
E
G
Yeah,
so
it's
the
open
boundaries
are
daily
temperature
and
salinity
and
sea
surface
side
from
that
we're
getting
from
glories.
It's
just
the
north
and
new
segments
of
the
map
and
then
gustavo
may
chime
in
into
the
second
part
of
the
question,
because
he
was
the
one
that
finished.
This
configuration
run
on
gustavus.
F
E
Yeah
thanks
for
the
talk,
I
want
to
encourage
you
to
do
what
you
said.
You'll
do
and
so
far
as
documenting
this
configuration,
we
were
just
having
a
chat
yesterday
at
lunch
in
princeton
and
commenting
that
to
open
up
mom6
to
general
usage.
It
really
would
be
valuable
to
have
more
very
detailed
configuration
documentation
how
you
actually
built
the
model
and
set
it
up
and
run
it
and
all
the
data
sets
and
everything.
E
So
I
know
that's
a
lot
of
work,
but
it
sounds
like
that's
something
that
you
want
to
do
and
just
make
it
happen
would
be
really
great.
I'm
going
to
put
into
the
chat
something
that
you
can
maybe
aspire
toward
it's
a
project
that
the
nemo
community
has
done
for
some
region
off
the
west
coast
of
the
u.s,
and
it's
an
amazing
documentation
of
that
project.
It's
something
that
you
know
could
be
something
to
maybe
emulate
a
bit
anyway.
Good
luck
and-
and
thanks
for
that.
G
Thank
you.
Thank
you
for
the
comment
and
yeah
I
I
think
this
is
really
important,
for
you
know
the
development
of
the
model
and
the
community,
and
you
know
I've
benefited
from
a
few
github
repos
that
are
very
detailed,
whether
from
enriquez
group
gfdl,
so
kind
of
like
paying
it
forward,
and
hopefully
people
can
use
it
whether
it
is
to
just
replicate
what
we're
doing
or
just
create
their
own
configuration
so
appreciate.
The
comment.
F
C
C
Short
notice,
so
I'm
I
have
a
question
actually
about
that.
You
may
have
touched
on
it
at
the
end
of
your
at
the
end
of
your
slides,
but
if
you
go
to
very
large
m,
so
you
have
lots
of
fast
steps
before
baroclinic
step
and
then
very
high
resolution
in
a
global
model
say
then
you
end
up
doing
a
lot
of
communication
and-
and
this
can
potentially
really
slow
down
your
throughput.
Do
you
did
you
experience
that
or
do
you
have
any
comments
about
that.
B
B
I
just
say
yes
you're
right
about
that,
since
you
know,
for
that
test,
is
we
fixed
the
fast,
the
time
step
size
for
the
fast
mode?
We
fix
that
then
use
larger
m
doesn't
mean
use
larger
time,
step
for
the
biokinetic
mode,
so
use
larger.
But
for
that
that's
a
global
ocean
test.
For
that
case,
you
have
100
layers
vertically
so
and
you
get
larger
mage
use
fixed
number
of
cores,
so
larger
may
larger
measure
doesn't
mean
each
core
has
more
computation.
B
That's
then,
you
can
see
for
larger
mesh.
The
efficiency
gets
better
and
the
better
you,
of
course,
if
your
mesh
is
too
small,
you
know
each
for
each
card.
Competition
cost
maybe
affect
a
lot
by
the
communications,
but
when
you
increase
scale
we
don't
see
much
also
also
for
that
global
version
test
we
see
the
paratopic
upper
clinic
so
really
dominate
the
overall
cost.
B
E
B
Your
questions
here
so
for
numerical
point
of
view,
if
you
want
to
reach
the
third
order,
you
know
that
that's
the
principle,
that's
more
like
a
a
multi-step
for
the
longer
could
I
just
single
step
with
marking
states?
Okay,
so
that's
the
difference
here,
so
it's
self-start,
you
don't
need
any
like
initially,
you
need
some
other
help
to
start
it.
So
when
we
try
to
say
how,
because
for
this
splitting
dynamics,
the
first
mode
is
only
2d,
so
you
really
want
to
use
smart.
B
You
can
use
large
small
time
steps
but
doesn't
use
much
computational
cost,
but
the
topic
is
3d.
You
really
have
to
use
large
time
steps
now
within
the
framework
of
longer
cuta.
