The Structural Design and Operational Behavior

1
The Structural Design and Operational Behavior of
a Specific SVAT Model
The particular land surface model (LSM) examined
here is the Mosaic LSM. Although this model has
some unique features, its description should
nevertheless give a sense for how typical SVAT
models work. References Koster, R., and M.
Suarez, Modeling the land surface boundary in
climate models as a composite of
independent vegetation stands, J. Geophys. Res.,
97, 2697-2715, 1992. Koster, R. and M. Suarez,
Water and Energy Balance Calculations in the
Mosaic LSM, NASA Tech. Memo. 104606,
Vol. 9., 1996.
SVAT stands for soil-vegetation-atmosphere
transfer. SVAT models include SiB and BATS.
2
MOSAIC LSM OVERALL STRUCTURE
Mosaic Strategy Using vegetation maps, the
heterogeneous vegetation cover within a grid cell
is subdivided into a mosaic of tiles.
Separate energy and water budgets are computed
over each (relatively homogeneous) tile. The GCM
atmosphere responds to the areally-weighted
fluxes.
Bare soil (9)
Deciduous Trees (35)
Needleleaf Trees (24)
Grassland (32)
TYPICAL TILE BREAKDOWN FOR A GCM LAND SURFACE
GRID CELL
3
Percent coverage of vegetation type within grid
cell
4
(No Transcript)
5
(No Transcript)
6
Structure of a Mosaic LSM tile Water Balance
precipitation
evaporation transpiration
INTERCEPTION RESERVOIR
throughfall
surface runoff
infiltration
SURFACE LAYER
ROOT ZONE LAYER
soil moisture diffusion
RECHARGE LAYER
drainage
7
Structure of a Mosaic LSM tile resistance network
Sensible heat network
Evaporation network
Not shown on this diagram is the
zero resistance associated with evaporation
from the canopy interception reservoir.
8
Operations performed at each time step
time step n
Turbulence subroutine computes Eo, Ho (and
tenden- cies) from Tsold and eaold
Update Seasonally- Varying Parameters
Reflectances computed gt Net shortwave, PAR flux
LSM computes energy and water balances
preprocessing
LSM call
9
Seasonally-varying parameters include
greenness fraction the fraction of
vegetation leaves that are alive
LAI the leaf area index roughness
length and other boundary layer parameters
root length We prescribe values to these
parameters, using one of two approaches
1. Assign values based on vegetation (or
soil) type and time of year. (This
is a necessary approach for many parameters.)
2. Assign geographically (and
seasonally) varying parameter values
from maps derived, e.g., from remote sensing
data.
LAI from tables (f(veg type))
LAI from satellite data
10
Preliminary Turbulent Flux Calculation
Without considering any energy or water balance
requirement, we can computepreliminary values
of evaporation (Eo) and sensible heat flux (Ho)
based on the values of surface temperature
(Ts-old) and canopy air vapor pressure (ea-old)
determined in the previous time step.
A function of Ts-old, ea-old, Tr, er, roughness,
etc.
Tr
ra-old
The surface temperature, Ts, is assumed to apply
to the canopy air, as well.
Ho
Ts-old
ea is the vapor pressure in the canopy air. We
treat ea as a prognostic variable, keeping track
of its value between time steps.
11
Under this framework, compute Eo, Ho, and their
tendencies. As will be seen later, these will
be necessary for the energy balance calculation.

Cpr

ra-old
12
Preliminary Reflectance Calculation
Reflectances (and thus net shortwave radiation
and photosynthetically active radiation PAR)
are assumed not to be affected by the energy and
water balance calculations, which means we can
compute them ahead of time. Shortwave radiation
is divided into four components a)
Visible direct radiation b) Visible
diffuse radiation c) Near-infrared
direct radiation d) Near-infrared
diffuse radiation. We calculate a reflectance
for each component using a simple empirical
formula that approxi- mates the results of the
full two-stream calculation. For full details,
see NASA Tech. Memo. 104538 (1991). Note in
the Mosaic LSM, albedo is not a function of
surface water content.
