►
From YouTube: Land 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, logistics & LMWG overview: Will Wieder
- FATES updates & highlights: Jacquelyn Shuman
- The CLM5 Parameter Perturbation Experiment (PPE): Daniel Kennedy
- CLM-FATES PPE: First steps in the calibration cascade: Adrianna Foster
- Simulating alpine tundra communities along an environmental gradient with CLM5 at Niwot Ridge LTER: Katya Jay
- A NEON Testbed for High Resolution Climate Data Assessment: Montasir Maruf
- Investigate the global impact of carbon-based logging in FATES: Shijie Shu
- Assessing climate change impacts on live fuel moisture and wildfire risk using a hydrodynamic vegetation model: Chonggang Xu
- Attribution of climate: CO2: fire suppression: and harvest on densification and composition shifts in Sierra Nevada mixed conifer forests: Polly Buotte
- Tree crown damage and its effects on carbon cycling in a tropical forest: Jessica Needham
- Canopy rainfall interception: Pierre Gentine
- New features and enhancements in CTSM-SNICAR snow albedo scheme: Cenlin He
- A new approach to soil initialization for studying sub-seasonal land-atmosphere interactions: Yelin Jiang
- Reduced terrestrial evaporation leads to increased water vapor over land: Marysa Lague
- Soil moisture modulates land-use change impact under global warming: Arshdeep Singh
- Implementing a dynamic urban scheme in the Community Earth System Model (CESM): Bowen Fang
- Applying the Data Assimilation Research Testbed (DART) towards improved CLM simulations of Earth System Carbon: Water and Energy Cycling: Brett Raczka
To learn more:
https://www.cesm.ucar.edu/events/workshops/2022/
A
Yeah
thanks
everyone
for
joining
the
land
model
working
group
at
the
cesm
workshop
today
and
we've
tried
a
number
of
kind
of
creative
formats
for
this
meeting
and
and
today
it's
gonna
be
speed
dating
because
we've
had
a
number
of
talks
and
we
elected
to
give
everyone
talks
who
requested
them.
Instead
of
having
a
few
people
relegated
to
posters.
So
so
it's
going
to
be
fast,
brace
yourselves
and
speakers.
A
So
I
wanted
to
go
ahead
and
get
started
and
offer
a
few
kind
of
just
overview
presentation
of
where
the
land
model
working
group
is
headed
in
the
next
year
for
the
most
part
like
where
we've
been
in
the
last
year
and
where
we're
headed,
but
really
leaving
most
of
the
science
highlights
for
the
talks
that
are
coming
up.
So
I
look
forward
to
the
presentations
that
we
have
coming
up
this
morning.
A
A
Is
this
new
neuopsy
coupling,
and
so
that
basically
replaces
the
mct
coupler,
which
is
coupler
seven,
so
you
can
think
of
neopsy
as
effectively
being
coupler
eight
and
the
reason
I
bring
this
up
is
this
mct
will
be
deprecated
soon
it'll
still
be
supported,
you
know
for
legacy
tags
or
the
model,
but
in
moving
to
new
opsi
there's
a
few
things
that
I
wanted
to
highlight.
One
is
that
esmf
libraries
are
required.
A
You
know.
Another
important
feature
is
that
the
mesh
file
replaces
the
domain
file
in
in
your
simulations,
and
then
the
control
of
datm
has
changed
extensively,
especially
with
things
like
user
nameless,
datm
streams
and
all
of
this
it.
It
really
does
facilitate
single
point
simulations
soon.
We'll
have
the
capability
to
to
make
regional
simulations
simpler,
but
we
and
I
recognize
that
we
still
need
on
the
land
model
side.
We
still
need
some
additional
documentation,
so
we're
working
on
that.
A
If
you're
interested
in
hearing
more
about
neuopsy
mariana
is
presenting
tomorrow
at
8
30
at
the
software
engineering
working
group
meeting,
but
there
kind
of
continues
to
be
a
number
of
posts
and
questions
and
frustrations
on
the
forums
and
on
github.
So
I
just
wanted
to
bring
this
up
that,
after
dev
tag,
63
we've
switched
to
neupopsy
in
terms
of
our
development
timeline.
You
know
in
2018
we
released
clm
5
and
soon
after
that,
ces
m2
and
then
kind
of
where
we
are
here
in
summer.
A
2022
they're
about
to
make
a
cesm
2.3
tag
relatively
soon,
I'm
not
exactly
sure
on
the
date
and
then
our
target
and
I'm
calling
the
cesm
2.x
it's
kind
of
a
prequel
to
cesm3,
but
that
by
next
summer,
we'll
be
testing
out
in
the
coupled
model.
What
will
become
ces
m3,
so
it
won't
be
csm3
won't
be
released
next
summer,
but
we're
going
to
start
testing
coupled
model
configuration
for
that.
So
in
the
meantime,
a
number
of
things
have
come
into
the
model
already.
A
So
that
includes
things
like
biomass
heat
storage,
this
roughness
length
parameterization,
that's
almost
the
main.
A
number
of
changes
to
phenology
fire
some
bug
fixes
the
addition
of
fate
sp,
as
well
as
space,
regional
and
point
capabilities
that
we'll
talk
about
or
hear
about
today,
as
well
as
the
neon
work.
A
So
in
the
meantime,
daniel
dave
and
katie
have
have
started
the
parameter,
perturbation
experiment
with
the
idea
that
we'll
learn
enough
about
the
parameterization
of
clm,
that
we
can
hopefully
tune
the
model
for
cts
m6
and
have
it
come
back
in
and
then
also
there's
a
new
branch.
That's
been
started,
which
is
ctsm
5.2,
which
is
just
for
surface
data
updates.
So
I'll
talk
a
little
bit
more
about
that
in
a
minute.
A
A
Then
fates
with
land
use
change,
fates
with
fixed
biogeography
and
then
a
kind
of
a
full
implementation
of
fates,
kind
of
tvd
when,
when
that
gets
integrated
into
ctsm
and
then
the
last
couple
of
things
on
here
are
a
switch
to
mimic
soil,
biogeochemistry
updates
to
crop
planting
dates
and
then
potentially
the
absolute
crop
model.
So
we'll
hear
about
some
crop
some
crop
updates.
A
Today,
too,
I
wanted
to
highlight
these
ctsm
5.2
dev
tag
that,
for
the
most
part,
just
mariana
started
working
with
us
or
you
know,
kind
of
building
the
capabilities
for
creating
mixer
data
with
esmf,
and
then
sam
is
continuing
that
I
don't
want
to
get
a
ton
into
the
the
details
here.
Other
than
creating
mixer
data
with
the
esmf
regretting
makes
it
much
quicker.
The
mapping
files
don't
have
to
be
created
as
separate
files,
and
it
facilitates
the
creation
of
variable
mesh
and
high
resolution
grids.
A
We've
also
got
this
new
transient
urban
datasets
and
gross
represented
unrepresented
land
use
transitions
that
peter's
been
working
on
and
then
as
well
as
dynamic
links
updates
to
soil
properties,
irrigation
methods,
glacier
fields
as
well
as
potentially
soil,
color
and
leaf
area
index.
So
the
idea
here
is
that
cts
m51
development
will
continue
and
then
we'll
merge
it
we'll
kind
of
test.
The
new
surface
data
sets
understand
how
they're
changing
answers
and
then
merge
those
back
to
the
trunk
kind
of
together.
A
There's
examples
for
how
to
run
supported
neon
tower
sites
as
part
of
the
neon
project,
as
well
as
photos,
global,
sp
simulations
and
then
also
a
workflow
for
generic
single
point,
simulations
that
uses
new
neuopsy,
and
so
all
of
this
is
kind
of
using
some
tools
that
are
in
ctsm
tools,
site
and
regional.
So
there's
two
new
python
scripts
there
that
negan
and
adriana
have
worked
on
that
are
run,
neon
and
subset
data
and
then,
in
this
tutorial,
there's
also
notebooks
on
how
to
looking
at
how
to
visualize
results
and
analyze
model
output.
A
So
this
is
all
these
notebooks
are
rendered
and
available
online
and
I'd
encourage
folks
to
take
a
look
at
those.
Those
are
meant
to
be
a
resource
so
that
you
know
if,
if
you
or
or
your
students
are
working
on
some
of
these
new
capabilities,
we
would
like
you
to
start
using
these
tools
and
then,
lastly,
I'd
like
to
request
that
you
ask
for
additional
topics,
and
so
you
know,
regional
cases
are
one
that
I
think
will
be
hopefully
rolling
out
before
too
long.
A
Just
in
the
kind
of
closing.
This
is
an
another
list
of
all
of
the
development
goals
that
we
have
for
cts
m6,
it's
relatively
short-
and
I
have
this
up
here,
mostly
as
a
resource,
so
that,
if
you're
curious
about
what's
coming
on
and
what's
coming
to
to
main
many
of
these
have
pull
requests
that
are
associated
with
them.
You
can
kind
of
follow,
along
with
your
favorite
favorite
development
plans
or
activities
that
are
happening
in
the
model.
A
C
B
B
I
want
to
emphasize
that
fates
is
very
much
a
community
effort
and
it
would
not
be
possible
without
the
large
team
and
the
very
many
outside
collaborators,
specifically
charlie,
coven,
rosie
fisher
and
ryan
knox.
So
this
is
not
just
a
reflection
of
my
work.
Okay,
there
we
go
so
fates
is
a
cohort
model.
It
fundamentally
does
vegetation
differently
than
the
standard
host
land
model
it
can
run
within
clm
or
elm.
B
Fates
represents
mixes
of
different
types
of
vegetation,
using
a
count
to
represent
the
density
of
a
particular
type
of
vegetation
of
a
particular
size
on
a
grid
cell.
You
have
multiple
times
since
disturbance,
and
so
you
can
have
within
a
grid
cell,
you
can
represent
a
newly
disturbed
patch,
all
the
way
up
to
a
mature
old
growth
forest,
and
so
you
have
different
collective
collections
of
vegetation
competing
with
one
another
in
space
that
changes
through
time,
and
so
fates
has
variability,
and
with
that
variability
you
get
different
interactions
with
the
atmosphere.
B
You
get
different
vulnerabilities
of
the
vegetation,
and
so
you
get
different
dynamics.
It
also
allows
us
a
more
realistic
connection
to
data.
This
is
important
because
if
you
look
globally
degradation
and
forest
change
are
an
important
component
driving
co2
emissions
as
well
as
land
atmospheric
fluxes,
forest
degradation
specifically
drives
emissions.
Deforestation
mainly
from
south
america
and
southeast
asia,
is
the
largest
contributor
for
deforestation.
B
Fires
from
temperate
and
boreal
regions
also
are
a
large
contributor,
and
so
fates
is
able
to
represent
both
of
these
processes,
which
are
inherently
size
related.
If
we
look
specifically
at
the
amazon
in
brazil,
the
amazon
has
been
a
net
carbon
emitter
for
many
years,
and
this
picture
shows
you
the
change
in
source
versus
sink
and
so
there's
a
difference.
B
B
Increased
productivity
and
altered
structure
has
the
capacity
to
change
the
way
the
boreal
regions
look
on
the
left.
We
have
increases
in
greening
shown
across
the
tundra.
This
increase
in
productivity
is
something
that's
going
to
change
the
height
and
types
of
vegetation
on
the
right.
We
have
the
extent
of
the
tundra
taiga
ecotone.
B
As
we
have
changes
in
productivity.
We
can
potentially
get
advancement
of
trees
into
this
area
shown
in
pink,
is
the
diffuse
tige,
and
so,
if
we
get
advancement
or
an
increase
in
the
density
of
trees,
that's
going
to
change
the
interaction
with
the
atmosphere
for
these
regions,
and
so
there
are
a
lot
of
instances.
B
B
One
thing
that
I
don't
have
a
specific
slide
on
is
our
updates
within
fire.
We
will
have
a
talk
about
live
fuel
moisture
today
and
but
I
I've
done
development,
adding
crown
fire
with
sam
levis
and
then
we're
doing
testing
with
polyviatt
to
add
a
more
mechanistic
implementation
of
crown
fire,
which
is
also
very
much
size
related.
B
So
globally,
we've
been
running,
fates
adriana
foster,
will
talk
about
reduced
complexity
modes,
early,
ppe
testing,
with
fates,
sp
fates
is
running.
You
can
run
fades
globally
with
fixed
biogeography,
we're
doing
diagnostic
testing
with
the
islam
framework.
We'll
hear
a
talk
today
about
a
global
implementation
of
land
cover,
land
use
change
and
then
regionally
I've
been
doing
testing
looking
at
fire
vegetation,
climate
feedbacks,
and
so
this
is
with
full
complexity
and
so
through
the
interaction
of
these
feedbacks.
We're
able
to
dynamically
assemble
a
realistic
ecosystem
with
reasonable
biomass
accumulation.
B
We
also
have
a
lot
of
updates
with
hydraulics
elon,
fang
and
jinyon
ding.
We
have
two
types
of
solvers
depending
on
the
level
of
accuracy
you
want
or
the
where
you're
testing
we
have
variable
fine
root,
layering,
updated
pressure
volume,
pressure
conductance
as
the
plants
grow
in
size.
You
get
a
variation
in
breeding
depths.
You
also
get
leaf
stone
model
humidity
that
drives
as
acts
as
a
control
on
transpiration.
B
B
You
can
see
that
moss
acts
as
a
control
on
soil
temperature
shown
here
in
the
top
and
then
moss
is
fundamentally
different
than
grass.
So
grass
is
shown
here
in
the
orange
line.
Moss
is
the
red
and
the
blue,
and
so
having
an
improved
representation
of
moss
compared
to
actual
data
will
be
important
for
the
model.
B
Marcus
longo
is
working
to
incorporate
and
test
drought
deciduous
phenology.
This
is
important
across
areas
such
as
the
amazon.
You
have
a
pronounced
rainfall
gradient
and
across
that
radiant.
You
have
variation
in
the
amount
of
evergreen
versus
deciduous
trees,
and
so
early
testing
indicates
that
you
can
get
coexistence
of
deciduous
and
evergreen
in
both
wet
and
dry
forests.
B
Julian
gao
from
berkeley
is
working
to
update
olamitry
and
phonology
for
grasses
in
california,
and
so
using
fitted
olametry
and
adjusted
phenology
they're
able
to
capture
the
magnitude
and
seasonal
variation
in
gpp
for
sites
in
california,
and
then
she's
supporting
this
with
grass
elementary
greenhouse
experiments.
That
will
expand
the
types
of
grasses
that
we
can
characterize.
B
Joshua
rady
has
implemented
a
vegetation
or
has
created
a
vegetation
management
module
that
can
function
with
faiths
and
it
has
more
specific
functionality
compared
to
the
existing
launch
logging
module,
and
so
this
is
an
example.
