Radar precipitation measurement in the Alps big improvements triggered by MAP Urs Germann, Gianmario Galli, Marco Boscacci MeteoSwiss, Locarno-Monti MeteoSwiss radar Monte Lema, 1625m
Can we measure precipitation in the Alps? 2 Gauge Errors because of high spatial variability, wind, snow drift, and systematic local gradients. Radar Large errors because of ground clutter, beam shielding, and other sources of error.
1 radar 3 300 km red colors >40mm/h
About 60 automatic gauges 4 300 km red colors >40mm/h
Strategy in the Alps 5 Radar estimate of surface precipitation rate is a weighted linear average of all - visible, - clutter-free, - profile-corrected, and - visibility-corrected radar measurements above the location of interest.
Recent modifications at MeteoSwiss 6 Since 1993 Automatic microwave calibration and hardware monitoring to guarantee stability Cavalli 1998 Summ 1997 Spring 1999 Spring 2001 Spring 2003 Introduce visibility map to account for partial and total shielding Pellarin et al 2002, Germann+Joss 2004 Improve operational 7-step clutter elimination decision-tree Joss+Lee 1995, Germann+Joss 2004 Introduce meso-beta profile for operational robust profile correction Germann+Joss 2002 Global bias correction to obtain agreement with gauges on large space-time scale; improve mesobeta profile algorithm Germann et al 2003 now Try local bias correction all Paper in in real-time preparation
7-step clutter elimination 7 83m gate leak 1999 OFF Noise level update clutter map Doppler width 1999 velocity+vicinity 1999 1-lag fluct. 2-lag fluct. 22% of area 5 mm/h Modification 1999: - less residual clutter - less erroneously eliminated weather ON clear-sky vertical gradient dynamic clutter map 1.3% of area 0.01 mm/h
Objective large-scale verification against gauges 8 Large data set: Daily totals of 1997-2004 at 58 (433) locations. "Objective" means: - meaningful + robust quality descriptors; - include gauges in whole Switzerland; -include all precipitation events; Gauge networks: red: automatic blue squares: manual
Quality descriptors 9 Bias: radar/gauge (accumulation over whole season) Scatter: variation of daily radar/gauge ratio cumulative contribution to total rainfall 16% percentile p e r f e c t scatter 84% percentile dry error (radar/gauge ratio) wet POD, FAR, ETS: skill scores for 0.3mm in 24h
Objective large-scale verification vicinity whole Switzerland Bias Bias Scatter POD FAR ETS Summer 1997 0.63 0.50 2.7 0.84 0.34 0.40 introduce visibility map Summer 1998 0.76 0.63 2.7 0.87 0.30 0.39 improve clutter elimination Summer 1999 0.41 0.34 3.6 0.73 0.08 0.53 introduce meso-beta profile correction Summer 2001 0.42 0.36 2.1 0.75 0.08 0.57 global bias correction; improve meso-beta profile Summer 2004 1.1 0.87 2.0 0.89 0.14 0.62 try local bias correction (training with 2003 data) Summer 2004 1.01 1.003 1.7 0.90 0.15 0.63 10 Factor Factor Factor
1999 versus 2004 (whole Switzerland) 11 Summer 1999 Summer 2004 (after correction of local bias as observed in 2003) too dry p e r f e c t too wet Bias: 0.34 (-76%) Scatter: 68% of rainfall within factor of 3.6 Bias: 1.003 (0%) Scatter: 68% of rainfall within factor of 1.7
Bias geography 12 Summer 1999 Factor 0.50 0.79 1.25 2.00 too dry perfect too wet
Bias geography 13 Summer 2004 after correction of local bias as observed in 2003 Factor 0.50 0.79 1.25 2.00 too dry perfect too wet
Local scatter 14 Scatter 2004 (after correction of local bias as observed in 2003) Ticino area: daily radar/gauge within factor of 1.4 for 68% of rainfall Alpine crest: daily radar/gauge within factor of 1.9 for 68% of rainfall
Debris floods and landslides: 29 August 2003 15 Radar accumulation of 4 most intense hours Good agreement between radar maxima and hydrogeological response. 100 mm Potential for better nowcasting? 80 km debris floods and landslides MeteoSwiss and IST-SUPSI
Applications in the Alps Recent initiatives for use of Alpine radar precipitation: 1 hydrologic model for prediction of sea level (research project with hydrologists) D-PHASE! Magadino, 16.10.2000 16 Lugano, 1951 2 3 4 automatic alert for heavy precipitation (operational) basis for training of heuristic probabilistic short-term forecast system for heavy precipitation (research) routinely delivery to hydro power companies (operational) low-level inflow from Doppler Lema radar, Ticino
Far too late for SOP in 1999 17... but right in time for D-PHASE ;-)
Summary 18 New algorithms and large-scale evaluation Radar + landslides 100mm in 4h Richness + risks 100 Heuristic nowcasting Heavy precipitation thank you!
Local bias 19 Summer 2004
Representativeness error of gauge 20 Background: Representativeness error of point observation for basin-average is large because of high spatial variability of precipitation. Idea: Use spatial variability of precipitation as seen by radar (variogram) to quantify expected representativeness error of gauge. Preliminary result: Using a basin of 26 km 2, a fictive gauge location, and a variogram climatology of Alpine precipitation we get: - Expected fractional error for hourly rainfall >18% in 50% of cases -Error >38% in 16% of cases -Error <6% in 16% of cases Bolliger and Germann, in preparation