Evaluation of Community Land Model Hydrologic Predictions

1

• Evaluation of Community Land Model Hydrologic
  Predictions
• Kaiyuan Y. Li1, Dennis P. Lettenmaier1, Ted J.
  Bohn1, and Christine Delire2
• Department of Civil and Environmental
  Engineering, Box 352700, University of
  Washington, Seattle, WA 98195, USA
• 2. Institut des Sciences de l'Evolution, cc 61,
  Université Montpellier II - 34095 Montpellier
  FRANCE

3
1
Introduction
Results and Discussions
The HAPEX-MOBILHY Soybean Site (43.7ºN 0.1ºW)
The ABRACOS Forest Site (10.1ºS 61.9ºW)
This soybean site was part of the larger
HAPEX-MOBILHY experiment held in southwest
France in 1986. Energy fluxes was measured in 15
min interval between May 28 and June 30. The 15
min measured fluxes are, in turn, averaged over
60 min to match the model time step. Soil
moisture was measured weekly (1.6m depth at 0.1m
interval) from January to December.

• The ABRACOS forest site was located in Reserva
  Jaru, Brazil. The surface energy
• fluxes were measured during two intensive field
  campaigns from August to
• October 1992 and from April to August 1993.
  Neutron probe measurements of the
• soil moisture down to 3.6 m were taken weekly
  from November 1991 to December
• A study by Wright et al. (1996) shows that
  trees are able to extract water
• from depths greater than 3.6 m because no
  significant decrease in forest
• transpiration was measured during the dry season
  when the moisture content within
• the 3.6 m of soil depth was low. In order to take
  this fact into account, Delire and
• Foley (1999) used a rooting depth of 10 m in
  their study of land surface model
• evaluation, and obtained good agreements between
  simulated and observed soil
• moisture content and energy fluxes. In this
  study, the soil depth for CLM is fixed
• to 3.43 m because a fixed soil layer scheme is
  implemented within CLM, while two
• different soil depths (3.43 and 10 m) were tested
  for VIC.

Confidence in representation and parameterization
of land surface processes in coupled
land-atmosphere models is strongly dependent on a
diversity of opportunities for model testing,
since such coupled models are usually intended
for application in a wide range of conditions
(regional models) or globally. Land surface
models have been increasing in complexity over
the past decade, which has increased the demands
on data sets appropriate for model testing. In
this study, we evaluate the Community Land Model
(CLM) by comparing with Variable Infiltration
Capacity (VIC) in terms of their ability to
reproduce observed water and energy fluxes in
off-line tests for two large river basins with
contrasting hydroclimatic conditions spanning the
range from temperate continental to arctic, and
for four point (column flux) sites spanning the
range from tropical to arctic. The two large
river basins are the Arkansas-Red in U.S.
southern Great Plains, and the Torne-Kalix in
northern Scandinavia. The column flux
evaluations are for a tropical forest site at
Reserva Jaru (ABRACOS) in Brazil, a prairie site
near Manhattan (FIFE), Kansas in central U.S.,
a soybean site at Caumont (HAPEX-Monbilhy) in
France, and a small grassland catchment at Valdai
in Russia. The objective of this study is to
test and evaluate the hydrologic performance of
CLM, particularly to identify the weaknesses, so
as to further enhance CLM hydrologic
predictions.
The Arkansas-Red basin
The Arkansas-Red basin is located in the southern
Great Plains of the United States, covering a
total area of 566,000 km2. This basin has been
highly sensitive to past droughts, and hence
testing of CLM over this basin is critical for
CLM-coupled modeling for U.S. future drought
prediction. In order to test the models ability
to reproduce observed runoff and its spatial
variability, below we show monthly streamflow
predictions for various subbasins with different
magnitudes of drainage area at different
locations. The subbasins are grouped into to
large basins (72,000 410,000 km2, left figure)
and small basins (440 11,000 km2, right
figure).
Model Description
2
Community Land Model (CLM)
From above figures, it can be seen that both CLM
and VIC simulated net radiation quite well, but
both poorly reproduced the observed ground flux.
VIC reasonably well reproduced both the latent
and sensible heat, while CLM underestimated
latent heat and overestimated sensible heat for
the growing season. However, in the non-growing
season (before sowing, Jan. Apr., and after
harvest, Oct Dec) CLM markedly overestimated
bare ground evaporation and hence underestimated
soil moisture content as compared to the
observation and VIC simulation. This
underestimation of soil moisture by CLM for the
period before planting (Jan. Apr.) subsequently
led to the underestimation of evapotranspiration
for the growing period of early May through late
August.
Both two figures show that VIC reproduced
observed runoff reasonably well in terms of both
magnitude and seasonality. CLM well captured the
runoff seasonality, but in general overestimated
wet season runoff.
Both CLM and VIC simulated the net radiation
quite well, but both poorly simulated the ground
heat flux. Both VIC-10m and VIC-3.43m reasonably
well reproduced the sensible heat and latent
heat, while CLM in general overestimated the
sensible heat and underestimated the latent
heat. The poor performance of CLM are
attributed to the large estimation of runoff and
poor simulation of soil moisture content. The
soil moisture content was very well estimated by
VIC-10m but poorly estimated by VIC-3.43m,
indicating that the 3.6m soil depth is too
shallow and the selection of 10 m soil depth is
appropriate. It should be noted that the
evapotranspiration by CLM is extremely low
(close to 0) in dry seasons, while the
corresponding soil moisture contents are much
higher than those simulated by VIC-3.43 in dry
seasons and hence no obvious seasonality for soil
moisture content was captured by CLM. This
indicates that soil-plant-water relationship and
soil moisture content are poorly represented in
CLM.
The FIFE Prairie Site (39.0ºN 96.5ºW)
The Torne-Kalix basin
Modeling Framework
The FIFE was conducted between 1987 and 1989 near
Manhattan, Kansas, in a tallgrass prairie.
Energy flux measurements were taken during the
growing season of 1987 and 1989. A neutron probe
measurements of the soil moisture content were
taken between 20 and 140 cm, and gravimetric
measurements were made at 2.5 and 7.5 cm depth
near surface.
The Torne-Kalix basin is located in northern
Scandinavia, with an area of 58,000 km2. This
basin was used to test how well the CLM model, in
particular the snow module, performs in high
latitude code region.

