1 / 40

JCSDA Meeting May 31, 2006 Greenbelt, MD

JCSDA Meeting May 31, 2006 Greenbelt, MD Biases and scaling in soil moisture and temperature data assimilation R. H. Reichle 1,2 R. D. Koster 1 M. G. Bosilovich 1 S. P. P. Mahanama 1,2 1 – Global Modeling and Assimilation Office, NASA 2 – GEST, University of Maryland, Baltimore County.

alijah
Download Presentation

JCSDA Meeting May 31, 2006 Greenbelt, MD

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. JCSDA Meeting May 31, 2006 Greenbelt, MD Biases and scaling in soil moisture and temperature data assimilation R. H. Reichle1,2 R. D. Koster1 M. G. Bosilovich1 S. P. P. Mahanama1,2 1 – Global Modeling and Assimilation Office, NASA 2 – GEST, University of Maryland, Baltimore County

  2. Outline

  3. Land data assimilation EnKF Observed precipitation, radiation Satellite surface soil moisture & temperature Model soil moisture & temperature Land data assimilation Land model “Optimal”soil moisture & temperature Data assimilation with the Ensemble Kalman filter (EnKF): Consider relative uncertainties in modeled and observed soil moisture.

  4. yk Propagation tk-1to tk: xki- = f(xk-1i+) + wki w = model error Update at tk: xki+ = xki- + Kk(yki - xki- ) for each ensemble member i=1…N Kk = Pk(Pk + Rk)-1 with Pk computed from ensemble spread Soil moisture assimilation xki state vector (eg soil moisture) Pkstate error covariance Rkobservation error covariance

  5. Outline

  6. Data sources

  7. Satellite vs. satellite bias (time avg. soil moisture) SMMR retrievals much wetter than AMSR-E retrievals. Magnitude of differences comparable to dynamic range.

  8. We found strong biases between AMSR-E and SMMR and are working with the AMSR-E algorithm team to address the issue. For assimilation, we are really interested in satellite vs. model biases.

  9. Satellite vs. model bias

  10. Satellite vs. model bias

  11. Satellite vs. model bias • SMMR and AMSR-E exhibit large and very different global and regional biases in all moments relative to the model. • Absolute soil moisture from satellites and model agree equally well (or poorly…) with ground observations  no agreed climatology. • For seasonal forecasts, need only normalized anomalies. •  Scale satellite data before assimilation into a model.

  12. Soil moisture scaling for data assimilation Assimilate percentiles.

  13. Soil moisture scaling for data assimilation (mean) ORIGINAL multi-year data sets (Satellite minus model) SCALED multi-year data sets (Satellite minus model) CDF scaling based on 1 year of satellite data Reichle et al. JHM 2004 Reichle & Koster GRL 2004 1 year of satellite data sufficient for considerable reduction in long-term bias.

  14. We’re ready to assimilate!

  15. Global soil moisture data sets GSMDB stations ~70 of 200 useful 1. Satellite retrievals (upper 1.25cm, 50-140km, ~3 days) AMSR-E (2002-05) Soil moisture retrievals not available under dense vegetation, near open water, in frozen soil. SMMR (1979-87) More AMSR-E data than from SMMR data. ASSIMILATE Number of data per month 2. Model data NASA Catchment Model (CLSM) forced w/ observation-corrected meteorological data. (upper 2cm, ~40…150km, 6h) 3. In situ data (upper 5…10cm and profile, point scale, hourly - 10 days) USDA SCAN stations ~17 of 122 useful VALIDATE

  16. Validation against in situ data Assimilation product agrees better with ground data than satellite or model alone. Modest increase may be close to maximum possible with imperfect in situ data. SMMR: Reichle & Koster, GRL 2005 AMSR-E: Reichle et al., in preparation

  17. Variance of normalized innovations Variance deficiency in dry climates, excess variance in wetter climates. Potential for improvement by (adaptively) tuning model error parameters. Reichle & Koster, GRL 2005

  18. Outline

  19. Land surface temperature (LST) assimilation Good news: Abundance of LST retrievals from infrared and microwave sensors on geostationary and polar-orbiting platforms (NOAA-xx, MODIS, GOES, METEOSAT, GMS,…) Problem 1: Satellite and model LST inconsistent in vertical. Problem 2: Satellite and model LST inconsistent in horizontal. Problem 3: Satellite LST sensor- or algorithm-specific.

  20. Solution: Once again, we scale the retrievals so that they are consistent with the model states. We do not make use of absolute temperature values; we make use of scaled temperature anomalies.

  21. Strategy for dealing with the inconsistencies Scale the data prior to assimilation. 1) Get time series mean μ and standard deviation σ for satellite Tskin (“T_sat”) and for model-based synthetic Tskin observations (“T_mod”), broken down by diurnal cycle and month. 2) Scale satellite Tskin (“T_sat”) into model climatology (std normal deviates): T_sat_scaled = σ_mod/σ_sat · (T_sat – μ_sat) + μ_mod 3) Assimilate scaled satellite Tskin (“T_sat_scaled”).

  22. Satellite skin temperature: - Int’l Satellite Cloud Climatology Project (ISCCP; 1983-2004) (NOAA-xx, GOES, METEOSAT, GMS,…) - 3-hourly, mapped to 1 deg lat-lon grid - clear-sky only! Demonstration of scaling approach Model: NASA Catchment land surface model on 1 degree lat-lon grid. (“off-line” – not coupled to atmospheric model) Surface meteorological forcing data: - Global Soil Wetness Project (GSWP-2; 1986-95) Assimilation: Ensemble Kalman Filter (developed at NASA/GMAO)

  23. A few days in July 1986 at Ft Peck, MT, USA Tskin [K] Tskin mean and dynamic range from satellite and model differ. Assimilation w/o scaling increases peak Tskin.