We
have
site
level
results
at
the
bottom
and
then
he's
applied
it
across
a
region
for
the
southeastern
united
states,
and
so
you
can
see
that
changes
in
co2
as
well
as
changes
in
the
forcing
scenario.
This
is
rcp
2.6.
B
B
Thank
you
I'm.
This
is
my
last
slide
and
so
on
the
fate
side,
greg
lemieux
has
updated
our
technical
documentation
and
our
user's
guide.
So
it's
searchable,
which
is
a
nice
change,
that's
available
at
the
website
listed
at
the
top
and
then,
as
will
said,
there
are
tutorial
materials
that
include
specific
activities
for
fates
and
that
website
is
listed
here
at
the
bottom.
B
D
B
Yeah,
that's
a
great
question
and
so
the
you
can
define
pfts
within
faiths
as
you
desire,
you
are
not
restricted
to
fates
as
their
pfts
as
they're
defined
by
either
clm
or
an
existing
implementation,
and
so
you
can.
You
can
increase
the
number
of
plant
functional
types
and
you
can
they
can
use
unique
elementary.
The
challenges
with
paleo
implementations
would
be
figuring
out
how
to
parameterize,
and
so
can
you
use
modern
plants
to
represent
these
ancient
plants.
But
yes,
absolutely
you
have
the
possibility
to
choose
your
own
adventure
with
fades
and
lara
kippers.
E
Yeah
great,
thank
you,
jackie,
and
I
see
that
there's
also
a
question
for
adrian.
It
looks
like
we
have
one
minute
before
daniel's
talk,
but
that
question
looks
like
it.
It's
potentially
more
directed
towards
will
wondering
the
about
the
difference
between
the
grid
and
the
mesh
files.
So
well,
I
don't
know
if
you
have.
A
An
answer
you
can
go
ahead
and
start
sharing
your
screen,
but
the
quick
answer,
adrian,
is
that
we're
we
have
a
get
regional
script
that
subsets
the
the
subsets
the
mesh
file
for
a
kind
of
standard
resolution
run.
That
is
almost
to
main,
I
think,
there's
still
a
few
little
testing
things
that
need
to
get
worked
out,
but
that
so
that
regional
capability
for
running
with
neuopsy
is
is
pretty
much
functional
and
almost
tested.
Well
enough
that
we
can
release
it.
E
A
F
Likewise,
we
want
to
explore
sensitivity
of
a
range
of
features
of
global
land
system
to
reasonable
uncertainty
and
model
parameter
values
for
land.
This
is
especially
important
because
we
know-
or
we
think
that
model
uncertainty
has
a
lot
to
do
with
parametric
uncertainty,
and
for
me,
one
of
the
things
I'm
especially
interested
in
is
water
use.
Efficiency.
F
You
can
see
that
clm
has
had
big
changes
in
water
use,
efficiency
from
model
to
model
or
from
across
the
different
versions
of
the
model.
So
why
is
this
important?
There's
a
really
large
influence
of
carbon
cycle
uncertainty
on
climate
projections,
so
this
is
a
plot
from
the
cmib-5
ensemble,
showing
the
spread
across
a
suite
of
climate
models
in
terms
of
the
integrated
terrestrial
carbon
sink
or
up
to
the
21st
century.
F
So
the
spread
among
the
models
is
approximately
a
hundred
years
of
current
fossil
fuel
emissions,
and
you
can
see
that
the
models
don't
actually
even
agree
whether
the
land
will
be
a
source
or
a
sink
through
to
the
21st
century
right.
So
this
set
of
models
has
the
land
as
a
sink,
and
this
set
of
models,
two
of
which
use
clm
4,
have
the
land
as
a
source
of
carbon.
F
So
what
I'm
interested
in
is
how
does
a
single
model
like
clm's
parametric
uncertainty,
compare
to
the
multi-model
spread?
If
we
sample
a
reasonable
set
of
parameter
values?
Would
we
get
a
small
uncertainty
cone
around
our
model
smaller
than
the
spread,
or
we
get
a
really
large
uncertainty
cloud,
that's
larger
than
the
spread
among
the
models.
F
Think
approach
to
setting
parameter
ranges.
No
one
person
really
knows
every
domain
of
the
model,
so
we
created
a
google
spreadsheet
and
worked
to
kind
of
fill
out
reasonable
ranges,
and
we
found
this
is
a
useful
way
to
kind
of
approach
this
and
having
the
infrastructure
to
read
directly
from
the
google
spreadsheets
has
been
nice
to
create
a
perturbation
experiment.
F
F
You
know
they
tried
automated
calibration
for
a
clm5
and
it
didn't
quite
work
and
then
they
were
left
in
a
large
kind
of
at
the
11th
hour,
and
so
when
I
say
the
model
is
expensive.
If,
if
you
know
you
work
with
other
components
of
cesm,
you
might
think
the
land
is
cheap
and
it
is
relatively
cheap
per
simulated
year,
but
the
g,
but
the
bgc
model
requires
greater
than
1
000
years
to
spin
up
the
geochemistry,
which
makes
it
expensive
per
simulation.
F
So
if
you
imagine,
you
were
going
to
spend
2
million
pe
hours,
that's
how
much
computer
time
you're
willing
to
spend.
You
would
only
get
about
five
simulations
of
the
bgc
model
if
you
were
running
at
one
degree,
resolution
which
is
not
enough
to
do
any
sort
of
meaningful
calibration
and
so
cost
management
was
a
big
part
of
the
work
that
we
had
to
do
so.
One
degree
simulation
requires
21,
000
grid
cells
and
then
two
degree
simulation
you
get.
F
So
this
was
without
this
we
weren't
going
to
be
able
to
do
any
sort
of
meaningful
parameter
experiments,
and
so
our
experiment
is
to
do
all
200
parameters,
perturbing
them,
one
at
a
time,
a
high
value
and
a
low
value
for
about
400
simulations.
Each
simulation
is
independently
spun
up
for
140
years,
using
that
cn
matrix
and
then
we're
doing
a
10
year,
steady
experiment
cycling
over
2005
to
2014
with
re-analysis
forcing
why
one
at
a
time.
F
So
it
yields
an
easy
to
interpret
data
set,
and
we
knew
that
many
of
the
parameters
will
have
very
small
effect
on
the
variables
we're
interested
in,
and
latin
hypercube
of
200
parameters
was
pretty
intimidating
and
what
I
mean
by
that
is
we
weren't
sure
that
we
would
get
meaningful
or
diagnosable
results
out
of
that.
So
we
wanted
to
start
with
one
at
a
time
and
then
proceed
onto
sort
of
a
multi-factor
experiment.
So
this
shows
across
the
ensemble.
You
know
approximately
400
members.
F
F
F
F
Okay
yeah,
so
let
me
just
go
through
this
and
then
I'll
take
questions.
Actually
I
can
skip
this
so
coming
out
of
this
we're
going
to
do
a
lii
calibration
experiment
where
we're
subsetting
25
parameters
for
a
multi-factor
ensemble
using
latin
hypertube
sampling
applied
to
transient
climate,
forcing
where
we'll
have
to
define
some
objective
function
for
tuning
the
model,
find
optimal
parameter
sets,
and
then
we
expect
to
run
an
ensemble
of
reasonable
parameterizations
into
the
future,
to
try
and
understand
the
uncertainty
and
then
so
that's
upcoming.
F
And
then
there
are
a
lot
of
coordinated
ppe
projects
that
are
all
super,
exciting
university
of
washington,
they're,
doing
lane
atmosphere,
interactions,
fates
adriana,
will
talk
next
and
then
there's
a
bunch
more
here
which
I
can
share,
but
we're
really
excited
and
hope
that
some
of
the
infrastructure
that
we're
developing
can
be
useful
for
coordinated
experiments.
So
I'll
leave
this
up
but
take
any
questions
there
are.
I
don't
know.
A
E
B
G
Okay,
great
cool
yeah,
thanks
yeah,
so
thanks
daniel,
I'm
gonna
follow
up
with
sort
of
a
definitely
right
now,
the
initial
stages,
but
following
on
daniel's
and
the
rest
of
the
clm
ppe
teams,
efforts,
I'm
doing
a
clm,
fates,
ppe
and
so
yeah.
So
jackie
gave
a
great
overview
of
fates,
but
just
to
remind
everyone,
fates
is
a
vegetation,
demographics
model.
It
operates
within
clm
or
you
can
also
run
it
with
an
elm.
G
But
this
means
that
it
breaks
vegetation
up
by
cohorts
of
similar
size
and
pft.
The
cohorts
grow
and
die
on
patches
of
a
similar
age,
and
then
we
have
atmospheric
fluxes
like
in
clm
that
are
simulated
very
quickly
and
then
the
plant
growth
and
allocation
occurs
on
a
daily
time
step,
and
so
why
does
this
matter
and
so
with
a
model
like
fates,
where
necessarily
in
the
business
of
capturing
ecological
processes
across
multiple
scales?
G
G
You
know
we
might
be
interested
in
some
kind
of
emergent
process
like
fur
waves
that
you
can
see
here,
and
this
is
a
result
of
all
of
these
multi-scale
interactive
processes
and
then
at
the
same
time,
we're
experiencing
a
boom
in
remote
sensing
technology.
Advancements
allowing
for
high
resolution,
observations
of
vegetation
characteristics
and
state.
G
And
so
we
have
things
like
jedi,
which
looks
at
either
biomass
or
stand
structure
ecostress
for
things
like
plant
stress
and
oco-3
for
stiff,
and
so
we
can
use
these
really
high
resolution
observations
to
calibrate
fates
at
higher
resolution
as
well,
however,
because
of
fate's
multi-scale
meter
and
the
fact
that
these
scales
often
interact
non-linearly
with
one
another.
It
can
be
really
difficult
to
calibrate
across
all
of
these
scales
at
once,
without
kind
of
hypertuning
into
some
rabbit
hole.
G
So,
for
example,
while
trying
to
calibrate
output
of
say,
gpp
or
transpiration,
we
also
have
to
ensure
we're
accurately
capturing
leaf
area,
stand,
structure,
recruitment
and
mortality
as
well
as
pft
composition,
and
so
ideally,
we'd
like
to
be
able
to
control
for
some
of
these
processes.
While
we
aim
to
better
parameterize
and
calibrate
others
and
in
fates,
we
do
have
the
ability
to
control
for
these
different
complex
processes.
G
So
the
face
team
has
been
working
on
developing
these
reduced
complexity
modes
for
fates,
so
with
satellite
phonology
mode,
much
like
in
the
clm
bgc
sp
mode.
We
use
observed
lai
over
time
for
each
pft
and
then
each
pft
consists
of
one
cohort
on
one
patch
and
so
here
we're
essentially
turning
off
growth,
disturbance
and
mortality
in
no
competition
or
no
comp
mode.
G
Pfts
are
given
a
fixed
area
to
grow,
but
do
not
compete
with
one
another
in
prescribed
biogeography
mode.
We
allow
for
growth,
disturbance
and
competition
between
pfts,
but
we
restrict
pft
establishment
and
growth
to
where
those
pfts
actually
occur
on
the
globe
and
then
finally,
a
full
face
mode.
With
all
of
these
levers
on
and
thanks
to
charlie
for
lending
me,
these
cool
cartoons
that
he
made,
and
so
these
different
modes
allow
us
to
calibrate
different
processes
and
parameters
within
the
model.
So
with
sp
mode.
G
You
know
we
can
test
photosynthesis
and
leaf
and
hydrology
parameters
in
no
comp
mode.
We
can
test
allomatry
and
allocation,
as
well
as
mortality
and
with
fixed
biogeography.
We
can
test
environmental
sensitivity
and
competition
for
resources,
and
so
the
fates
team
and
namely
rosie
fisher
and
charlie
coven
have
been
developing
this
concept
of
a
calibration
cascade
to
test
various
parts
and
parameters
within
the
model.
G
Excuse
me
so
initially
I
conducted
similarly
a
one-at-a-time
mid
max
parameter,
ensemble,
so
running
a
simulation
with
a
min
and
a
max
for
each
parameter
value.
I
also
used
this
sparse
grid
to
save
on
computation
time.
So
again,
these
400
grid
cells
were
chosen
based
on
representativeness
and
responsiveness
to
climate
and
environmental
change,
and
then
for
each
simulation.
G
Now,
because
we're
in
sp
mode,
I
actually
expect
most
of
these
parameters
to
have
absolutely
no
effect
on
model
output,
and
I
really
only
expected
those
parameters
related
to
photosynthesis,
leaf
traits
and
radiation
to
be
impactful
in
this
mode.
But
again
we
kind
of
just
wanted
to
do
a
sanity
check
to
make
sure
that
this
was
the
case
so
of
these
170
parameters,
iran
only
36
have
an
impact
in
sp
mode
and
as
for
impact,
I
was
looking
at
history:
output
of
gpp,
transpiration
cell
temperature,
soil,
moisture
and
albedo.
G
Excuse
me
so,
as
I
expected
most
of
the
important
parameters
had
to
do
with
leaf
traits
and
radiation
with
some
oleometry
parameters
showing
up,
though
these
were
very
minimally
impactful
and
then
just
note.
I
did
have
one
fail,
then
I'm
still
working
on
the
leaf
stem
orientation
index
failed
when
I
tried
to
increase
it.
G
C
G
Excuse
me
similarly,
here
is
grid
cell
average
annual
latent
heat
flux,
with
the
standout
parameters
related
to
model,
slope
and
intercept,
and
this
all
makes
quite
a
bit
of
sense.
So
gpp
is
most
impacted
by
photosynthesis
and
leaf
traits
laden
heat
flux
is
really
highly
influenced
by
model
conductance,
but
I
wanted
to
have
a
metric
that
would
show
the
effect
of
each
parameter,
but
I,
but
that
was
not
impacted
by
the
units
of
the
history
variable
or
the
units
of
the
parameter.
G
G
So
now
what
okay?
So
essentially,
the
idea
would
be
now
to
implement
calibration
using
these
top
n
parameters,
I'm
not
sure
how
many
we
should
start
with
at
first,
but
this
is
something
I'll
be
working
on
with
feedback
from
the
fates
team
and
the
clm
ppe
team,
and
the
idea
would
be
to
calibrate
against
outputs
of
gpp
albedo,
transpiration
and
other
fluxes.
G
However,
this
will
greatly
increase
the
number
of
simulations
needed
if
each
pft
parameter
combo
is
considered
separately.
So
one
thing
that
the
clm
ppe
is
looking
into
is
seeing
where
pfds
overlap
and
whether
they
can
run
you
know
completely
non-overlapping
pfts,
at
the
same
time,
to
save
on
computation.
So,
for
example,
we
may
be
able
to
calibrate
arctic
plants
and
c4
grasses
at
the
same
time,
and
so
ultimately,
this
is
just
the
initial.
E
G
G
H
Sorry
I
forgot
to
unmute
myself:
there
you
go.
Okay,
thanks.
