Advances in Land Surface Modelling and Data Assimilation at EC Focus upon improved analysis of snow cover CanSISE Workshop : 30 October – 1 November, 2013 Victoria, BC Marco Carrera, Stéphane Bélair, Nathalie Gauthier, Bernard Bilodeau, Dorothée Charpentier, Chris Derksen, and Libo Wang Environment Canada
Page 2 – MAIN ASPECTS for LAND SURFACE MODELING and ASSIMILATION Surface characteristics databases space-based remote sensing Modeling surface processes interactions with atmosphere external high-resolution system Assimilation CaLDAS (Canadian Land Data Assimilation System) Saskatchewan Northern Territories Toronto Central Quebec
Page 3 – The CANADIAN LAND DATA ASSIMILATION SYSTEM (CaLDAS) ISBA LAND-SURFACE MODEL OBS ASSIMILATION xbxb y (with ensemble Kalman filter approach) x a = x b + K { y – H(x b ) } K = BH T ( HBH T +R) -1 with CaLDAS INOUT Ancillary land surface data Atmospheric forcing Observations Land surface initial conditions for NWP and hydro systems Land surface conditions for atmospheric assimilation systems Current state of land surface conditions for other applications (agriculture, drought,... Screen-level (T, Td) Surface stations snow depth L-band passive (SMOS,SMAP) MW passive (AMSR-E) Multispectral (MODIS) Combined products (GlobSnow) T, q, U, V, Pr, SW, LW Orography, vegetation, soils, water fraction,...
Page 4 – CMC Operational Snow Analysis (Global, Regional, and LAM) Uses an Optimum Interpolation (OI) methodology to combine a first- guess snow field with snow depth observations (Brasnett 1999). Simplistic “snow-model” where previous snow-analysis is used as background and operational precipitation is used to compute the new snow accumulation. ECMWF has recently revised their operational snow analysis to make use of the OI scheme (instead of Cressman) with the weighting functions of Brasnett (1999).
Page 5 – Strategies used at EC to improve snow analyses 1.GEM-Surf : High-resolution near-surface and land surface forecast system. Offline system forced with hourly forcing from CMC deterministic prediction system NWP 2.CaLDAS-EnsOI : Combine Ensemble Kalman filter (EnKF) approach for soil temperature and moisture with ensemble optimal interpolation (OI) for snow depth. 3.CaLDAS-Snow : EnKF approach to assimilate SWE retrievals from microwave passive measurements into GEM-Surf.
Page 6 – High resolution GEM-Surf system Components and validation ATMOS MODEL Atmospheric forcing at first atmos. model level (T,q,U,V) 3D integration GEM-Surf 100 m 2D integration Downscaled to 100m Atmospheric forcing at surface.(S,PR, P0) Adaptation of T,U,V,q,P0 corrected for difference in elevation between forcing model and GEM-Surf. PR phase is adjusted to be coherent with new T. LOW-RES HIGH-RES Land surface characteristics specified using High-Res external databases
Page 7 – High resolution GEM-Surf system Developed for high-resolution environmental prediction over Canada →Can run at resolution of the most detailed surface characteristics database available. →Integrates only land and near-surface processes →Computational cost of GEM-Surf much less than that of 3D atmospheric models Good forecast of land surface conditions tied to representation of local surface characteristics such as orography, vegetation type, soil type, snow coverage. External model of the land surface and near-surface which evolve separately of the 3D operational forecast model in term of resolution and time-step. SWE – GEM-3DSWE – GEM-Surf Characteristics Carrera et al. 2010, Bernier et al. 2011, Leroyer et al
Page 8 – August 25, 2014 DOWNSCALING LAI: SOURCES of INFORMATION 30m1km10km200m TARGET RESOLUTION LCC-2000 Land Use / Land Cover (types) MODIS 10-year NDVI climatology Biome-BGC 10-year Clim from runs (LAI) Grass Forests
Page 9 – August 25, 2014 LAI (m 2 m -2 ) LAKE ONTARIO GREATER TORONTO AREA Light grey: pavement fraction Dark brown: building fraction
