Assimilation of remotely sensed hydrological datasets in the North American Land Data Assimilation System

(NLDAS)

Sujay V. Kumar1, Christa D. Peters-Lidard1, David Mocko1, Rolf

Reichle2, Ben Zaitchik3, Yuqiong Liu1, Kristi Arsenault1, Youlong

Xia4, Michael B. Ek4, George Riggs5, Ben Livneh6, Michael Cosh7

1 – Hydrological Sciences Laboratory, NASA/GSFC, Greenbelt, MD

2 – NASA Global Modeling and Assimilation Office, Greenbelt, MD

3 – Johns Hopkins University, Baltimore, MD

4 – Environmental Modeling Center, NOAA, College Park, MD

5 – Cryospheric Sciences Branch, NASA/GSFC, Greenbelt, MD

6 –Cooperative Institute for Research in Environmental Sciences, Boulder, CO

7 – USDA ARS Hydrology and Remote Sensing Laboratory, Beltsville, MD

2

NLDAS – a real-time and retrospective data assimilation system to provide accurate land surface states for NWP models and end-use applications

NLDAS is a successful collaboration project that has produced nearly 34 years of hourly 1/8th degree surface forcing and land surface model output over CONUS and parts of Canada/Mexico

The new phase of the NLDAS project builds upon NLDAS phase-2 and includes the following:

Assimilation of remotely sensed terrestrial hydrological datasets.

New versions of the land surface models (Noah.3.3, Catchment Fortuna 2.5, VIC 4.1.1, SAC HTET/SNOW17).

Conduct the simulations using a common platform using the NASA Land Information System (LIS).

Comprehensive model evaluation using the Land surface Verification Toolkit (LVT).

North American Land Data Assimilation System (NLDAS)

3

Experiment Setup Model domain: Continental United States (CONUS) at 1/8th degree spatial resolution, including parts of Canada/Mexico (25-53° N; 125-67° W) Forcing data: NLDAS-phase II (NLDAS2) meteorological forcing data. Hourly precipitation includes CPC’s daily PRISM Corrected gauge analysis, downward shortwave radiation bias-corrected using GOES SRB shortwave data, all other fields derived from the NCEP North American Regional Reanalysis (NARR) data. Land surface model: Noah LSM version 3.3 and Catchment version Fortuna 2.5, includes a 15-year spin-up, followed by a 33 Data assimilation method: Ensemble Kalman Filter (EnKF) and Ensemble Kalman Smoother (EnKS) Time period: Jan 1, 1979 to 1 Jan 2012.

4

Soil moisture data assimilation

Soil moisture retrievals are available from low-frequency (C, X, and L-band) active and passive microwave data (SMMR, TMI, AMSR-E, WindSat, SMOS, SMAP, …) In the NLDAS project, we use the following product for assimilation

Essential Climate Variable (ECV) soil moisture product (Liu et al. 2012, Wagner et al. 2012) from ESA; uses C-band scatterometers (ERS-1/2 scatterometer, METOP advanced scatterometer) and multi-frequency radiometers (SMMR, SSM/I, TMI, AMSR-E, Windsat) used in model simulations from 1979-2002.

Land Parameter Retrieval Model (LPRM; Owe et al. 2008) retrievals of AMSR-E used in model simulations from 2002 to 2011.

1990 2000 20101980

SMMR

SSM/I

TMI

Aqua

Corriolis

SCAT

METOP-A

Nimbus-7

DMSP; F8-F16

AMSR-ETRMM

Windsat

ASCAT

ERS-1&2

SMOS

SMAP

Data flagged for light and moderate vegetation, no precipitation, no snow cover, no frozen ground, no RFI are used in data assimilation. The observations are scaled to the LSM’s climatology using CDF matching A 12-member ensemble is used in assimilation. A spatially distributed observation error standard deviation (between 0.02-0.12 m3/m3) specification was employed.

