Do Global Models Properly Represent the Feedback Between Land and Atmosphere?An Observational Study from GLACE
Paul Dirmeyer1, Randy Koster2 and Zhichang Guo1
1Center for Ocean-Land-Atmosphere Studies (COLA), Calverton, Maryland, USA 2NASA Goddard Space Flight Center, Greenbelt, Maryland, USA
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 3
GLACE Informing GABLS/GLASS
• How little we can do to verify coupled land-atmosphere behavior in global weather/climate models
• Evidence for the need for co-located observations of land (subsurface and surface), atmosphere (near-surface through PBL) and fluxes between them
• The power of the plural – the value multi-model approaches
Things to watch for in this presentation:
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 4
Participating Modeling Groups
Not all models
listed are part of this
study. Some are
latecomers or did not supply all necessary
output fields.
Institute GCM Land Model
BMRC - Australia BMRC CHASM
U. Tokyo - Japan CCSR MATSIRO
Env. Canada CCCma CLASS
COLA – USA COLA SSiB
CSIRO – Australia CSIRO -CC3 & -CC4
NASA/GSFC/CRB – USA GEOS-CRB HySSiB
GFDL – USA GFDL LaD
Hadley Centre – UK HadAM3 MOSES2
SNU – Korea SNU LSM
NCAR – USA CAM3 CLM2
NOAA/NCEP – USA GFS OSU & NOAH
NASA/GSFC/GMAO – USA NSIPP Catchment
UCLA – USA UCLA SSiB
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 5
Experiment Design• All groups
integrated their global models for June-August with specified SST.
• In the control case (W), land surface state variables evolve freely and initial conditions for each ensemble member vary widely (e.g., from 1 June of different years of an AMIP simulation).
• One ensemble member is used as the source of land state variables to be specified in every member of the test cases…
W Simulations: Control integrations - establish a time series of surface conditions
(Repeat without writing to obtain simulations W2 –16)
16-member ensembles for June through August
time step n
Step forward thecoupled AGCM-LSM
Step forward thecoupled AGCM-LSM
Write the valuesof the land surface prognostic variablesinto file W1_STATES
Write the valuesof the land surface prognostic variablesinto file W1_STATES
time step n+1
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 6
Test Cases
• In case R, all land state variables are replaced at each time step of integration.
• In case S, only sub-surface soil moisture is replaced.
R Simulations: Run a 16-member ensemble, with each member forced to maintain the same time series of land surface prognostic variables.S Simulations: Run a 16-member ensemble, with each member forced to maintain the same time series of subsurface soil moisture prognostic variables
time step n
Step forward thecoupled AGCM-LSM
Step forward thecoupled AGCM-LSM
Throw out prognostic soil moisture; replace with values from W1_STATES
time step n+1
Throw out prognostic soil moisture; replace with values from W1_STATES
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 7
All simulations in ensemble respond to the land surface boundary condition in the same way
(coupling strength) is high intra-ensemble variance is small
Simulations in ensemble have no coherent response to the land surface boundary condition
is low intra-ensemble variance is large
We defined a diagnostic variable Ω that describes the impact of the surface boundary on the generation of precipitation.
Diagnostic Analysis
Ω = (16σ2<X> - σ2
X ) / 15 σ2X ,
where σ2X is the intra-ensemble
variance of X and σ2<X> is the
corresponding variance of the ensemble-mean time series – averaged over six-day intervals.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 8
Global Land-Atmosphere Coupling Experiment
Koster, R. D., P. A. Dirmeyer, Z. Guo, G. Bonan, E. Chan, P. Cox, H. Davies, T. Gordon, S. Kanae, E. Kowalczyk, D. Lawrence, P. Liu, S. Lu, S. Malyshev, B. McAvaney, K. Mitchell, T. Oki, K. Oleson, A. Pitman, Y. Sud, C. Taylor, D. Verseghy, R. Vasic, Y. Xue, and T. Yamada, 2004: Regions of strong coupling between soil moisture and precipitation. Science, 305, 1138-1140.
