The South America Land Data Assimilation System (SALDAS) Luis Gustavo G. De Goncalves [email protected] ESSIC University of Maryland NASA Goddard

The South America Land Data Assimilation System(SALDAS)

Luis Gustavo G. De [email protected]

ESSIC University of MarylandNASA Goddard Space Flight Center

LIMA Water Cycle Capacity Building workshopNovember 30 – December 4, 2009

You are going in the right general direction…

…but when you open your eyes, you correct towards the center of the road

(Land surface model)

(Observations)

Data Assimilation:Sleepy Driver Analogy

Courtesy Matt Rodell

Weight matrix K depends on model uncertainty.Extended KF:

Linearized dynamic matrix equation for P.

Horizontal error correlations too expensive.EnKF:

Use sample covariance from ensemble.

Nonlinear propagation of uncertainty.Can account for horizontal error

correlations.

Update: xi

(+) = xi(-) + K [ Z - H xi

(-)]

K = P Ht [ H P Ht + R]-1

X = soil moisture state (i-th ensemble member)(-) before update (+) after update

Z = remote sensing dataH = measurement operatorK = weight matrixP = state error covariance

(model uncertainty)R = measurement error cov

(meas. uncertainty)

Ensemble Kalman Filter

Slide credit: Paul Houser


Goal: combine local observations and parameters with NASA advanced hydrological modeling expertise and capabilities to improve Global and SA NWP, climate and water management through collaboration with various centers (government, universities and research institutes)

•Missions•Land Information System

NASA/GSFC•Local resources (observations)•Human Resources•Central to South America

CPTEC/INPE

SALDAS

REGIONALWeather and ClimateWater Management

GlobalHydromet Databanks

CapacityBuilding

South American Land Data Assimilation System (SALDAS)


Background

• SALDAS project seeks to provide accurate, near-real-time and retrospective land surface states over South America

• Quality of land surface model (LSM) output is closely tied to the quality of the meteorological forcing data used to drive the model

• Model and observation-based data used to create high-quality forcing data used by Noah, SSiB, SiB2, CLM2, MOSAIC, and VIC (to be tested) LSMs– Retrospective (2000-2005, CPTEC)– Real-time (2002-Present, CPTEC)

GRACE Data Assimilation

three snow layerssurface excessroot zone excess“catchment deficit”

Catchment LSM (Koster et al., 2000):

z

• The Gravity Recovery and Climate Experiment (GRACE) satellite mission produces monthly maps of Earth’s gravity field with enough precision to infer changes in total terrestrial water storage (groundwater + soil moisture + surface water + snow) over large areas (> 160,000 km2)• Catchment land surface model• Ensemble Kalman smoother data assimilation


Results have higher resolution than GRACE alone, better accuracy than model alone.

GRACE TWS anomalyJanuary 2003 – June 2006

GRACE Assimilating Catchment LSM TWS anomaly, mm

January 2003 – June 2006

From scales useful for water cycle and climate studies…

To scales needed for water resources and agricultural

applications


Zaitchik, Rodell, and Reichle, J. Hydromet., 2008.

Upper Mississippi

colu

mn

wat

er (m

m)

Ohio-Tennessee

colu

mn

wat

er (m

m)

Missouri

colu

mn

wat

er (m

m)

Lower Miss-Red-Arkansascolu

mn

wat

er (m

m)

Modeled Water Storage

Model-GRACE Assimilation

GRACE Water Storage

LDAS models produce continuous time series; near-real time capable.

Mississippi River sub-basins

Daily estimates are critical for

operational applications

Monthly GRACE data anchor model

results in reality


LDAS models produce continuous time series; near-real time capable.GRACE Data Assimilation

LPB Implementation plan 2005

Schematic representation of the land surface modeling into the NASA Land Information System (LIS), a multi-LSM distributed framework where SALDAS in built upon. (S.V. Kumar, et al, 2006)


NASA Land Information System (LIS)

• 3-Hourly files• 1/8th and 1/10th Degree over

Equator• Quality controlled, adjusted for

terrain height• Modeled and observation-based

fields

Spatial Domain


Forcing Specification

Atmospheric Forcing Fields• Model-Based Estimates of Standard Climate Station Data