  24. A few days in July 1986 at Ft Peck, MT, USA Tskin [K] Tskin mean and dynamic range from satellite and model differ. Assimilation w/o scaling increases peak Tskin. When assimilating w/o scaling, model produces excessive sensible heat flux. Latent heat flux also increases when soil moisture is available. Sensible heat flux [W/m2] Latent heat flux [W/m2]

  25. Surface energy balance: G = Rnet - LE - H - “assimilation flux” G = Ground heat flux Rnet = Net radiation LE = Latent heat flux H = Sensible heat flux Assimilation flux = Added energy flux such that model pulls close to Tskin observations. Ideally small and white noise in time. “Assimilation flux”

  26. Surface energy balance: G = Rnet - LE - H - “assimilation flux” G = Ground heat flux Rnet = Net radiation LE = Latent heat flux H = Sensible heat flux Assimilation flux = Added energy flux such that model pulls close to Tskin observations. Ideally small and white noise in time. “Assimilation flux” Excessive, non-white “assimilation flux” when assimilating w/o scaling.

  27. Impact on root zone moisture (Bondville, IL, USA, 1986-1995) Tskin [K] Latent heat flux [W/m2] Root zone soil moisture [m3/m3] Even though effects on monthly (incl. diurnal!) average Tskin are small, assimilation w/o scaling impacts latent heat, and eventually root zone soil moisture. SUMMARY: Without scaling assimilation has negative impact.

  28. Conclusions No agreed global climatology of (absolute) surface soil moisture. Severe mismatch between modeled land surface temperature and satellite skin temperature. Scaling needed for assimilation. Assimilation of satellite data improves soil moisture estimates. Without scaling, assimilation of LST produces worse estimates. Immediate future tasks: Improve data assimilation: - Quality control. - Spatially variable model and observation error parameters. - Adaptive tuning of model and observation error parameters. - Implement operational land initialization for seasonal prediction (AMSR-E). - Multi-variate soil moisture and temperature assimilation. - Do improved land initial conditions lead to better seasonal forecasts?

  29. THE END.

  30. Monthly average assimilation flux (1986-1995) SUMMARY: Without scaling assimilation has negative impact.

  31. Global soil moisture data sets GSMDB stations ~70 of 200 useful 1. Satellite retrievals (upper 1.25cm, 50-140km, ~3 days) AMSR-E (2002-05) Soil moisture retrievals not available under dense vegetation, near open water, in frozen soil. SMMR (1979-87) More AMSR-E data than SMMR data. Number of data per month 2. Model data NASA Catchment Model (CLSM) forced w/ observation-corrected meteorological data. (upper 2cm, ~40…150km, 3-6h) 3. In situ data (upper 5…10cm and profile, point scale, hourly - 10 days) USDA SCAN stations ~17 of 122 useful

  32. Satellite anomaly minus model anomaly EnKF anomaly minus model anomaly Impact of SMMR assimilation – July 1982 Assimilation product lies “between” SMMR and model. There are interesting dynamical effects. Reichle & Koster, GRL 2005

  33. Koster et al., Science, 2004 Anomaly time series Reichle et al., JHM, 2004, also showed that… …satellite and model anomalies agree where soil moisture is important for seasonal forecasts!

  34. A few days in July 1986 at Ft Peck, MT, USA Tskin [K] Tskin mean and dynamic range from satellite and model differ. Assimilation w/o scaling increases peak Tskin. When assimilating w/o scaling, model produces excessive sensible heat flux. Sensible heat flux [W/m2]

  35. Satellite vs. satellite bias (time avg. soil moisture) Maybe the 80’s were much wetter than 2002-05…? Perhaps a little, but model estimates suggest much smaller differences.

  36. Satellite vs. satellite bias (time avg. soil moisture) Maybe the 80’s were much wetter than 2002-05…? Perhaps a little, but model estimates suggest much smaller differences.

  37. Soil moisture scaling for data assimilation For “new” sensors: - cannot use time series from historic satellites, - long time series not immediately available! Solution: Approximate CDF from many 1-year time series at grid points within some distance from point of interest.

  38. Soil moisture scaling for data assimilation (std) ORIGINAL multi-year data sets (Satellite std minus model std) SCALED multi-year data sets (Satellite std minus model std) CDF scaling based on 1 year of satellite data Reichle et al. JHM 2004 Reichle & Koster GRL 2004 1 year of satellite data sufficient for considerable reduction in long-term bias.

  39. Validation against in situ data Assimilation product has improved phase of annual cycle. Reichle & Koster, GRL 2005

  40. Soil moisture mission planning Commonly, soil moisture mission planners require a measurement accuracy of ~0.04 m3/m3 (“4%”) in absolute soil moisture. Time-invariant errors contribute to RMSE but do not affect anomaly estimates. Observing System Simulation Experiment (OSSE) result: For a large part of the Red-Arkansas river basin, satellite retrievals might be useful (R>0.5) even though their absolute errors exceed 0.04 m3/m3. For modeling and forecasting applications, satellite retrievals might be more useful than previously assumed. Crow et al., GRL 2005

More Related