The GLACE project showed that while the 12 participating models differ in their land-atmosphere coupling strengths (the change in , or between cases S and W), certain features of the coupling patterns are common to many of the models. These features are brought out by averaging over all of the model results.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 9
Arid Humid
W→ET ET→P
W2
W1
P
ET
Arid regime: ET (mostly surface evaporation) very sensitive to soil wetness variations, but the dry atmosphere is unresponsive to small inputs of water vapor.
Humid regime: Small variations in
ET affect the conditionally
unstable atmosphere (high
moist static energy), but deep-rooted
vegetation (transpiration) is not
responsive to nominal soil wetness
variations.
Coupled Feedback Loop
In between, soil wetness sensitivity and conditional instability both have some
effect.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 10
The Current Study
We have, in the results of GLACE, a multi-model-based estimate of the strength and spatial variation of land-atmosphere coupling, and its relationship to state variables and fluxes within global models. Can we confirm or refute the GLACE results using the observational record?
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 11
The Observational Quandary
Three major impediments to validating the GLACE results:
• The parameter is a handy construct for model comparisons and analysis, but is not a physical quantity. It is an artifact of ensemble model simulations. There is no direct way to calculate a field of , never mind , from observations.
• It is very difficult to infer feedbacks from the observational record. This is one of the main reasons we use models, where we can control the parameters of experiments, generate very large sample sizes for statistical testing and separate signal from noise.
• We lack global measurements of soil moisture & surface fluxes, which are key elements of the coupling pathway. Thus, at best, we can only validate the behavior of global models over a small number localities.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 12
What in situ Data Are There?
In order to compare the model representation of land-atmosphere coupling strength to the real world, we need:
• Complete observations of land surface state variables, near surface atmospheric states, and fluxes between land and atmosphere.
• A long enough period of record to provide a large sample that both spans the range of variability of these variables and provides for adequate statistical significance of the results.
• Data in the same season as the GLACE experiments: June, July and August.
There are very few sources of observational data that can meet all these requirements. Two are identified for this study.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 13
ARM/CART
• DOE operates the Atmospheric Radiation Measurement (ARM) program; in particular, the Southern Great Plains site consists of a Central Facility and a number of Extended Facilities
Elevation (m
MSL)
across a large area of Oklahoma and southern Kansas (map at right - nine stations have sufficient data for comparison with the models).
Kansas
Oklahoma
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 14
FLUXNET
• The FLUXNET network of micrometeorological tower sites (table right). We have drawn upon the long-term archive at the Oak Ridge National Laboratory DAAC.
• European sites do not measure soil moisture – of limited use.
FLUXNET Sites
Latitude
Longitude
Surface
Bondville40.006
N88.292
W
Corn/soybean rotation
Little Washita
34.960 N
97.979 W
Grass, rangeland
Bayreuth50.161
N11.882 E
Needleleaf evergreen
Hyytiala61.847
N24.295 E
Needleleaf evergreen
Loobos52.168
N5.744 E
Needleleaf evergreen
Tharandt50.964
N13.567 E
Needleleaf evergreen
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 15
Closed Energy Balance
FLUXNET
ARM/CART
No GHF
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 16
Coupling Strength ~ Goodness of Fit For the coupling of soil moisture to ET, we hypothesize that the goodness of fit of a curve relating soil wetness and evaporative fraction should be proportional to ∆NLH (the change in coherence of normalized latent heat flux (NLH) from case W to case S). The goodness of fit parameter g = s/R where:
2/1)( 2
ii
i iniin
n
NLHNLH
s
)min()max( ii NLHNLHR
Best fit through 20 bins (i) with equal number of points (blue lines in next slide).
Range in y of the best fit.
Make no a priori assumption about the functional relationship of NLH on SW.Low g means good fit, high g poor fit.
NLH = LH/NetRad
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 17
Goodness of Fit (Case W)
NLH vs. SW for nine models for the ARM Central Facility. The points are 6-day means from all ensemble members. The red dots are for the member of W chosen as the basis for the test cases R and S. The relationship from observations is above.
