Assimilation of MODIS and AMSRE Land Products into the NOAH LSM

1
Assimilation of MODIS and AMSR-E Land Products
into the NOAH LSM
Paul Houser, Xiwu Zhan, Alok Sahoo, Kristi
Arsenault, Brian Cosgrove
Water Cycle Research Making a Difference
http//crew.iges.org
2
Project Motivation and Goal

• Land surface information improves weather and
  climate prediction
• Near-real-time land observations (MODIS, AMSR-E)
  are available
• Few satellite land products are used in
  operational weather and climate prediction
• Lack of proven operational land assimilation
  methods have been a limit
• GOAL Implement Kalman Filter to assimilate land
  satellite data products into the Noah land
  surface model installed in the Land Information
  System (LIS)

3
Objectives

• Identify relevant MODIS AMSR data products
• Implement the Kalman Filter in LDAS/LIS
• Examine the efficiency and benefits of
  assimilating the satellite data products into the
  NOAH LSM.

Progress

• Satellite data products selected
• Data assimilation technique implemented
• Results of soil moisture data assimilation
• Results of using MODIS land cover data

4
Data Sets for Testing the DA Features in LIS

• SGP99 TMI observations Resampled and calibrated
  soil moisture retrievals from TRMM Microwave
  Imager (TMI) for the SGP99 domain for the days
  from 8 to 20th of July, 1999.
• 2. Synthetic 0-2cm SM from Mosaic or Noah LSMs
  simulated soil moisture Mosaic and Noah LSM in
  LIS and the NLDAS atmospheric forcing data set
  for the days from 18 June to 20 July, 2002
• 3. NASAs AMSR-E soil moisture data product
  Level 3 soil moisture retrievals since 18 June,
  2002
• 4. MODIS land cover replace AVHRR with MODIS LC
• 5. MODIS snow cover nudging model snow
  cover/depth/ SWE with MODIS and in situ (SnoTEL)
  snow data
• 6. MODIS LST update 4 layer soil temperature
  with MODIS LST using Kalman filter DA could also
  use GOES LST?

5
AMSR-E Surface Soil Moisture Retrievals
Version B02
Version B00
Version B01

• AMSR-E SM algorithm changes
• Newest algorithm starts on 2/15/05
• Some areas do not have retrievals
• Will be assimilated at 0.25 grids
• May directly assimilate TB data if retrievals are
  suspect.

6
AMSR-E Model Soil Moisture Evaluation
Averaged soil moisture plot from 17 sites
(SMEX03-Georgia) over AMSR-E 1/4 degree grid.
Noah (10 cm and 5 cm layer SM), CLM (4.5 cm
layer, layer 1 layer 2), SCAN (just one station,
5 cm), AMSR-E (2 cm layer), SMEX03 (6 cm layer).
7
AMSR-E Soil Moisture Evaluation
8
Land Information Systems (LIS)

• LIS is restructured from LDAS using
  object-oriented programming and parallel
  computing to run various land surface models for
  various domains and resolutions
• The O-O programming technology allows adding new
  feathers (such as data assimilation) to LIS as
  new plug-ins without major modifications.

9
LIS Flow Chart and Data Assimilation
Implementation
Domain Init read card, define grids, allocate
mem, tile spec, etc.
LSM Init tile allocate, LSM comp registration,
etc.
Base Obs Forcing Init standardize format, mem,
etc.
DA Init algorithm/data registration, etc.
LSM Setup wire forcing input, static param,
output struct, et.
Read Restart LSM state variable initial value,
etc.
Tick Time
Done
Read Base and Obs Forcings
Transfer Grid Forcings to Tiles
Run LSM on Tiles
Write LSM Output
Write Restart
10
Data Assimilation Algorithms Implemented
Direct Insertion (DI) replace LSM states when
corresponding observation data are available
(also includes snow rule-based updating) Xa
Z (or conditional updating) Extended Kalman
Filter (EKF) correct LSM states by weighing
model forecasts and observations with their error
covariance K PHT/HPHT R P (I-KH)P Xa
Xb KZ h(Xb) Ensemble Kalman Filter
(EnKF) correct LSM states by weighing model
forecasts and observations with their error
covariance Pym S(Xy-µy)(Xm-µm)/n K Pym/(Pmm
R) Xa SXb K(Z Xb)/n
11
SGP99 TMI SM Data Assimilation with Mosaic LSM
Wet Start
TMI
DI
EKF
12
SGP99 TMI SM Data Assimilation with Mosaic LSM
Wet start, 0-2cm Layer
No DA DI EKF TMI Obs
o

• For wet start case, KF DA advantage is more
  significant.

