GPSRO Usage at Deutscher Wetterdienst (DWD)

Deterministic models assimilating RO

  • ICON: non-hydrostatic grid-point model, icosahedral-triangular grid.
  • Hybrid height-based vertical coordinates.  Horizontally/vertically self-nesting.
  • Horizontal resolution:
    • Global domain: R03B07 (2949120 grid-points, ~13 km average mesh size), 90 levels, lid at 75 km.
    • Europe nest (two-way): R03B08 (659156 grid-points, ~6.5 km average mesh size), 60 levels (~22.5 km).
  • 8 forecast runs per day:
    • to 180 hours at 00 UTC, 12 UTC (main forecast runs),
    • to 120 hours at 06 UTC, 18 UTC
    • to 30 hours at 03 UTC, 09 UTC, 15 UTC, 21 UTC (boundary conditions for LAM)

Data assimilation method

  • 3D-Var PSAS with a window size of 3 hours (synoptic time T +/- 1.5h).
  • Cutoff-time for main run analysis: T + 2h 14min.
  • Cutoff times for late assimilation runs (from synoptic time):
    • 00, 12: 11h11m
    • 03, 15: 08h36m
    • 06, 18: 06h01m
    • 09, 21: 04h36m

RO quantity assimilated

  • Bending angle.
  • 1D operator, based on implementation by Michael Gorbunov.
  • Uses MSIS-90 climatology above model top.
  • No tangent point drift. We use an effective tangent point derived from rays between 3 km and 15 km impact height, not the nominal tangent point.
  • Vertical thinning depends on impact parameter:
    • 3 – 14 km: 400 m
    • 14 – 15 km: 500 m
    • 15 – 20 km: 700 m
    • 20 – 45 km: 1000 m
  • Thinning is complemented by superobbing (with exponential smoothing).
  • No bias correction

Observation error characteristics

  • Assigned observation error depends on impact height h and latitude (the errors vary smoothly in latitude between these bounds):
    • Tropics:
      • 1.5% of observed bending angle for h > 12 km,
      • 20% at 0 km,
    • Polar regions:
      • 0.9% of observed bending angle for h > 9 km,
      • 10% at 0km,
  • Additive representativity error near (statically modeled)  cold-point tropopause: 2% in tropics, 1.2% in polar regions.
  • Global absolute lower limit of 2*10^-6 rad.
  • Errors provided in the BUFR will be ignored.
  • Observations errors are assumed uncorrelated. (Diagonal R)

Impact

  • Adjoint sensitity studies are not possible with ICON or the former GME.
  • Only forecast impact studies were done before operational implementation and published later (H. Anlauf et al., Atmos. Meas. Tech., 4, 1105-1113).