Accounting soil moisture assimilation for hydrologic predictions
1.
2.
3. Soil moisture
• Water contained in the vadose zone of subsurface
• Major component of soil hydrology that influences the exchange of heat and
moisture between the atmosphere and land surface
• Forecasts of runoff, flood, groundwater recharge and evapotranspiration
4. Applications of soil moisture
Soil
Moisture
Drought
monitoring
Climate
science
Flood
forecasting
Groundwater
recharge
Land
Atmosphere
processes
Ecological
status
Agronomy
5. Soil moisture
measuring
equipment's
Thermostat weight
method
Measure the electrical resistance
Hydrogen atoms that
changes with time
Changes in frequency of
signals due to the
dielectric properties of
the soil
Changes in the
dielectric properties of
the soil at microwave
frequencies
Soil moisture tension
No large-area coverage
Expensive and problematic
Remote Sensing for Soil Moisture Measurement
Measure soil moisture with high frequency, high resolution and at multiple depth
Errors by soil type
Landscape roughness
Vegetation cover
Inadequate coverage in both space and time
6. Data assimilation
• Sequential time-stepping procedure, in which a previous model forecast is
compared with newly received observations
• Transform diverse and incomplete observations to gridded estimates that can
easily be used and interpreted
• Produces quantitative information on model error, forecast skill, and
observational errors, all of which allows us to improve models
Soil moisture assimilation
• Integration of remotely-sensed soil moisture information into the hydrologic
model
7. Data assimilation techniques
Assimilation Technique Salient Features Sources
Kalman filter (KF) Propagation and update of state error
covariance and mean for a linear
stochastic system
Kalman (1960)
Extended Kalman filter (EKF) Propagation of state error covariance with
linearized model
Smith et al. (1962)
Ensemble Kalman filter (EnKF) Propagation of state error covariance and
mean by ensemble integration
Evensen (1994)
Ensemble Transform Kalman Filter
(EnTKF)
Estimating the effect of observations on
forecast error covariance
Bishop and Toth
(1998)
Ensemble Transform
Kalman smoother (EnTKS)
Maintaining the covariance matrix is not
feasible computationally for high-
dimensional systems
Evensen and van
Leeuwen (2000)
Maximum Likelihood Ensemble
Filter (MLEF)
Optimizes a non-linear cost function
along with a Maximum Likelihood
Zupanski (2005)
8. Kalman filter (KF)
• Recursive data processing algorithm
Doesn’t need to store all previous measurements and
reprocess all data each time step
• Generates optimal estimate of desired quantities given
the set of measurements
For linear system and white Gaussian errors, estimate
based on all previous measurements
• Sophisticated sequential data assimilation method
Group of data viewed as a whole
• Predict the error statistics forward in time in the model
estimate
Framework of KF
Framework of EnKF
9. Framework of soil moisture assimilation procedure
Input
(P’)
API
Output
(P*)
Input
(Φ)
Model
Output
(Runoff)
Output
(S)
Crow and Ryu (2009)
Where,
P’ = Estimate of daily rainfall
Φ = Remotely sensed surface soil moisture
P*= Corrected Rainfall
API = Antecedent Precipitation Index
S = Predicted soil moistureEnKF
Input
(P’)
Input
(P’)
Model
Input
(Φ)
Output
(Runoff)
Output
(S)
Model
Kalman filter
Stage (I) Rainfall correction
Stage (II) State correction
11. Use of models for hydrologic predictions
Models Functions Sources
Sacramento Soil Moisture
Accounting (SAC-SMA)
• Distribute applied moisture in various
depths and energy states in the soil
• Stream flow simulation
• Percolation characteristics
National Oceanic and Atmospheric
Administration (NOAA), National
Weather Service (NWS)
NWS Hydrologic Laboratory’s
Research Distributed Hydrologic
Model (HL-RDHM)
• Soil moisture and overland flow
• Channel routing algorithms
• Soil temperature and snow
National Oceanic and Atmospheric
Administration (NOAA), National
Weather Service (NWS)
Advanced Microwave Scanning
Radiometer-Earth Observing
System (AMSR-EOS)
• Surface water
• Sea surface temperature, near-surface wind
speed, radiative energy flux,
• Cloud properties, Ice and snow
National Aeronautics and Space
Administration (NASA)
Soil and Water Assessment Tool
(SWAT)
• Quality and quantity of surface and ground
water
• Soil and water dynamics
• Sediment and pollutant yields
USDAAgricultural Research
Service (USDA ARS) and Texas
A&M AgriLife Research, Texas
University
12. Improved soil moisture data products using multiple satellite
observations within a constraint model-data assimilation framework
• Study area: National water and climate centre, Virginia, USA
• Model used: Ensemble Kalman filter (EnKF) and Surface Energy
Balance (SEB)
• Measured soil moisture at depth: 5, 15, 45, 60 and 100 cm
Barrett et al. (2011)
13. Variation of soil moisture with depth