Globallyr2 for ∆NLH vs. g(NLH) =
0.33but
r2 for ∆LHF vs. g(LHF) = 0.53
g=0.319
g=0.102 g=0.058
g=0.247 g=0.151 g=0.221
g=0.280
g=0.855g=0.194
g=0.201
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 20
Models Don’t Behave Like Obs.
Models rarely have these attributes.
•Obs show better fit for NLH than LHF, only 3½ models do.
•Obs show better fit for SHF than LHF, 1 model does.
•Obs: SHF has better fit than NSH, only 1½ models do.
g(*,SW) LHF SHF NLH NSH
Obs (ARM) 0.427 0.232 0.201 0.248
CCCma 0.253 0.337 0.319 0.342
COLA 0.186 0.260 0.194 0.204
CSIRO-CC3 0.988 0.743 0.855 1.045
GEOS-CRB 0.085 0.176 0.102 0.118
GFDL 0.103 0.151 0.058 0.126
HadAM3 0.284 0.371 0.280 0.308
CAM3 0.291 0.335 0.221 0.228
GFS/OSU 0.223 0.295 0.247 0.245
NSIPP 0.146 0.219 0.151 0.153
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 21
Why the Differences Between Models and Obs?
One possible explanation is that the GCMs emphasize a different factor controlling surface heat flux than does the real world. For example the Penman-Monteith equation and similar relationships have two main terms; • One based on potential evapotranspiration (effectively net radiation)• One based on the humidity gradient between the land surface and near-surface air.We lack complete information (namely aerodynamic resistance) that would allow us to directly compare the relative magnitudes of each term for each model and for observations. We can, however, compare the main components of each term among the models and observations.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 22
Cause of Model Behavior
• All of the models suffer from a tendency to simulate excessively warm temperatures and unrealistically low daytime relative humidity at least over the ARM region.
Categorical frequency of occurrence of net radiation (top), the difference between actual and saturation specific humidity (middle) and temperature (bottom) over the ARM region for observations (bars), and the mean of the GCMs (markers). Vertical lines span the range of models for each bin.
Net Radiation
Temperature
q Deficit
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 23
Reversed Relationship Is GlobalThe stronger dependence on soil wetness of latent heat than evaporative fraction predominates in models over most of the globe (blue areas in map of multi-model g(LHF,SW)/g(NLH,SW) below). All models have a global mean value of this ratio <1, and 7 of 9 models have a majority of the land surface covered by values <1. Thus, according to this analysis, most models appear to have a “reversed” relationship between soil wetness and surface fluxes – in contrast to nature, soil moisture in models appears to be tied more strongly to evaporation than to evaporative fraction.
ARM SiteARM Site
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 24
Betts Analysis
Betts (2004) found a strong relationship between surface properties and lifting condensation level (LCL) in ERA40.
GLACE model relationships vary – the table on the next slide shows r2 between SHF and LCL and estimated mean PBL heating rates. European FLUXNET sites are included since soil wetness is not needed for these calculations.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 25
Betts Analysis
OB
S (S
HF)
OB
S
(SH
F+
GH
F)
BM
RC
CC
Cm
a
CO
LA
CS
IRO
-CC
3
GEO
S-C
RB
GFD
L
Had
AM
3
CA
M3
GFS
/O
SU
NS
IPP
Mod
el A
vera
ge
AR
M
r2 0.59 0.63 0.03 0.70 0.72 0.21 0.37 0.74 0.65 0.78 0.36 0.49 0.51Htg rate
3.7 4.1 0.7 6.1 3.2 1.9 5.2 3.7 2.1 3.2 2.8 3.3 3.2
Bon
d-
ville
r2 0.27 0.22 0.03 0.58 0.89 0.60 0.55 0.73 0.70 0.69 0.57 0.00 0.53
Htg rate
4.1 4.8 -0.4 3.9 4.9 2.2 3.6 3.2 3.1 3.4 3.4 3.0
Little
W
ash
ita
r2 0.59 0.65 0.03 0.70 0.58 0.16 0.37 0.72 0.69 0.76 0.37 0.53 0.49
Htg rate
2.9 4.1 0.7 6.1 2.5 1.6 5.7 4.0 2.7 3.7 3.2 3.6 3.4
Bayre
ut
h
r2 0.40 0.51 0.01 0.31 0.09 0.52 0.00 0.00 0.60 0.61 0.18 0.15 0.25
Htg rate
5.8 5.8 0.4 5.8 -2.0 2.2 4.0 3.9 5.3 3.5 2.9
Hyytia
la
