1
Improvement of Land Surface Parameters and States:
Diagnosing Forecast and Model Deficiencies
Michael Barlage (NCAR)Xubin Zeng (UA), Patrick Broxton (UA), Fei Chen (NCAR)
12th JCSDA Science Meeting – 22 May 2014
2
Introduction
Temperature biases in the Noah model can reduce the number of satellite observations that are assimilated
In addition, snow melts too quickly and to correct for this, water is added during data assimilation, which results in too much melt
Advances in model structure aim to improve surface temperature and snow simulation, increase atmospheric assimilation, and increase land surface assimilation
This presentation documents deficiencies in the current forecast system and attributes part of them to model structural deficiencies
3
Noah LSM Deficiencies
Flagstaff WRF/Noah v3.2 T2m simulation (green) compared to METAR observations(black)
• Cold bias during the day results from capped surface temperature at freezing
• Bias recovers during the night
• When snow is gone, bias is low
Challenges with Noah LSM StructureD
ays
in F
ebFE
B1
F
EB13
FEB
20
FEB
27
KFLG Forecast Bias (ºC)
Forecast Hour (Initialized at 12Z daily)
Challenges with Noah LSM Structure• May 2007 temperature time series for a single location in Arctic System
Reanalysis (3D-Var, land assimilation of vegetation, snow and albedo)
• Observations in blue, analysis in red and model forecast in green
• Pre-snowmelt period cold bias exists, assimilation helps
• Significant cold bias exists during melt period (up to 15°C)
• Post-melt period performance is quite good
Challenges with Noah LSM Structure
NCEP operational NAM model 24-hour forecasted snow minus analysis snow shows excessive melting during the entire month of March
• Simultaneously with low temperature biases, snow continuously gets assimilated during spring
• Due to model structure, this snow melts during the 24 hours until the next assimilation cycle
• This reinforces the cold bias and inserts more water into the system, potentially causing adverse effects to hydrology prediction
6
Snow Water Equivalent in GFSSWE (kg/m^2) (Forecast – Analysis) *White areas: SWE <10 kg/m^2 in forecast and analysis
7550250-25-50-75
For GFS, compare the forecast with a lead time of 4 days with the coincident analysis for each day in 2013
7
Noah LSM in NCEP Eta, MM5 and WRF Models(Pan and Mahrt 1987, Chen et al. 1996, Chen and Dudhia 2001,Ek et al., 2003)Noah-MP LSM in WRF and NCEP CFS (Yang et al., 2011; Niu et al., 2011)
Snow (x,y)
Reality
Tcan(x,y,z)
Tsnow(x,y,z)Tbc(x,y,z)
Challenges with Noah LSM Structure
Tg(x,y) Snow
Noah
Snow
Noah-MP
Tcan
Tsnow(z)Tbc Tg
Tskin
Single surface temperature
Multiple surface temperatures and distinct canopy
Noah and Noah-MP LSM Structure Comparison
• Six-month simulations using coupled atmosphere-land model from March – June 2010
• Compare only grids with 100% snow cover and evergreen needleleaf trees
• When temperatures are below freezing, Noah-MP is warmer but consistent with Noah
• When temperature approaches freezing, Noah temperature cannot get much above freezing
Noah and Noah-MP LSM Structure Comparison
• Compare to daily MODIS/Aqua land surface temperature at 13:30 overpass
• Noah peaks near-freezing
• Noah-MP is warmer than MODIS by 2-4K but distribution is much better than Noah
Snow Water Equivalent simulated by six LSMs
Noah and Noah-MP LSM Structure Comparison
Noah and Noah-MP can produce similar snow through a modified snow albedo formulation
Chen, et al. 2014
Diurnal cycle of surface albedo: Niwot Ridge Jan, Mar, and Jul 2007
11
AmeriFlux Obs, MODIS, Noah, VIC, SAST, LEAF, CLM, Noah-MP
Albedo: 0.66 (Cline, 1997) over snow, 0.34 (MODIS, Jin et al., 2002),
• Large variation among modeled winter albedo • Noah: larger seasonal variations• Noah-MP: drop during March spring melt
Noah and Noah-MP LSM Structure Comparison
Chen, et al. 2014
Monthly daytime min, max and mean absorbed SW and sensible heat (W/m2) for Jan, Mar, May and Jul
• Comparison of observed (O), Noah (N), and Noah-MP (M).