Jackie
all
right,
like
jackie,
said,
we'll
be
switching
gears
a
little
bit
to
talk
about
some
single
point:
work,
simulating
alpine
tundra
communities
at
nywon
ridge
and
before
I
get
into
that,
I
just
want
to
quickly
acknowledge
my
funding
sources
and
my
co-authors,
who
have
contributed
to
this
work
both
from
cu
and
from
ncar,
and
I've
had
a
lot
of
help
on
the
modeling
side.
So
that's
been
really
great.
H
For
those
of
you
who
aren't
familiar
with
nylon
ridge,
it's
a
long-term
ecological
research
site
located
in
the
colorado
front
range
and
the
site
has
a
long
history
of
measurements,
going
back
to
the
1980s,
including
things
like
vegetation,
snow
depth,
soil
properties,
freshwater
lakes
and
a
lot
of
other
things,
and
it
also
has
a
flux
tower.
That's
been
active
since
2008,
as
you
can
see
in
this
photo
here
known
as
t-van,
and
this
because
it's
a
high
alpine
environment.
It
has
a
lot
of
complex
topography.
H
So
in
this
photo
here
you
can
see
our
representative
hill
slope
at
nywatt
ridge,
which
is
this
area
known
as
the
saddle,
and
it
has
three
distinct
tundra
vegetation
communities
as
a
result
of
differences
in
exposure
through
wind
and
aspect
and
then
resulting
differences
in
snow
accumulations.
So,
in
the
background
here
in
the
snow
covered
area,
you
can
see
what's
known
as
the
moist
meadow,
and
this
is
an
east
facing
area
that
accumulates
a
lot
of
snow
and
this
photo
is
taken
in
late
june.
So
you
can
see
the
snow
is
melting
out
pretty
late.
H
Just
to
kind
of
give
a
quick
overview
of
our
model
setup
we
have
like.
I
said
we
have
a
lot
of
site
input,
data
coming
from
nylock
ridge,
so
things
like
precipitation
from
the
saddle
and
meteorology
from
the
flux
tower,
as
well
as
radiation
data
from
an
ameriflex
tower,
and
then
we
also
have
three
different
parameterizations
to
kind
of
simulate
these
three
different
communities.
H
H
Meadow
is
on
top
wet
meadow
in
the
middle
and
dry
metal
on
the
bottom,
and
you
can
see
that
we
have
this
gradient
of
snow
accumulation
where
the
moist
meadow
has
a
pretty
large
snowpack
and
then
the
dry
meta
has
almost
no
snow
in
the
winter,
but
a
little
bit
more
snow
in
the
spring,
which
is
because
it
tends
to
get
wetter
heavier
snow
that
doesn't
blow
away
as
much
in
the
spring
and
then
also
just
to
note.
The
observations
we
have
from
this
from
the
snow
depth
are
really
variable
between
years
and
over
time.
H
So
now,
looking
at
soil
temperature,
these
plots
are
showing
new
soil
temperature
kind
of
along
this
gradient
of
snow
accumulation.
So
from
moist
to
dry
communities
and
overall
you
can
see
the
differences
between
the
communities,
so
the
dry
meadow
soils
are
a
lot
warmer
during
the
growing
season
compared
to
the
other
two
and
colder
in
the
winter.
So
that's
due
to
the
buffering
effects
of
snow,
especially
in
the
moist
meadow.
H
H
So,
as
a
result
of
these
physical
differences,
we
see
differences
in
growing
season,
length
and
productivity
in
these
different
plant
communities.
This
top
pot
is
showing
you
leaf
area
index
so
with
wet
metal
and
blue
dry
metal
and
orange
and
moist
meadow
in
green.
You
can
see
the
dry
metal
has
the
longest
growing
season.
The
moist
meadow
has
the
shortest,
and
then
this
bottom
plot
is
showing
above
ground
npp
again
with
observations
from
some
biomass
with
harvest
and
black.
H
H
So
we
know
that
aspect
is
really
important
in
the
alpine,
so
we
were
interested
in
looking
at
what
would
happen
if
we
simulated
these
communities
on
north
and
south
facing
aspects
and
to
provide
a
little
bit
more
context.
They're
conducting
a
turf
transplant
field,
experiment
in
the
coming
years,
up
at
nywat
ridge,
where
they're
going
to
be
digging
up
blocks
of
tundra
and
transplanting
them
across
macro
and
microsite
gradients.
H
So,
in
the
interest
of
time,
I'm
just
going
to
kind
of
show
some
soil
results
from
this
aspect.
Work.
The
soil
temperature
is
on
the
top
and
soil
moisture
on
the
bottom.
In
these
plots
and
in
all
these
plots,
the
north
aspects
have
the
darker
colored
lines
and
again
going
from
moist
to
wet
to
dry
meadow,
and
you
can
see
that
overall,
the
south
aspects
are
significantly
warmer
and
drier,
especially
in
the
dry
meadow
on
average
over
the
over
the
year.
H
But
one
thing
that's
interesting
to
note
is
that
the
timing
of
when
these
communities
are
or
when
that
south
aspect
is
warmer
kind
of
differs
by
community.
So
you
can
see
in
the
voice
meadow
during
the
winter
there
there
are
very
few
differences,
but
then,
during
the
growing
season
the
south
aspect
is
much
warmer,
whereas
in
the
dry
meadow
it's
the
other
way
around
where
in
the
winter
the
south
aspects
are
warmer.
But
then
the
differences
are
smaller
during
the
growing
season.
H
So
this
kind
of
suggests
that
there
are
historical
differences.
We
know
that
there
are
historical
differences
in
snow
accumulation
in
these
communities
and
that
that
will
moderate
how
these
communities
are
experiencing
warming
in
the
future
across
different
aspects
and
kind
of
showing
the
buffering
effects
of
snow.
B
H
C
C
H
I
I
I
I
I
I
I
I
Here
and
the
axial
axis
is
normalized
grouping
square
and
the
radial
axis
is
correlation,
so
in
the
first
diagram
is
cr5
and
we
can
see
that
clm
high
resolution
is
has
a
better
correlation
than
clm
one
degree
in
india.
Five
in
the
first
diagram
high
resolution
correlation
is
0.98,
whereas
their
load
resolution
is
0.97
in
the
second
diagram
is
the
meter
tube
here
you
can
see
that
clm
high
resolution
for
relation
is
0.93,
whereas
the
cll
and
low
resolution
is
0.92
and
the
third
one
is
the
levy
here.
I
This
is
the
temporal
average
for
all
the
17
equal
regions,
and
that
is
only
for
you
at
a
pipe
here.
We
can
see
that
the
all
the
or
the
three,
the
three
regions,
which
are
three
four
and
eight,
these
three
regions
simulated
well
in
clm
higher
resolution
than
the
cnn
low
resolution
in
clm
low
resolution.
I
I
I
So
these
are
residence
time,
spatial
comparison
with
all
the
four
models
and
every
five.
So
here
we
can
see
that
the
dry
regions
like
southwest
united
states,
gets
a
relatively
higher
residence
time
in
clm
higher
resolution
than
the
clm
low
resolution.
If
you
see
these
regions,
these
regions
get
higher
residence
time
than
their
clm
low
resolution
and
also
echo
regions.
Eight
got
higher
higher
the
soil,
marginal
residence
time
value
than
the
clm
low
resolution.
I
So
this
is
a
normalized,
routine
score
and
correlation
comparison
for
the
residence
time.
Here
we
can
see
that
clm
high
resolution.
You
can
probably
get
better
correlation,
which
is
0.98,
whereas
clm
low
resolution
got
point
nine
four,
as
a
correlation,
so
clm
high
resolution
performs
really
well
in
soil
moisture
residence
time
compared
to
the
other
models.
I
B
B
J
Yeah
so
so,
let's
start
yeah
hi,
I'm
shujiyashu,
currently
a
postdoc
working
in
lawrence
berkeley
national
lab
today.
My
topic
will
be
the
global
impact
of
carbon
based
logging,
which
we
recently
implemented
into
fates
models
and
I'd
like
to
first
thank
collaborator.
Jennifer
holm,
my
supervisor
and
also
the
whole
face
team
without
which
is
help
could
not
make
this
work
happen
and
also.
I
would
like
to
thank
the
sponsor
from
the
department
of
energy
and
interesting
project.
J
So
what
why
are
we
going
to
implement
the
biomass
c
based
phase
logging?
The
first
most
important
thing
is
the
biomass
carbon
based
harvest
rate
data
is
more
accurate.
Comparing
to
the
area
based
harvest
rate
data
set,
which
right
now
most
model
is
using
area
based,
but
within
the
phase
models,
the
good
thing
is
in
the
fades.
J
The
basic
ideas
as
this
workflow
is
showing
first
we're
reading
comparing
to
the
highest
rating
air
reflection,
we're
reading
the
highest
rate
in
carbon
biomass,
and
then
we
assume
the
biomass
is
equally
distributed
in
each
cohort
and
then
calculate
the
fraction
of
the
area
based
on
the
fraction
of
the
highest
rate
to
the
total
forest
carbon,
which
is
available
to
be
harvested
in
the
phase.
And
then
after
we
calculated
this
fraction.
J
J
So
for
this
global
cases
the
spin-up
will
be
from
the
zero
for
150
years
under
a
new
function
called
fixed
biogeography
sitting
which
each
which
sells
can
only
allow
those
pfts
described
in
the
biogeographic
map.
So
so
this
can
constrain
very
weird,
like
tropical
forests
residing
into
a
temporary
region
happen,
and
this
has
been
introduced
by
jackie
and
will
earlier
so
for
this
thing
up,
because
from
the
simulation
1890s,
we
already
have
the
secondary
force
exists.
But
when
we
spin
up,
we
don't
generate
enough
secondary
force.
J
So
we
separate
the
spin
up
phases,
transition
case
transient
cases
into
two
phase
for
the
phase,
one
which
starts
from
1850
to
80
90
40
years
to
just
introduce
certain
area
to
be
secondary
force
based
on
the
harvest
rate
described
by
the
h2,
and
our
official
simulations
for
the
transient
cases
will
start
from
1890
to
check
the
harvest.
Carbon
area
for
both
primary
secondary
forests.
J
And
the
first
is
a
brazil
side
case.
It's
actually
two
by
two
degree
and
to
the
left
is
showing
the
validated
harvest
car
harvested
carbon
comparison
for
phase
and
aih2.
This
will
check
the
consistency
between
the
data
set
and
the
actual
carbon
calculator
and
phase
and
to
the
right
well
most.
More
importantly,
is
we
want
to
check
once
we're
reading
this
carbon
amount?
J
How
this
land
area
of
the
secondary
forest
looks
like
when
comparing
the
phase
to
the
aiuc
term
series,
as
you
can
see
here,
the
calculated
secondary
force
fraction
is
actually
a
little
bit
less,
comparing
to
the
highest
error
fraction
by
h2.
This
is
mostly
due
to
a
higher
biomass
in
the
current
site
compared
to
the
biomass
dataset
used
by
lh2
and
a
quick
validation
of
the
site
level.
J
Let's
see
the
how
the
global
case
is
working
for
the
left
is
showing
the
secondary
force
area
in
the
united
fraction
of
each
grid
area
in
the
year
2009
and
to
the
right.
Its
primary
force
harvested
carbon,
as
you
can
see,
for
growth
of
the
secondary
forest.
Most
of
places
are
having
consistent
area.
When
you
compare
air
h2
to
the
fates
except
for
in
in
the
northern
region,
the
phase
seems
to
have
more
area
transferring
into
the
secondary
forest.
Mostly
due
to
there.
J
There
will
be
a
less
vegetation
carbon
available
for
harvest
in
that
region
growing
there
and
to
the
right.
We
have
the
primary
forest
harvested
carbon
and
most
regions
they're
comparable
to
each
other,
and
the
only
places
having
a
little
bit
trouble
is
also
the
same
places
where
we
have
the
less
of
the
more
area
transfer
into
secondary
forest,
and
you
have
less
carbons
being
harvested
also
in
there,
and
this
is
showing
the
curve
of
the
to
the
left
of
the
global
harvest.
J
Carbon
and
to
the
right
is
the
harvested
area
of
the
comparison
between
the
phase
model
simulator
and
the
air
issue.
Data
set
showing
so
to
the
left.
We
can
see.
The
phase
estimation
is
almost
can
cover
what
primary
forest
described
by
ls2
data
set,
but
for
the
secondary
forest
harvested
patches
we
can
see
the
fates
only
can
capture
a
part
of
that
and
we
will
show
later
in
the
global
map,
where
we
don't
have
enough
carbon
to
be
harvested
and
to
the
right.
J
We
were
more
concerned
about
when
we
transferred
to
the
harvested
area
where
the
c
carbon
biomass
base
look
like,
and
we
can
see.
There
are
large
differences
between
the
secondary
forest
increment
described
as
the
green
color
line,
showing
calculated
by
fades,
comparing
to
the
harvested
primary
force
area
described
by
the
ah2
datasets.
J
But
but
when
we
have
another
standard,
we
think
about
only
with
the
harvested
biomass
density.
Very
large
means
where
those
regions
have
more
current
being
harvested.
We
have
threshold
0.02
kilogram
company
square
per
year.
Then
we
will
see
those
area
with
most
of
the
current
being
harvested
have
a
very
equivalent
area
which
is
showing
that
the
large
difference
is
actually
mostly
caused
by
those
regions
with
relatively
sparse
vegetation,
carbon.
J
And
here
is
showing
a
harvest
depth
between
the
primary
and
the
secondary
forests.
The
harvest
step
is
can
be
explained
at
the
carbon
amount
not
harvested
successfully
due
to
less
biomass
than
the
harvest
rate
described
by
the
dataset
to
the
left
is
showing
the
primary
patches
and
to
the
right
is
showing
secondary.
As
we
can
see,
most
regions
have
a
certain
kind
of
shortages
of
the
of
the
carbon
available,
and
we
also
have
several
standard
of
discussing
the
causes
of
these
kind
of
shortages.
J
J
One
is
we
don't
have
any
cohorts
survived
for
those
who
sales
and
reason
number
two
have
low
growth
rate,
which
means
the
change
of
the
secondary
biomass
is
lower
than
the
harvest
rate,
as
we
calculated
and
third
third
reason
is:
we
have
very
tiny
area
of
the
secondary
forest
which
we
define
as
less
than
one
percent
of
the
brazil,
and
here
certain
sites
we
extracted
from
the
global
map.
The
first
site
says
places
near
sahara
desert
which
we
don't
have
any
cohort
survive.
This
is
showing
the
curve
of
the
spin-up.
J
As
you
can
see,
there's
certain
time
we
have
this
seafoam
grasses
and
also
evergreen
needle
leaf
evergreens
existed
for
certain
period,
but
as
spin-up
going
on
it
disappeared.