13
Now that the preliminary calculations are done,
its time to call the Mosaic LSM itself. The
parameter list includes inputs, updates, and
outputs INPUTS - Vegetation type and
time step length - GCM weather
rainfall, wind speed, vapor pressure in air,
etc.) - Seasonally-varying parameters
- Eo, Ho, and tendencies - Radiation
quantities (with shortwave reduced by
pre-calculated albedo) UPDATES (i.e., prognostic
variables) - Surface/canopy temperature
Ts - Deep soil temperature Td -
Canopy vapor pressure ea - Water contents
of three soil layers - Water content of
interception reservoir - Snow water
equivalent OUTPUTS/DIAGNOSTICS -
Evaporation E and sensible heat flux H -
Surface runoff and drainage out of the column
- Anything else that might me of interest
14
INSIDE THE MOSAIC LSM
First step Compute rc-eff, a single surface
resistance that accounts for all
evaporation pathways (transpiration, bare soil
evaporation, interception loss, snow
evaporation). Assuming no snow, we assume that
the tile is covered by a wet fraction (over
which the interception reservoir is full) and a
dry fraction (over which it is empty).
Note similarity to electrical resistance network
calculation
Wet fraction Dry fraction We find the single
effective resistance reff (for the entire
surface) that would give the same evaporation as
the dry area evap- oration (computed with
rdry-eff) added to potential evaporation from the
wet area.
15
The energy balance calculation has two unknowns
DTs and Dea. It thus needs two equations. The
first one has been seen before
Sw Lw Sw Lw H lE G
basic energy balance
In the Mosaic LSM, we assume
Snowmelt (M) is a special case, to be treated
later. Emissivity 1, so that the
upward longwave radiation sT4 The
ground heating, G, is composed of two terms
heating of the surface
system (CpDTs/Dt) and a heat flux into the deep
soil (assumed proportional
to Ts - Td). Ho, Eo, and their
tendencies are provided by the GCM as
described above, so that we
can assume
H Ho
DTs
Dea

Dea
DTs

E Eo
16
Assuming Ts Ts-old DTs and ea ea-old Dea,
we can show that the energy balance equation
reduces to Qo 4sTs-old3 dH/dT ldE/dT
Cp/Dt b DTs dH/dea ldE/deaDea, where
Qo Sw - Sw Lw - sTs-old4 - Ho - lEo -
b(Ts-old -Td), and b deep soil
heat flux proportionality constant.
Equation 1
How do we get the second equation?
er
Assume that evaporative flux from canopy air to
reference level...
E
ra
ea
...equals the flux from the saturated surface
(within stomates, etc.) to the canopy air. (That
is, the canopy air cant build up moisture.)
rc-eff
E
es(Ts)
17
In other words, Eo (dE/dT) DTs (dE/dea) Dea
(0.622r/ps) (es(Ts) - ea) / reff
Flux from canopy air to reference level
Flux from surface to canopy air
Expanding, and neglecting 2nd order terms, gives
Eo - (0.622r/ps) (es(Ts-old) - ea-old) /
reff-old (1/reff-old)
(0.622r/ps) des/dT - (dE/dT) reff-old -
Eo(dreff/dT) DTs (1/
reff-old ) (-0.622r/ps) - (dE/dea) reff-old -
Eo(dreff/dea) Dea
Equation 2
The two equations are solved for DTs and Dea.
Afterward, snowmelt is accounted for, if
necessary. Note that the equations simplify in
cases of dewfall or snow evaporation, for which
we assume reff 0.
18
Next compute water fluxes (i.e., solve the
various water balance equations).
1. Evaporation. With DTs and Dea computed, the
evaporation rate for the time step is known. We
remove this moisture from the interception
reservoir, the surface soil layer, and the root
zone soil layer in amounts consistent with the
resistances. 2. Moisture transport between soil
layers. We use a discretized version of Darcys
law for unsaturated flow. (See water balance
lecture. Some features differ e.g., we use an
upstream hydraulic conductivity.) Moisture
flow from surface layer to root zone layer
accounts slightly for subgrid heterogeneity.
