ASSIMILATING DENSE PRESSURE OBSERVATIONS IN A HIGH-RESOLUTION WRF ENSEMBLE USING AN ENKF Luke Madaus -- Wed., Sept. 21, 2011
Overview Motivation Background and previous research Treating the model Treating the observations Future work
Past problems Weather models still poorly predict the timing and intensity of significant weather events Images from Phil Regulski Eckel and Mass (2005) – For short-range forecasts,, important to capture variability at small scales using very high resolution Data assimilation can try to introduce small-scale features – if variables assimilated are chosen judiciously
Why pressure? Less sensitive to representativeness error Widely available observations Has far-reaching meso- and synoptic-scale relevance Also can provide information in the vertical (Dirren et al 2007)
Previous studies Surface pressure obs used to do early 20 th century reanalysis (Whitaker et al 2004) Increases in both temporal frequency and spatial density of pressure observations lead to decreased errors (Anderson et al 2005) Large covariances between pressure and geopotential height through hydrostatic balance (Dirren et al 2007) Pressure/altimeter observations shown to increase accuracy of modeled MCSs and cold pools (Wheatley and Stensrud 2009)
Fundamental Question We’ve seen that pressure observations contain information relevant to synoptic scales… …but to what extent can pressure observations be used to describe phenomena on the mesoscale?
Fundamental question To investigate: Use a large ensemble capable of resolving mesoscale features Need observations at a density sufficient to represent the same scales of variability we are trying to model To what extent can pressure observations be used to describe phenomena on the mesoscale?
Previous Studies ExperimentAnderson 05Dirren 07Wheatley 09Madaus 11 Model usedGFDL/GCMWRFWRF-ARW V2WRF-ARW V3.2 Grid spacing6 degrees45-km30-km4-km Domain areaGlobeW US, E Pac.CONUSPac. NW Vertical levels Num Ens. Mems # P obs assim (~ASOS)
Previous studies Anderson Et. Al 2005 – “30 latitudes, 60 longitudes”
Previous studies -- high-resolution? Wheatley and Stensrud, km grid spacing
High-resolution Current Setup 4 km grid spacing Quasi-explicit resolution of: Some convective processes Small-scale boundaries Some localized orographic effects Need observed data to match! Weisman et al. 2008
Data sources ASOS All potential obs
Data sources TOTAL – observations hourly across Pac. NW ASOS – Canada and US (100) Weather Underground (650) AWS Schoolnet (80) CWOP (250) RAWS (5) Oregon RWIS (10) Pendleton WFO Network (15) Land/Sea Synop (30) Other (50)
Experiments in progress Test case – July 25, 2011
Sample Impacts Control assimilation – only QC obsAll obs raw assimilation
Bias Removal Bias in observations can throw off model estimate of state Covariances with other variables can amplify the impact of bias
Possible solutions MADIS RSAS Analysis Grids Successfully used to quality check MADIS observations (CWOP, AWS, etc.) (Miller and Barth 2002) Could be applied in a manner similar to Mass and Ferber (1990) Preliminary success – around 90% of biased obs “corrected” What about pressure tendency? Should not be affected by bias Not investigated as something to assimilate
Short-term future work Investigate these areas: Look into how the ensemble responds to pressure assimilation with respect to mesoscale variability How much of the full state vector can pressure capture? What is the structure of covariances? How does model variability on small spatial scales reflect the introduction of more pressure observations? How does the ensemble’s mesoscale accuracy change with variations in spatial density and temporal frequency of observations? How can the EnKF assimilation system be appropriately adjusted to reflect the properties of these observations? Continue bias mitigation efforts
Medium-term future work Additional case studies Hopefully a storm this fall… Begin discussions with forecasters at Seattle WFO Look into assimilating pressure tendency observations to see their impact Application to NCAR visiting graduate student program Developing DART code
Long-term future work If pressure assimilation can help resolve mesoscale variability, it has great potential to not only help predict, but also to inform Use this as a tool to investigate convective initiation High-density pressure assimilation could allow detailed low-level structure to be represented in models This could help to clarify key features leading to convective initiation
Acknowledgements Advisors – Cliff Mass and Greg Hakim Phil Regulski Rahul Mahajan Mark Albright Jeff Anderson and Nancy Collins at NCAR
References Anderson, J., B. Wyman, S. Zhang, T. Hoar, 2005: Assimilation of surface pressure observations using an ensemble filter in an idealized global atmospheric prediction system. J. Atmos. Sci., 62, Dirren, S., R. Torn, G. Hakim, 2007: A data assimilation case study using a limited-area ensemble filter. Mon. Wea. Rev., 135, Eckel, F. A., C. Mass, 2005: Aspects of effective mesoscale, short-range ensemble forecasting. Weather and Forecasting, 20, Mass, C. and G. Ferber, 1990: Surface pressure perturbations produced by an isolated mesoscale topographic barrier, part 1: general characteristics and dynamics. Mon. Wea. Rev., 118, McMurdie, L., C. Mass, 2004: Major numerical forecast failures over the northeast Pacific. Weather and Forecasting, 19, Miller, P. and M. Barth, 2002: RSAS Technical Procedures Bulletin. MSAS/RSAS. Web. Accessed: Sept. 12, Weisman, M., C. Davis, W. Wang, K. Manning, J. Klemp, 2008: Experiences with 0-36-h explicit convective forecasts with the WRF-ARW model. Weather and Forecasting, 23, Wheatley, D. and D. Stensrud, 2010: The impact of assimilating surface pressure observations on severe weather events in a WRF mesoscale ensemble system. Mon. Wea. Rev., 138, Whitaker, J., G. Compo, X. Wei, T. Hamill, 2001: Reanalysis without radiosondes using ensemble data assimilation. Mon. Wea. Rev., 132,
Pressure tendency Covariances not as strong—less impact than raw pressure (Wheatley and Stensrud 2009) Pressure tendency requires continuity of observation Not currently supported in the DART EnKF assimilation framework
Different Parameterization WSM-3 microphysicsWSM-5 microphysics
How bad is bias? Before Bias Correction 535/1100 error >1.5mb After Bias Correction 50/1100 error > 1.5mb
Pressure tendency What about pressure tendency as a way to avoid bias? Bias not present in this representation of pressure obs
EnKF assimilation Pressure (hPa) 1010 hPa 1009 hPa Ensemble Observation New Estimate