global modeling and assimilation office nasa goddard space flight center phone: 301-614-5693

Rolf Reichle, Gabrielle De Lannoy, Clara Draper, Bart Forman, Randy Koster, Qing Liu & Ally Toure

Global Modeling and Assimilation OfficeNASA Goddard Space Flight CenterPhone: 301-614-5693Email: [email protected]

AMSR-E Science Team MeetingAsheville, NC

28-29 Jun 2011

Update on AMSR-E soil moisture and snow data assimilation

Outline

MERRA (atmospheric) reanalysis• Baseline skill of global estimates

without land data assimilation

Soil moisture assimilation• AMSR-E and precipitation corrections• AMSR-E and ASCAT

Snow assimilation• AMSR-E and MODIS

MERRA: Recently completed GEOS-5 re-analysis

NASA/GMAO re-analysis

• 1979-present (updated w/ ~1 month latency), global,Lat=0.5º, Lon=0.67º, 72 vertical levels

• MERRA-Land: Enhanced product for land surface hydrological applications (Reichle et al., J. Clim., 2011)

Precipitation MERRA MERRA + GPCPv2.1

Surface Tair, Qair, SW, LW, etc. MERRA MERRA

Catchment model version MERRA Fortuna

Data product MERRA MERRA-Land

Gauge & satellite

estimates

Latest GEOS-5 model version

Soil moisture validation (2002-2009)

MERRA-Land has • better soil moisture

anomalies than MERRA (attributed to GPCP-based precip. corrections), and

• is comparable to ERA-Interim.

Skill (pentad anomaly R)vs. SCAN in situ observations

Improvements also in canopy interception, latent heat flux, and runoff (not shown).

Reichle et al. JCLIM (2011) submitted.

Anomalies ≡ mean seasonal cycle removed

Skill metric: Anom. time series corr. coeff. R

Snow depth validationMERRA-Land v. CMC snow analysis

Pentad anomaly R (2002-2009)

CMC snow analysisDensity [stations/10,000 km2]

Not shown: MERRA and MERRA-Land have similar skill.Similar result for comparison against in situ data (583 stations) and for snow water equivalent (SWE).

MERRA -Land shows good skill without station-based snow analysis.

Low R values in areas without in situ obs.

Outline





Key limitations of satellite soil moisture retrievals

Surface layer (0-5 cm)

“Root zone” layer (0-100 cm)

Satellite soil moisture retrievals:1) are sensitive to moisture and

temperature only in a 5 cm surface layer (under 0-5 kg/m2 vegetation),

2) have limited coverage in time and space, and

3) are subject to measurement errors.

Models suffer from precipitation uncertainties and other errors Use data assimilation to merge retrievals into land surface model.

Objective

What are the relative contributions of precipitation observations and

(AMSR-E) soil moisture retrievals to the skill of soil moisture estimates in a land data assimilation system?

Combine into 12 experiments:

Experiments

Four different precipitation forcing datasets:• MERRA

Reanalysis, 0.5 deg• MERRA+CMAP• MERRA+GPCPv2.1Pentad, 2.5 degGlobalSatellite + gauges• MERRA+CPC Daily, 0.25 degCONUSGauges

Separately assimilate two different AMSR-E soil moisture retrieval datasets:• NSIDC• LPRM (X-band)

NASA/GMAO land data assimilation system, including the GEOS-5 Catchment model

At the pentad (daily) and 2.5 deg (0.25 deg) scale, the corrected re-analysis precipitation matches the correcting observations.

MERRA MERRA +NSIDC MERRA +LPRMMERRA+GPCP MERRA +GPCP +NSIDC MERRA+GPCP +LPRMMERRA+CMAP MERRA +CMAP +NSIDC MERRA+CMAP +LPRMMERRA+CPC MERRA +CPC +NSIDC MERRA+CPC +LPRM

1) USDA SCAN stations (only 37 of ~120 suitable, single profile sensor, surface and root zone soil moisture), and

2) four AMSR-E CalVal watershed sites (RC, WG, LW, LR; distributed sensors, only surface soil moisture)

Validating in situ observations

x SCAN+ CalVal

Anomalies ≡ mean seasonal cycle removed

Skill metric: Anom. time series corr. coeff. R

Soil moisture skill increases with• precipitation

corrections and • assimilation of

surface soil moisture retrievals.Improved root zone

soil moisture!