3. Assign precipitation water to
reservoirs. Assume a uniform precipitation depth
within a prescribed fractional wetted area, and
allow a fraction of this storm area to consist of
previously wetted leaves. Surface runoff and
infiltration are computed from resulting
throughfall.
19
The Mosaic LSM, like any other SVAT model, has a
drawback -- it requires realistic values for
numerous parameters Soil Layer
capacities Porosity
Saturated soil matric
potential Saturated soil
hydraulic conductivity
Soil pore size distribution index
Bedrock slope
Surface heat capacity Vegetation
Surface type (one of 10 generalized types)
Leaf area
index
Greenness fraction
Roughness height
Vegetation height
Unstressed canopy resistance
parameters (6)
Vapor pressure deficit stress parameter
Temperature
stress parameters (3)
Leaf water potential stress parameters
(5)
Subcanoy aerodynamic resistance parameters (2)
Other Storm fractional area
The vegetation type assigned to the tile defines
the values used for most of these parameters
Note that some of the parameters cannot be
directly measured
20
The Mosaic LSMs structure allows the breakdown
of total evaporation by vegetation tile...
and the breakdown of each tiles evaporation by
component.
21
How do we evaluate the performance of such an LSM?
Online approach test GCM output against
observations.
Advantage The coupling effects can be studied,
and various sensitivity tests can be
performed. Disadvantage The model forcing
(precipitation, radiation, etc.) can be
wrong, so validating the land surface model can
be very difficult. (Garbage in -- Garbage out)
Example from GISS GCM/LSM The Amazon river is
poorly simulated, but we cant tell if this
is due to a bad LSM or poor precipitation from
the GCM.
22
Better approach Offline forcing (one-way
coupling)
Forcing Data
Advantage Land surface model can be driven with
realistic atmospheric forcing, so that the impact
of the LSMs formulations on the surface fluxes
can be isolated. Disadvantage Deficient
behavior of the LSM may seem small in offline
tests but may grow (through feedback) in a
coupled system. Thus, offline tests cant get at
all of the important aspects of a land surface
models behavior.
Output File
P, radiation, Tair, etc.
E, H, Rlw , diagnostics
LSM
PILPS model intercomparisons (to be discussed in
a later lecture) have largely focused on such
offline evaluations.
23
Mosaic LSMs behavior in PILPS 2c (a study based
in the Red-Arkansas River Basin).
Forcing data covering several years for each of
61 1o X 1o grid cells in the Red Arkansas Basin
were provided to participants.
24
Parameter values make a difference! In a recent
study, it was found that the apparently poor
behavior of the Mosaic LSM in an offline study
using Oklahoma measurements was associated in
part with an inaccurate setting of the ground
heat capacity.
Ground heating rates improve when the right heat
capacity is used.
Ground heating rates for mosaic are way off,
throwing off the other fluxes.
Robock et al., JGR, 108, D22, 8846,
doi10.1029/2002JD003245, 2003.
25
Coupled System Analysis Analysis, using Mosaic
LSM, of what makes a SVAT model act
differently from the standard Bucket model...
26
Sensitivity test Addition of vapor pressure
deficit stress
Precipitation differences with VPD stress minus
w/o VPD stress
Inclusion of vapor pressure deficit stress leads
to large decreases in rainfall in some regions.
Why? Stomatal suicide -- a serious positive
feedback in the coupled system
Higher VPD stress leads to reduced evaporation
Reduced humidity leads to higher VPD stress
Reduced evaporation leads to precipitation
Reduced evaporation leads to reduced humidity
27
Sensitivity tests Removal of temperature
stress removal of interception loss mechanism
Precipitation (top) and evaporation (bottom)
differences with interception loss allowed
minus w/o interception loss allowed
Precipitation differences with temperature
stress minus w/o temperature stress
28
Precipitation (top) and evaporation (bottom)
differences standard formulation minus
bucket-type formulation
Sensitivity tests Coupling strategy
A simulation was performed with a
pseudo-bucket, one that used a bucket-style
coupling to the atmosphere but was carefully
controlled to reproduce the Mosaic LSMs
long-term surface energy budget in offline
simulations. In the plots, large differences are
seen in simulated evaporation and precipitation
rates. These differences result strictly from
feedbacks between the land and the atmosphere.