r2 0.55 N/A 0.46 0.30 0.61 0.81 0.58 0.60 0.76 0.58 0.25 0.36 0.53
Htg rate
4.5 N/A 3.7 4.9 8.6 4.1 7.7 5.4 5.7 7.0 6.2 7.3 6.1
Loob
os
r2 0.51 0.57 0.27 0.38 0.63 0.39 0.46 0.32 0.65 0.62 0.21 0.17 0.41Htg rate
6.0 7.2 -5.3 7.6 4.0 3.5 6.8 10.7 3.2 5.3 7.8 -14. 2.9Th
a-
ran
dt
r2 0.39 0.49 0.01 0.66 0.49 0.38 0.00 0.23 0.45 0.61 0.14 0.15 0.31
Htg rate
3.1 3.7 0.4 7.5 3.6 2.2 5.5 2.9 3.4 4.3 3.5 3.7
Models usually underestimate the strength of the relationship between SHF and LCL. PBL heating rates are rarely within 0.5° of observed mean rates. Low values of r2 suggest models that do not represent the relative importance of SHF as a source of boundary layer heating (or cooling) compared to other thermodynamic processes.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 26
LCL vs. Soil Wetness
Observed ARM relationship agrees with Betts’ theory of soil wetness controls on SHF. The models are all over the place.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 27
W2
W1
P
ET
Coupled Feedback Loop
Everything so far has concerned the terrestrial branch of the loop – what about observational validation of the behavior of GLACE models’
precipitation?
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 28
Links to Precipitation
Potential evidence for land-atmosphere coupling over the central U.S. has been found in the observational record of precipitation, based on lagged autocorrelation of pentad precipitation (map below; Koster et al. 2003) and categorical monthly precipitation (Koster & Suarez 2004).Is there a similar relationship in theGLACE models?
July1
166 10 21 26
Pre
cipi
tatio
n
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 29
Pentad Model PrecipitationAveraged over the conterminous U.S.
and grouped by month, some
models, especially the multi-model average, show a magnitude and
time evolution of lagged auto-
correlation similar to observations
(dotted lines).
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 30
Month-to-Month Persistence
(Koster & Suarez 2004) showed a tendency for persistence of anomalous precipitation in NH mid-latitudes that using pentile rankings (wettest 20% of months were usually followed by wet months, etc.). We repeat the investigation with quartiles (right) and find the models are slightly weaker than observations at showing persistence of wettest (purple) and driest (hatched) 25% of cases. Other influences (e.g. SST impacts) may also play a role.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 31
Summary
• There exist few locations with long records of observations of the necessary data to verify weather and climate models’ coupling behavior between land and atmosphere.
• In these locations, GCMs show stronger dependence of LHF on soil moisture than observations suggest, and weaker links to SHF or evaporative fraction.
• Systematic errors in surface temperature and humidity may contribute to the incorrect dependencies.
• These problems may also lead to excessive boundary layer growth and incorrect PBL heating rates.
• Nevertheless, the models (averaged together) capture observed lagged relationships of monthly rainfall.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 32
Summary
• GLACE results cannot be disproved by the poor validation of individual models, but there is certainly room for improvement in the parameterization of model “physics”.
• The multi-model approach is further supported by the results of this validation study – the multi-model mean performs better than most models in all circumstances, and is often best.
• Long-term co-located measurements of soil wetness, surface fluxes and near-surface meteorology should be distributed around the globe in order to aid model development and assess the potential for SW as a predictor for climate via land-atmosphere feedback.
20 Sep 2005 Dirmeyer - Joint GABLS/GLASS Workshop - DeBilt, NL 33
• Thank you!
This work was conducted under support from National Aeronautics and Space Administration grant NAG5-11579.