1 ARPA-SIM, Bologna, Italy and CIMA, Savona, Italy. Improving the radar data mosaicking procedure by means of a quality descriptor Fornasiero, A., Alberoni, P.P., Amorati, R., Marsigli, C.
Quality Descriptor (ERAD, 2004) Q indexUnit Anomalous P.[]: 0, 1, 2 distance[m] Beam Block.[dB] vol [dB] PIA[dB] Q d = quality before correction Q c = quality of the correction [0, 1] err fract > 0 err fract < 0 Fornasiero A. et al, 2005 : Effects of propagation conditions on radar beam-ground interaction: impact on data quality, ADGEO Fornasiero A., 2006 : On the uncertainty and quality of radar data, PhD thesis.
Issues 1.definition and testing of radar data composition methods taking into account data quality 2.verification of quality definition consistency with data reliability
The compared methods QUALITY-BASED APPROACHES MAX_Q: maximum quality AVE_Q: quality-weighted average CLASSIC APPROACHES: MAX_Z: maximum reflectivity MIN_DIST: minimum distance AVE_DIST: r -2 weighted average Gattatico San Pietro Capofiume Short pulse areas
Case study – 24 May gat quality spc reflectivity gat reflectivity spc quality
MAX_ZMAX_QAVE_Q gat-spc weight gat-spc reflectivity
Scores – tp (10 h) threshold (mm)1246 observations om=1.76 mm
Case study – August gat quality spc reflectivity gat reflectivity spc quality
MAX_Z MAX_QAVE_Q gat-spc reflectivity gat-spc weight
Scores – tp (18 h) threshold (mm) observations om=11.9 mm
Concluding.. Quality information improves precipitation estimate in radar composits in convective cases, respect to traditional composition methods The wider is the spectrum of error sources enclosed within the quality descriptor, the more accurate is the composed precipitation field, even if some errors are not corrected AVE_Q is preferable with respect to other method especially when there is a lack of informations in Q The distance-based methods seem to be preferable respect to MAX_Z It is necessary to test the method in stratiform cases, after inserting VPR-related quality component into the Q function
Appendix Radar data resampling Data comparison Quality components Radar precipitation verification Data correction Addition of Q comp.
Addition of crucial quality components produces relevant changes in Max_q method om=1.76 mm
Data Correction Doppler filter Choice of the minimum elevation that is not affected by clutter and with a beam blocking rate lower than 50% Topographical beam blocking correction, based on a geometric optics approach Anomalous propagation clutter suppression Fornasiero, A., Bech, J., and Alberoni, P. P. Enhanced radar precipitation estimates using a combined clutter and beam blockage correction technique. pp SRef-ID: /nhess/
az az_min az_max 250 m Radar data resampling
Data comparison -radar data are resampled in a 1kmx1km grid -the observation is compared with the nearest radar measure -the precipitation is accumulated from the beginning to the end of the event -raingauges sampling interval=30 min. -only raingauges with the complete dataset (nmeasures=nhours*2) are considered -radar cumulated rain in 1 hour is calculated as weighted average of min 3, max 5 measures 1 KM
Quality components (1/3) CLUTTER Qd = 0 if clutter is present from VCT Qc = 0.5 Q* = 0.5 Qd =0.8 if the test is not applied BEAM BLOCKING Qd = f(BB)= 1-(BB/BB max ) 1/1.5 with BB MAX =50% Qc = f(BB)*f( q err)*f( D trs)*f( D rrs) f( q err)= 1- q err 1/1.5 pointing error f( D trs)= e - D trs/ D T time distance from radios. D T= 4 h f( D rrs)= e - D rrs/ D R space distance from radios. D R= 50 KM derived from Bech et al., 2003
DISTANCE Qd= e - b r from Koistinen and Puhakka, 1981 adj-factor clima = r/g=1-err fraz è e - b r clima FOCALIZATION/DIVERGENCE ERROR Qd = 1 – ( D Vol/Vol) 1/1.5 Vol = volume variation respect to standard propagation Quality components (2/3)
ATTENUATION Qd = 1 – (ATTENUATION RATE) 1/1.5 Quality components (3/3) Burrows and Attwood, 1949 =5cm, T=18°C
Radar precipitation verification (1/2) Categorical: only one set of possible events occurs Discrete predictand: takes only one of a finite set of possible values... is conducted as verification of categorical forecasts of discrete predictands
yesno yesab nocd raingauges obs > thr radar obs > thr scoreformulawhat does it represent?bpvwpv BS=bias score(a+b)/(a+c)how often event is forecasted respect to observed 1 HR = hit rate(a+d)/nfraction of correct forecasts10 TS=Threat score or critical succes index a/(a+b+c)number of correct ‘yes’ forecasts/ total forecasts or observed (HR after removing correct ‘no’ forec.) 10 POD=probability of detection a/(a+c)prob. event forecasted, given that it occurred 10 FA=false alarms ratiob/(b+c)proportion of forec. that failed01 HSS= Heidke skill score....(HR-HR random )/(1-HR random )1 -- rmseN (1/n (f i -o i ) 2 )/om rmse nomalized respect to mean observed field 0 biasN (1/n (f i -o i ))/om bias nomalized respect to mean observed field 0 “forecast”