Developing Synergistic Data Assimilation Approaches for Passive Microwave and Thermal Infrared Soil Moisture Retrievals

1
Developing Synergistic Data Assimilation
Approaches for Passive Microwave and Thermal
Infrared Soil Moisture Retrievals
Christopher R. Hain SPoRT Data Assimilation
Workshop May 5, 2009
2
Soil Moisture and Data Assimilation

• Soil moisture is one of the most important
  variables within a land surface model and a
  correct representation is advantageous for
  accurate predictions of sensible and latent heat
  fluxes.

• Satellite-based retrievals of soil moisture
  using both passive microwave (PM) and
  thermal-infrared (TIR) sensors can begin to
  alleviate this lack of ground-based observations.

• Various studies have shown that the assimilation
  of such retrievals with the use of data
  assimilation techniques such as the ensemble
  Kalman filter (EnKF) can provide a more accurate
  representation of soil moisture in land surface
  models (Reichle et al. 2002a,b Crow et al. 2005
  Drusch 2007 Reichle et al. 2007).

3
Atmosphere Land Exchange Inversion Model (ALEXI)

• The differential in surface radiometric
  temperature is computed from GOES Sounder Band 8
  (11.0 micron) during the time period from 1.5
  hours after sunrise to 1 hour before local noon.

• The time-differential Trad drives the change in
  surface temperature and subsequent boundary layer
  growth.

4
Atmosphere Land Exchange Inversion Model (ALEXI)

• ALEXI provides daily estimates of surface fluxes
  (LE H G Rnet) over cloud-free pixels at
  spatial resolutions greater to or equal to the
  native resolution of the TIR data.

• ALEXI provides a valuable signal of surface soil
  moisture conditions over bare soil, root-zone
  soil moisture conditions over dense vegetative
  canopies, and a combination of surface and
  root-zone signal over mixed cover pixels.

• Hain et al. (2009a,b) showed that retrievals of
  a soil moisture proxy using ALEXI surface flux
  estimates agreed well with averaged 0-75 cm soil
  moisture observations over the Oklahoma Mesonet.
  The average error between ALEXI and the
  observations was on the order of 15 to 20, or
  about 0.04 m3 m-3.

5
AMSR-E PM Soil Moisture Retrievals

• PM sensors such as AMSR-E provide retrievals of
  surface soil moisture in the first few
  centimeters depending on the wavelength of the
  sensor.

• This study will make use of the VUA retrieval
  products (Owe et al. 2001 2007) which uses one
  dual polarized channel (either 6.925 or 10.65
  GZz) for the retrieval of both surface soil
  moisture and vegetation water content, while LST
  is derived separately from the vertically
  polarized 36.5 GHz channel.

• This differs from the official NSIDC AMSR-E
  retrieval products because of the use of a higher
  frequency band for retrieval of LST, and the
  parameterization of vegetation water content,
  leaving only the retrieval of soil moisture to be
  optimized (Rudiger et al. 2009).

6
Soil Moisture Retrieval Specifications

• PM and TIR retrievals of soil moisture provides
  a unique opportunity to synergistically exploit
  the strengths of each methodology in a data
  assimilation framework.

ALEXI AMSR-E
Sensing Depth Variable with rooting depth First 2 centimeters
Spatial Resolution 5-10 km 30 - 55 km
Repeat Times Daily 1-2 Days
Data Availability April 1 2000 - current July 1 2002 - current
Flagged Pixels Any cloud cover Precipitating clouds
Average Continental US Coverage 25 to 30 60 to 80
7
Land Information System (LIS) Soil Moisture
Estimates

• The model-based soil moisture product will taken
  from a 11-year long (1997-2008) Land Information
  System (LIS) simulation.

• The LIS will provide soil moisture estimates at
  four layers of the soil profile (0-5 cm, 5-40 cm,
  40-100 cm, and 100-200 cm).

LIS Configuration
Spatial Resolution 10 km
Atmospheric Forcing NLDAS
Land Surface Model NOAH v2.7
8
Intercomparison of LIS NOAH, ALEXI and AMSR-E

• Several studies have performed quantitative
  comparisons between modeled surface soil moisture
  and PM retrievals of surface soil moisture.

