Surface Soil Moisture Assimilation with SWAT Eunjin Han, Venatesh Merwade School of Civil Engineering, Purdue University, West Lafayette, IN, USA Gary C. Heathman USDA-ARS, National Soil Erosion Research Laboratory, West Lafayette, IN, USA August 4, 2010 Seoul, Korea
Bacground Importance of soil moisture (SM) in hydrology, agriculture, meteorology and environmental studies Why surface soil moisture data assimilation? Limitations of field measurements and remote sensing data - Temporal and spatial limitation (field measurement) - Shallow sensing depth and large grid size (remote sensing) Data assimilation: better prediction of SM profile by combining observed surface SM data and hydrologic models 2
Research Objectives Application of data assimilation techniques to a watershed scale hydrologic model to improve soil water content SWAT + Ensemble Kalman Filter (EnKF) To investigate how near surface SM assimilation affects runoff and streamflow prediction through synthetic eperiment To investigate how the spatial variability of inputs affect the potential capability of data assimilation (DA) techniques 3
Study Area Upper Cedar Cree Watershed, IN Environmental Monitoring Networ by USDA-ARS, NSERL UCCW:198 m2 4
Previous Research 1D point-scale soil moisture profile estimation (AS1 and AS2) Application of two DA methods (EnKF and direct insertion) into the Root Zone Water Quality Model with field measured surface SM Comparison of simulation results (simulation period : April to October, 2007) Site Depth (cm) W/O assimilation Direct Insertion EnKF R RMSE MBE R RMSE MBE R RMSE MBE A S 2 5 0.6904 0.0908 0.0782 0.8913 0.0491 0.0418 0.9038 0.037 0.0279 20 0.32 0.0495 0.0241 0.573 0.0311 0.0045 0.6119 0.0286-0.0054 40 0.1087 0.0725 0.0559 0.3362 0.0541 0.0407 0.3827 0.0461 0.0335 60 0.4124 0.0333 0.0217 0.46 0.031 0.0194 0.4847 0.0279 0.0148 Daily assimilation of surface soil moisture improved soil moisture estimation in the upper dynamic layers (5 and 20cm depth) while less certain improvement in the deep layers (40cm and 60cm depth) Watershed scale application of the EnKF with a (semi-) distributed model (SWAT) 5
Forecast step Methodology (Source: Reichle et al. 2002) ( ) w f + = +1 Nonlinear System model v H y + = w : Process noise, P(w)~ (0, Q) v : Measurement noise, P(v)~(0,R) ( )( ) e T P P = * Error Covariance Matri: y : Measurement ( ) i i i w f 1 1 1 + + = T D D N P 1 1 = ( ) ( ) [ ] = N D,, 1 = = N i i N 1 1 N i,, =1 Update step [ ] i i i i v H y K + + = + [ ] 1 + = T T R H P H H P K Data assimilation Ensemble Kalman Filter 6
Why synthetic eperiment? SWAT: HRU based semi distributed model Coarse resolution of currently available remote sensed soil moisture data Uncertainties due to model and observation are nown Better understanding on the effect of DA Eperiment set up True state Rainfall data from NCDC and NSERL raingauge networ prior (no assimilation) Rainfall data only from NCDC Intentionally poor initial condition Methodology EnKF Same set up as prior Updated soil water content using the observed surface soil moisture (~5cm) Our imperfect nowledge of the true system 7
Methodology NCDC(Angola) Rainfall gauge stations NCDC(Butler) NCDC(Garret) 8