– Temperature ( 2 m assuming grass)– Specific Humidity ( 2 m assuming grass)– U East-West Wind Component (10 m assuming grass)– V North-South Wind Component (10 m assuming grass)– Surface Pressure ( 0 m assuming grass)

• Observation-Based Data– Downward Shortwave Radiation– Precipitation


Terrain Height Adjustment


• ETA temperature, pressure, humidity and longwave radiation adjusted for differences in ETA versus LDAS terrain height

• Temperature and pressure corrected using standard lapse rate• Specific humidity and longwave radiation corrected by holding relative humidity

constant

• Corrections of up to 3.7K, 60hPa, 2.24W/m2, 0.06 Kg/kg


Forcing File Creation

• Observations not always available, so CPTEC/SARR and CPTEC/ODAS data used as base– Current stage:– SARR, 6 hourly, 40km, 2000-2004 (planned 1979-2008)– Next step:– ODAS, 3 and 6 hourly, 20km, 2002-present (planned 20Km)

• Spatially interpolated to 1/10th degree• Temporally interpolated to 3-hourly data• To be quality controlled using ALMA ranges

CPTEC ETA South American Regional Reanalysis SARRCPTEC ETA Operational Data Assimilation System ODAS

Spatial Distribution of GPCC/GTS Raingauges

Spatial Distribution of CPTEC/INPE Raingauges

(a) (b)

Spatial Distribution of GPCC/GTS Raingauges

Spatial Distribution of CPTEC/INPE Raingauges

(a) (b)

Precipitation

Automated Surface Stations

Courtesy: INMET/INPE

Observation-based forcing

• Model-based data subject to model error, so observations used when possible

• Radiation– GOES-CPTEC downward shortwave– GOES-CPTEC PAR (not implemented yet)– GOES-CPTEC skin temperature (not implemented yet)

• Precipitation– CPC daily gauge data– Combined TRMM-raingauge– Sub daily automated surface stations network– Rain gauges from various SA agencies (~5x > GTS)


• GOES data processed at CPTEC/DSA (Divisao de Satelites Ambientais: Environmental Satellites Division) to create 1/25 degree, hourly, instantaneous surface downward shortwave radiation, PAR and skin temperature fields– Interpolated to 1/10th degree

GL 1.2 GOES downward shortwave radiation (W/m2)

SARR downward shortwave radiation (W/m2)

Merged SALDAS downward shortwave radiation (W/m2)


Radiation Forcing Processing Statistics

Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec0

10

20

30

40

50

60

70

80

90

100

110

Shortwave radiation data frequency GL1.2 Model to 2000

%

Months

9Z 12Z 15Z 18Z 21Z monthly rel. freq.


10

20

30

40

50

60

70

80

90

100

110


%

Months



10

20

30

40

50

60

70

80

90

100

110


%

Months



10

20

30

40

50

60

70

80

90

100

110


%

Months



Observed RadiationMean difference

SARR - GL1.2 (W/m2)15Z - January 2004

Mean GL1.2 (W/m2)15Z - January 2004

Mean difference SARR - GL1.2 (W/m2)

15Z - June 2004

Mean GL1.2 (W/m2) 15Z - June 2004

Observed Radiation

Precipitation• Make use of TRMM and raingauges analysis data to form best available

product—a temporally disaggregated 3-hourly value

Temporal Disaggregation Process Description: Use of rain gauge to correct satellite bias precipitationSpatial Resolution: 0.25 degreeTemporal Resolution: total daily precipitationDomain: South AmericaMethodology: Use TRMM sub-daily precipitation pulses to dissagregate the total daily

amounts.

