Radio occultation (RO) and its use in NWP Chris Burrows Sean Healy (ECMWF), Jordis Tradowsky (Bodeker Scientific), John Eyre (Met Office), András Horányi (ECMWF), James Cotton (Met Office) 10 May 2016
Contents Introduction to radio occultation (RO) Impact of assimilating RO data Bias correction of radiosondes using RO Future instruments Summary
Introduction to radio occultation (RO)
Radio occultation geometry Bending angle Fig from ROPP user guide. Impact parameter During a rising/setting occultation event, the tangent point (TP) sweeps up/down through the atmosphere/ionosphere. This results in a profile of bending angle (α) as a function of impact parameter (a). These observations are “unbiased”.
Distribution and number of obs Coverage of RO from Metop-A for one month. Distribution is uniform, and multiple instruments minimise swath and LST effects for any model cycle. Total observations (profiles) received in a month. All instruments sum to 75840 for this period.
Impact of assimilating RO data
RO data denial experiment Removing RO from a global NWP system degrades forecasts against a number of metrics. The following slides contain a small selection of results. (Both the experiment and control used VarBC.)
500 hPa height compared to ECMWF NH SH Removing RO increases RMSE
T+72 temperatures compared to ECMWF NH SH Removing RO increases RMSE
T+72 wind speed compared to ECMWF NH SH Removing RO increases RMSE
Data denial experiments (incomplete OSE) Data denial experiments (incomplete OSE). Metric is change in Met Office “NWP Index” Verified against obs Verified against “own” analyses Verified against ECMWF analyses
FSOI (plots from James Cotton) Impact of observations on T+24h forecast skill. (moist energy norm integrated from surface to 150hPa). Impact of RO is relatively small compared to ECMWF. This is partly due to the norm cutoff. After sondes, RO has the largest impact per sounding (i.e. per profile).
Where does RO influence the NWP forecast? This is the vertically-integrated moist energy norm of the difference between a 1hr forecast with all observations in the analysis and the 1hr forecast with no RO. Effectively this is the difference between analysis increments with/without RO. Each blob corresponds to a radio occultation observation.
Indirect impact VarBC was implemented at the Met Office in March 2016. In VarBC, “anchor” measurements (RO, radiosondes, etc.) attempt to constrain the biases of the system. This allows RO to have an indirect impact. [For a quantitative comparison of direct vs indirect impact see Aparicio and Laroche, 2015, Mon. Wea. Rev., 143, 1259-1274.]
O-B statistics No VarBC VarBC Backgrounds pull closer to RO with VarBC applied. Likely due to the RO itself, and perhaps radiosondes.
T+6 Bias against refractivity (COSMIC-1) No VarBC VarBC
Bias correction of radiosondes (for greater consistency between anchor observations)
Method of diagnosing radiosonde temperature biases ROM SAF visiting scientist Jordis Tradowsky from Bodeker Scientific spent two months at the Met Office investigating radiosonde biases using RO. The result was a vertical bias correction profile for each radiosonde site for four different solar elevation angles. Sun et al 2011 (and others) have used co-locations to achieve this. We use model backgrounds to avoid co-location errors A linear temperature retrieval was used to provide additional control (mainly to remove dependence on prior information). α denotes bending angle. Jordis’s report: http://www.romsaf.org/visiting_scientist.php Characterisation of radiosonde temperature biases and errors using radio occultation measurements. Submitted to JAMC (Tradowsky, Burrows, Healy, Eyre).
A few results First guess mean temperature departures for RO (blue), radiosonde (red) and correction (black). The green line is an extension of the black, to cover all standard pressure levels. Site in Norway. Type: RS92 Site in Russia. Type: MARZ 2-2
Solar elevation angle > 22.5° There is “clustering” of particular bias corrections in various regions. To first order, this effect is likely to be due to the radiosonde types.
Solar elevation angle > 22.5° An assimilation experiment with these corrections is underway There is “clustering” of particular bias corrections in various regions. To first order, this effect is likely to be due to the radiosonde types.
Future instruments (EDA study) Slides from: András Horányi (ECMWF), Sean Healy (ECMWF), Axel von Engeln (EUMETSAT), Yago Andres (EUMETSAT) In brief, the spread in the ensemble of data assimilations (EDA) is a proxy for the analysis and forecast error statistics. This is used to assess the impact of future observations by assimilating simulated observations. It is an alternative to OSSEs.
Scaling of GNSS RO impact (F. Harnisch) Temperature analysis at 100 hPa ~ 50 % of the impact of 128 000 profiles ~ 25 million bending angles per day today Large improvements up to 16 000 profiles per day Even with 32 000 – 128 000 profiles still improvements possible → no evidence of saturated impact up to 128 000 profiles (although the additional impact per observation is decreasing)
COSMIC-2 Polar + (EPS-SG +COS2-EQ, extra- tropics) - TEMPERATURE COSMIC-2 EQ + COSMIC-2 POLAR No RO EPS-SG EPS-SG ~2800 occs per day (2 sats measuring GPS + Galileo) +COSMIC2-EQ ~10500 (total) +COSMIC2-POLAR ~18000 (total) (EPS-SG+ COSMIC-2 EQ)
Summary RO data perform well in forecast impact experiments and FSOI With VarBC, the mean RO forecast departures are reduced RO may act as a reference in a hierarchical bias correction system Increasing data volumes can be expected to improve analyses and forecasts
Questions