Applications of GRACE data to estimation of the

Applications of GRACE data to estimation of the water budget of large U. S. river basins Huilin Gao, Qiuhong Tang, Fengge Su, Dennis P. Lettenmaier Dept. of Civil and Environmental Engineering, University of Washington GRACE hydrology workshop Nov. 4 th, 2009 U N I V E R S I T Y O F WASHINGTON

Outline 1. Background and motivation 2. Research strategy 3. Evaluation of remotely sensed precipitation, evapotranspiration (ET), and terrestrial water storage (TWS) 4. Testing the ability to close the water budget solely from remote sensing 5. Further evaluation of GRACE terrestrial water storage change (TWSC) over the west coast 6. Conclusions U N I V E R S I T Y O F WASHINGTON 1

Background ∆S = P –R– ET 1. Importance for understanding water budget at continental scale 2. Limitations of observations and modeling 3. The opportunities brought by remotely sensed water budget terms, especially GRACE TWS (Rodell et al. , 2004; Tang et al. , 2009) 4. Challenges to remote sensing products (Sheffield et al. , 2009) U N I V E R S I T Y O F WASHINGTON 2

Background ∆S = P –R– ET 1. Importance for understanding water budget at continental scale 2. Limitations of observations and modeling 3. The opportunities brought by remotely sensed water budget terms, especially GRACE TWS (Rodell et al. , 2004; Tang et al. , 2009) 4. Challenges to remote sensing products (Sheffield et al. , 2009) Motivation Ø Over major river basins across the CONUS, how well can estimates of terrestrial water budget terms derived entirely from remote sensing be used to close the terrestrial water budget? Ø Which remotely sensed terms have the largest/least uncertainty, and is it possible to close the water balance by selecting a suite of satellite products with superior performance? U N I V E R S I T Y O F WASHINGTON 2

Research Strategy ? R (observed) = P – ∆S – ET (remote sensing) Research Domain – Continental U. S. ü High quality precipitation from gridded gauge measurements - help evaluate P ü LSM outputs using quality forcings - help evaluate ΔS and ET Precipitation ET ΔS Runoff Remote sensing TMPA CMORPH PERSIANN MODIS based by UM, PU, UW GRACE by CSR; GFZ; JPL Inferred Observed/ Modeled Gridded gauge data *VIC output Observed runoff *VIC output: Variable Infiltration Capacity model forced by gridded gauge precipitation 3

Hydrological Regions and River Basins in the U. S. Arkansas-Red (AR) California (CA) Colorado (CO) Columbia (CB) U N I V E R S I T Y O F WASHINGTON East Coast (EA) Great Lakes (GL) Great Basin (GB) Gulf (GU) Lower Mississippi (LM) Rio Grande (RG) Upper Mississippi (UM) Missouri (MO) Ohio (OH) 4

Annual Precipitation (2003~2006) U N I V E R S I T Y O F WASHINGTON 5

Seasonal Precipitation • Orographic effects are poorly represented by the remote sensing products • Remotely sensed precipitation is biased high over the central CONUS • TMPA precipitation performs the best among the three U N I V E R S I T Y O F WASHINGTON 6

Annual Evapotranspiration (2003~2006) U N I V E R S I T Y O F WASHINGTON 7

Seasonal Evapotranspiration • Remotely sensed ET accounts for irrigation contribution • It is difficult to validate remotely sensed ET at the continental scale • Over most regions, UM ET tends to provide the smallest values, and UW ET is closest to VIC estimate U N I V E R S I T Y O F WASHINGTON 8

Dynamic Range of TWS (2003~2006) U N I V E R S I T Y O F WASHINGTON (acknowledgement to Dr D. P. Chambers for smoothing method) 9

Seasonal TWS • Dynamic range of VIC TWS is larger than GRACE over the western hydrologic regions • Dynamic range of VIC TWS is smaller than GRACE estimates in much of the Mississippi basin • GRACE products from different data centers are similar in their differences with VIC U N I V E R S I T Y O F WASHINGTON 10

Inferred Runoff v. s. Observed Runoff (I) 3× 3× 3=27 ensemble members ? R (observed) = P – ∆S – ET (remote sensing) U N I V E R S I T Y O F WASHINGTON 11

Inferred Runoff v. s. Observed Runoff (II) 3× 3=9 ensemble members ? R (observed) = P – ∆S – ET (remote sensing) U N I V E R S I T Y O F WASHINGTON 12

Amplitude of Seasonal TWS mean of the three GRACE datasets maximum and minimum of the three Are the biases from VIC or GRACE? U N I V E R S I T Y O F WASHINGTON 13

Satellite/Observation based TWSC ∆S = P –R– ET KL 03 KL 04 PRISM gauge reliable Satellite validated PRISM: Parameter-elevation Regressions on Independent Slopes Model Blodgett Tonzi Ranch Vaira Ranch 14

ET Validation • • KL 03, KL 04 Ameri. Flux towers METRIC (Mapping Evapotranspiration at high Resolution and with Internalized Calibration) Landsat estimates (a) (b) Observed and estimated ETday at flux tower KL 04 (irrigated site) U N I V E R S I T Y O F WASHINGTON (Details about this ET algorithm and its applications are available through Tang et al. , JGR, 2009) 15

TWSC (mm) TWSC intercomparisons U N I V E R S I T Y O F WASHINGTON 16

Conclusions • Water budget closure at the scale of large continental river basins is not currently possible on the basis of satellite data alone, even with a combination of the best products; • Among the remotely sensed budget terms, precipitation has the largest error; • ET estimation errors are the second most important, and notwithstanding their coarse spatial resolution, GRACE TWSC errors are of smaller magnitude than the other two sources; • GRACE water storage change appears to be underestimated along the west coast. 17

Thanks!!! Questions?