Soil moisture at 100 cmSoil moisture at 60 cm
Soil moisture at 5cm
Soil moisture at 5 cm
Soil moisture at 45 cm
Barrett et al. (2011)
14. Soil moisture assimilation system based on Ensemble Kalman filter
• Study area: Tibetan Plateau, China
• Model used: Ensemble Kalman filter (EnKF) and Simple Biosphere
(SiB2)
• Selected site: MS3478, MS3608 and MS3637
• Measured soil moisture at depth: 4, 20 and 100 cm
Hunang et al. (2008)
15. Assimilation of in situ soil moisture observations at different depth
Hunang et al. (2008)
AssimilationObservationSimulation
AssimilationObservationSimulation
• Assimilation results improve the estimation of soil moisture in surface layer and
root zone, and differ from the simulated soil moisture at deep layer
• Assimilated soil moisture are close to observed soil moisture when bias errors in
model operator
16. Prediction error statistic of simulated and assimilated soil moisture
Site Layer Simulated soil moisture Assimilated soil moisture
RMSE Average error RMSE Average error
MS3478 Surface 0.0226 0.0153 0.0164 0.0024
Root zone 0.0316 0.0295 0.0211 0.0166
Deep layer 0.0146 - 0.0141 0.0153 - 0.0155
MS3608 Surface 0.0311 0.0286 0.0152 0.0095
Root zone 0.0763 0.0751 0.0582 0.0569
Deep layer 0.0151 - 0.0139 0.0113 - 0.0097
MS3637 Surface 0.0148 0.0402 0.0133 0.0035
Root zone 0.0129 0.0164 0.0561 0.0537
Deep layer 0.0078 - 0.0007 0.0006 - 0.0045
Hunang et al. (2008)
17. Improving estimated soil moisture through assimilation of
AMSR-E soil moisture retrievals with an Ensemble Kalman filter
• Study area: Washita watershed, Oklahoma, USA
• Model used: AMSR-E and ARS
• Measured soil moisture at depth: 5, 25, 45 and 100 cm
Key words
• Control run (Control): baseline performance of the NOAA Land surface model
• Daily accounting (DA) and Daily Accounting Mass conservation (DA Mass Con): simulation
features assimilation of AMSR-EOS model
• AMSR-EOS: Advanced Microwave Scanning Radiometer-Earth Observing System
• USDAARS: USDAAgriculture Research Service
Li et al. (2012)
18. Comparison of basin averaged daily soil moisture
Li et al. (2012)
At 5 cm depth At 25 cm depth
• AMSR-EOS capturing better in situ soil moisture than ARS measurement, Control
exhibiting strong correlation with the ARS measurements and DA and DA MassCon
both reduced the bias and RMSE over Control
• DA and DA MassCon decreased soil moisture at 25 cm depth and reduced the bias
and RMSE over control
19. At 45 cm depth At 100 cm depth
Li et al. (2012)
• DA lowered the soil moisture at 45 and 100 cm depth
• DA MassCon increased the amount of soil moisture at 45 and 100 cm depth, capture
soil moisture removed from the top two layers
• DA MassCon improved over Control at 100 cm with reduced bias and RMSE
20. Monthly surface runoff (upper layer) and base flow (lower layer)
Li et al. (2012)
• Control, DA and DA MassCon predicted very
similar surface runoff
• DA produced driest soil moisture profile, no
prolonged precipitation periods
• DA predicted the lower base flow while DA
MassCon predicted largest base flow
• Control, DA and DA MassCon base flow
significantly increased in the winter months,
more correction in soil moisture
21. Improving hydrologic predictions of a catchment model via assimilation of soil
moisture
Chan et al. (2011)
• Study area: Cobb Creek watershed, Oklahoma, USA
• Model used
Observation: Remote Sensing
Open loop: Soil and Water Assessment Tool (SWAT)
Assimilation: Ensemble Kalman filter (EnKF)
• Measures soil moisture at depth: 5, 25, 45 and 100 cm
22. Observed vs open loop and real data EnKF Volumetric soil moisture
Chan et al. (2011)
• EnKF and Open Loop displayed similar soil moisture trend than observed soil
moisture
• EnKF effectively update Open Loop soil moisture and EnKF soil moisture profile
required greater improvement
23. Chan et al. (2011)
• Persistence positive biasness in the Open Loop soil moisture, difficulties in real data
and modelling error
24. Chan et al. (2011)
• EnKF soil moisture differ from the Open Loop soil moisture, soil moisture is stable
and influenced by moisture in surface layer
25. Observed vs open loop and real data EnKF stream flow
Chan et al. (2011)
• Assimilation of soil moisture data had limited success at 5 cm and 25 cm depth,
unsuccessful in improving stream flow predictions
26. Conclusions
• Ensemble Kalman Filter (EnKF) is effective technique for assimilating soil moisture
observations into numerical models
• Assimilation results slightly differ from the simulation results in deeper soil profile
because soil moisture in deeper layer is stable and influenced by moisture in surface
layer
• Improvement was maximum in top layers and marginal for lower layers because the
data assimilation was unable to correct the presence of biasness in lower soil moisture
layers
• Soil moisture assimilation technique was able to reduce error in soil moisture
measurement