13
SGP99 TMI SM Data Assimilation with Mosaic LSM
Wet start, 2-148cm Layer
No DA DI EKF

• KF DA uses the correlations between the different
  soil layers in the Mosaic LSM.

14
SGP99 TMI SM Data Assimilation with Noah LSM
ReStart
TMI
DI
EKF
15
SGP99 TMI SM Data Assimilation with Noah LSM
Dry start, 0-10cm Layer
No DA DI EKF TMI Obs

• SM DA with Noah LSM needs special treatment for
  using 0-2cm SM obs to update 0-10cm top soil
  layer SM of the LSM
• Directly using 0-2cm SM for the 0-10cm SM of Noah
  LSM may be misleading.

o
16
SGP99 TMI SM Data Assimilation with Noah LSM
Dry start, 10-40cm Layer
No DA DI EKF

• Second layer SM of Noah LSM did not get updated
• There is no SM correlation between the Noah LSM
  soil layers?

17
EnKF Assimilation of Synthetic SM Data
Open-Loop LSM Run
Synthetic Obs
EnKF Assimilation
Noah
Mosaic

• Twin experiment successful

18
EnKF Assimilation of AMSR-E SM Retrievals
LSM Run
AMSR-E
EnKF Assimilation
Noah
Mosaic

• Very low variability in resulting soil moisture
  field
• We just dont believe that this product is closer
  to the truth due to problems with AMSR soil
  moisture

19

• CDF Matching AMSR-E to Model Simulations
•  
• Purpose LSM simulations accuracy is uncertain
  AMSR-E retrievals are biased with low variation.
  Data assimilation requires unbiased model and
  observation what to do when we dont know the
  truth?
• The cumulative distribution function (CDF)
  matching method used in Reichle Koster (2004)
  can scale the AMSR-E retrievals to the scales of
  the simulations of LSMs to be used for
  assimilation

20
EnKF Assimilation of Scaled AMSR-E SM Retrievals
AMSR-E SM
LSM Run
Scaled AMSR-E
EnKF Assimilation
Noah
Mosaic
21
Impact of MODIS LC Data on Noah LSM Simulations
MODIS V4 UMD land cover
MODIS V3 UMD land cover
AVHRR UMD land cover
Rio Grande River Basin in New Mexico Below
Elephant Butte Dam
Arsenault et al. 2005
22
Differences between (1) AVHRR run and (2)
MODIS-V3 May 30, 2002 (18 Z)
Latent Heat Flux (W m-2)
Top 10 cm Soil Temperature (Celsius)
Sensible Heat Flux (W m-2)
These figures show the differences in latent heat
flux, sensible heat flux and the top layer soil
temperature for the Noah LSM.
Arsenault et al. 2005
23
Latent Heat Flux (W m-2) AVHRR run MODIS3 run
Albuquerque, NM area May 30, 2002 (18Z)
24

• Summary
•  
• Three data assimilation algorithms (DI, EKF,
  EnKF) have been implemented in the Land
  Information System (LIS) to assimilate soil
  moisture observations into Noah LSM
• The LIS DA capability has been tested with
  difference soil moisture observation data sets
• MODIS land cover shows large impact on LSM
  Temperature and heat fluxes
• Future Work
• Further quantitative evaluation and optimization
  of Noah DA
• Expand validation of assimilated soil moisture
  results.
• Optimize ensemble perturbation procedures
• Finalize AMSR-E scaling philosophy
• Explore brightness temperature assimilation
  (CRTM)
• Expand to soil moisture and snow cover
  assimilation

25
Impact of Assimilating MODIS Snow Cover Data
21Z 17 January 2003
SNOTEL and Co-op Network SWE (mm)
Rodell et al., 2003
Enhanced MODIS Snow Cover ()
IMS Snow Cover
Control Run Mosaic SWE (mm)
Assimilated Mosaic SWE (mm)
Mosaic SWE Difference (mm)