• Make use of TRMM and rain gauges analysis data to form best available product—a temporally disaggregated 3-hourly value

Temporal Disaggregation Process

Data interpolated to 1/8th degree

TRMM-3B42RT total daily ingestion at 0.25º

combined with rain gauges

Temporally dissagregate raingauge to 3-hour using 3B42RT hourly weights,

conserving the total daily precipitation

3B42RT 3-hourly precipitationpulses at 0.25º

Note:*3B42RT data used to derivetemporal disaggregation weights*Sum of hourly data values equals original daily TRMM/gauges total

Note:•3B42RT data used to derive temporal disaggregation weights•Sum of hourly data values equals original total daily TRMM/gauge•Additive/Ratio Bias correction (Vila et al, 2009) JHM


Implement data assimilation techniques (EKF, EnKF) for low data density regions Participants: CPTEC, NASA, COLA, GMU, University of Arizona, Utah University, IPH,

Universidade de Vicosa, Universidade de Santa Maria

Streamflow data assimilation (use of ANA monitoring capabilities for determining average soil moisture over low data density basins)Soil moisture data assimilation (combine AMSR-E Aqua satellite estimates with ground based observations over areas with low vegetation cover) PCD’sSkin temperature data assimilation (remote sensed skin temperature, DSA derived, combined with surface observations) PCD’s

SENAMHI (Peru)?


Eta

CPTECRPSAS

Observations Error Stats

LSM

LSM4DDA

EKF/EnKF

Observations Error Stats

Eta

LSM

An operational implementation of a land surface data assimilation into the CPTEC-INPE ETA/PSAS/LTEKF atmospheric data assimilation cycle

… Timestep1

Timestep2

Timestep …3

SALDAS SALDAS SALDAS

Modeling Activities Summary

LDAS Applications

LDAS ODAS

Uncoupled Land Data

Assimilation

CoupledEarth SystemModel with

Atmospheric,Land and

Ocean Data Assimilation

ADAS

Current

Future


La Plata Basin: high spatial and temporal resolution runs help to improve the understanding of the hydrological and meteorological processes over the region. Under the LIS framework, SALDAS can be set to run at up to 1Km spatial resolution with hourly output.

La Plata basin volumetric soil moisture at 1Km resolution (January 2000)

La Plata basin integrated total runoff (Kg/m2) at 1Km resolution (January 2000)


High resolution satellite-based 1Km crop mapping over the LPB

Courtesy: Mutlu Ozdogan

Four LSM’s (NOAH, CLM, MOSAIC and VIC) were run from 1990-2005 using GDAS and NASA’s MERRA as atmospheric forcing.

We are interested on seeing how different models under the same conditions simulate the same period of time.

The focus is on the models spread and mean for simulated water balance variables: surface and subsurface runoff and evaporation and how they compare with input precipitation.

We concentrate on three major basins in South America: Amazonas, Tocantins and La Plata, each with its unique characteristics.

South American Land Data Assimilation System (SALDAS)Multi-model ensemble


AM

TC

PL

Three major South American hydrological basins analyzed: Amazonas (blue), Tocantins (green) and La Plata (purple). High resolution simulations (1Km) are being performed at the Parana sub-basin for the La Plata Basin as shown in the figure details.


-4e-005

-2e-005

0

2e-005

4e-005

6e-005

8e-005

0.0001

0.00012

0.00014

0 2 4 6 8 10 12 14

Kg

/m2

/s

Month

Mean Monthly Water Balance 1990-2005Precipitation

Surface RunoffSubsurface RunoffEvapotranspirationChange in Storage

0

1e-005

2e-005

3e-005

4e-005

5e-005

0 2 4 6 8 10 12

Kg

/m2

/s

Month

Mean Monthly Surface Runoff 1990-2005CLM

NOAHMOSAIC

VICAverage

0

1e-005

2e-005

3e-005

4e-005

5e-005

6e-005

0 2 4 6 8 10 12

Kg

/m2

/s

Month

Mean Monthly Subsurface Runoff 1990-2005CLM

NOAHMOSAIC

VICAverage

2.5e-005

3e-005

3.5e-005

4e-005

4.5e-005

5e-005

5.5e-005

6e-005

0 2 4 6 8 10 12

Kg

/m2

/s

Month

Mean Monthly Evapotranspiration 1990-2005CLM

NOAHMOSAIC

VICAverage

AM AM

AM AM


Summary

• Model and observation based data merged to create robust, accurate 1/10th degree 3-hourly forcing data set– CPTEC-SARR/ODAS/ETA data serves as base– CPTEC-DSA/GOES, TRMM/raingauge data used to

augment data set• Common set of forcing integral to SALDAS LSM

intercomparisons• Five years archived, with continuing production• Validation effort proceeding


Summary

• High resolution simulations up to 1Km over selected regions (e.g. La Plata basin) with soil moisture and soil temperature DA

• Improvement of ancillary data over regions outside Brazil based on in situ and remote sensing observations (SENAMHI?)