Finally,
so
we
need
to
think
about
how
we
can
maintain
certain
amount
of
cohort
for
those
those
kind
of
area
and
the
second
one
is
the
low
growth
rate
which
we
extracted
from
one's
site,
close
to
siberia
regions,
and
we
can
see
this
vegetation.
J
J
Oh
sorry,
so,
finally,
we
we
maintain
have
these
kind
of
challenges
through
this
work.
Right
now
we
have
a
mismatch
of
the
primary
force.
Distribution
and
also
the
secondary
forest
is
substantial
even
from
beginning
when
you
think
about
initialization
position,
and
we
have
to
think
about
different
traits
being
applied
when
we
regrow
these
signal
force.
Thank
you.
B
C
B
J
A
A
Maybe
the
high
level
comment
and
I'm
curious
in
the
kind
of
parameter
calibration
space
it
like
you,
have
to
have
the
right
biomass
to
get
the
right
carbon
to
get
the
right
amount
of
secondary
forest
in
a
grid
cell,
which
just
makes
it
really
tricky
to
understand
how
to
do
the
parameter
calibration
and
I'm
curious
where
land
use
change
falls
in
that
calibration
cascade.
Is
that
something
that
happens?
A
Does
land
use
change
happen?
After
does
it
come
in
like
after
you
do
the
the
the
kind
of
allocation
parameterization
that
comes
in
with
fixed
biogeography
or
those
two
things
happen
together
because
you
kind
of
have
to
get
yeah,
and
I
don't
even
know
how
to
ask
the
question
that
I'm
struck
by
the
complexity
of
the
problem
in
watching.
J
J
J
Yeah
there
has
certain
region
which
you
have
too
high
biomass,
but
the
health
is
rate
relatively
low,
so
you
can
only
have
very
tiny
area
to
become
secondary
force
patches
and
a
certain
small
area
of
the
secondary
force
patches
later
in
those
regions,
you
have
higher
secondary
force
hardest
rate,
and
then
you
cannot
fulfill
that.
This
is
one
very
big
question
I
have
found
right
now
in
this.
E
G
G
Get
that
going
well
and
then
fixed
biogeography
is
kind
of
when
we
bring
in
all
these
other
dynamics
and
yeah.
We
didn't
talk
about
disturbance,
but
obviously
disturbances
like
fire
and
land.
You
know,
logging
are
really
important,
so
I
think
that's
probably
going
to
be
an
iterative
process
where
we
bring
in
these
mortality
impacts.
G
E
J
Oh
so,
and
they
said
you
have
two
different
house
rate
for
the
secondary
forest,
secondary
forest
young
forest
and
secondary
mature
forest
right
now,
in
the
face,
we
have
our
definition
for
this
maturity,
which
rhinoprinter
by
default,
setting
to
90
years,
so
the
patches
survive,
the
secondary
forest
patches
survive
for
90
years.
There
will
be
mature,
secondary
forests
before
that
will
be
young,
secondary
forest,
and
right
now,
for
my
simulation,
I
said
that
to
40
years.
J
B
She
do
thanks
for
fielding
extra
questions,
so
I
contacted
chung
gang
and
he
is
trying
to
join.
We
do
have
a
break
scheduled
for
1005,
so
we
will
not
skip
our
break,
but
I
the
we've
done
development
on
the
fire
side,
and
so
it
was.
I
was
optimistic
to
let
you
all
hear
about
this,
but
I'm
not
sure
how
long
we'll
wait
so.
B
E
B
A
B
C
I
was
just
going
to
say
that,
because
luh2
actually
prescribes
both
the
carbon
amounts,
this
is
the
shinji.
It's
just
that
you
could
just
you
could
actually
actively
use
both
you
don't
need
to
use
just
one
or
the
other.
It's
like
you
could
combine
the
two
together
to
work
out
the
carbon
extraction
densities
and
ensure
that
you
get
the
secondary
forest
area
as
luh2
would
have
expected,
and
would
that
address
some
of
the
issues
he's
finding.
J
Oh
yeah
yeah,
I
I
thought
about
like
when
you
initialize,
you
use
the
because
I
I
know
h2
has
the
second
fourth
fraction,
so
you
you,
you
can
superimpose
that
that
data
set
in
there
first
add
the
area
and
then
use
the
carbon
amount.
That
has
the
harvest
rate
to
drive
the
model
yeah
you
certainly
can
can
get
those
be
being
consistent
to
each
other.
But
when
you
superimpose
the
area,
I
guess
you
don't
simulate
the
dynamic
process
when
you
grow
from
zero
to
the
the
grown
forest.
C
J
C
K
C
B
J
B
Yeah,
it's
an
important
part
of
you
know
how
that
needs
further
testing
yeah,
but
the
work
that
you've
done
demonstrating
it
globally
is
a
important
step
forward
for
fades.
I'm
going
to
pass
it
back
to
will
and
then
we
will
try
to
check
in
with
chung
gang
to
hear
about
live
fuel
moisture
so
we'll
take
it
away.
Thank
you.
A
That's
awesome
yeah,
so
we've
got
a
just
a
five
minute
bio
break
scheduled.
But
if
you
need
more
time,
please
take
it
kind
of
one
of
some
of
the
feedback
that
we
often
get
in
these
meetings
is
that
we
more
kind
of
informal
time
to
chat
would
be
nice
to
have,
and
so
we've
scheduled
20
minutes
in
here
to
have
virtual
coffee
break
rooms.
So
you
know
go
stretch.
Your
legs
have
a
little
break,
come
back
with
some
tea
or
coffee.
A
If
you'd
like
and
then
when
you
do
come
back
from
the
break,
you
can
just
jump
into
a
breakout
room.
Hopefully
everyone
will
be
able
to
move
themselves
there
and
just
you
know,
stop
in
and
chat
with
people.
So
if
you
see
someone,
you
want
to
talk
to
grab
a
big
grab
with
breakout
room
and
go,
go
chat
there
and
hopefully
roberta
folks
are
able
to
do
that
on
their
own
and
kind
of
move
about
freely.
So
but
then
we'll
come
back
at
10
30
mountain
time
for
paulie's
presentation.
A
C
G
A
C
I
A
A
A
A
C
A
A
L
And
these
pictures
here
are
showing
what
you
know
what
has
happened
in
the
forest.
It
went
from
pre-colonization
times
to
from
a
pine
dominated
low-density
surface
fire
regime
type
of
forest
and
there's
been
a
lot
of
infilling
density
has
increased
and
a
lot
of
that
increased
density
has
come
from
shade,
tolerant
species.
L
So,
thank
you
all
very
much,
and
I'd
also
like
to
say
that
the
reason
why
we're
doing
this
wasn't
we
didn't
start
out
to
do
an
attribution
study.
We
want
to
use
fate
so
we're
using
clm
fates
and
we
want
to
use
it
to
project
what
are
the
implications
of
management
that
might
happen
today
for
future
forest
change
and
fire
severity
and
fire
intensity.
L
But
to
do
that,
we
really
have
to
believe
that
the
model
can
actually
represent
this
forest
well
and
so
there's
really
two
prerequisites
for
so.
The
attribution
study
is
like
a
step
on
the
way
to
doing
that
and
the
prerequisites
for
the
attribution
study
are.
We
have
to
be
able
to
accurately
stimulate
the
pre-colonization
forest
structure
and
composition
and
the
fire
regime.
L
So
again,
we're
working
just
in
the
mixed
conifer
forest
of
the
sierra
nevada
and
pfts
are
are
defined
to
represent
three
fairly
different
ecological
strategies
and
they're.
They
differentiate
in
terms
of
shade
tolerance,
drought,
tolerance
and
fire
resistance.
We
have
a
sort
of
general
pine
type,
that
is
the
least
shade,
tolerant,
but
the
most
fire
resistant.
L
Instead
of
needing
to
go
through
the
different
complexity
modes,
that
adriana
was
talking
about.
The
strategies
we
used
would
be
much
more
challenging
at
a
global
scale,
but
here
they
were
so
for
the
pre-colonization
simulations
did
spin
up
with
early
1900s
climate
holding
co2
constant
at
285
parts
per
million,
then
checking
for
equilibrium
conditions,
which
I'm
also
not
going
to
talk
about
today
and
the
site.
L
L
So
here
we
use,
prescribed
ignitions
spatially,
explicit,
prescribed
ignitions
to
replicate
the
observed
area,
burn,
created,
a
harvest,
time
series,
an
area-based
harvest
based
on
harvest
records
and
running
with
transient
climate
and
co2.
Then
benchmarking,
the
regional
simulation
against
observed,
changes
in
species,
composition
and
density,
and
some
sanity
checks
on
water
use.
Efficiency
to
kind
of
gauge
is
the
co2
response
reasonable.
L
L
So
some
results.
This
top
figure
is
showing
over
time
the
change,
the
difference
from
the
baseline
in
percent
units
of
percent,
and
this
is
percent.
The
change
in
percent
pine
averaged
over
the
spatial
domain
for
each
year
and
the
bottom
is
showing
what
is
this?
The
average
of
this
time
series
just
to
more
easily
see
which
drivers
are
the
most
important
and
pretty?
L
L
L
E
Yeah
great
thanks
paulie,
if
you
yeah
you
stop
sharing,
and
so
if
there
are
questions,
if
people
want
to
ask
them,
I
see
that
there's
already
one
in
the
chat.
Do
you
have
some
hypothesis
about
why
co2
has
a
negative
effect
on
pine
and
jessica?
If
you
want
to
start
screen
sharing
while
paulie's
answering
that
question,
that
would
be.
L
I
I
haven't
had
much
time
to
look
at
this
yet,
but
I
I
think
it's
a
competitive.
I
don't.
I
don't
think
it's
that
co2
itself
is
is
hurting
pine.
I
think
that
it's
it's
it's
the
competitive
dynamics
with
cedar
and
fur
that
are
decreasing,
pine's,
competitive
ability,
but
that's
a
very
quick
early
answer
because
I
really
haven't
had
much
time
to
dig
into
it.
Yet.
L
Yeah,
so
am
I
again,
I
don't
know
and
but
there's
another
student
whose
name,
oh
jackie
or
lara,
if
you
can
help
me
with
who
is
really
starting
to
dig
into
the
snow
dynamics
in
in
the
some
of
these
simulations
as
well.
So
again
I
don't
know,
but
it
does
in
real
life.
So
hopefully
it's
playing
out
in
the
model
as
well.
E
M
So
the
rationale
for
this
work
was
that
crown
damage
is
a
really
important
process
in
tropical
forests.
It's
a
really
strong
predictor
of
mortality,
and
it
has
a
big
impact
on
the
carbon
cycling.
So
this
is
a
figure
from
a
paper
from
two
years
ago,
three
years
ago,
showing
structural
equation.
Modeling
that
found
that
crown
form
was
a
really
important
predictor
of
mortality,
and
then
this
paper
used
data
from
across
six
tropical
forests
and
found
that
crown
loss.
M
So
in
fates
we
we
implement
crown
damage
as
a
reduction
in
cohorts
crown
area
and
also
their
crown
biomass.
So
that's
shown
in
in
panel
a
here.
We
have
these
canopy
trees
shown
in
light
green.
They
have
this
reduction
in
their
crown
area
and
biomass
and,
as
a
result,
these
understory
trees
shown
in
the
dark
green
here
get
promoted
into
the
canopy
and
we
basically
split
up
cohorts.
M
So,
starting
with
this
first
question,
we
know
that
trees
can
survive
for
many
years
or
even
decades,
with
really
quite
severe
crown
loss,
and
these
altered
dolometries
could
change
canopy
structure,
light
levels
through
the
canopy
competitive
dynamics.
So
that's
what
we
were
trying
to
trying
to
address
with
this
first
question,
so
we
we
tested
this
with
four
configurations
of
fates.
M
We
do
also
see,
though,
that
there's
a
reduction
in
biomass
in
the
damage
only
simulations
and
similar
for
canopy
high
threshold.
So
this
is
the
minimum
size
that
a
cohort
would
need
to
be
to
be
in
the
canopy
and
again,
the
largest
decreases
come
from
these
simulations
that
have
additional
mortality.
M
So
we
then
looked
at
competitive
dynamics,
so
in
this
simulation
we
have
two
pfts
they're
identical
except
the
following
damage.
The
first
pft
will
allocate
all
of
its
available
carbon
to
regrowing,
its
crown
so
getting
back
on
to
regular
ollometries
and
the
second
pft
will
allocate
all
available
carbon
to
dbh
growth.
M
Okay.
So
how
does
damage
change
demographic
rates?
So
we
suspected
that
this
would
be
quite
sensitive
to
carbon
metabolism.
Damage
is
basically
reducing
the
photosynthetic
capacity
of
a
tree
whilst
leaving
respiration
of
the
stem
and
the
below
ground
tissues
intact.
And
so
you
can
imagine
that
at
some
point
as
if
the
crown
is
so
is
reduced
so
much
that
photosynthesis
can't
meet
respiratory
demand,
then
we
might
start
to
see
increases
in
mortality
from
carbon
starvation.
M
M
M
We
also
find
that
damage
leads
to
a
reduction
in
growth
rates,
so
this
is
crown
loss
against
diameter
growth
per
year.
The
the
green
here
are
observations
from
barrow
colorado
island,
and
then
we
have
the
the
low
root
respiration
configuration
of
fates
and
the
high
root
respiration
configuration
and
as
kind
of
as
expected,
the
high
root
respiration
configuration
of
phase
has
much
lower
growth
rates,
but
both
configurations
do
capture
that
reduction
in
growth,
with
increasing
crown
loss
that
we
see
in
the
data.
M
Okay,
so
this
is
really
the
first
kind
of
exploration
of
this
new
functionality.
I
think
there's
a
lot
more
that
we
can.
We
could
still
do
with
this,
and
one
thing
I'm
interested
in
is
how
how
disturbances
that
overlap
in
space
and
time
could
change
final
mortality
rates.
So
we
can
imagine
that
a
tree
that
is
damaged
in
one
disturbance
event
might
be
more
susceptible
to
mortality
in
following
disturbances,
and
we
don't
really
capture
that
at
the
moment.
M
The
effect
of
damage
on
stand
structure
was
mostly
due
to
altered
mortality,
but
we
do
find
that
damage
itself
decreased
above
ground
biomass
and
also
altered
the
canopy
height
structure,
pfts
that
allocate
to
regrowing
their
crown
were
able
to
out-compete
pfts
that
allocated
carbon
to
their
diameter
growth
damage
caused
decreases
in
growth
rates
and
increases
in
carbon
starvation
mortality,
but
these
results
were
very
sensitive
to
carbon
metabolism
and
future
work
will
really
delve
into
how
damage
influences
post-hurricane
forest
recovery
and
with
that
I'll,
just
thank
funders
for
their
support.
Thank
you.