Page 10 – Ancillary land surface data Offline Snow Model Atmospheric forcings Surface stations snow depth CaLDAS SD Analysis Snow data assimilation currently operational at EC Orography, vegetation, soils, water fraction,... Optimal Interpolation Assimilation based on optimal interpolation Background (first guess) snow depth is given by a simple offline snow model Observations from SYNOP and METAR are used (Brasnett, 1999) T, hu, winds Precipitation Radiation
Page 11 – Ancillary land surface data GEM-Surf Atmospheric forcing Observations CaLDAS SD Analysis The Canadian Land Data Assimilation System (CaLDAS) Orography, Vegetation, Soils, Water Fraction,... Assimilation EnKF Temp, Humidity, Winds, PR, and Radiation Assimilation EnsOI (SD) xbxb y EnKF x a = x b + K { y – H(x b ) } K = BH T ( HBH T +R) -1 with CaLDAS-Snow : BH T covariance between Snow Depth and SWE HBH T : Model Error Variance
Page 12 – Forcing: RUN 00Z 6-12hr forecasts Forcing: RUN 12Z 12-18hr forecasts 00Z06Z12Z18Z Forcing: RUN 00Z 12-18hr forecasts Forcing: RUN 12Z 6-12hr forecasts 18Z Forcing: RUN 12Z 6-12hr forecasts GEM-Surf 6h forecasts CaLDAS: General strategy PR :spatial shift and additive perturbation using CaPA methodology Radiation :spatial shift coherent with PR shift TT :additive Gaussian perturbations
Page 13 – Canadian Precipitation Analysis (CaPA) Generation of an Ensemble of Precipitation analyses Optimum Interpolation Assimilation Perturbed Precipitation Gauge Observations Measurement + Errors of Representativity Spatially perturbed Model forecasts of precipitation Ensemble of precipitation analyses RDPA (Regional Deterministic Precipitation Analysis)
Page 14 – EnKF x a = x b + K { y – H(x b ) } K = BH T ( HBH T +R) -1 with CaLDAS-Snow : BH T covariance between Snow Depth and SWE HBH T : Model Error Variance CaLDAS-EnsOI Combine ensemble Kalman filter approach for soil temperature and moisture with ensemble optimal interpolation (OI) for snow depth Examine impact of CaLDAS initial conditions in forecast mode for the Global Deterministic Prediction System (GDPS)
Page 15 – Improved forecasts of temperature at 2m with CaLDAS Operational 2m temperature vs CaLDAS- Screen with ensemble of OI analyses Winter results – 2m temperature (00Z runs – Northern Canada) Bias STDE CaLDAS OP
Page 16 – CaLDAS-Snow: General strategy Snow in CaLDAS: Ensemble Kalman Filter Observations: SWE retrievals from AMSR-E or GlobSnow (1/day) Control variable: snow mass Other snow variables: snow density and snow albedo cycled Background (first guess): GEM-Surf 6h prediction Number of members: 24 Assimilation step: 6h Ensemble spread obtained by perturbing the atmospheric forcing, the observations and the analysis Observation errors are constant in time and in space: 20mm Bias errors (systematic) are removed from SWE observations.
Page 17 – Canadian experiments for snow depth Test period : 1 November 2006 – 21 April 2007 Computational domain : Central part of Canada Observation network : 50 surface stations (Environment Canada) Hudson Bay Grid spacing : 15 km SWE retrieval algorithms developed from field campaigns for this area The SWE AMSR-E data were re-gridded on the 15km CaLDAS-Snow grid
Page 18 – CMC-OP Open Loop ENS-OI OBS Improved analysis of snow depth with CaLDAS (ENS-OI) Evaluation against surface snow depth observations. Operational snow depth analyses (OI) vs CaLDAS ensemble of OI analyses Mean snow depth Bias STDE
Page 19 – CaLDAS with AMSR-E vs open loop EnKF Assimilation experiment with SWE from AMSR-E Open Loop CaLDAS-AMSR-E Mean snow depth Bias STDE OBS CaLDAS-AMSRE not as good as ENS-OI shown earlier
Page 20 – Preliminary GlobSnow Open Loop CaLDAS-GS OBS Mean snow depth Bias STDE GlobSnow (CAREFUL… GlobSnow experiments not leave-one-out type)
Page 21 – Summary : Current Status CaLDAS has been accepted into CMC operations as an experimental product running offline on a global domain at 25 km resolution. January 2014 : final tests will begin for CaLDAS coupled with the new upper-air assimilation system. Goal is to transfer this coupled system into operations in CaLDAS is being coupled with the Global Ensemble Prediction System (GEPS) at 50-km resolution. Individual CaLDAS members providing the surface initial conditions to the GEPS.