How
you
design,
okay,
one
mode
is
large
time
says
one
more
is
smart.
How
do
you
design
a
scheme
in
this
framework?
You
get
overall
high
order.
That's
the
tricky
question.
B
That's
why
we
test
the
current
market
rate
called
empire
se,
no
more
than
first
order
accurate.
We
use
benchmark
tests
to
verify
that.
That's
why
we
say
the
idea
of
the
question.
Is
they
use
many
kind
of
prediction
correction,
but
when
you
check
the
truncation
arrow,
it's
it's
first
order,
it's
stable,
but
only
give
you
first
order.
Okay,
that's
what
I'm
saying
so.
B
Yeah
we
have
to
do
three
parts.
That's
two!
That's
true,
okay,
but
I
want
to
say
things
is
use
because
the
use
of
different
time
step
how
to
design
a
high
order.
Skin
is
really
a
mathematic
question.
You
know
not
that
trivial
say:
okay,
I
just
you
small,
sometimes
go
ahead.
Do
the
command
together,
you
don't
have
accuracy.
That's
what
I
want
to
say
here.
F
Can
you
see
me
speed
yep,
so
I'm
going
to
talk
about
these
titles
of
fully
couple
simulations
with
csm
on
six,
and
this
work
has
been
done
in
collaboration
with
a
lot
of
people
from
the
ocean
section,
but
also
from
amp,
especially
in
a
ci
state,
bailly,
marika,
holland
and
also
I
want
to
acknowledge
our
soft
engine.
The
main
software
engineer
working
on
this,
which
was
marianna
bertostein,.
F
F
It
increases
the
ocean
mass
and
therefore
it
increases.
It
can
also
decrease.
Actually
the
ocean
heat
content.
So
the
idea
of
adding
or
removing
mass
into
the
ocean
has
implication
for
heat
as
well,
and
if
you
now
look,
if
you
have
enthalpy
fluxes,
your
heat
balance
at
the
surface
will
be
slightly
different.
That
will
use
would
used
to
with
the
old
pop
model
or
models
that
use
a
virtual
assault,
flux
in
which
the
netflux
will
be
the
the
the.
F
F
The
way
we
have
defined
at
this
time,
the
enthalpy
fluxes
are
as
follows
for
for
liquid
in
green,
and
they
all
follow
the
same
idea
of
is
a
temperature
times
a
mass
times,
the
heat
capacity
I'm
using
the
seawater
heat
capacity.
You
know
whether
the
sea,
water
versus
the
fresh
water
is
the
right
one
and
we
we
haven't
decided.
F
We
use
the
sea
water
at
this
time,
but
for
liquid
components
we
compute
the
maximum
between
the
sea,
surface
temperature
and
zero,
and
what
we
are
doing
is
because
the
other
components
are
not
providing
the
temperature
associated
with
the
mass,
whose
enthalpy
we
are
calculating.
We
need
to
guess
that
temperature
and
that's
why
we
apply
these
limiters.
So
in
the
case
of
with
liquid
mass,
we
apply
a
maximum,
so
the
max
access
to
zero
for
frozen.
We
do
a
minimum
between
ssg
and
zero
and
then
vapor
it
comes
at
ssg.
F
F
To
show
how
this
is
adds
to
the
balance
to
the
heat
balance
after
a
100
year
run
within
fully
coupled
mode,
I'm
showing
you
the
net
heat
budget
from
the
mediator.
So
the
mediator
prints
this
table
every
now,
and
then
we
can
ask
how
frequent
that
happens.
So
I'm
showing
that
the
need
the
need
need
ops,
not
hit
budget
after
100
years.
F
I
want
you
to
pay
attention
to
the
new
terms,
which
are
the
enthalpy
terms,
there's
six
of
them
here
on
the
left
and
notice
that
only
the
ocean
column
has
non-zero
values
at
this
time
and
you
can
get
an
idea
of
of
the
magnitude.
This
is
in
watts
per
meter
square,
so
the
two
dominant
terms
are
rain
and
evaporation
followed
by
the
liquid
runoff.
F
So
there
are
the
rain
and
evaporation
are
ordered
to
each
watts
per
meter
square
and
then
the
if
you
sum
all
these
terms,
it's
being
added
into
the
sensible
heat
for
the
atmosphere
model.