5

Snow data assimilation

Multi-sensor snow products are available – snow depth retrievals from passive microwave (SMMR, SSM/I, AMSR-E) and near IR (MODIS). Passive microwave sensors can measure during cloudy and nighttime conditions, but have coarser spatial resolution, problems in areas with dense vegetation and open water. In the NLDAS project, we use three passive microwave based snow products for assimilation

SMMR (spans 1978-1987), SSM/I (spans 1987-2002) and AMSR-E (spans 2002-2011); SMMR and SSM/I retrievals are based on the Chang et al. (1987) and AMSR-E retrievals are based on the improved retrieval algorithm from Kelly et al. (2009)

Remote sensing observations are bias-corrected using the Cressman method using the in-situ observations from Global Historical Climate Network (GHCN). A 12-member ensemble is used in assimilation.

snow depth from obs background field (from snow depth EDR) weight function, which is a function of horizontal ( ) and vertical difference ( )

6

Terrestrial Water Storage (TWS)

data assimilation

Data assimilation scheme: Ensemble Kalman Smoother (EnKS) modified to accommodate monthly terrestrial water storage anomalies from GRACE. DA-TWS employs the assimilation of 1° x 1° resolution gridded TWS anomaly fields directly, without first averaging them over large river basins. Assimilation updates the model soil moisture and snow states.

GRACE measures changes in total terrestrial water storage, including groundwater, soil moisture,

snow, and surface water.

Animation of monthly GRACE terrestrial water storage anomaly fields. A water storage anomaly is defined here as a

deviation from the long-term mean total terrestrial water storage at each location.

7

Evaluation of NLDAS outputs

Soil moisture:

USDA Soil Climate Analysis Network (SCAN); 123 stations chosen after careful quality control (used for evaluations between 2000-2011)

Four USDA ARS experimental watersheds (“CalVal” sites) (used for evaluations between 2001-2011)

Snow depth:

Global Historical Climate Network (GHCN) – used for evaluations between 1979-2012.

Canadian Meteorological Center (CMC) daily snow depth analysis – used for evaluations between 1998-2012.

Snow Data Assimilation System (SNODAS) products from the National Operational Hydrologic Remote Sensing Center (NOHRSC) – used for evaluations between 2003/10 – 2012)

Streamflow:

Gauge measurements from unregulated USGS streamflow stations (1981-2011) and naturalized streamflow data used in Koster et al. (2010) (1979-2011).

All model verifications and analysis generated using the Land surface Verification Toolkit (LVT; Kumar et al. 2012)

8

Soil moisture DA:

evaluation of soil moisture fields

Statistically significant improvements in surface soil moisture and root zone soil moisture as a result of soil moisture DA Anomaly R increases, Anomaly RMSE reduces and unbiased RMSE reduces with soil moisture assimilation.

ARS CalVal (surface soil moisture)

Open loop (no DA)

LPRM DA

Anomaly R 0.84 +/- 0.02 0.86 +/- 0.02

Anomaly RMSE (m3/m3) 0.021 +/- 0.001 0.019 +/- 0.001

ubRMSE (m3/m3) 0.024 +/- 0.002 0.022 +/- 0.002

SCAN (surface soil moisture)

Open loop (no DA)

LPRM DA

Anomaly R 0.67 +/- 0.02 0.67 +/- 0.02

Anomaly RMSE (m3/m3) 0.037+/- 0.002 0.036 +/- 0.002

ubRMSE (m3/m3) 0.043 +/- 0.003 0.041 +/- 0.003

SCAN (root zone soil moisture)

Open loop (no DA)

LPRM DA

Anomaly R 0.60 +/- 0.02 0.59 +/- 0.02

Anomaly RMSE (m3/m3) 0.032 +/- 0.002 0.030 +/- 0.002

ubRMSE (m3/m3) 0.041 +/- 0.003 0.039 +/- 0.003

9

Soil moisture DA:

Evaluation of streamflow

NIC_R

NIC_NSE

The improvements are expressed using an Normalized Information Contribution (NIC) metric that measures the skill improvement from DA as a fraction of the maximum possible skill improvement

Overall improvements in all skill metrics (RMSE, R and NSE) are observed in streamflow estimates after data assimilation Skill improvements from soil moisture assimilation are mostly over parts of the Mississippi, Missouri and Arkansas-Red basins and parts of Southeastern U.S.