Methodology True state preparation Data source - DEM: 30m - Soil: SSURGO database - Landuse : NLCD2001 Watershed delineation & HRU creation using ArcSWAT - 25 subbasins - Total 206 HRUs Model calibration using measured streamflow - Warm-up period (April 2003 April 2006) - Calibration period (May 2006 May 2008) - Validation period (June 2008 May 2010) - Calibrated 18 parameters based on sensitivity analysis and literature (Cn2, Esco, Surlag, Timp, Ch_K2, Sol_Awc, Blai, Alpha_Bf, Sol_Z, Rchrg_Dp, Smtmp, Epco, Ch_N2, Revapmn, Slope, Sol_Alb, Canm, Sol_K) 9
True state preparation (contd.) Methodology R 2 R NS 2 Calibration (May 2006 May 2008) 0.46 0.44 Validation (June 2008 May 2010) 0.42 0.37 Streamflow (cfs) (m 3 s -1 ) 50 40 30 20 10 observed Simulated(Validation) 0 May 10 Apr 10 Mar 10 Feb 10 Jan 10 Dec 09 Nov 09 Oct 09 Sep 09 Aug 09 Jul 09 Jun 09 May 09 Apr 09 Mar 09 Feb 09 Jan 09 Dec 08 Nov 08 Oct 08 Sep 08 Aug 08 Jul 08 Jul 08 Jun 08 10
Methodology Implementation of the EnKF with the SWAT Daily update Ensemble size: 50 Observed data for the l st layer (~5cm) generated from the true simulation by adding observation error (0.007) Ensemble of initial SM profile Begin daily loop Read in weather info Loop for each HRUs per subbasin Call subbasin - Call surface (compute surface RO) - Call percmain (perform soil water routing) - Call etpot (compute ET) - Call crpmd (compute crop growth) - Call gwmod (compute GW contribution) Ensemble of forecasted SM profile Average of ensemble SM profile Call route (channel routing) SM Observation available? Yes Data assimilation algorithm (EnKF) 11 (Source: Reichle et al. 2002) No Day=Day+1 Ensemble of updated SM profile
Preliminary Results [mm d -1 ] 60 40 20 0-20 -40 Difference in watersehd average precipitation rates *Difference= True PCP from all raingauges interpolated PCP from the only NCDC gauges 08-6-1 08-7-21 08-9-9 08-10-29 08-12-18 09-2-6 09-3-28 09-5-17 09-7-6 09-8-25 09-10-14 09-12-3 10-1-22 10-3-13 10-5-2 Date (yy-mm-dd) Streamflow (cfs) (m 3 s -1 ) 40 35 30 25 20 15 10 Comparison of streamflow True state "prior" (no assimilation) EnKF 5 0 08-6-1 08-7-21 08-9-9 08-10-29 08-12-18 09-2-6 09-3-28 09-5-17 09-7-6 09-8-25 09-10-14 09-12-3 10-1-22 10-3-13 10-5-2 Date (yy-mm-dd) 12
Preliminary Results 40 Daily streamflow eceedance curve Statistical results 35 30 R 2 R 2 NS prior (no assimilation) 0.447-0.118 EnKF 0.514 0.407 Streamflow (cfs) (m 3 s -1 ) 25 20 15 True state 10 "prior" (no simulation) EnKF 5 0 0 20 40 60 80 100 % of time streamflow is equal or eceeded Better representation of soil water contents through surface soil moisture assimilation can improve rainfall-runoff modeling - Effective especially for improving the overestimated streamflow due to the wrong high rainfall 13
Conclusions/Future wors Inaccurate simulation results due to the limited nowledge on the forcing data can be improved partially by updating surface soil moisture condition (EnKF data assimilation) Future wors To investigate how the surface soil moisture assimilation affects each hydrologic components and water balance (Total water content in the soil, ET, lateral flow etc.) To eplore how to use currently available remote sensing (soil moisture ) data for HRU-based SWAT model for data assimilation To eamine better assimilation approaches (e.g. assimilating streamflow observation) 14
Reference Reichle, R. H., Waler, J. P., Koster, R. D., and Houser, P. R. (2002). "Etended versus Ensemble Kalman Filtering for Land Data Assimilation." In: Journal of Hydrometeorology, American Meteorological Society, 728. 15
More Info: ehan@purdue.edu 16