So
it
shows
up
here
in
the
sum
of
the
sensible
heat
and
that's
what
we
we
do
as
a
short-term
fixer,
because
the
other
components
are
not
accounting
for
the
end
of
the
fluxes
and
to
accommodate
for
that
right.
Now,
we're
just
using
this
this,
what
we
call
fixer.
F
F
The
magnitude
of
you
know
the
global
mean
of
of
these
terms,
but
I
think
it's
also
structured
to
look
at
a
maps
of
this
different
terms,
so
I'm
now
doing
a
30-year
average,
and
I
want
to
highlight,
for
example,
although
the
liquid
runoff
mean
is,
is
relatively
small,
0.4
approximately
0.4
watts
per
meter
square
locally.
It
can
be
very,
very
large,
it
can
be
up
to
245
watts
per
meter
square
and
that's
in
a
point
right
outside
the
amazon
river,
where
the
mouse
is.
F
F
Now,
moving
to
the
fully
couple
simulations
that
we
have
done
so
we're
using
the
pre-industrial
control,
the
components
are
mm6
cam6,
finite
volume,
size,
5,
cln5,
the
community.
I
should
model
version
2,
which
is
no
evolve,
so
it
doesn't
change
its
the
shape
of
the
ice
sheets
and
then
the
runoff
model
is
mozart.
F
I'm
showing
now
the
ocean
model
configuration
so
on
the
middle
middle
panel
we
have
pop
and
on
the
right
we
have
mom
this.
This
table
appeared
yesterday
on
frank's
talk
and
we've
shown
before
just
to
highlight
the
difference
between
the
two
models.
I'm
going
to
go
quickly
over
the
the
main
differences.
So
we
have
a
slightly
higher
resolution
grid
with
mum.
Six,
it's
a
two
thirds
of
degree
in
comparison
with
a
nominal
one
degree.
F
In
pop,
we
are
using
two
different
coordinate
systems
in
more
armor,
using
a
z-star
with
65
levels
and
also
a
hybrid
between
z-star
and
isobic.
Now,
with
75
levels
in
terms
of
vertical
mixing,
we're
using
kpb
vsd
mix,
while
the
fully
copper
simulations
with
pop
also
have
the
language
mixing
enable
we
do
have
the
ability
to
turn
on
the
lag
more
mixing
when
coupling
with
the
wave
model
and
that's
something
that
albert
belarge
and
myself
are
working
on
in
conjunction
with
brandon
hi,
caller
gfgl,
we'll
present
in
a
in
a
later
date.
F
F
A
F
Penetration
using
the
monitor
scheme-
and
at
this
time
we
are
providing
a
climatology
chlorophyll
while
we
ma
in
pop
that
is
done
with
the
omen
2003
scheme
and
it's
using
a
prognostic
chlorophyll
scheme,
and
I
won't
go
on
to
that.
Actually,
you
can
be
yourself,
okay,
so
now
going
into
the
experiments,
we
I'm
going
to
present
three
different
experiments,
two
of
which
have
the
same
vertical,
coordinate,
z,
star.
So
it's
what
we
call
bmom
z,
star
and
then
in
green
is
the
b
mom
gm
tuning
z
start,
which
was
done
at
a
later
time.
F
So
we
first
performed
the
two
experiments
on
the
top
one,
our
other
one,
with
hybrid
and
then
and
looking
at
results.
We
decided
to
do
a
tuning
simulation
where
we
tuned
the
gm
coefficient
to
improve
the
acc
transport.
As
I
will
show
in
a
second
all
of
the
simulations
have
worked.
We
did
a
tuning
on
in
cam
to
lower
the
rest
om,
so
we
play
with
this
club
gamma
coefficient.
We
set
it
to
a
smaller
value
than
the
default
value
and
then
in
the
bmom
gm
tuning
z-star
scheme
we
play
with
the
geometric
alpha
parameter.
F
F
We're
looking
at
now
a
global
mean
temperature
and
salinity
plots
so
starting
with
the
ocean
potential
temperature
throughout
the
entire
simulations.
Those
are
time
series
in
in
red.
We
have
the
results
from
pop
pre-doser
control
as
well,
and
just
to
highlight
that
the
pop
has
been
initialized.
The
ocean
and
ci's
have
been
initialized
from
previous
spun
up
simulations,
as
described
in
goku's
paper
from
2020.
So
that's
why
there
is
a
difference
between
the
two.