Precipitation corrections v. retrieval assimilation

Different precipitation forcing inputs

Skill v. SCAN in situ obs

AMSR-E retrievals

No assimilation

AMSR-E assimilation

Liu et al. JHM (2011) doi:10.1175/JHM-D-10-05000.

(reference)


Extract average skill contributions of precipitation corrections and retrieval assimilation:

Precip. corrections and retrieval assimilation contribute:• roughly evenly and• largely independently to skill improvement.

Results from single sensor per watershed (SCAN data) are consistent with those from distributed CalVal in situ sensors.

R = skill improvement over reference model integration


Liu et al. JHM (2011) doi:10.1175/JHM-D-10-05000.

Outline





Model and Assimilation

Active vs. passive microwave soil moisture retrievals

Model: Catchment modelASCAT: Radar; C-band, TU ViennaAMSR-E: Radiometer; X-band, LPRMPeriod: 2007-2010

Best result for assimilation of data from both sensors.

AMSR-E better in dry conditions.ASCAT better in more humid conditions.

Model and Retrievals

Soil moisture error std-dev (Triple Co-location)Three estimates: AMSR-E, ASCAT, model.1. Assume independent errors.2. Convert data to standard-normal deviates.3. Obtain dim.-less error std-dev estimate.4. Scale error std-dev into units of model variability.

AMSR-E (LPRM-C)

m3/m3

m3/m3

m3/m3

ASCAT model

AMSR-E minus

ASCAT

Modelsoil

moisture variability

Red: AMSR-E better in south-west deserts but also in eastern US.

Blue: ASCAT better in central US.

Dominant pattern in err-std-dev is soil moisture variability.

Outline





Assimilation of MODIS and AMSR-E snow observations

Noah land surface model (1 km resolution)100 km X 75 km domain in northern Colorado

Multi-scale assimilation of • AMSR-E snow water equivalent (SWE)

and • MODIS snow cover fraction (SCF).

Validation against in situ obs from COOP (Δ) and Snotel (▪) sites for 2002-2010.

AMSR-E MODIS

MODIS SCF also improves timing of onset of snow season (not shown).

Assimilation of MODIS snow cover fraction (SCF)

MODIS SCF successfully adds missing snow,

MODIS

SCF assim.

Noah

… except during melt season.

12 Oct 09 18 Nov 09 7 Jan 10 2 Mar 10 10 Apr 10 31 May 10

De Lannoy et al. WRR (2011) submitted.

Assimilation of AMSR-E snow water equivalent (SWE)

AMSR-E does not observe thin snow packs no improvement at start and end of snow season.

Some improvement at lower elevations with shallow snow pack (COOP sites).Problematic in mountains with deep snow packs (SNOTEL sites).

AMSR-E

SWE assim.

Noah

12 Oct 09 18 Nov 09 7 Jan 10 2 Mar 10 10 Apr 10 31 May 10


SWE anomalies (40.38N, 106.66W)

AMSR-E snow water equivalent (SWE)

SWE 1 Mar 2004

AMSR-E MERRA cm

Example: Anomaly time series at SNOTEL site.

AMSR-E SWE retrievals and MERRA differ a lot in absolute terms.


Example: Snapshot over North America

AMSR-E snow water equivalent (SWE)

AMSR-E v. MERRA MERRA v. CMC

Jan

Feb

Mar

all months

AMSR-E SWE retrievals and MERRA differ a lot in terms of anomaly correlation, while MERRA is consistent with CMC (ultimately based on in situ obs. of snow depth).

AMSR-E SWE retrievals not ready for assimilation.

Anomaly R (2002-09)

Tedesco et al., IEEE/TGARS, 2010: (DOI:10.1109/TGRS.2009.2036910)

Novel retrieval approach using dynamic snow depth from land surface model, but …“… the novel retrieval approaches, […] do not outperform [snow depth] estimates from the land surface model alone ...”

“… it is possible to simulate […] brightness temperatures by means of [electromagnetic] models driven with inputs derived from snow pit measurements.”

“ … [but such approaches,] at the global scale, face tremendous difficulties with modeling key processes such as grain size evolution.”

AMSR-E and global snow models

∫−

=t

t0d

)snowdepth()Tair()Tground(cTGI(t) τ

τττ

Model:

Use TGI as surrogate for snow grain size?