E
E
B
M
A
M
M
When
we,
when
I
initially
ran
the
model
with
the
default
parameterization,
we
didn't
see
any
carbon
starvation
mortality,
which
was
a
little
unexpected,
so
we
were
kind
of
thinking
about
how
like
we
know
that
trees
do
sometimes
die
of
carbon
starvation
mortality,
so
kind
of.
Why
were
we
not
seeing
that
in
fates
and
what
were
the
kind
of
parameters
controlling
that.
C
E
M
Oh
yeah,
that's
that's
a
really
good
point.
So
in
this
implementation
we
we
left
kind
of
the
below
ground
elementaries
intact,
but
I
think
that
actually
the
data
does
suggest
that
if
a
tree
has
some
crown
damage,
it
would
actually
reduce
allocation
to
roots,
and
there
would
be
kind
of
we
probably
wouldn't
actually
see-
maybe
the
carbon
starvation
that
we
saw
in
the
model
if
we'd,
if
we'd,
had
that
reduction
in
biomass
following
following
crown
loss.
M
E
N
Yeah,
great
thanks
is
it
full
screen
just
to
check?
Yes,
it
is
great
okay,
perfect,
so
it's
mostly
presenting
the
work
of
a
postdoc
in
my
group.
His
name
is
julian
and
he's
been
working
on
canopy
interception
and
that's
been
supported
by
schmidt
future
and
also
by
our
recent
nsf
and
science
and
technology
center
called
leap
which
is
at
the
interface
between
artificial
intelligence
and
physics.
N
So
we
wanted
to
kind
of
revisit
that
and
trying
to
think
what
could
could
actually
happen
at
the
global
scale,
and
another
motivation
is
that
a
lot
of
the
kind
of
interception
models
are
actually
relatively
empirical
and
a
lot
of
them
are
actually
based
on
this
gash
model
from
back
in
the
90s
and
really
strongly
depend
on
knee
fairy
index
and
the
total
leaf
and
stem
area
that's
going
to
be
covered
by
vegetation
and
also
on
some
parameter,
which
is
the
storage
capacity.
So
how
much
you
can
think
of
that?
N
Maybe
what
if
we
could
actually
fit?
Let's
say
a
model
or
a
statistical
model,
away
from
a
precipitation
event
and
then
another
one
during
precipitation
event.
And
then,
if
we
were
to
take
the
difference
in
between
then
that
could
actually
encapsulate
basically
cannot
be
interception
that
could
also
account
for
fast
soil
evaporation.
But
overall,
that's
what
we
would
call
kind
of
interception
and
so
the
design
we
had.
We
took
a
rainy
period.
N
You
know
across
all
of
the
fluxnet
sites,
so
we
did
that
over
each
particular
site
and
at
an
hourly
time
scale
and
we
really
defined
two
models,
one
that
was
away
from
the
precipitation
event
so
further
than
12
hours
back
in
time.
And
then
we
expanding
the
expanded
that
model
back
into
the
period
when
we
had
rain.
And
then
we
looked
at
the
difference
between
the
model
with
rain
and
without
rain,
and
that
was
our
best
estimate
of
canopy
interception.
N
We
use
fluxnet
sites
across
the
globe
so
spanning
different
climates,
different,
functional
types.
We
aggregate
everything
to
the
hourly
time
scale.
We
removed
everything
that
was
below
zero
degrees
celsius
because
we
didn't
want
to
cover
snowfall,
because
we
know
that
it's
going
to
be
a
lot
more
time
dependent.
So
that's
something
we
wanted
to
avoid
and
and
then
we
used
the
life
air
index
from
others
as
a
driver
as
well,
and
when
we
look
at
the
training
and
and
test
set
for
those
two
different
models
with
rain
and
without
trains.
N
At
the
top,
you
have
the
rainless,
so
without
rain
experiment
at
the
bottom.
You
have
the
case
with
rain.
They
tend
to
do
a
very
good
job
on
the
training
data,
but
also
on
the
test
data,
so
excluding
points
where
we
are
trained
the
model,
so
we,
but
it
is
kind
of
built
confidence
into
the
quality
of
the
model,
and
we
tried
also
in
periods
when
we
really
away
from
the
from
the
precipitation
event
and
that
works
really
well
as
well.
N
A
sanity
check
that
we
had
was
to
check
that,
and
we
did
that
more
systematically
across
sites,
but
to
check
that.
Indeed,
when
you
have
the
model,
that's
trained
with
rain
you're,
actually
getting
more
mass
right,
more
water,
more
evaporation,
and
that's
basically,
what
is
shown
here.
So
the
dark
blue
here
is
the
model
with
rain
and
the
other
one
is
without
tray,
and
you
can
see
that
there's
a
systematic
bias
in
between,
which
is
what
we
expect
right.
N
N
So
I
don't
have
time
to
get
into
that,
but
we
are
actually
planting
from
memory
as
a
water
storage
component,
and
indeed
we
can
see
that
when
there's
rain
actually
there's
this
increase
in
evapotranspiration,
which
is
this
kind
of
interception
that
we
are
picking
up
even
when
we
we
have
rain
at
night
and
then
later
in
the
day
when
radiation
is
available,
that's
really
when
kind
of
the
interception
would
actually
help
you
and
then
we
started
investigating
the
drivers
of
that.
So
thinking
about
precipitation
during
the
rain
event.
N
N
N
And
then
we
could
come
up
with
global
estimates
of
that
and
we
compare
that
to
other
estimates
and
of
course
you
would
expect.
I
mean
there's
indeed
some
at
higher
level
index
your
and
higher
precipitation
dependence.
We
are
actually
seeing
that
increase
in
canopy
interception.
That's
what
you
would
expect,
but
it's
actually
instructive
to
look
at
the
ratio
between
kind
of
interception
and
precipitation
and
there
we
can
see
that
it's
not
just
lei.
N
N
We
can
derive
a
global
mean
upscaling
of
the
ratio
of
canopy
interception
to
mean
precipitation
and
it's
roughly
nine
percent,
and
that
can
be
used
as
a
constraint
across
climates
and
plant
functional
types
for
clm,
and
I
haven't
had
the
time
to
show
that.
But
we
also
looked
at
the
trend
and
kind
of
interception.
N
We
could
show
that
there's
been
quite
a
bit
of
decrease
because
rain
events
are
less
frequent
and
more
intense
and
we've
that's
been
reported
for
quite
some
time
since
2000
and
that
drives
a
global,
declining
kind
of
interception
and
the
ratio
of
our
precipitation.
So
with
that,
I'm
happy
to
take
any
questions.
E
N
E
N
C
Up
here
there
was
that
was
really
great,
and
it's
really
fun
to
sort
of
see
how
you're
bringing
this
these
ideas
of
like
machine
learning
into
informed
parameterizations.
Is
this
something
that
would
directly
go
into
a
model,
a
land
service
model
like
clm
or
because
you
were
actually
talking
about
it,
sort
of
physically
based,
and
you
would
actually
it's
consistent
with
the
penman
monteith
equation.
Or
would
you
just
use
this
to
sort
of
help
guide
and
tune
cln.
N
I
would
that's
a
great
question.
I
would
see
that
in
two
steps
to
be
pragmatic
I
would
say
first,
let's
use
that
as
a
tuning
and
that
would
using
the
existing
parameters,
so
that
would
be
one
way
to
actually
partition
things
potentially
better
and
then
the
second
step
is
because
I
didn't
talk
about
that,
but
we
are
basically
resolving
the
water
storage
capacity
of
the
canopy.
So
that's
something
we
could
actually
embed
in
the
model
as
well
and
that's
something
we
could
actually
resolve,
and
that
seems
relatively
simple,
so
yeah.
N
That
would
be
the
ultimate
plan
I
would
say,
but
we're
trying
to
dig
a
little
bit
more
into
the
physics
as
to
why
I
mean
I
didn't
talk
about
that
that
much,
but
wind
is
also
very
important,
because
that
can
shake
the
canopy
and
we
can
see
that
actually
some
experiments
with
accelerometers
as
well,
where
you
can
see
the
shaking
and
we
can
see
that
it's
also
decreasing
the
wave
operation,
so
there's
more
penetration
through
the
canopy.
So
there's
some
fun
like
very
non-linear
feedback.
E
O
O
So
this
is
really
a
collaborative
work
under
support
of
the
noaa
cpt
project.
So
as
we
know
that
the
snow,
albedo
and
snowpack
is
very
important
in
terms
of
it
can
regulate
the
land
service,
energy
and
water
balance,
especially
when
there
is
some
perturbation
in
the
snowpack
and
albedo,
it
will
trigger
the
positive
feedback,
as
you
can
see
here,
to
either
cooling
or
warming
the
land
banner
service.
So.
O
Here,
I'm
briefly
introduced
the
the
major
model
physics
included
in
the
current
climate
models,
including
the
cesm
ctsm
system.
So
first
we
have
these
snow
physics.
Physical
properties,
which
can
be,
can
be
used
to
interact
with
the
solar
radiation,
and
then
we
will
have
these
snow
relative
properties
computed.
And
finally,
we
can
calculate
the
snowpack
albedo
through
the
ready
transfer
scheme
and
specifically
for
the
snow,
physical
properties.
O
Once
we
have
the
the
snow
pure
snow,
air
beetle,
then
one
thing
that
can
a
complicated
picture
is
the
light,
absorbing
aerosols
like
dust
suit
or
black
carbon,
as
well
as
some
of
the
light
absorbing
organic
carbon
when
they
deposit
onto
the
snowpack.
They
can
reduce
the
snow
albedo
by
by
modifying
these
mixed
snow,
aerosol
mixture,
optical
properties
through
these
different
relative
refractive
index,
size,
distribution,
concentration
and
the
mixing
states.
O
This
is
what
we
we
currently
do
for
the
snowpack
arabido
calculations.
We
have
this
mi
calculations
through
the
lookup
tables,
to
reading
the
snow,
optical
properties
and
also
the
aerosol
optical
properties.
Remember
here,
the
crm
assumes
the
sphere
and
snow
shape
and
then
for
this
online
sneaker,
coupled
with
the
crm
ctsm.
O
We
also
have
the
snow
aging
process
to
alter
the
snow
green
size
which
further
affects
these
snow
optical
properties,
and
if
we
have
this
snow,
optical
and
and
optics
and
the
aeros
optics,
then
by
assuming
the
aerosols,
no
mixing
states,
then
we
can
calculate
the
mixture
optical
properties,
so
by
default.
Crm
assumes
the
external
mixing,
which
means
that
we
just
linearly
added,
add
all
these
snow
and
aerosol
optical
properties.
O
O
So
here
is
the
comprehensive
list
of
all
the
new
features
and
enhancements
listed
on
the
first
column,
the
left
column.
I
highlighted
the
the
key
updates
by
red
and
on
the
middle
column,
then
you
can
see
the
new
name
list,
controls
added
to
the
crm
ctsm
to
accommodate
these
new
features
and
enhancement
and
on
the
right
column.
The
last
column
shows
the
original,
oh
and
the
current
ctsm
snicker
name
list
or
features.
So.
First,
we
updated
the
ice
optical
properties
to
include
multiple
different
options
and
specifically
the
default.
O
Which
is
more
updated
compared
to
the
original
ones
and
more
cracked
based
on
the
recent
measurements.
We
also
updated
the
aerosol
optical
properties,
we
included
the
particularly.
We
include
multiple
dust
options
or
types,
so
the
default
one
is
set
to
the
sahara
dust,
which
is
in
the
middle
of
the
range,
and
this
is
also
updated
compared
to
the
older
one.
We
also
update
the
downward
solar
spectra,
which
includes
multiple,
like
six
different
downward
solar
spectras.
Originally,
we
only
have
this
middle
layer
to
the
winter.
O
O
And
also,
we
included
the
non-spherical
snow,
green
parameterizations,
the
black
carbon
or
suit
snow
internal
mixing,
that's
just
no
internal
mixing
schemes
and
for
the
snow
shape,
particularly,
we
included
four
different
representative
snow
shapes
and
we
set
the
hexagonal
plates
shape
to
be
default
compared
to
the
sphere
ones.
Based
on
the
observations
and
finally,
we
also
include
a
hyper
spectral
capability,
which
includes
the
10
nanometer
spectral
resolution
kit
for
the
calculations.
O
So
in
the
next
few
slides
I'm
going
to
show
you
the
sensitivity
of
by
using
these
new
features
and
enhancements
in
a
global
setup
for
the
ctsm
crm
simulations
here
is
the
the
the
brief
setup
for
these
global
simulations.
We
are
doing
this
for
one
degree
and
use
this
sp
setup
and
the
first
five
years
for
spinning
up
in
the
last
five
years
for
analysis
so
quickly
here,
I'm
showing
this
new
updated
ice
reflecting
index
or
optics
compared
to
the
old
one
shown
as
this
blue
line
here.
O
So
essentially,
we
have
a
stronger
ice,
visible
absorption
and
there-
and
hence
we
have
this
in
slightly
lower
albedo
in
the
new
ice
objects,
and
here
I'm
showing
you
the
global
sensitivities.
So
as
you
can
see
the
major
difference
I
on
the
are
in
the
high
latitudes,
but
it's
it's
very
small,
it's
less
than
0.01.
O
So
but
but
we
are
going
to
go
in
towards
the
correct
direction
or
accurate
directions
and
then
the
next
one
is
to
show
you
the
impact
of
these
aerosol
optics.
For
for
this
is
not
a
real
difference.
So
let
me
see
so
here,
I'm
showing
you,
the
r
sky,
direct
and
diffuse
albedo
difference
by
using
the
new
simulation,
a
new
data
set
minus
the
old
data
set
in
the
simulation.
O
As
we
can
see
here.
First,
you
notice
these
lower
snow
orbital
over
most
of
the
northern
hemisphere.
This
is
caused
by
the
stronger
aerosol
absorption
due
to
the
updated
aerosol
optics,
and
then
we
have
this
lower
snow
albedo
and
here
this
map,
showing
you
the
enhanced,
aerosol
albedo
reduction
due
to
the
updated
optics
and
then
what
what's
interesting
is
that
this
very
high
sensitivity
in
the
high
latitudes
caused
by
these
updated
downward
solar
spectral
weights.
We
are
still
using
the
mid
layer
to
the
winter
as
the
old
one
or
default
crm,
but
slightly
updated.
O
The
fraction
of
the
different
bands
shown
as
here
the
blue
one
is
the
other
one.
The
old
spectra
and
the
red
one
is
a
new
spectrum.
Even
for
these
tiny,
this
tiny
difference
there
you
can
see
here
we
have
up
to
point
0.002
or
0.03
difference
in
the
albedo,
oh
and
here
to
confirm
our
explanation.
We
also
conducted
the
standalone
version
of
the
snicker
simulations
by
using
the
different
spectra,
and
you
do
see
this
difference
in
this
diffuse
snow
epidural.