NIC_RMSE

10

Snow DA:

Evaluation of snowdepth

The assimilation of gauge-corrected snow EDR provides improvements to the snow depth fields, primarily over the peak winter and spring melt periods.

The gauge corrected snow EDR data shows similar skill to that of CMC and SNODAS.

Selected GHCN stations

RMSE Bias

Open loop (no DA)

SNOW-DA CMC SNODAS

RMSE (mm)

174.0 +/- 8 114.0+/- 8 158.0+/-8 154.0+/- 8

Bias (mm) -84.1+/- 8 -31.6 +/- 8 -66.0+/- 8 33.9 +/- 8

Snow DA:

Evaluation of streamflow

NIC_RMSE

NIC_R

NIC_NSE

NIC_RMSE

NIC_R

NIC_NSE

Overall degradation in RMSE and NSE are observed in streamflow estimates after data assimilation Most improvements are observed over the Missouri and Upper Mississippi basins, degradations over Northwest and Colorado headwater basins.

The improvements are expressed using an Normalized Information Contribution (NIC) metric that measures the skill improvement from DA as a fraction of the maximum possible skill improvement

AP

A

ALA

RED PUE

RAL LEE OHI

UPM GUN POT DEL

SBB GRE

RAN

WIL ICE

MUS GAR FTP

12

Soil moisture and snow DA:

Evaluation of streamflow at major basin outlets

Streamflow is compared at major basin outlets against naturalized streamflow data (used in Koster et al. 2010, Nat. Geosc.)

Snow DA improvements noted over high latitude basins (Green river, Musselshel, Snake River, Missouri River) Soil moisture DA improvements over Apalachicola, Alabama, Willamette, Missouri at Garrison Degradations observed over the largest basins (Ohio, Upper Mississippi).

NIC_RMSE

NIC_R

NIC_NSE

TWS DA: Soil moisture evaluations

Surface (SCAN) OL DA-TWS

Anomaly R 0.51 +/- 0.02 0.55 +/- 0.02

Anomaly RMSE 0.050 +/- 0.002 0.050 +/- 0.002

Unbiased RMSE 0.061 +/- 0.002 0.060 +/- 0.002

Root zone (SCAN) OL DA-TWS

Anomaly R 0.50 +/- 0.02 0.54 +/- 0.02

Anomaly RMSE 0.035 +/- 0.002 0.034 +/- 0.002

Unbiased RMSE 0.045 +/- 0.002 0.043 +/- 0.002

Surface (ARS)

OL DA-TWS

Anomaly R 0.64 +/- 0.02 0.69 +/- 0.02

Anomaly RMSE 0.029 +/- 0.002

0.028 +/- 0.002

Unbiased RMSE 0.039 +/- 0.002

0.036 +/- 0.002

DA-TWS provides statistically significant improvements in both

surface and root zone anomaly R in comparisons to SCAN and ARS in-situ

data.

The improvements noted in other metrics (anomaly RMSE and

unbiased RMSE) are not always statistically significant.

TWS DA: Evaluation of streamflow

Overall improvements in all skill metrics (RMSE, R and NSE) are

observed in streamflow estimates after data assimilation

NIC_RMSE

NIC_R

NIC_NSE

The improvements are expressed using an Normalized Information Contribution (NIC) metric that measures the skill improvement from DA as a fraction of the maximum possible skill improvement

Pg. 15

TWS DA Streamflow evaluations (at major basin outlets)

NIC_RMSE

NIC_R

NIC_NSE

AP

A

ALA

RED PUE

RAL LEE OHI

UPM GUN POT DEL

SBB GRE

RAN

WIL ICE

MUS GAR FTP

Improvements in RMSE are observed for most basins, including the largest basins – Ohio, Upper mississippi R and NSE improves over several basins (Missouri, Lee, Fort Peck)

NIC_RMSE

NIC_R

NIC_NSE

16

The assimilation of remotely sensed soil moisture, snow and terrestrial water storage datasets have been completed. The results show promise in improving soil moisture, snow, streamflow estimates.

The use of LVT allows rapid evaluation of key components of the terrestrial water and energy components.

The ongoing work will also focus on the concurrent assimilation of all datasets and the generation of multi-model, DA-based ensemble averages.

Summary