6 th SMOS Workshop, Lyngby, DK Using TMI derived soil moisture to initialize numerical weather prediction models: Impact studies with ECMWF’s Integrated Forecast System Matthias Drusch ECMWF Acknowledgements: E.F. Wood and H. Gao (Princeton University)
6 th SMOS Workshop, Lyngby, DK Outline 1.Motivation and introduction 2.Operational OI analysis vs Open Loop experiments - Forecast impact - Soil moisture validation against OK Mesonet 3.Operational OI analysis vs TMI nudging experiment - Bias correction - Soil moisture validation - Forecast impact 4.Summary and Outlook
6 th SMOS Workshop, Lyngby, DK ECMWF long-term forecasts (from ENSEMBLES project) volumetric soil moisture [%] 2 m temperatures [º Celsius] (monthly averages for North America) [W m -2 ] turbulent surface fluxesfractional cloud coverage [%] Soil moisture has an impact on the atmosphere and the weather forecast. Systematic errors in the land surface scheme result in a (dramatic) dry down with summer values close to the permanent wilting point. The corresponding 2 m temperature forecasts show a strong warm bias exceeding 10 K during summer and 5 K during winter. The model has to be re-initialized with analysed soil moisture to prevent from drifting into an unrealistic state.
6 th SMOS Workshop, Lyngby, DK General introduction A well posed analysis is a better estimate of the true state than either the modelled background information or the observation data sets available. - initial state for a numerical weather forecast - reference against which to quality check other observations - pseudo observation for e.g. satellite retrieval algorithm development sequential, intermittent assimilation analysis observations shortrange forecast medium-range forecasts
6 th SMOS Workshop, Lyngby, DK Data Assimilation Experiments 1. CTRL OI (Optimum Interpolation) based on screen level analyses for the top three model soil layers. 2. OL (Open Loop) without any soil moisture analysis. 3. NUDGE (Nudging) experiment using the TMI Pathfinder soil moisture product. Common features: - Full atmospheric 4DVar analysis using ~ 10 6 observations / 6h (reflecting the operational configuration). - Model version CY29R1. - T511 spectral resolution, 60 vertical levels. - ‘Early delivery’ set up with 10-day forecasts from 00 and 12 UTC. - Study period from 1 June to 31 July 2002.
6 th SMOS Workshop, Lyngby, DK Soil moisture increments (CTRL OI) [mm] Accumulated root zone soil moisture increments for June 2 to July 30, Analysis increments are a sizeable part of the terrestrial water budget.
6 th SMOS Workshop, Lyngby, DK Forecast skills Temperature at 1000 hPa grey: OI black: OL solid: North America dotted: Europe dashed: East Asia Root-mean-square error E areaHeight [hPa] 24 h72 h120 h168 h216 h Northern Hemisphere Europe East Asia North America Significance levels The proxy ‘observations’ are efficient in improving the turbulent surface fluxes and consequently the weather forecast on large geographical domains.
6 th SMOS Workshop, Lyngby, DK Validation against OK Mesonet observations
6 th SMOS Workshop, Lyngby, DK Validation of forcing data area averages for Oklahoma (72 stations) daily precipitation model forecast (OI) observations total amount of rainfall: June87.3 mm modelon19 days 87.8 mm observationson 9 days July 110. mm modelon26 days 79. mm observationson20 days daily downward shortwave radiation model forecast (OI) observations Correlation: 0.85 Bias: Wm -2
6 th SMOS Workshop, Lyngby, DK Validation of soil moisture area averages for Oklahoma (72 stations) surface soil moisture model forecast (OI) observations model forecast (OL) Too quick dry downs (model problem). Too much precip in July (model problem). Too little water added in wet conditions (analysis problem). NO water removed in dry conditions (analysis problem). root zone soil moisture model forecast (OI) observations model forecast (OL) Precipitation errors propagate to the root zone. Analysis constantly adds water. The monthly trend is underestimated. The current analysis fails to produce more accurate soil moisture estimates.