Gracias

Luis Gustavo G. De [email protected]

ESSIC University of MarylandNASA Goddard Space Flight Center

LIMA Water Cycle Capacity Building workshopNovember 30 – December 4, 2009

http://ldas.gsfc.nasa.gov/SALDAS/

ftp://ftp1.cptec.inpe.br/jgerd/SALDAS/Assimila.html



ftp://ftp1.cptec.inpe.br/jgerd/SALDAS/Assimila.html

Integrating NASA Earth Sciences Research results into Decision Support Systems for Agriculture and Water Management in South America

Ernesto H. Berbery3, Eric F. Wood4, Maria Assuncao F. Silva Dias5, Osvaldo Luiz L. De Moraes6 and Jurandir Zullo7

&

Latin America Water Resources and Capacity Building

David L. Toll2, Ted Engman2

Luis Gustavo G. de Goncalves1,2

1Earth System Science Interdisciplinary Center, University of Maryland2Hydrological Sciences Branch, NASA Goddard Space Flight Center3Department of Atmospheric and Oceanic Science, University of Maryland4Princeton University5Centro de Previsao do Tempo e Estudos Climaticos/Instituto Nacional de Pesquisas Espaciais6Universidade Federal de Santa Maria7University of Campinas

Paul Simon Water Act for the Poorlegislation that requires special attention to be given to

semi-arid regions such as Northeast Brazil, western Argentina and Atacama


NASA Applied Sciences starting FY10

Use of NASA remote sensing and modeling products combined with surface observations at various scales to improve decisions support systems in agriculture, drought and water resources management for South America (SA)

Build upon a partnership between NASA and various U.S. and international agencies and universities to contribute to the dissemination of NASA Earth Sciences research results within that continent

Provide valuable information based on NASA Earth Sciences products to South American national agencies and other end-users

OUTLINE

http://sucuri.cpd.ufsm.br/








Development of methods to estimate climatic risks of regional individual crops, intercropping and cattle-farming integration systems

The “Agricultural Zoning for Climatic Risks” (AZCR) is a program developed in Brazil since 1995, to provide guidance information for agricultural practices.

THE AZCR

Cattle farming systems are adding up quality to the grain production

and yield of pastures, mitigating the environmental impacts and reducing

the pressure on the Amazonian forest









“Incorporating high quality observations of surface conditions from NASA satellites and other observing systems to quantify continental scale daily ET at 5~20Km spatial resolution and water storages and fluxes at basin scale will enable enhancement of decision support tools for agriculture, human use and disaster mitigation over South America.”

EOS multi-sensor (AIRS, CERES and MODIS) based

evapotranspiration

The Gravity Recovery and Climate Experiment (GRACE) estimates of monthly variations in terrestrial water storage (TWS)

The Tropical Rainfall Mission Measurement (TRMM) derived precipitation

AMSR-E and TRMM Microwave Imager (TMI) soil moisture

NASA’s Land Information System (LIS) and the South American Land Data Assimilation System (SALDAS)

Land Surface Modeling Framework

CPTEC/INPE atmospheric models

INMET surface observations network









PRINCETON UNIVERSITY Remote Sensed ET Estimation

Data Type Variable Unit Source Platform Resolution

Surface Meteorological

Data

Air temperature

Pressure

U-Wind

V-Wind

Vapor Pressure

C

KPa

m/s

m/s

KPa

AIRS

AIRS

CERES (GMAO)

AIRS

AQUA

AQUA

AQUA

AQUA

AQUA

25 km*

25 km*

20 km

20 km

25 km*

Radiative Energy Flux

Incident SW Incident LW

W/m2

W/m2

CERES

CERES

AQUA

AQUA

20 km

20 km

Vegetation Parameters

Emissivity

Albedo

LAI

Veg. Type (MODIS UMD

Classification)

-

-

-

-

MODIS

MODIS

MODIS

MODIS

AQUA

AQUA

AQUA

TERRA

5 km

5 km

1– 5 km

20 km

Processing the Data streams

*Processed to 25-km at Princeton“Fast response” variables put into global, hourly UTC files“Slow” MODIS land cover variables are in 8-day or 16-day files

