Use of radar data in ALADIN Marin Juraek

Use of radar data in ALADIN Marián Jurašek marian. jurasek@shmu. sk Slovak Hydrometeorological Institute

Contents of presentation – current work with radar and ALADIN data in ALADIN’s countries – future: “Research plan for radar data assimilation in ALADIN” Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Current status – assimilation of radar data not yet developed for ALADIN – usage of radar data: – for ALADIN verification – together with ALADIN in hydrological models – together with ALADIN in nowcasting application – all work is done only on national level Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

AUSTRIA – archiving 1 h cumulated precipitation from 10 minutes radar data in lat-lon grid – archiving precipitation fields based on surface observations in the same grid – mainly used for ALADIN convective rainfall forecast visual verification Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

AUSTRIA (2) – first study made for eastern alpine domain – period: summer 2003 – first results: – ALADIN precipitation forecast is not selective enough in space – early bias with regard to the onset of precipitation – in some areas model generates convective precipitation on almost every day during the summer season Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

AUSTRIA (3) Hourly precipitation rate estimated by radar measurement Use of radar data in ALADIN Hourly precipitation rate interpolated from the local obs. network EWGLAM/SRNWP/COST 717, Lisbon 2003

AUSTRIA (4) 17 th August 2003, 17 UTC Convective cloudiness and precipitation prognosed by ALADIN Use of radar data in ALADIN Convective precipitation (hourly rate) prognosed by ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

AUSTRIA (5) – results leaded to the experiments with modified trigger functions in the ALADIN deep convection scheme. – larger project connected with integrated flood forecasting system will probably start next year – combination surface precipitation observations, radar and ALADIN Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC – ALADIN data used for generating radar image forecast – wind field data - crucial point of the radar echo prediction – tested 3 different methods for wind field calculation – COTREC – Wavelet Transform Decomposition – ALADIN geopotential field at 700 h. Pa Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (2) COTREC – comparison of two consecutive radar images using some similarity criteria – smoothing of final field WAVELET – similar to COTREC, but radar image is decomposed to subspaces using the wavelet transformation – calculation of decomposition similarity criteria at several different detail levels – smoothing of final field like in COTREC Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (3) ALADIN – cloud motion is controlled by air mass flow at approx. 35 km above sea level – it corresponds with geopotential at 700 h. PA (cca 3 km) – ALADIN data interpolated to the image size and to the resolution of radar data – motion field calculated from geostrophical approximation Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (4) COTREC Wind field calculated by COTREC method Use of radar data in ALADIN Forecasted radar image by COTREC method EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (5) WAVELET Wind field calculated by WAVELET method Use of radar data in ALADIN Forecasted radar image by WAVELET method EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (6) ALADIN Wind field calculated by ALADIN method Use of radar data in ALADIN Forecasted radar image by ALADIN method EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (7) Results of comparison: – all methods improve radar information – for all methods, similar decrease in forecast quality with forecast time – in most cases, the ALADIN method is slightly worse – ALADIN method needs only one radar image – ALADIN method has the smallest hardware requirements Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

CZECH REPUBLIC (8) Conclusion: – in operational use only COTREC and ALADIN methods – forecasted radar image generated every 10 minutes – forecasted for +10 min +20 min. . . +90 min Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

FRANCE – study of flood event in southern France from 8 th to 10 th September 2002 – visual verification (comparison) of cumulative radar rain with cumulative ALADIN rain forecast – main goal: to see the general evolution of the precipitation Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

FRANCE (2) 48 h cumulated precipitation measured by radar Use of radar data in ALADIN 48 h cumulated precipitation forecasted by ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

FRANCE (3) – space-time interpolation of precipitation fields – pattern matching applied on radar data – attempt to prepare radar data for non-visual verification – radar data space filtered to ALADIN resolution (cca 10 km) – spectral study of radar and ALADIN precipitation Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

FRANCE (4) – result of spectral study: forecast error is like white noise, no difference between frontal and convective precipitation – spectral filtering used for separating large- from small scale precipitation – discrepancy between radar and model data separated to: – large scaling error – large scale geometrical deformation error – small scale residual Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

FRANCE (5) – Computation of optimal deformation operator: – correlation method - numerically too expensive, already for 200 x 200 points field – incremental variational technique – highly non-quadratic pattern matching cost function è problem with optimisation if good first guess not available Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

HUNGARY – radar and ALADIN data as input for hydrological model DIWA (DIstributed WAtershed) running outside of Hungarian Meteorological Service – input for model: – – ALADIN precipitation forecast ALADIN min/max temperature ECMWF forecasts ( as ALADIN + ensemble) calibrated 12 h cumulated precipitation from radar measurements Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

HUNGARY (2) Example of calibrated 12 h radar precipitation Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Example of hydrological model simulation

HUNGARY (3) Verification of 72 h hydrological model forecast over 110 days period Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Conclusion of first part – none radar data assimilation to ALADIN – usage of radar data together with ALADIN data is not coordinated – only some applications on national level Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Research Plan – in June 2003 prepared proposal of workplan for research of radar data assimilation for ALADIN – radar data - essential for mesoscale assimilation – available radar data: – reflectivity – instantaneous rainrate – cumulated rainfall – doppler wind , wind shear, turbulence – vertical wind profile – quantities from multiple polarisation measurements Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Which radar data? – each has advantages and disadvantages – not all available on all radar sites – reflectivity seems to be available on most European sites – not common form of data: – “PPI” images – volume data Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

The philosophy – learning from satellites – remote sensing process is complex and nonlinear è we should assimilate quantity close to measured (reflectivity instead of rainrate) è derived quantities contain hardly correctable errors – observation operator for simulation of reflectivity for each radar è development of system for radar data against model data monitoring Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

The philosophy (2) – biases and big errors cannot be handled by 3 D/4 DVar è software for detection and removal of corrupted data è study of space- and time structure of biases between simulated and observed data for bias correction è each radar processed independently è thinning of too dense data consistently with the resolution of the analyses Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

The philosophy (3) – very accurate modelling of physical process of observation è precisely interpolating / averaging model variables along the radar beam path – physical part of observation operator should be prepared by radar specialists – observation operator should relatively independent from model Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Things To Do – get samples of very good quality radar reflectivity data – get idea of fields needed to simulate reflectivity – get simple (to start with) reflectivity simulation formulae – specify obs. operator by list of necessary model fields, information about observation Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Things To Do (2) – carefully specify the technical implementation of previous things – implement radar data into the ODB processing – implement direct interpolation of model fields – convert model field to the reflectivity, compute and store difference with observation – study monitoring statistics Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Things To Do (3) – code tangent linear and adjoint of obs. operator – simulate one radar pixel – study the impact of reflectivity assimilation to forecast – run several cycles of data assimilation to see the cumulative effect – retune preprocessing and analysis parameters Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003

Thank You for your attention! marian. jurasek@shmu. sk Use of radar data in ALADIN EWGLAM/SRNWP/COST 717, Lisbon 2003