• One vegetation layer, ten soil layers and up to
  5 snow layers
• Represented as nested subgrid hierarchy in
  which grid cells are
• composed of multiple land units, snow/soil
  columns and PFTs
• Designed in such a way that model components
  from other
• land surface models can be incorporated into
  it.

We have noticed that a better agreement between
the observed and simulated soil moisture
contents by CLM was obtained in a study by Dai et
al. (2003), in which the bottom of soil column
(at 3.43m) was assumed as granite bedrock
without drainage. However, there seems to be no
evidences that support this assumption.
Variable Infiltration Capacity (VIC)
The Valdai Grassland Site (57.6ºN 33.1ºE)
Valdai site is a small catchment (0.36 km2) in
central Russia, mainly covered by grassland.
The results shows that the CLM and VIC
performed similarly in prediction of snow depth,
but the runoff peak of CLM came earlier than
observation for the period of 1967-1973,
indicating that snow melted earlier in CLM than
observations in some years. CLM and VIC
performed similarly in the simulation
of evapotranspiration except that CLM
underestimated the peak value. Soil moisture
contents were poorly simulated by CLM with
less obvious seasonality than observations.
Above two figures show the runoff predictions and
observations for 14 subbaisns. Generally, VIC
better captured the runoff magnitude and
seasonality as compared to CLM for all the 14
subbasins. CLM overestimated the runoff peak
value for most subbasins except for Raktfors,
Pajala Rumphus and Pallo, where the runoff peak
value was largely underestimated. For most basins
in most time, the runoff peak of CLM came
earlier than did VIC and observation, implying
that in CLM snow tends to melt earlier than
observation.
VIC simulated all four energy fluxes quite well,
while CLM simulated net radiation, sensible and
ground heat flux reasonably well but
underestimated latent heat. This underestimation
of the latent heat by CLM can be attributed to
the larger runoff simulation. The above left
figure shows that both surface runoff and base
flow of CLM are larger than those of VIC.
4
Conclusions
The left figure shows that on average, the runoff
peak and hence the snow melting of CLM came one
month earlier than simulations. Due to the offset
of the large underestimation at a few subbasins,
the mean runoff peak value of CLM appeared to be
only slightly higher than observed and similar to
VIC.

• CLM tends to overestimate runoff in some
  places
• CLM tends to melt snow earlier than observed
• CLM poorly simulates soil moisture contents
  with less obvious seasonality
• In general, VIC performs better than CLM in
  terms of hydrologic predictions
• Improvements of CLM are expected by
  incorporating some aspects of VIC
• hydrologic parameterizations into CLM.

Both CLM and VIC well simulated soil moisture at
surface layer (0 10 cm), but under-estimated
the soil moisture content at 10 140 cm depth.
Because VIC well simulated energy fluxes and the
simulated runoff is quite small, there appears to
be no reason why simulated soil moisture content
are markedly lower than the observation. We
suspect that it might be attributed to the
possible overestimation of soil moisture content
from the neutron probe measurements.

• Infiltration and runoff spatial variability
• Multiple vegetation classes in each grid cell
• Energy and water budget closure at each time
  step
• Non-linear baseflow generation
• Precipitation variability represented with
  elevation bands in
• complex terrain