Assimilation of Doppler radar observations for high-resolution numerical weather prediction Susan Rennie, Peter Steinle, Mark Curtis, Yi Xiao, Alan Seed
Introduction Numerical Weather Prediction (NWP) is continually advancing The use of radar data to improve the forecasts is growing What are the applications for radar data in NWP? What are the challenges?
RNDSUP 2004-2009 SREP 2010-2014 Since SREP And there are more coming The Rise of Doppler
Australian Numerical Weather Prediction ACCESS Australian Community Climate and Earth System Simulator Based on UK Met Office software Includes atmospheric forecast model, observation processing system and variational analysis package Used for BoM NWP operationally since 2010
Evolution of ACCESS City scale Australian Parallel Suite releases Version Resolution Assimilation Date operational APS0 0.05 (~5 km) No 2010 APS1 0.036 (~4 km) No 2013 APS2 0.015 (1.5 km) No 2016 APS3 0.015 (1.5 km) Yes 2018 APS4 0.015 (1.5 km)? Yes TBD Comparable to radar observation resolution
Why is this now possible Improved computing power to handle large grids and large observation volume Improvements to the formulation of atmospheric modelling Improved communication: observations in, forecasts out Improved data assimilation
Role of ACCESS-City Forecasts at short lead times 3+ hours Take over from nowcasting/persistence Severe, intense weather in short time scales Skill for 12-24 hours Small domain: spin up vs. advection Nowcast storm warning
The Forecast Demonstration Project Sydney domain, September-December 2014 Test out new forecasting products in a pseudo-operational environment Included the prototype for APS3 ACCESS-City 1.5 km resolution, Rapid Update Cycle Hourly assimilation (including radar!) Forecasts out to 12-36 hours every hour
SREP/FDP Sydney variable grid BoM ACCESS model for capital cities Boundary conditions from ACCESS-G at 25 km res or ACCESS-R at 12 km res Grid varies from 4 km to 1.5 km 70 vertical levels, to 40 km Every 10 th grid shown Radar observations near T 20, T, T+20
Doppler radar observations What do we get? 1º increments 250 or 500 m range bins Max range 150-300 km Nyquist velocity 26 to 52 m s -1. 6-10 minute scans 14 PPI scan elevations This is data overload! QC and thinning required Limit correlated observations (esp. spatial proximity) Limit range due to location uncertainty Limit frequency of use
Echo from precipitation and insects Yarrawonga 7/11/2010 940 UTC (just after sunset) Velocity from precipitation and insects Moths and locusts Difficult to separate these Plan is to use both, but handle separately
First step is to de-alias observations Preparation of radial winds
Radar echo classification Three types of echo for radial velocity applications 1: Okay to assimilate (strong precipitation echo) 2: Maybe okay to assimilate (weak weather echo, insects, smoke) 3: Do not assimilate (ground clutter, sea clutter, birds ) For Forecast Demonstration Project Used a Naïve Bayesian Classification combined with thresholds, map-based probabilities Assimilated precipitation echo and insect echo.
Coverage and Classification
Assimilation of radial winds Observation Processing System, precedes assimilation Comparison with model background Observation operator used to create model-equivalent values Consistency check by comparing with observations Can t assimilate observations largely different from model background
Observation processing Superobbing Spatial averaging of observations reduce correlations and errors Thinning Usually to coarser than model resolution Reduce correlations, data density
4D-Var Data Assimilation J x 0 Best model state that fits obs in 4D = x x b T B 1 x x b + x x a x b Observations N i=0 y i h i x i Background forecast Analysis Where: J is the cost function x is the model state y is the observation h is the observation operator that transforms the model state into observation space B is the background covariance matrix R is the observation covariance matrix T R 1 y i h i x i t 0 t N time
Assimilation in ACCESS-Sydney domain Observations from multiple radars within Sydney domain Modification to model wind field after assimilation
Coarser horizontal resolution More sensitive to clear air Number of observations assimilated Precipitation Clear Air (insects) 30% of obs contributed from insect echo 0.9 2.4 4.2 5.6
Speed bias (model obs) Larger bias in elevations with most ground or sea clutter Clutter not filtered on-site and not classified correctly Not enough data because only part of scan in domain
Noisier velocity contributed by difficulties in onsite dual PRF dealiasing Mean Obs-Bkg by scan
Mean of assimilated innovations These have undergone OPS QC and superobbing No major differences between precipitation and insect observations after all QC Bias from undersampling
Altitude (m) Meridional wind speed (m s -1 ) Observation impact and limitations southerly Observations at the time of an approaching southerly change Most radar observations are above the height of the change, except near the radar. Latitude northerly
Other radar assimilation applications BoM may use these in future Already developed by UK Met Office, collaborative partner Radar reflectivity Direct assimilation of reflectivity Can provide information about where it isn't raining Clutter would be very detrimental
Other radar assimilation applications Radar refractivity Atmospheric refraction mostly a function of humidity Change in refractivity determined from change in phase of echo from fixed target Finally a use for ground clutter! Frédéric Fabry, 2004: Meteorological Value of Ground Target Measurements by Radar. J. Atmos. Oceanic Technol., 21, 560 573. Nicol, J. C., Illingworth, A. J. and Bartholomew, K. (2014), The potential of 1 h refractivity changes from an operational C-band magnetron-based radar for numerical weather prediction validation and data assimilation. Q.J.R. Meteorol. Soc., 140: 1209 1218
Conclusions We can successfully assimilate Doppler radar wind observations into the high-resolution NWP system Quality control is essential Ongoing issues include De-aliasing onsite (dual PRF errors) Offsite de-aliasing Echo classification, particularly removal of ground and sea clutter
Future More radar network upgrades Await the arrival of dual polarisation -- It will solve everything! Trials for the impact of radar observations on the forecast It should make the forecast better, but no guarantees, especially if the QC is not adequate. For data assimilation 1 bad observation can negate the benefit of many (50-100) good observations