29
Main conclusions from coupled sensitivity
analysis (Koster and Suarez, Advances in Water
Resources, 17, 61-78, 1994.) 1. Of the
environmental stresses that increase canopy
resistance, -- temperature stress is
not significant -- vapor pressure
deficit stress is significant, partly due to
feedback. 2. Of the main differences between the
two model types, the presence of the
interception reservoir in the SVAT model has the
largest effect on evaporation rates. 3. The
incorporation of a bucket model structure appears
to have an effect on precipitation rates in the
tropics and subtropics, perhaps due to the
damping of diunal and synoptic-scale variability
in land surface control. The
differences, in any case, reflect land-atmosphere
feedback. (In a later study, with the same LSM
but a modified AGCM, the impact on the general
circulation was found to be reduced.)
30
COMPUTER LAB RUNNING A LAND SURFACE MODEL (A
Retropective!) This model is designed to
simulate a tropical forests response to
prescribed atmospheric forcing over a repeated
full seasonal cycle. The relevant files
are Model gm_model.f (Includes driver
written in FORTRAN.) Forcing file TRF.DAT.30
(Includes rainfall rates, radiation forcing,
etc., at a 30 minute time step over a full annual
cycle. Model automatically interpolates to a 5
minute time step.) Initialization file
input/lsm_input.dat (Includes parameter values to
change for class experiments.) How to run the
model 1. Create input and output directories
below the current directory. (This assumes a
UNIX system.) 2. Place lsm_input.dat in the input
directory. 3. Find a directory that can
comfortably hold trf.dat.30.diur (1.4 Mb) 4.
Compile the program gm_model.f 5. Modify the
model parameters in lsm_input.dat as
appropriate. 6. Run the program. 7. Four output
files will be produced in the output directory
mosaic.trf.mon.xxxx (4.5 Kb)
mosaic.trf.dat.xxxx (388 Kb)
mosaic.trf.tra.xxxx (12.9 Kb for 3-year run)
mosaic.trf.123.xxxx (291 Kb) where
xxxx is the label for the particular
experiment. 8. For new experiments, start at
instruction 5.
31
INPUT FILE/land/koster/pilps/TRF.DAT.30
This is the forcing data modify path as
necessary. VEGETATION IDENTIFIERtrf
Leave as is EXPERIMENT IDENTIFIERgp7 By
changing this according to your own system of
codes, you control the labeling of the output
files of different experiments.  TIME STEPS  
T.S. LENGTH   DIAGS   1ST FORCING    ALAT   
534529           300.   
2880              0     -3.
534529 (365x3 31) x 24 x 12 1
of time steps in 3
years 1 January 1 time step.
300 number of seconds in the 5 minute time
step. DIAGS, 1ST FORCING,
ALAT do not need to be changed. NUMBER OF
TILES          1                TYPE  
FRACTION                  1        1.0
Type 1 tropical forest
Fraction 1 means a
homogeneous cover  INITIALIZATION         
TC      TD      TA     TM                  
     300.0   300.0   300.0   300.0
TC Initial
canopy temperature
TD Initial deep soil temperature
TA Initial
near-surface atmospheric temperature
TM Initial assumed
first forcing temperature               
WWW(1)   WWW(2)  WWW(3) CAPAC   
SNOW                0.5000  0.5000 
 0.5000   0.5       0.
WWW(i) Initial degree of
saturation in soil layer i
CAPAC Initial fraction of
interception reservoir filled
SNOW Initial snow amount
32
 EXPERIMENT 1    HEAT CAPACITY     WATER
CAPACITY FACTOR     TURBULENCE FLAG          
70000.                        