O
O
Okay,
thank
you
so,
and
or
we
also
have
enhanced
organic
carbon
but
reduced
black
carbon
relative
forcing
so
next,
I'm
showing
you
the
sensitivity
to
the
multiple
different,
multiple
dust
types
due
to
the
there
are
different
optics.
As
you
can
see
here,
they
have
quite
different
absorption
for
the
light
for
solar
radiation
and
there's
no
a
bit
of
difference
between
the
green
land
dust
and
the
colorado
dust
as
shown
here.
So
they
are
there.
O
They
do
show
the
noticeable
and
important
difference
in
the
erbil
reduction
caused
by
dust
over
almost
all
the
not
northern
hemisphere.
So
another
one
is
the
effect
of
the
updated,
more
accurate,
ready,
transfer
server.
As
you
can
see
here,
the
old
server
shows
very
high
over
estimates
as
an
ir
band,
but
the
new
server
substantially
reduced.
These
overestimates,
especially
for
the
irbns,
as
well
as
some
of
the
visible
bins.
O
O
Okay,
so,
oh,
I
would
very
quickly
show
you
the
impact
of
the
higher
snow
habit
for
the
non-spherical
snow
or
because,
if
we
assume
the
different
shapes
for
the
snow
greens
non-spherical
shapes,
then
we
will
have
higher
snow
albedo
due
to
their
optical
properties.
Difference,
and
here
I'm
showing
you
the
snow,
a
bit
of
difference
over
the
globe
for
our
sky
situation.
You
can
see
it.
This
airbuild
difference
can
be
up
to
0.05.
O
E
Yeah
thanks
and
then,
if
anybody
has
questions
for
someone,
please
put
them
in
the
chat.
Tell
them
if
you
can
stop
sharing
next,
we
have
yellen
jam
yeah.
So
go
ahead
and
looks
like
we're
just
seeing
your
desktop
background
there
you
go
yeah
and
then
remember
to
unmute
yourself
before
you
start
talking.
P
Okay,
thank
you.
Firstly,
thanks
ankar
for
the
opportunity
to
share
our
current
studies
and
I'm
john
phd
ph.d
student
from
university
of
connecticut.
This
study
is
advised
by
dr
guriwa,
and
my
topic
today
is
a
new
approach
to
soil
initialization
for
studying,
subsidiary
and
atmospheric
interactions.
P
So
temperature
also
plays
a
very
important
role
in
the
land
atmosphere
interactions.
In
order
to
better
understand
the
law
of
the
non-atmospheric
interactions,
tons
of
sensitivity
tests
have
been
carried
out
during
the
past
20
years.
Previous
studies,
they
typically
firstly
identify
a
rectangle
region
as
a
targeted
area,
then
impose
the
soil
anomalies
to
the
targeted
region
only
and
then
for
the
soil
conditions
elsewhere,
they
are
still
the
same
as
control.
P
They
have
two
disadvantages
for
the
conventional
approach.
First,
it
will
create
sharp
contrast
across
the
boundary
of
the
targeted
areas
for
the
soil
initializations,
and
then
it
will
result
in
unrealistic,
special
patterns
of
soil
condition.
That
is,
contamination
scale,
for
example,
taking
the
conversion
approach,
this
figure
shows
a
special
pattern
of
soil
temperature
in
response
to
the
warming
in
the
targeted
area,
and
we
can
see
the
surrounding
areas.
They
have
the
cooling
signal,
while
the
targeting
area
area
has
a
warming
signal,
that's
not
realistic
in
realistic.
P
The
soil
temperature
is
characterized
with
a
certain
degree
of
special
correlations.
We
can
see
in
the
regression
coefficient
map.
The
surrounding
regions
here
should
follow
a
certain
degree
of
of
change
in
response
to
the
warming
in
the
targeted
areas
to
account
for
the
realistic,
a
realistic,
special
correlation
of
soil
conditions,
we
propose
a
new
approach
to
soil
initializations.
P
We
name
it
as
a
slope
approach.
Here
we
use
a
soil
temperature
as
an
example.
Firstly,
we
need
to
calculate
the
specialty
average,
also
a
temperature
over
the
targeted
region,
for
example
in
odessa
region
and
then
derive
the
linear
regression
slope,
as
I
j
between
the
spatial
average
and
the
third
temperature
at
every
grid.
Point
everywhere
in
south
america
then
derive
the
temperature
anomaly
over
the
target
regions.
With
the
other
t
equal
to
t
emission
minus
control
for
each
grid.
P
Point
else
are
we
derive
the
soil
temperature
anomalies
with
delta
tij
equal
to
sigma
delta.
T
here
is
to
constructed
to
the
initial
condition
for
the
soil
temperature
we
can
see
here.
These
two
figures
are
all
realistic:
soy
nationalization
with
spatial
continuing
content
continued,
and
these
two
figures
show
the
similar
special
patterns
for
the
soil
initialization
targeting
different
regions.
P
P
After
that,
we
have
multiple
wall
monster,
simulations
hot
cool,
wet
and
dry
experiments.
They
employed
the
slow
approach
with
different
soil
visualizations,
for
example,
on
the
hot
experiment,
is
initialized
with
99
percent
soil
temperature
and
for
the
soil
moisture
for
this
experiment.
It
directly
comes
through
on
control.
P
P
P
P
we
can
see
the
soil
memory
for
all
moisture
is
almost
one
month
and
for
the
temperatures
you
can
see
like
10
days
within
the
targeted
region.
The
temperature
availability
in
soil
anomaly
are
qualitatively
similar
between
two
soil
naturalization
approaches.
Just
the
signal
of
the
slope
approach.
Here
we
can
see
the
red
line
a
little
bit
stronger
than
the
blue
line
and
in
the
conventional
approach
we
compared
the
simulated
temperature
from
the
slope
and
the
conversion
approach.
P
In
addition,
we
compare
the
special
patterns
with
the
head
pattern,
from
the
loan
simulation
and
for
era
5.
For
on
the
analysis,
we
can
see
the
warming
signal
over
both
the
targeted
region,
and
this
is
running
in
both
loan
simulation
and
for
eri
five,
we
analyzes
and
also
unlike
the
commission
approach.
There
is
no
sharp
contrast
for
our
slope
approach.
P
Here
are
also
the
similarity
between
experimentation
in
different
regions
means
slow
approach,
but
that's
for
the
variable
soil
moisture.
You
can
see
the
symmetry
in
soil,
moisture,
initialization
the
qualitatively
as
imaging
soil,
moisture
length
and
for
the
similarity
in
hydrometer
responses
to
the
soil,
moisture
change.
P
In
summary,
we
proposed
a
new
approach
to
soil
initialization
for
studied
subsidiary
knowledge,
atmosphere,
interaction,
the
slow
approach
is
superior
to
the
conventional
proteins
three
perspectives.
First,
it
has
the
realistic
spatial
patterns
of
the
soil,
temperature
and
moisture,
and
it
will
not
create
the
sharp
contrast
across
the
boundary
of
the
area
and
it
can
increase
the
computation
efficient.
P
E
E
Q
Super
duper
yeah,
so
this
should
be
probably
fairly
short.
It
was
just
a
neat
little
result
that
I
stumbled
across
while
playing
with
the
model,
and
this
is
looking
at
how
reductions
in
terrestrial
terrestrial
evaporation
actually
lead
to
increased
water
vapor
over
land.
This
is
all
sort
of
for
the
modern,
looking
earth.
Q
Q
And
this
is
you
know.
This
is
something
that
our
models
have
been
able
to
capture
for
a
very
long
time.
So
this
is
a
figure
from
a
paper
in
1982
where
they
have
on
the
top
panel,
a
planet
with
really
wet
soils
on
the
land
surface
and
the
bottom
panel
is
really
dry.
Soils
and
the
contours
are
showing
you
july
surface
temperatures.
Q
So
really
all
you
see
from
this
is
that
the
contours
are,
you
know
much
more
spaced
out
and
much
less
high
numbers
when
you
have
really
wet
soils,
and
then
the
land
gets
really
hot.
When
you
have
your
dry
soils,
so
less
evaporation
means
hotter
land
and
everyone's
kind
of
on
board
with
that
for
for
the
modern
earth.
Q
But
what
I
really
wanted
to
dig
into
here
was
looking
at
how
reductions
in
land
evaporation
actually
change
atmospheric
water
vapor-
and
this
is
an
important
thing
to
understand,
because
we
care
about
how
changes
in
water
vapor
impact
things
like
relative,
humidity
and
vpd.
These
matter
for
vegetation
on
the
land.
Surface
water
vapor
is
a
precursor
to
cloud
formation
and
precipitation,
and
it's
also
a
great
greenhouse
gas.
Q
So
it
has
a
lot
of
implications
for
sort
of
surface
and
planetary
climate
and
there's
reasons
you
can
imagine
reducing
land
evaporation
could
alter
atmospheric
water
vapor
in
both
directions.
So
you
might
expect
reducing
evaporation
from
land
to
lead
to
less
water
vapor
over
the
continents,
because
you're
literally
putting
less
water
from
the
land
up
into
the
atmosphere,
but
you
could
also
potentially
expect
it
to
lead
to
increases
in
water
vapor,
because
the
vapor
pressure
of
water
increases
as
temperature
increases.
Q
So
what
I'm
going
to
show
you
are
the
results
of
some
really
end
member
simulations
of
possible
land
evaporation.
So
we're
going
to
look
at
one
that
I'm
calling
desert
land
here
and
this
is
sort
of
making
land
always
dry
everywhere.
So
you
can
imagine
this
as
land
that
is
just
really
really
efficient
at
draining.
So
like
a
gravel
land
surface
that
as
soon
as
the
water
rains
onto
it,
it
immediately
drains
away
and
goes
off
into
runoff.
Q
So
you
can't
really
have
any
water
available
to
evaporate
from
the
land
surface
and
we
can
compare
that
to
the
other
extreme
which
I'm
going
to
call
swap
land,
which
is
where
we're
forcing
the
land
to
be
always
saturated
with
water
everywhere,
regardless
of
how
much
precipitation
it
gets
regardless.
If
it's
topography-
and
you
can
imagine
this
as
a
planet
that
has
sort
of
a
super
fancy-
in-ground
irrigation
system
and
we're
doing
these
simulations
with
cesm.
Q
So
to
start
off,
we
can
just
look
at
what
the
change
in
latent
heat
flux,
so
evaporation
was
at
the
land
surface,
and
this
is
the
difference
between
our
desert
and
our
swamp
simulation.
Now
the
brown
colors
are
showing
you
reductions
in
evaporation
from
the
surface,
so
brown
over
land
is
good.
It
means
we
set
up
our
simulations
right,
so
we're
turning
off
terrestrial
evaporation
and
we
see
that
we
do
get
increases
in
evaporation
over
the
ocean
regions.
Q
But
then,
what's
kind
of
neat
and
what's
surprising
when
I
first
looked
at
it,
was
that
we
get
these
really
big
increases
in
total
atmospheric
water
vapor
over
most
regions
of
the
planet,
basically
everywhere,
except
the
subsiding
branches
of
the
hadley
cell,
and
that's
what
these
sort
of
teal
colors
are
showing
you
on
the
panel
on
the
right.
So
this
is
the
the
column
integrated
water
vapor,
so
the
water
vapor
at
every
level
stacked
up
and
while
I'm
showing
you
the
column,
integrated
values
here.
Q
What's
kind
of
neat
is
that
the
water
vapor
here
is
increasing
not
just
near
the
surface
or
just
a
loft.
It's
really
increasing
throughout
the
whole
column,
which
I'll
show
you
later.
So
what
that
means
is
if
we
turn
off
evaporation
from
the
land
surface,
you
get
an
atmosphere
with
more
water
vapor
in
it
over
the
land.
Q
So
that's
kind
of
weird
the
increased
moisture,
though
it
basically
has
to
be
coming
from
the
ocean,
because
we
turned
off
evaporation
from
the
land
surface.
But
what's
neat
is
that
it's
reaching
right
into
the
middle
regions
of
the
continent.
It's
not
just
increasing
sort
of
at
the
coasts
where
you
might
expect
it
to
be
able
to
infect
easily
in
so
there's
two
potential
pathways
that
immediately
come
to
mind
for
how
more
water
vapor
is
ending
up
over
our
continental
regions.
The
first
option
is
just
that,
because
we
turned
off
land
evaporation.
Q
Our
planet
is
warmer
like
we
would
expect.
So
if
we
have
a
generally
warmer
planet
that
might
lead
to
more
evaporation
from
the
ocean
and
then
we
would
have
higher
water,
vapor
and
sort
of
our
free
troposphere
over
the
ocean
regions
which
might
just
be
infecting
over
the
continents,
and
if
that
was
hap.
If
that
was
the
main
driver
of
what
was
happening,
we
would
expect
still
our
air
that's
sort
of
near
the
surface
over
the
continents
to
be
fairly
dry
and
then
for
the
air
higher
up
in
the
troposphere
to
be
anomalously
wet.
Q
The
second
option
is
that,
as
we
heat
our
land
surface,
we
have
this
change
in
circulation,
where
we've
got
more
uplift
over
the
land
surface
and
that's
pulling
moist
air
from
the
oceans
onto
the
continents.
And
if
that
was
the
case,
we
would
expect
more
moisture
near
the
surface
over
the
continents,
especially
near
the
coasts,
but
maybe
not
necessarily
aloft,
and
it
would
also
be
kind
of
hard
to
drag
it
all
the
way
into
the
middle
of
say,
like
eurasia.
Q
But
if
we
just
look
at
an
average
cross
section
of
what's
happening
to
specific
humidity.
So
this
is
like
literally
the
grams
of
water
in
the
air.
At
every
level.
This
is
a
zotal
mean
just
over
land
regions
from
the
south
pole
on
the
left
up
to
the
north
pole
on
the
right
going
from
the
surface
up
to
basically
the
tropopause.
Q
What
we
see
is
that
most
regions
are
green
except
again
right
in
those
sort
of
subsiding
branches
of
the
hadley
cell
near
the
surface.
That
means
that
we're
getting
more
water
vapor
in
our
atmosphere
throughout
pretty
much
the
whole
column,
which
means
we're
probably
getting
a
combination
of
both
of
these.
These
options,
where
we've
just
got
more
water
vapor
in
the
atmosphere
from
the
ocean,
but
we're
also
getting
these
changes
in
circulation
that
are
leading
to
an
increase
in
moisture
near
the
surface.
Q
It's
important
to
note,
though,
that,
even
though
we
have
literally
more
water
in
the
atmosphere,
we're
still
effectively
getting
drying
over
our
land
surface,
so
the
left
panel
now
is
showing
you.
The
change
in
relative
humidity,
just
overland
regions
sonali
averaged
and
in
particular,
what's
important
is
that
we
do
get
these
really
strong
red
colors
so
decreases
like
20
to
50
percent
in
the
relative
humidity
over
the
land
surface
in
the
lower
troposphere
when
we
turn
off
evaporation.