6 th SMOS Workshop, Lyngby, DK TMI Pathfinder Data Set (%) July 2 nd, 1999 (Gao et al. 2006) Data set produced by: Dept. Civil and Environmental Engineering, Princeton University, NJ Basis: brightness temperatures at GHz horizontal polarization Method: physical retrieval based on land surface microwave emission model and auxiliary data sets from the North American Land Data Assimilation Study project Output: surface soil moisture [cm 3 cm -3 ],
6 th SMOS Workshop, Lyngby, DK Corrected TMI soil moisture volumetric surface soil moisture [%] for 06/06/2004 the modelled first guess original TMI Pathfinder data corrected TMI data set (bias correction through CDF matching)
6 th SMOS Workshop, Lyngby, DK Nudging set up Delayed cut-off atmospheric 4D-Var (12 h) AN FC AN FC TMI sampling period (daily) soil moisture analysis 1/4 2/41/42/4 10-day forecasts
6 th SMOS Workshop, Lyngby, DK Validation of soil moisture area averages for Oklahoma surface soil moisture Nudging / satellite data remove water effectively and produce a realistic dry down. Nudging the satellite results in the most accurate surface soil moisture estimate. root zone soil moisture The information introduced at the surface propagates to the root zone. The monthly trend is well reproduced using the nudging scheme. Satellite derived soil moisture improves the soil moisture analysis and results in the most accurate estimate.
6 th SMOS Workshop, Lyngby, DK Forecast skill correlation (observation / fc)bias Nudging OL OI rH T T The impact of the satellite data on the forecast quality (of screen level variables) is neutral (correlation). The biases obtained from the nudging experiment are slightly higher when compared against the OI and lower when compared against the OL.
6 th SMOS Workshop, Lyngby, DK Forecast – observation differences CTRLOpen LoopNUDGE RH 2m [%] T 2m [%] The nudging experiment performs best in the south-western and central parts of the study area, which are characterized by ‘low vegetation’ (short grass) and ~ 15 % of bare soil.
6 th SMOS Workshop, Lyngby, DK Impact on weather parameters CTRLNUDGEOpen Loop surface soil moisture [%] at 18 June, 12 UTC latent heat flux [Wm -2 ] mean over 18 June 12 UTC to 00 UTC sensible heat flux [Wm -2 ] mean over 18 June 12 UTC to 00 UTC planetary boundary layer height [m] at 19 June 00 UTC total cloud coverage [0-1] at 19 June 00 UTC
6 th SMOS Workshop, Lyngby, DK Soil moisture increments [mm] accumulated increments over June and July 2002 Optimal Interpolation (2 m T and RH) Nudging (TMI soil moisture)
6 th SMOS Workshop, Lyngby, DK Summary The OI analysis using 2 m temperature and precipitation is efficient in Improving the turbulent fluxes and consequently the weather forecast on large geographical domains. The quality of the resulting soil moisture profile is not improved. The OI analysis is not able to correct for the underestimation of the seasonal cycle in root zone soil moisture and for the effects of erroneous precipitation forecasts. However, it prevents the system from drifting into a too dry state. Surface soil moisture is a strong constraint for the NWP model. The surface scheme is able to propagate the information introduced in the top layer to the root zone. Soil moisture analysed from the satellite data is most accurate. There is a clear impact of soil moisture on weather parameters. The forecast skill is neutral (rms) to slightly negative (rH bias).
6 th SMOS Workshop, Lyngby, DK Summary (continued) The best soil moisture product does not necessarily result in the best NWP forecast. New (satellite) observations help to identify model errors and to improve physical models. In the end, the forecast will benefit from a better soil moisture product. In-situ observation are of fundamental importance for the development of model / data assimilation systems. It is important to observe soil moisture AND fluxes, screen level variables and meteorological forcings.
6 th SMOS Workshop, Lyngby, DK Oklahoma data sets 2002
6 th SMOS Workshop, Lyngby, DK TMI soil moisture transformation / bias correction r 2 = 0.66 r 2 = 0.69 r 2 = 0.01 r 2 = 0.18 CDF matching reduces systematic errors: The bias has been removed and the dynamic range has been adjusted. The random error may increase. transfer funcion 03/ /2002 x‘-x x Bias: % Bias: %