1.                  
0 Heat capacity is in J/oK. If water capacity
factor is 0.5, then the default capacity is
halved if it is 2, then the default capacity is
doubled, etc. Turbulence flag you wont need
this.  EXPERIMENT 2    INTERCEPTION
PARAMETER   PRECIP. FACTOR                      
1.               
1. Interception parameter you wont need
this. Precip. factor factor by which to multiply
all precipitation forcing.  EXPERIMENT 3   
ALBFIX    RGHFIX   STOFIX        0       
  0         0 ALBFIX If this is 1, you
are using tropical forest albedo. RGHFIX If this
is 1, you are using tropical forest roughness
heights STOFIX If this is 1, you are using
tropical forest water holding capacities.  EXPER
IMENT 4    FRAC. WET     PRCP CORRELATION       
  0.3                   0. FRAC. WET
The assumed fractional coverage of a storm
equivalent here to the assumed probability that a
rainfall event will be applied to the land
surface model. PRCP CORRELATION Imposed
time-step-to-time-step autocorrelation of
precipitation events.
33
EXPERIMENT 1 CHANGE IN MODEL PARAMETERS
Background The heat capacity of the soil
surface has an important effect on the land
surface models surface energy budget
calculations. Presumably, the higher the heat
capacity, the more slowly the surface temperature
will change under a given forcing, leading to a
smaller amplitude of the diurnal temperature
cycle. This could have profound effects on the
annual energy balance. The water holding
capacity of the soil has an important effect on
the annual water balance and thus on the annual
energy balance. A larger water holding capacity,
for example, means that high precipitation rates
in the spring can more easily lead to high
evaporation rates during a subsequent dry
summer. Possible experiments .Modify the heat
capacity. You may have to modify it by an order
of magnitude or so to see significant effect on
the energy budget terms. .Modify the water
capacity factor. For starters, try 0.5 and
2. Questions to answer (choose 1) 1. How does
varying the heat capacity affect the diurnal
energy balance, in particular the amplitude of
the diurnal temperature cycle? How large does
the change have to be to see an effect? Is the
effect in the expected direction? 2. How does
varying the heat capacity affect the annual
energy balance? 3. How does varying the water
holding capacity affect the diurnal and annual
energy and water budgets? Does a higher capacity
imply a larger annual evaporation?
34
(Recall from earlier lecture)
The choice of the heat capacity can have a major
impact on the surface energy balance.
Low heat capacity case
High heat capacity case
-- Heat capacity might, for example, be chosen so
that it represents the depth to which the
diurnal temperature wave is felt in the soil. --
Note that heat capacity increases with water
content. Incorporating this effect
correctly can complicate your energy balance
calculations.
35
Effect of Heat Capacity Change
Diurnal Cycle of Surface Temperature Black Cp
70000 J/kg (control) Purple Cp 700000
J/kg Yellow Cp 7000 J/kg
36
Effect of Heat Capacity Change
Annual Cycle of Surface Temperature Black Cp
70000 J/kg (control) Purple Cp 700000
J/kg Yellow Cp 7000 J/kg
37
Effect of Water Holding Capacity
Recall from earlier lecture
38
Effect of Water Holding Capacity Change
control value x1
control value x2
control value x5
control value x 1/2
39
EXPERIMENT 2 CHANGE IN MODEL INITIALIZATION
Background All models require a spin-up
period to remove the effects of initialization.
In other words, the initial conditions imposed in
a model may be inconsistent with the preferred
model state, and this inconsistency may lead to
energy and water budget terms that are
unrealistic they reflect the inappropriate
initial conditions imposed rather than the model
parameterizations or the atmospheric forcing.
The length of the spin-up period is a function of
the model (in particular its heat and moisture
capacities) and the forcing. Possible
experiments Initialize the soil moisture
reservoirs to complete saturation set WWW(1),
WWW(2), and WWW(3) to 1. Initialize the soil
moisture reservoirs to be completely dry set
WWW(1), WWW(2), and WWW(3) to 0.0001. Initialize
the soil moisture reservoirs to be completely
dry, and double the water holding capacity set
WWW(1), WWW(2), and WWW(3) to 0.0001, and set the
water capacity factor (from experiment 1) to 2.