• This study attempts to perform a quantitative
  comparison between modeled soil moisture and PM
  soil moisture retrievals consisted with the
  sensing depth signal from ALEXI.

• Root-zone soil moisture from AMSR-E is estimated
  using the surface soil moisture retrievals and an
  exponential filter as shown by Albergel et al.
  (2008)

9
Intercomparison of LIS NOAH, ALEXI and AMSR-E

• Soil moisture datasets from various sources
  typically exhibit very different mean values and
  variability.

• These differences must be removed by
  transforming each dataset through statistical
  methods such as anomaly or CDF matching before
  any data assimilation is performed.

• In this study, 14-day composites of ALEXI and
  AMSR-E soil moisture are scaled into a
  representative value based on the statistical
  properties of the model-based LIS soil moisture.

10
Yearly Spatial Anomaly Correlation
11
Yearly Spatial Anomaly Correlation
r NOAH/ ALEXI NOAH/ AMSR-E ALEXI/ AMSR-E
2003 0.68 0.42 0.38
2004 0.48 0.23 0.13
2005 0.68 0.55 0.50
2006 0.65 0.57 0.47
2007 0.74 0.58 0.56
2008 0.48 0.44 0.40
Average 0.62 0.47 0.41
12
Time Series Anomaly Correlation (2003-2008)

• ALEXI performs better than AMSR-E over most of
  the eastern US, which is consistent with high
  vegetation cover.
• AMSR-E and ALEXI both perform well over the
  central US and western US.
• This analysis highlights the potential of dual
  assimilation of each retrieval to provide added
  skill over a vast majority of the United States.

r(99 CI)0.27 CONUS-average
NOAH/ALEXI 0.58
NOAH/AMSR-E 0.46
ALEXI/AMSR-E 0.41
13
Estimation of Retrieval Error with Triple
Collocation

• The goal of intercomparison studies before
  implementation of data assimilation is to allow
  the user to make a more educated quantification
  of the error structure for each soil moisture
  dataset.

• An estimated value of RMSE can be computed using
  a triple collocation error estimation technique
  (Janssen et al. 2007 Scipal et al. 2009) once
  the soil moisture datasets have been scaled to
  a consistent distribution/climatology and
  assuming each soil moisture dataset has
  uncorrelated errors.

14
Western Texas Lat 34 N Lon -100 W
Northern Alabama Lat 34.5 N Lon -87.5 W
Eastern Montana Lat 47 N Lon -107.5 W
15
Triple Collocation RMSE ( error dynamic range)
16
Implications for EnKF Data Assimilation

• It has been shown the soil moisture retrievals
  both from ALEXI and AMSR-E provide significant
  skill over a large portion of the United States.

• It should be noted that the LIS simulation is
  exploiting a very dense precipitation monitoring
  network, which further validates the
  relationships observed with each soil moisture
  retrieval, which rely on no antecedent
  precipitation information.

• This study also highlights the potential with
  respect to the use of a data assimilation system
  which can implement multiple soil moisture
  retrievals.

• Therefore, it can be hypothesized that the
  neither retrieval will exceed the added skill
  attained by the potential dual assimilation
  technique in a respective single assimilation.

17
Implications for EnKF Data Assimilation

• Based on error estimation using the triple
  collocation technique, there was no observed
  large systematic differences in RMSE between each
  of the three soil moisture datasets.

• Data assimilation work has begun testing the
  impact of each soil moisture retrieval in a
  single assimilation framework, along with future
  work with a dual assimilation framework.

• Quantification of assimilation impact will be
  assessed using ground-based soil moisture
  observations from the SCAN network and the
  Oklahoma Mesonet over the study period of
  2003-2008.

• Additionally, data-denial experiments are being
  formulated in attempt to quantify skill of
  assimilation when using degraded or poor
  observations of parameters such as precipitation.

18
r NOAH ALEXI VUA USDA NSIDC
NOAH 1 0.62 0.45 0.29 0.04
ALEXI 0.62 1 0.40 0.16 -0.03
VUA 0.45 0.40 1 0.23 -0.03
USDA 0.29 0.16 0.23 1 0.60
NSIDC 0.04 -0.03 -0.03 0.60 1