Q
But
if
you
can
see
them,
they're
sort
of
showing
anomalous
flow
of
air
onto
the
continents
near
the
surface
and
specifically,
if
we
just
look
at
what's
happening
in
the
equator,
because
it
tells
a
pretty
tidy
story
and
we
just
look
at
a
cross-section
of
that
anomalous
vertical
motion.
What
we
see
is
that
we've
got
this
really
strong
anomalous
vertical
motion
over
the
continents,
that's
being
driven
by
the
land
getting
really
hot.
So
it's
basically
dry
convection.
Q
Here,
that's
driving
this
as
opposed
to
sort
of
moist
conduction
that
we
typically
think
of
as
happening
in
the
tropics.
So
it's
this
anomalous
heating
at
the
surface
that
drives
really
strong
upwards
motion
that
pulls
in
moist
air
from
the
continents,
and
it
just
shoots
it
all.
The
way
up
through
the
full
depth
of
the
troposphere.
Q
But
what's
kind
of
neat
is
that
only
part
of
this
surface
heating
and
this
anomalous
upwards
motion
is
driven
by
the
warming?
That's
directly
due
to
less
evaporation
and
to
tease
that
apart,
we
can
do
something
called
a
radiative
kernel,
decomposition
which
this
bar
graph
is
showing
you,
and
this
bar
graph
has
a
ton
going
on,
so
you
can
ignore
most
of
it,
but
it's
basically
a
way
of
saying
how
much
extra
energy
do
we
get
into
the
surface
the
positive
values
as
a
result
of
shifts
in
evaporation?
Q
That's
the
first
one
or
changes
in
water,
vapor
clouds,
temperatures
so
on.
Really
all
you
need
to
care
about
are
two
pieces
of
this:
the
fact
that
less
evaporation
means
more
energy
to
heat
the
surface.
That's
this
middle
gray
bar
and
the
fact
that
less
evaporation
also
means
much
less
cloud
cover
over
the
land
surface.
That
means
more
short
wave
radiation
into
the
land
surface
and
that
leads
to
actually
even
more
heating
of
the
land
than
that
initial
reduction
in
evaporation
did
on
its
own.
Q
So
really
all
I
did
here
was
we
looked
at
how
suppressing
land
evaporation
can
actually
drive
increases
in
water
vapor
over
both
land
and
ocean
regions,
and
it's
because
this
heating
of
the
land
surface
induces
an
anomalous
atmospheric
circulation
that
brings
in
water
from
the
ocean,
but
only
part
of
that
heating
is
due
to
having
a
hotter
surface.
The
rest
of
it
is
actually
due
to
reductions
in
cloud
cover.
Q
E
Q
No,
you
didn't
miss
it.
I
didn't
show
it
in
the
interest
of
time.
You
get
huge
reductions,
your
participation
over
most
land
areas,
with
the
exception
of
the
tropics,
where
you
get
increases
along
the
west
coast
of
south
america,
not
over
the
amazon,
because
we've
effectively
killed
precipitation
recycling
in
our
desert
simulation
big
increases
in
tropical
africa
and
big
increases
over
the
maritime
continent,
but
outside
of
the
tropics,
you
get
big
reductions
in
rainfall.
C
Q
Yeah,
I've
definitely
considered
doing
it.
I
have
not
explicitly
done
it
for
the
southwest
us.
What
I'm
really
curious
in
is,
if
you
sort
of
incrementally,
did
this
reduction
in
evaporation
if
there
is
a
place
where
it
like
peaks
or
shifts
direction
like
it's
not
obvious
to
me
that
I
think
initially,
you
could
reduce
evaporation
that
might
lead
to
less
water
vapor
and
only
once
you've
reduced
it.
So
much
that
you
can
drive
this
massive
anomalous
circulation.
E
D
D
D
Okay,
land
use
change
is
increasing.
So
what
you
see
in
the
left-hand
side
is
the
crop
percentage
increase
in
the
present
climate
in
the
top
and
the
future
climate
in
the
bottom
and
the
time
series
plot
that
is
shown
on
the
right
hand,
side
shows
the
land
you
change
or
the
percentage
cropland
in
the
in
the
two
scenarios.
So
in
the
historical
case,
both
are
the
same,
but
as
we
go
into
the
future
climate,
most
of
the
land
use
chains
are
taking
place
in
the
in
the
middle.
D
D
There
have
been
some
literature
in
terms
of
how
the
land
use
change
impact
can
change
in
the
future
climate,
and
we
have
heard
about
reduction
in
the
snow
cover
and
that
can
impact
the
land
use
change
change
impact
on
climate.
There
is
a
co2
physiological
response.
We
heard
about
the
yeah,
feeding
back
to
the
cloud
cover
and
precipitation,
and
also
due
to
dynamic
vegetation
changes.
D
So
there
have
been
individual
study
which
have
used
the
individual
climate
model
and
to
come
up
with
with
a
slightly
different
answer,
and
here
the
goal
was
to
to
analyze
the
land.
You
change
using
the
new
set
of
the
experiment
and
in
terms
of
the
multi-modal
assessment
and
see
if
you
can
get
some
consistent
response
and
also
we
would
separate
the
local
impact
from
the
total
impact
by
comparing
the
climate
change
impacting
the
two
neighboring
region
where,
in
one
region
there
is
a
it
has
seen
the
land
you
change
in
other
region.
D
So
in
here
is
the
multimodal
mean
from
the
six
sim
f6
climate
model.
In
the
leftmost
column,
you
see
the
total
impact,
that's
basically
the
difference
between
all
forcing
run
and
then
no
algorithm,
and
we
see
the
widespread
cooling
impact
in
the
in
the
at
least
in
the
middle
latitude
region,
due
to
a
landing
change
in
the
historical
climate.
D
It
occurs
both
in
terms
of
the
annual,
as
well
as
the
jj
season
when
we
separated
the
local
impact,
that
is
in
the
middle
column.
That
also
saw
a
cooling
impact
due
to
land
use
change
in
the
in
the
middle
latitude
and
also
parts
of
the
tropical
region
in
the
future
climate.
We
cannot
do
the
total
impact,
because
there
is
not
much
land,
you
change
between
rcp
8.5
and
then
ssp
555
and
ssp
126.
So
we
did
only
local
impact
in
the
future
climate
and
what
we
find
that
it's.
D
It's
converting
into
from
the
cooling
impact
to
a
overall
warming
impact
in
the
mid
latitude
region.
If
you
see
the
yeah
right
most
column
here
and
then
in
in
the
tropical
region,
like
india,
the
cooling
impact
get
strengthened
in
the
future
climate.
So
we
wanted
to
understand
why
it
is
changing
the
sign
from
the
historical
climate
to
the
future
climate.
D
D
You,
you
see
the
total
impact.
You
see
historical
local
impact.
I
think
the
main
story
is
yeah
when
we
go
from
the
historical
local
to
the
future.
Local
impact
yeah.
Most
of
the
models
are
a
consistent
increase
in
the
temperature
for
north
america
and
eurasia
and
consistent
decrease
in
temperature.
For
for
the
tropical
region,
like
india
and
across
both
annual
as
well
like
the
jj.
D
We
we
investigated
in
in
terms
of
its
driver,
one
is
related
to
how
much
albedo
is
changing
due
to
land
use
change
and
that
mostly
follows
the
land.
You
change
trade
trajectory.
That
means,
if,
if
land
use
change
is
not
increasing,
you
would
not
see
much
increase
in
the
albedo,
whereas
land
you
change,
increases
in
india.
So
you
see
the
increase
in
the
albedo.
D
However,
the
another
factor
that
comes
into
the
play
in
the
middle
attitude
is
the
soil,
moisture
drawing
that
is
shown
in
the
bottom
row
and
so
almost
right
out
in
both
north
america
and
eurasia
and
euchange
region,
and
it
increases
in
the
india
land.
You
change
region,
and
that
is
mainly
due
to
increase
in
the
precipitation
in
india
region.
D
So
we
use
a
simple
equation
to
understand
different
drivers.
The
dte
is
the
total
evaporative
cooling
or
the
warming.
If
the
small
moisture
dries
and
then
there
are
some
constant
terms
that
we
derived
using
the
multi-model
mean
value
in
in
out
of
six
climate
model
and
then
acid,
the
small
moisture.
So
one
thing
we
suspected
that
the
salmon
acid
is
coming
in
the
denominator.
So
if,
if
it
is
decreasing,
then
it
might
have
a
increasing
impact
on
on
the
temperature.
D
D
We
also
replace
v,
that
is,
the
vapor
pressure
deficit
using
the
historical
climate
value,
and
it
has
some
impact,
but
its
impact
was
not
as
large
as
as
the
ds
impact
and
yeah.
Initially,
we
started
with
say
assumption
that
this
is
s
that
is
causing
more
of
the
warming,
but
in
this
formula
it
was
the
ds
that
was
causing
most
of
the
warming
impact
in
this
future
climate.
D
D
E
D
C
Sanjeev
nice,
I
feel,
like
I
asked
the
same
question
last
time.
This
ball
asked
again
because
I
don't
remember
the
answer.
You
know
if
you're,
if
you're
looking
at
like,
say
the
north
america,
one
and
you're
saying
that
the
impact
of
land
use
change
is
reducing
right,
less
of
a
cooling.
So
are
you
talking
about?
C
E
R
So
in
this
discussion,
I'll
give
a
brief
introduction
to
the
background
and
motivation
of
the
project
and
then
go
into
a
little
bit
of
technology.
Details
about
the
implementation
of
the
transient
urban
scheme,
followed
by
a
recent
set
of
simulation,
some
results.
There
are
satisfied
ways
and
some
next
steps.
R
So
first
some
background,
ctsm
or
crm
has
been
a
very
powerful
tool
for
land
simulation,
and
one
highlight
we
found
especially
important.
Is
it's
a
transient
land
use
mechanism
where
some
land
use
types,
including
natural
vegetation,
including
crops,
could
change
throughout
the
simulation.
This
allows
for
a
more
flexible
and
hopefully,
more
accurate
representation
of
the
landscape.
R
R
We
know
urban
has
a
relatively
small
area,
however,
with
the
rapid
urbanization
in
the
future.
The
combined
effect
of
the
addition
of
impervious
urban
surface
and
the
loss
of
the
natural
or
vegetated
surface
could
combine
and
actually
make
a
difference
in
terms
of
the
regional
climate
and
more.
We
also
want
to
model
how
the
global
climate
actually
impacts
the
more
and
more
urbanized
cities
and
how
we
can
deal
with
that
with
mitigation
adaptation
strategies,
and
none
of
those
can
be
studied
if
the
urbanization
level
is
kept
static,
as
was
the
case
prior
to
our
implementation.
R
R
However,
the
model
doesn't
take
that
data
format.
It
would
rather
divide
urban
into
three
categories,
so
tall
building,
districts,
high
density
and
medium
density
and
each
density
type
will
have
different
properties.
So
we
need
to
manually
assign
density
types
to
these
urban
regions
and
we
also
had
to
reconcile
some
urban
urban
groups
with
other
land
use
change.
R
This
is
very
straightforward,
but
the
only
one
issue
here
is
when
urban
first
appear
in
a
great
cell.
We
need
to
initialize
it
so
that
it
can.
Some
of
the
state
variables
can
start
from
a
relatively
realistic
condition,
so
to
simplify
things.
Initially,
we
we
try
to
initialize
urban
everywhere,
but
that
would
cause
a
burden
in
performance
and
memory.
R
So
later
we
have
an
improved
approach
where,
when
we
created
the
land
use
time
zero
file
in
the
previous
step,
we
add
a
variable
called
percentage
of
urban
max
which
corresponds
to
the
maximum
of
urban
fractions
throughout
the
whole
simulation
for
a
given
grid
cell.
So
with
that
variable
we
can
tell
where
urban
exists
or
will
grow,
and
we
can
initialize
urban
only
for
those
places.
R
R
R
So
and
with
some
treatment
for
the
carbon
for
missing,
we
also
ensure
the
same
for
mass
conservation,
so
we
we
ensure
conservation
of
mass
and
energy
here
and
finally,
we
made
this
game
for
the
community
and
we
understand
there
are
multiple
excellent
urbanization
projection
data
set
and
we
want
to
make
it
as
simple
as
possible
for
other
users
to
supply
their
own
urbanization
data.
When
we
started
this
project,
we
compiled
the
data
set
all
manually,
but
later
we
found
out
there's
a
existing
infrastructure
called
make
surf
data
map
tool
with
some
modification.
R
R
So
that's
a
quick
summary
of
the
workflow
and
we
recently
tested
the
implementation
with
some
simple
simulation
and
I'd
like
to
share
with
you
some
some
simple
results.
So
first
the
setup
of
the
simulation,
it's
also
very
straightforward.
We
use
a
pair
of
eye
mode
simulation,
so
land
only
simulations
and
the
only
difference
between
the
two
cases
would
be
with
or
without
dynamic
urban
scheme.
R
To
clarify
we
use
the
new
code
after
modification
for
both
cases
and
for
the
static
urban
we
just
fit
the
the
model
with
a
land
use
time
series
that
has
that
urban
won't
change
throughout
the
simulation
and
for
the
dynamic
urban
case.
It
of
course
updates
annually,
and
here
is
a
quick
virtualization
of
dynamic
urban
and
the
static
curve.
R
This
is
a
region
near
new
york
and
it
experienced
the
rapid
organization
and
in
this
center
plot,
the
urban
fraction
is
shown
as
the
the
size
of
the
dots
and
the
urban
two
meter
daily.
Average
temperature
is
the
color
of
the
dot
and
with
the
color
bar
on
the
right.
So
we
generally
see
a
warming
trend
with
urban
sprawl,
which
is
expected,
and
the
vertical
axis
of
those
dots
is
the
difference
between
dynamic
urban
and
static
herb,
so
in
terms
of
the
great
cell
average
surface
temperature.
R
They
are
not
fully
coupled
since
only
the
land
component
is
fully
active,
so
we
must
acknowledge
that
we
won't
be
able
to
capture
any
two-way
interaction
between
land
and
atmosphere,
and
the
urbanization
effect
we
show
here
is
rather
limited,
and
that
leads
to
our
conclusion.
Next
steps,
so
simply
so
in
conclusion:
the
we
implemented
a
transient
urban
scheme
and
it's
tested
to
work
great,
and
we
believe
it
has
some
great
potential
for
further
investigation
and
for
next
steps.
R
Yeah
yeah
and
also
carl
jung
from
el
university.
She
contributed
a
lot
and
she's
currently
working
on
the
fully
couple
simulation
and
we
hope
this
presentation
is
of
some
help
to
anyone
who
is
interested
in
this
new
feature
and
please
let
us
know
if
anything
can
be
improved
and
thank
you
for
your
time.
E
R
Currently,
the
the
urban
part
of
the
model
is
static,
it's
not
dynamic,
but
it's
already.