Complete drydown. Set WWW(1), WWW(2), and
WWW(3) to 1, and set the precip. factor to 0.
(This turns off all precipitation.) Note for
these experiments, you may want to increase the
number of time steps. (You wont know if you
need to until you run them.) If n is the number
of years you want the model to run, set the of
time steps to (365n)31)24121. Questions
to answer (Choose 1) 1. How does the transient
model response differ in the drydown and wet-up
simulations (1 2)? 2. How does doubling the
water holding capacity affect the wet-up
period? 3. How long does complete drydown take
(simulation 4)? Is equilibrium ever really
achieved? Can you define a time scale for the
drydown?
40
First 5 years
First year, including equilibrium cycle
41
EXPERIMENT 3 CHANGE IN MODEL BOUNDARY CONDITIONS
Background GCM deforestation experiments have
examined how replacing the Amazons forest with
grassland can affect the regional climate. In a
land surface model, forest and grassland are
distinguished from each other only by the values
used for various parameters. The experiments
below examine deforestation in an offline
environment. (Of course, deforestation effects
in a fully coupled GCM environment may be
different.) Possible experiments .Perform a
control simulation, using TYPE 1 (tropical
forest). .Replace the tropical forest with
grassland set TYPE4. .Replace the tropical
forest with grassland, but maintain tropical
forest albedo set TYPE4 and ALBFIX1. .Replace
the tropical forest with grassland, but maintain
tropical forest roughness set TYPE4 and
RGHFIX1. .Replace the tropical forest with
grassland, but maintain tropical forest water
holding capacity set TYPE4 and
STOFIX1. .Replace the tropical forest with
grassland, but maintain tropical forest albedo,
surface roughness, and water holding capacity
set TYPE4, ALBFIX1, RGHFIX1, and
STOFIX1. Questions to answer (choose 1) 1. What
is the effect of deforestation on the annual
energy and water budget? What effect does it
have on diurnal cycles? 2. How do albedo change,
roughness change, and storage change contribute
to the tropical forest / grassland differences?
Which effect is largest? 3. Are the impacts of
albedo change, roughness change, and storage
change linear? E.g., do the changes induced by
these three parameters alone add up to the
changes seen in simulation 6?
42
Main effect of deforestation Black forest,
purple grassland
Black forest purple grassland yellow
grassland with forest albedo
Black forest purple grassland yellow
grassland with forest storage
43
EXPERIMENT 4 CHANGE IN MODEL FORCING
Background The precipitation forcing, which
comes from a GCM, need not be assumed to fall
uniformly within the GCMs grid cell area. If
the typical areal storm coverage is, say, only
half the grid cells area, then one can consider
an alternative interpretation that whenever the
GCM provides precipitation for a grid cell, the
probability that it occurs at a given point
within the cell is ½, and when it does occur
there, the GCMs precipitation intensity is
doubled. A further consideration is the temporal
autocorrelation of storm events, i.e., the
probability that a point gets wet during one time
step given that it was wetted in the previous
time step. Possible experiments .Perform a
control simulation. .Perform simulations that
assume a fractional storm coverage of ranging
from .1 to .9 (i.e., set FRAC. WET x, where x
ranges from .1 to .9). .Perform simulations that
assume a fractional storm coverage of .1 and a
time step to time step autocorrelation that
ranges from .1 to .9. (i.e., set FRAC. WET0.5
and PRCP CORRELATIONx, where x ranges from .1 to
.9). Questions to answer (Choose 1) 1. How does
runoff ratio (runoff / precipitation) change with
the assumed fractional coverage? 2. How do
runoff ratios change when temporal
autocorrelations are included?
44
Storm area 0.01
Effect of fractional storm area
Effect of correlation in storm position. (Storm
area .1)
Correlation 0.9
Correlation 0.