AMS2014_Assimilating GRACE-TWS

Improving River Discharge Estimation by Assimilating
GRACE Terrestrial Water Storage (TWS) Retrievals into a
Distributed Hydrological Model:
Water budget analysis in the Upper Zambezi River
Basin (UZRB) and the Northern Kalahari Aquifer
(NKA) in Southern Africa
Jing Tao, Ana P. Barros
Dept. Civil & Environmental Engineering
Pratt School of Engineering
Duke University, USA
Water Resources in Southern Africa
Congo
Zambezi
Okavango
(Schefuß et al., 2011, Nature)
AHP: Angola High Plateau
UZRB: Upper Zambezi River Basin
Large Aquifer Systems
Aquifer Storage Type
Aquifer Thickness
Congo
Intracratonic
Basin
Northern
Kalahari Aquifer
Southeast
Kalahari
Aquifer
(WHYMAP, 2008)
Groundwater Storage
(BGS, 2011)
(BGS, 2011)
Aquifer Productivity
(BGS, 2011)
Depth to Water Table
Terrestrial Water Storage Anomalies (TWSA)
AHP
Std. of 10-year
GRACE TWSA
UZRB Basin-averaged
(June in each year)
Spatial-Temporal Rainfall Variability
•10-year (20032012) averaged
seasonal (left) and
monthly (right)
precipitation in
Southern Africa,
generated from
TRMM 3B42 (V7)
3-hourly data.
•Rainy season lasts
from Nov. to next
Mar., with
extremely large
spatial-latitude
gradient.
DJF
MAM
JJA
SON
5
Scientific Questions/Objectives:
1) What is the reason/source for the increase in terrestrial
water storage over the last decade? How does water in
the subcomponents of the terrestrial system change?
2) The surface-subsurface water interaction is important for
the long term sustainability of aquifers. Does the NKA
gain/lose water?
3) What is the inter-annual and intra-annual variability of
the surface-groundwater interaction?
4) How can one quantitatively monitor groundwater and
water resources in a region virtually without ground
measurements?
6
Upper Zambezi River Basin(UZRB)
Hydrological Basins in Southern Africa
UZRB
Congo
Zambezi
Limpopo
Orange
West
Coast
Drainage Area
≈515,175km2
Coupled Surface-Groundwater Hydrology
Model (3D-LSHM)
The model is implemented at
hourly and 5km resolution:
- Atmospheric forcing datasets were
extracted from ERA-Interim (6hourly, about 0.703deg).
- Precipitation from TRMM 3B42.
- Land surface attributes from MODIS.
- Soil texture was extracted from the
Harmonized World Soil Database
(HWSD).
(Tao&Barros, 2013, J. Hydro)
(Tao&Barros, 2014, HESS)
Results – Open Loop Simulations
Headwater
Catchment
Headwater
Catchment
Downstream
Valley
Uncertainty of Streamflow Observations???
Uncertainty in the Forcing – Rainfall
Uncertainty in Initial Model Conditions
Uncertainty in Model Parameters
Terrestrial Water Storage Change: Estimation vs. Observation
2005
2007
Terrestrial Water Storage Change: Inter- &Intra-annual variability
2004
2005
Wet Year
(2004)
Dry Year
(2005)
Annual Rainfall (mm)
Terrestrial Water Storage Change: Inter- &Intra-annual variability
Barotse (Balozi)
floodplain
2004
2005
Wet Year
(2004)
Dry Year
(2005)
Methodology – Data Assimilation
EnKF (Ensemble Kalman Filter) Updating:
x j ( ti' | Z i ) = x j ( ti' | Z i −1 ) + K  z − M  x j ( ti' | Z i −1 ) , ω j ( ti' ) , ti'  


Kalman Gain:
( )=
(
+
)
- error cross covariance between states and model prediction
- error covariance of the model prediction
- error covariance of the observations
EnKS (Ensemble Kalman Smoother) Updating:
X
=X
+
[
−
X
]
T- Assimilation/Smoothing Window
* Retrospectively Updating (Monthly Window)
13
HDAS - Hydrologic Data Assimilation System
Time-varying input u(t)
(forcing data, e.g.
precipitation, radiation, etc.)
Specified
(mean)
States: soil moisture at each layer and water table.
Measurements: terrestrial water storage.
Perturbation
Spin up
3DLSHM
States x(t) (e.g. soil
moist, water table)
Measurement System
(GRACE Monthly TWSA)
Perturbation
with error ω
Specified
(mean)
Time-invariant
input (e.g. soil
parameters)
Mean and
covariance of true
inputs and output
measurement
errors
Output
Measurement, zi,
(e.g. TWS, etc.)
Data
Assimilation
(EnKS)
Output
Results
Updating
states
Key Ref:
(Zaitchik, et al., 2008.
J. Hydrometeorol.)
Results – Assimilating GRACE TWS
Open-Loop:
With TWS Assimilation:
15
Results – Assimilating GRACE TWS
Headwater
Catchment
better
Headwater
better
Catchment
Downstream
Valley
worse
16
Summary & Discussion
•1) What
Soil water
dominates the
water
storage
is the reason/source
forterrestrial
the increase
in terrestrial
change
in a over
shortthe
time.
water
storage
last decade? How does water in
subcomponents of the terrestrial system change?
• the
With
such a short simulation period and without
2) The
surface-subsurface
water interaction
is important
for
incorporating
the trans-boundary
flux
from adjacent
the
long term
the NKA the
aquifer
(e.g.sustainability
Congo), it of
is aquifers.
difficultDoes
to determine
gain/lose water?
current scenario/phase of the aquifer.
• Surface-Groundwater interaction is vigorous and
3) What
is the
inter-annual
andlarge
intra-annual
variability
highly
nolinear
showing
inter- &
intra of
the
surface-groundwater
variability,
but stayinginteraction?
active all year along in the
floodplain.
4)
can onehydrological
quantitativelymodel
monitorand
groundwater
• How
Combing
satellite and
water resources in a region virtually without ground
observations through data assimilation technique is
measurements?
very promising, but ground measurements are highly
17
needed for validation.
Acknowledgements:
This research was supported by a NASA
Earth Systems Science Fellowship.
Question?
18
Reference:
Zaitchik, B.F., Rodell, M., Reichle, R.H., 2008. Assimilation of GRACE
terrestrial water storage data into a Land Surface Model: Results for the
Mississippi River basin. J. Hydrometeorol., 9(3): 535-548.
Schefuss, E., Kuhlmann, H., Mollenhauer, G., Prange, M., Paetzold, J., 2011.
Forcing of wet phases in southeast Africa over the past 17,000 years. Nature,
480(7378): 509-512.