The
infrastructure
is
there
and
from
what's
already
there,
the
urban
is
divided
into
three
density
types.
According
to
a
data
set
coming
with
the
model,
I
believe
it's
supplied
by
jack's
network
and
we
just
we
just
base
our
base,
our
urbanization
data
set
on
that
existing
model,
and
we
will
we
will
ensure
the
partitioning
of
different
density
types
resemble
what's
already
there.
So
that's
the
that's.
The
basic
idea
of
this
petitioning.
E
S
Perfect
thanks
tonica,
and
so
yes,
I'm
an
associate
scientist
in
the
data
assimilation.
Research
section
at
ncar
and
today
I'll
be
talking
about
some
of
the
results
using
our
ensemble
data
simulation
software
coupled
with
clm
just
want
to
thank
the
rest
of
our
dart
group
and
also
the
university
collaborators
who's.
Some
work
that
I'll
be
presenting
today.
S
One
way
to
try
to
induce
reduce
that
uncertainty
is
to
try
different
modes,
so
clm,
sp
and
satellite
phenology
mode,
which
is
essentially
prescribing
the
leaf
area
and
vegetation
directly
into
the
model
reducing
that
uncertainty.
I've
heard
this
today
that
there's
also
a
mode
to
prescribe
in
soil
moisture
as
well
seal
on
dart
so
ceiling,
coupled
with
the
da
system
kind
of
goes
a
step
beyond
this,
and
so
essentially
you
can
constrain
clm
simulations
with
any
observed
land
surface
property,
preferably
one.
S
That's
simulated
closer
closely
simulated
by
cln
and
it
also
produces
uncertainty,
estimates
and
those
are
very
useful
depending
upon
but
who's.
Looking
at
your
results,
and
so
why
is
this
so
important
and
interesting?
Well,
there's
been
an
explosion
of
satellite
observations
over
the
last
decade,
and
so
this
gives
us
kind
of
unprecedented
opportunity
to
look
at
different
land
surface
features
so
anything
from
solar
induced
fluorescence,
land,
surface
temperature,
biomass,
soil,
moisture.
This
is
something
that
we
can
observe
from
remote
sensing
platforms.
S
I
will
note
you
usually
have
to
use
caution
when
using
these
as
well,
because
there's
a
lot
of
science
that
goes
in
from
what
they're
measuring,
which
is
you
know,
spectral
radiance
bands
and
then
converting
those
into
land,
surface
properties
or
interpreting
those
and
so
as
an
end
user.
You
have
to
be
careful
in
terms
of
what
the
uncertainty
estimates
are
and
what
they're,
what
they're
actually
measuring,
but
these
are
often
used
in
tandem
to
complement
our
ground-based
networks.
S
So
many
are
familiar
with
the
neon
and
ameriflux
eddy
covariance
tower
sites,
and
so
these
two,
these
together,
give
us
a
lot
of
opportunity
to
constrain
your
system.
It's
a
little
bit
more
about
the
bayesian
approach
with
clm.
Dart
is
general
combination
of
these
earth
system.
Observations
with
some
uncertainty,
giving
us
a
likelihood
a
prior
forecast
generated
from
clm,
also
with
an
uncertainty
combining
those
together
and
getting
an
answer
somewhere
in
between.
So
it's
not
a
prescription
approach
and
so
a
little
bit
more
detail.
S
We
drive
our
ensembles
through
the
cam,
the
atmospheric
model
and
cesm,
and
so
through
dart,
and
so
this
ensemble
right
now.
So
if
you
want
to
know
more
about
that,
kevin
rader
is
providing
a
poster
on
that.
We
have
as
many
as
80
different
realizations
of
the
atmosphere
that
probe
the
uncertainty
in
the
atmospheric
state
and
that's
used
to
drive
the
uncertainty
in
clm.
This
gives
us
a
forecast
uncertainty.
We
run
that
until
we
come
into
contact
with
an
observation
in
time.
S
So
I
should
say
this
is
looking
at
a
variety
of
different
resolution
types,
a
variety
of
different
observation
products,
but
one
consensus
we've
found
across
the
three
is
that
when
you
assimilate
in
lai
in
above
ground,
biomass
you're,
reducing
the
simulated
biomass
within
clm.
So
here
I'm
just
showing
you
you're,
showing
about
a
30
reduction
in
the
simulated
letharian
biomass,
30
and
70
reduction
in
the
arctic
boreal
domain
in
a
20
26
reduction
across
the
globe.
And
so
you
could
see
a
lot
of
that
reduction
is
within
forested
regions.
S
So
just
digging
a
little
bit
closer
into
two
of
those
analyses:
the
western
united
states
and
the
arctic
boreal
region.
Here,
I'm
comparing
clm
and
cln5
dart
against
an
independent
benchmark,
so
the
fluxcom
machine,
learning,
benchmark
and
looking
at
the
seasonal
cycle,
the
component
fluxes
and
then
in
this
work
here,
looking
a
little
bit
closer
using
the
island,
benchmarking
software
so
evaluating
clm,
the
free
run
and
with
dart
against
a
metric
such
as
bias
and
rmse,
showing
this
relative
scale
where
the
better
values
are
the
colder,
the
bluer
colors
and
showing
improvement.
S
This
improvement
also
goes
into
the
hydrological
cycle,
so
getting
a
better
estimate
or
a
more
consistent
estimate,
with
the
benchmark
of
about
papa
transpiration,
latent
heat
and
sensible
heat.
Another
thing
to
look
at
is
that
we're
not
getting
more
consistency
in
metrics
that
are
historically
challenging
to
to
estimate
and
that's
net
ecosystem
production,
for
example,
so
just
isolating
that
this
generally
or
net
ecosystem
exchange
did
not
come
into
more
consistency
with
the
the
independent
benchmarks.
S
You
would
generally
expect
on
the
higher
elevations
to
be
more
of
a
sink
of
carbon
with
lower
elevations
would
be
a
source,
and
here
we're
not
generally
seeing
that
pattern,
and
then
we
definitely
see
that
consistency
in
the
arctic
boreal
region,
the
net
ecosystem
exchange,
the
soil,
carbon
and
also
the
snow
water,
equivalent
kind
of
become
more
deprecated
or
lose
the
consistency
with
the
independent
benchmarks
and
so
sort
of
what
that
does
is.
It
motivates
assimilating
in
additional
data
streams
to
better
constrain.
What
is
going
on
so
soil,
moisture
and
snow
or
snow
water
equivalent?
S
We
soil
carbon
data
to
better
get
at
ecosystem
respiration.
We
have
the
flux
tower
network
that
we
can
depend
on
and
also
kind
of,
the
newer
data
sets
of
solar
induced
fluorescence,
which
are
tightly
linearly
related
to
to
gpp,
but
for
interest
of
time,
I'll.
Just
I'll
just
talk
about
the
soil,
moisture
and
snow
where
so.
This
is
some
of
the
soil.
S
Moisture
assimilation,
work,
we've
done
across
china,
and
so
the
free
run
on
the
left
and
seal
on
with
dart
on
the
right
and
showing
increased
correlation
with
the
air
five
soil
moisture
benchmark
here,
and
so
I
should
say,
the
observations.
We're
assimilating
in
is
a
composite
observation
set
of
smos
smap
and
the
amstrade
satellite
platforms
as
well,
and
so
showing
improved
correlation
at
the
surface.
S
S
Here
we're
hoping
to
to
investigate
how
how
much
further
this
goes
down,
and
so
the
interesting
part
here
is
that's
what's
in
the
root
zone,
and
so
this
should
have
impacts
on
the
carbon
site,
carbon
cycle,
and
so
what
we
want
to
go
do
going
forward
is
investigate.
You
know
what
this
adjustment
does
to
the
gpp,
the
carbon
cycle
and
soil
respiration,
if
you're
interested
in
carbon
cycling
perspective,
also
I'm
kind
of
stealing
a
slide
from
the
earth
system.
S
And
so
what
we
want
to
bring
this
to
is
we're
adapting
this
into
the
newer
clm5
and
creating
a
repartitioning
scheme.
So
we
can
more
directly
bring
in
data
sets
of
snow
water,
equivalent,
adjust
the
column,
total
column
of
snow
water
equivalent
and
then
re-partition
that
across
the
different
snow
layers.
S
And
so
that's
what
we're
mostly
focused
on
now
and
of
course,
that
has
implications
for
albedo
and
surface
energy
balance
and
just
wrapping
up
in
order
to
get
a
larger
user
group,
we're
kind
of
investing
more
in
tutorial
material,
and
so
we
had
kind
of
some
more
simpler
material.
That's
kind
of
investigating
some
of
the
tools
of
dart
using
a
simple
toy
model.
E
E
E
K
K
Most
of
the
models
may
only
consist
the
dead
fuel
moisture,
but
they
do
not
consider
as
likely
moisture.
So
we
feel
like
it's
important,
that
we
consider
lifeview
moisture,
mainly
because
it's
going
to
play
a
really
important
role.
For
example,
in
terms
of
crown
fire
when
the
fire
the
flames
high
and
then
get
to
the
crown
and
then
the
last
few
months,
you
get
really
important
for
the
cheese
but
at
the
same
time
for
the
shop
processing
where
the
shrub
is
slow.
K
So
that
live
view
marsh
can
also
play
really
important
role
in
terms
of
five
spread
and
the
fire
burn
area.
As
well,
and
there's
a
paper
back
in
2009
by
denis
and
maltese,
they
have
now
analyzed
the
burn
cumulative
burn
area
for
the
california
chapel
system
and
against
live
fuel,
moisture
content.
We
can
see
that
it's
a
very
non-linear
dynamic
threshold
behavior
here.
K
Seventy
percent
of
the
live
fuel,
moisture
content
there
we
see
a
substantial
increase
of
the
cumulus
burning
area
there,
so
at
indicating
that's
really
important
that
we
potentially
should
incorporate
the
live
fuel,
moisture
content.
So,
in
this
study,
as
mentioned
very
beginning,
we
are
using
the
doe
sponsored
faith.
K
We
used
specifically
using
a
module
called
the
hydrodynamic
or
phase
hydro
here
on
the
hydrodynamic
model,
really
trying
to
track
the
water
transportation
from
the
rhizosphere
to
the
absorbing
new
transporter,
the
stem
and
also
the
leave
water
t-shirt
as
well.
So
the
models
solving
all
this
based
on
the
water
content
here.
So
in
this
case,
we
able
to
track
the
water
storage
at
different
tissues
so
in
this,
and
we
are
able
to
use
that
to
link
to
the
live
fuel
moisture.
K
So
it
allows
us
to
estimate
how
much
fuel
moisture
going
to
change
under
future
climate
change.
So
we
still
looking
at
trigger
hypothesis
here
we
there's
multiple
factors
going
to
impact
livestream
moisture
in
the
future,
and
warming
clearly
going
to
reduce
life
mulch
because
of
higher
transpiration
loss,
but
the
same
time
the
co2
fertilization
they're,
going
to
preserve
the
water
storage
in
the
leaves
and
t-shirts,
mainly
because
the
co2
normally
reduces
the
model
conductance.
K
So
that's
one,
the
hypothesis,
we
feel
like
we
test
the
hypothesis
that
only
have
strong
impact
and
then
second
hypothesis,
really
thinking
about
the
seasonal
cycle
here.
We're
thinking
about
the
reduction
in
precipitation
during
the
spring
are
going
to
make
a
longer
five
seasons
if
you
use
a
live
few
moisture
content
based
on
75
as
a
criteria
here
and
then
finally
testing
the
third
hypothesis
really
looking
at
the
combined
effect
of
warming
precipitation.
I
think
it's
larger
than
the
individual
impact
add
together.
So
this
is
innovative
component
here.
K
So
so
this
is
before
we
apply
this
model.
We
do
calibrate
the
model,
so
that
model
captures
the
observed,
lifeview
moisture
in
the
field,
and
this
is
the
monthly
measurement
of
the
stun.
I
just
mentioned
that
this
is
the
stun
wrench
for
our
study
side
a
lot
of
measurement
there,
and
this
is
the
red
line
indicating
the
model
simulation
and
that's
the
blue
line
indicating
the
fuel
observation.
K
We
do
use
some
ensemble
simulation
here
to
estimate
the
best
primary
combination
that
best
fits
observations
here
and
if
you
want
to
point
out
the
beginning,
the
model
tracked.
The
water
content
in
the
leaves
and
also
the
stain,
but
they
really
did
not
target
self-life
emotional
life.
Emotion,
is
the
combination
of
leaving
the
small
branches
there.
So
we
use
a
empirical
equations
here
to
estimate
the
life
view,
moisture
content
based
on
these
water
content
simulated
by
the
model
here
so
overall,
this
is
just
the
one.
K
Again,
looking
at
this
one
specific
tft,
we
see
that
the
five
season
lens
can
increase
from
one
135
days
to
155
days,
end
of
the
century
and
then
rcp
4.5.
This
is
going
to
increase
too
close
to
145
days
so
last,
but
we
still
have
looking
at
10
to
20
days
increase
in
terms
of
high
risk
fire
season
in
the
future.
The
chat
pro
specifically
represented
by
stone
ranch
the
jet
pro
ecosystem
and
then
another
important
part
here
we're
looking
at
what's
the
rate
of
importance
of
different
climate
drivers.
K
So
what
we're
doing
here
is
basically
the
including
or
excluding
a
certain
climate
drivers,
for
example,
co2,
precipitation,
humidity
temperature
and
combination
temperature
in
community
and
what
we
found
that
we're
looking
at
there's
no
co2,
then
the
number
of
days
actually
are
going
to
be
higher
because
the
co2
can
reduce
the
milk
conductance,
no
temperature,
that
the
number
of
days
could
the
sizes
and
lines
can
be
lower,
and
it
clearly
can
see
that
the
first
hypothesis
is
cracked
based
on
the
stimulation.
The
warming
had
much
stronger
impact
compared
to
co2
itself.
K
K
Body
fat
actually
is
not
much
larger
than
the
individual
effect
here.
So
with
that,
I
think
we
would
like
to
thank
everybody
on
this
paper
and
thanking
the
funding
thoughts
from
the
youth
lab
to
feed
program
where
collaboration
between
university,
california
and
the
national
arts,
and
also
the
ng
topic
project
for
the
the
faith
model,
and
also
thank
you
for
all
the
authors
here
yeah.
K
C
B
A
Yeah
thanks
jackie
and
thanks
everyone
for
sticking
around
the
rest
of
the
workshop's,
exciting.
There's
a
couple
sessions
on
climate
justice
both
this
afternoon
and
I
think
on
thursday
afternoon
that
are
kind
of
an
exciting
avenue
for
us
to
start
pursuing
and
there's
a
nice
example,
at
least
in
in
in
both
of
those
of
work
that
the
land
model
group's
contributing
to
thanks
for
joining
us
today
and
look
forward
to
hearing
more
from
you
in
the.