}

7 GB raw RS data/day

Processing the Princeton Remotely-Sensed ET Product (beta version)

necessary upscaling & downscaling

Use the calculated EF (soon from both Aqua and Terra) to scale SRB diurnalinsolation data to obtain daily (weekly, monthly) LE and ET [mm/day]

Penman-MonteithInstantaneous Retrieval

(1:30pm local)

SEBSInstantaneous Retrieval

(1:30pm local)

apply quality flag filters

Determination of cell-specific, instantaneous EFEF = LE / (Rn – G )

Figure. Mean seasonal maps of ET from RS-PM and VIC for 1984-2000.

ET Estimates (Mexico): Comparison with VIC LSM

Jan

Apr

Jul

Oct

RS-PM ET VIC ET VIC – RS-PM








NASA Water Resources Program Project Support for Activities in Latin America

“Initiate capacity building {and end to end projects} programs to develop tools for using remote sensing data in support of water management, and to show the value of Earth observations generally in water resource management. The program will be initiated in Latin America” {GEO Tasks WA-06-07 & DI-07-01}

The NASA Water Resources Program may assist:

1) Assist with workshop support including the training of students and travel for US visits.

2) Support graduate students & post-docs3) End to End Projects with Decision Support Systems









Collaborative work for the past 4 years (mostly unfunded) with the Brazilian Center for Weather Forecast and Climate Studies (CPTEC - an equivalent no NOAA/NCEP in SA) a division from the Brazilian National Institute for Space

Research (INPE - an equivalent to NASA in SA). CPTEC/INPE is a lead and reference institution in Latin America for operational and research modeling.

This collaborative work includes promoting interaction between Latin America students and researchers and US institutions as well as Capacity Building

Funded Activities

Recently Funded

La Plata basin integrated total runoff (Kg/m2) at 1Km resolution (January 2000)

NASA THP - La Plata Basin combine NASA products with local observations to improve understanding of the hydrological and meteorological processes over the region NDVI1981-2000 trends: surrogate for

primary production from NOAA-AVHRR images. Red: decrease Blue: increase. [Courtesy of Jobbagy.]

IAI - La Plata Basin Land Use/Land Change due to natural and anthropogenic causes

NASA LBA Ecology – LSM Intercomparison Project leverage on 8 flux sites in the Amazon region. SALDAS forcing used for wall-to-wall intercomparison.

NASA Applied Sciences Program proposal to improve decisions support systems in agriculture, drought and water resources management for South America

Amazon Observational Network of eddy flux tower sites (red dots) on a map of vegetation types in Amazônia and Brazil.

mm Wm-2(a) (b)

SALDAS – CTR precipitation SALDAS – CTR latent heat flux

mmmm Wm-2Wm-2(a) (b)

SALDAS – CTR precipitation SALDAS – CTR latent heat flux

Regions with significant impacts on latent heat flux when comparing initial conditions for operational NWP models models with well-balanced SALDAS fields




2006 Enrique Rosero (Ecuador) Amazon LSM parameter calibration

2007 Rafael Rosolem (Brazil)Amazon Hydrology & Calibration

2009 Mario Quadro (Brazil)La Plata Precipitation Recycling

Debora Roberti (Brazil)Wet Lab for regional soil moisture and ETIn support to agriculture

2008 Joao Mattos (Brazil)Operational Land-Surface DA

2009 Claudia Ramos (Brazil)AMSR-E Soil Moisture DA

Visiting Researchers and Students to NASA/GSFC/HSB



Capacity Building Meetings and Workshops

Public: graduate students and young scientists

Two weeks hands-on training in hydrology, meteorology, ecological modeling and data assimilation

More than 100 registrations from North and South America, Africa, Europe and few from Asia…

To be held at the ITAIPU hydropower plant at the border of Brazil, Argentina and Paraguay

Inter-American Institute

PI: Hugo Berbery (UMD)

Documents

The South America Land Data Assimilation System (SALDAS) Luis Gustavo G. De Goncalves [email protected] ESSIC University of Maryland NASA Goddard