• Noah has less absorbed solar radiation resulting in colder surface and lower (or negative) sensible heat flux
Noah and Noah-MP LSM Structure Comparison
100200300400500
-100 0100200300
O
N
M
Jan 2007
O
N
M
O
N
MO
N
M
Mar 2007
Chen, et al. 2014
Noah-MP uses a two-stream radiative transfer treatment through the canopy based on Dickinson (1983) and Sellers (1985)
• Canopy parameters:– Canopy top and bottom– Crown radius, vertical and horizontal– Vegetation element density,
i.e., trees/grass leaves per unit area– Leaf and stem area per unit area– Leaf orientation– Leaf reflectance and transmittance for
direct/diffuse and visible/NIR radiation• Multiple options for spatial
distribution– Full grid coverage– Vegetation cover equals prescribed
fractional vegetation– Random distribution with slant shading
SWdn
shaded fraction
Advantages with Noah-MP LSM Structure
• Over a Noah-MP grid, individual tree elements can be randomly distributed and have overlapping shadows
• Noah-MP albedo is calculated based on canopy parameters
• Noah prescribes snow-free and snow-covered albedo from satellite climatology
SE Minnesota in Google Maps
Advantages with Noah-MP LSM Structure
• Using prescribed vegetation fraction varying from 5% to 100% as radiation fraction
• Increasing vegetation fraction increases snow, decreases albedo
Advantages with Noah-MP LSM Structure
• Using randomly distributed shadows as radiation-active fraction through use of sun angle and canopy morphology
• Complex interaction between vegetated and shadowed fraction and canopy/snow radiation absorption
Advantages with Noah-MP LSM Structure
Fractional coverage of each land cover Type (%) - Alaska
TypeBU-IGBP+
tundra_1kmMODIS5.1
0.5km
MODIS5.1 Fill by surrounding 0.5
degree forest typeEvergreen Needleleaf 14.07 4.82 12.09Evergreen Broadleaf 0.00 0.00 0.00Deciduous Needleleaf 0.43 0.37 3.67Deciduous Broadleaf 0.07 0.03 0.03Mixed Forest 4.77 2.48 3.94Woody Savanna 0.00 12.04 0.00
MODIS5.1 Fill by surrounding 0.5 degree forest typeBU-IGBP+tundra_1km
MODIS5.1 0.5km
Everg
reen N
eedlelea
f
Deciduous B
road
leaf
Deciduous N
eedlel
eaf
Deciduous B
road
leaf
Mixe
d Forest
Closed Sh
rublan
dOpen
Shru
bland
Woody Sav
anna
Sava
nnaGras
sland
Perman
ent W
etlan
d
Croplan
dUrb
anCro
pland/N
atural
Snow/Ic
eBarr
en
Not Lan
dWooded
Tundra
Mixe
d Tundra
Bare G
round Tu
ndra
Continue to Develop Satellite Land DatasetsMODIS 500m land cover climatology delivered to WRF v3.6 (Broxton et al., 2014)
Marks- Individual value of year2001 2002 2003 2004 2005 2006 2007 2008
Original Pixel Data Final Smooth Climatology
Lines- black: median- yellow: tile climo (Savanna)
1. Remove suspect data2. Fill missing data3. Smooth
Continue to Develop Satellite Land Datasets
• WRF satellite-derived input datasets tend to produce too little vegetation outside of the tropics
• Using fraction of photosynthetic absorbed radiation as a vegetation proxy
May
Continue to Develop Satellite Land Datasets
20
• Monthly mean WRF radiation June 2010 (Dudhia scheme)
• ERA-Interim monthly mean radiation June 2010
• Up to 100 W/m2 difference
• 20 – 40% too high
Can’t always blame the LSM
21
Domain 1 TEMPERATURE (°C)Mean error RMSE
U* fix +0.027 +0.011VEGFRA -0.144 -0.032Radiation -0.364 -0.061OP25 -0.482 -0.088
D1
Domain 3 TEMPERATURE (°C)
Mean error RMSE
U* fix -0.053 -0.054
VEGFRA -0.138 -0.018
Radiation -0.241 -0.061
OP25 -0.420 -0.090
D3
Continue to Develop Satellite Land Datasets
• Noah model land data assimilation– Favorable to directly assimilate (use) “bulk” land surface properties
• Albedo• Green vegetation fraction (via NDVI or EVI)• Leaf Area Index (LAI)
– Bulk surface treatment causes problems when heterogeneity is necessary (e.g., snow and vegetation)
• Noah-MP model land data assimilation– Increased prognostic states for assimilation
• LAI through dynamic vegetation model• Albedo needs to be treated differently (parameter estimation)• Vegetation fraction: what does it mean in the model?
– More available states that can inform surface emissivity models• Prognostic LAI, partition of canopy water into ice/liquid
• Both models use similar soil moisture treatment for soil moisture assimilation
Relationship to Land Data Assimilation
23
ConclusionsTemperatures: significant biases can occur in the current Noah model, many of these are due to structural limitations when heterogeneities exist at the surface.
Snow: In Noah, generally there is too little snow during the spring. Assimilation replaces this snow, it immediately melts and gets added to the soil or surface runoff.
Are we reaching a structural limit in the Noah model? Do we need to move toward a more process-based model to capture important states that can informed satellite assimilation?
Before the eventual (hopefully) update of the operational LSM, can we exploit the benefits of a more complex model, e.g., in a system such as LIS?