Snow temperature gradient index (TGI)

Works best in cold, dry snow conditions.

one grid cell

ΔT B

[K]

R(TGI, ΔTB)

oneseasonno anomalies!

AMSR-E: ΔTB = TBv18–TBv36

Josberger and Mognard, 2002

Multi-regression approach not likely to work. Physically-based radiative transfer model may (or may not) work.

AMSR-E TB differences vs. snow model statesR(SWE, TBv10–TBv18)

Feb

Similar results for R(SWE*TSURF, ΔTB) [not shown].

R(SWE, TBv10–TBv36) R(SWE, TBv18–TBv36)

R(TGI, TBv18–TBv36)R(TGI, TBv10–TBv36)R(TGI, TBv10–TBv18)

Feb

Summary and outlook

Soil moisture assimilation• Improved surface and root-zone soil moisture from

assimilation of AMSR-E and ASCAT surface soil moisture retrievals

• Precipitation corrections and retrieval assimilation contribute independent and comparable amount of information

• There are differences between AMSR-E products

Snow assimilation• AMSR-E SWE retrievals not ready for assimilation• Still trying to find a way to use AMSR-E TB for SWE

assimilation into global models

Thanks for listening!

Questions?

EXTRASLIDES

Soil moisture data assimilation

Satellite retrievals of surface soil moisture

Soil moisture data product: Surface and root-zone soil moisture

Data Assimilation

Surface meteorology, incl. precipitation

Land model

Assimilating satellite soil moisture retrievals into a land model driven with observation-based forcings yields:– a root zone moisture product (reflecting satellite soil moisture data), and– a complete and consistent estimate of soil moisture & related fields.

Key limitations of model soil moisture

Land model estimates of soil moisture depend critically on the quality of the precipitation forcing data.(Re-)analysis precipitation forcing suffers from:1) short-term errors in

diurnal cycle, and 2) long-term bias.

Correct (re-)analysis precipitation with global gauge- and satellite-based precipitation observations to the extent possible.

Reichle et al. J Clim (2011)

submitted

Precipitation

Correct MERRA precipitation with global gauge- and satellite-based precipitation observations to the extent possible.

Reichle et al. J Clim (2011) submitted

Synoptic-scale errors

1981-2008 [mean=-0.03 mm/d] Aug 1994 [mean=-0.04 mm/d]

Long-term bias

MERRA – GPCPv2.1

Precipitation corrections

MERRAReanalysisHourly0.5 deg

For each pentad and each 2.5 deg grid cell, the corrected MERRA precipitation (almost) matches GPCP observations.

GPCP v2.1Satellite + gaugesPentad2.5 deg

MERRA + GPCPv2.1(hourly, 0.5 deg)

Rescale MERRAseparately for each pentad

and 2.5 grid cell

Runoff

GPCP corrections yield significant improvements in 3 large basins.MERRA and MERRA-Land (0.5 deg) better than ERA-Interim (1.5 deg).Not shown: In all cases the revised interception parameters yield improved runoff anomalies (albeit not significant).

Validation against naturalized streamflow observations from 9 “large” and 9 “small” basins (~1989-2009).

Streamflow skill (3-month-average anomaly R)

MERRA MERRA-Land

Feb

2004

Apr

200

4

Categorical analysis of snow cover fraction vs. MODIS

MERRA SCF agrees well with MODIS SCF observations.

False alarm rate increases in MERRA-Land.

MOD10C2, aggregated to monthly avg. SCF

small errors large errors

Observation error estimation via Triple Co-location

GEOS-5 (model)

Obtain error std-dev from 3 independent estimates:

Estimated anomaly error std-dev of surface soil moisture (2007-10)

ASCAT (active MW)

Units: [m3/m3] Consistent with model anomaly variability.

Active retrievals better in more vegetated areas.

Passive retrievals better in more arid conditions.

LPRM retrievals better than NSIDC.

Work in progress!LPRM-X NSIDC

AMSR-E (passive MW) AMSR-E (passive MW)

GEOS-5 anom std-dev

• active MW• passive MW• modeling

global modeling and assimilation office nasa goddard space flight center phone: 301-614-5693

Documents

merra land

land snow depth

amsre soil moisture

merraland estimates

better soil moisture

situ surface

ascat snow assimilationamsre

situ data