Verification of LAMI (Local Area Model Italy) using non-gts Data over Mountainous Regions Elena Oberto (*), Stefano Bande (*), Massimo Milelli (*) (*) ARPA Piemonte, Torino, Italy
LAMI model Non-hydrostatic Limited Area Model (Italian version) developed in the framework of the COSMO (Consortium for Small-Scale Modelling) project between Germany, Poland, Switzerland, Greece and Italy. Technical aspects: Domain: 50 /2 /32 /24 (234*272 grid points) Resolution: 0.0625 (7.5 Km) Vertical layers: 35 Forecast time: +48h (+72h available since Dec 02: not taken into account because of the poor statistics) Model runs: 00,12 UTC Boundary conditions: GME (DWD) Initial conditions: GME (nudging version available since Dec 02: not taken into account because of the poor statistics)
Objectives Verification of precipitation above 1000 m: LAMI model output compared to observations over the western alpine chain. This is a study of high resolution model reliability in case of complex orography in perspective of the XX Olympic Winter Games in 2006. Period considered: Oct 02- Feb 03. Standard schemes of precipitation verification: contingence tables for different thresholds and statistical indices like BIAS, ETS, FAR and HRR. Very dense non-gts network of rain gauges (126) in the north-west part of the Alps (Piemonte, Liguria, Valle d Aosta, Ticino) above 1000 m. Method: comparison between station point and grid point (the one with the closest elevation among the 4 surrounding grid points).
Verification of vertical profile: The new radiosounding of Cesana Pariol (1545 m), placed in the Olympic area, is used to compare the observed and forecasted vertical temperature profiles (at 00UTC every day) An other radiosounding in our region is placed near Cuneo Levaldigi Airport (installed in 1999, since 1 year it is a GTS station): we perform the same vertical temperature profile verification to have a comparison with a station in a non-mountainous area. Mean error (BIAS) and Root Mean Square Error for each level (averaged levels every 25hPa) of the temperature vertical profile (00UTC LAMI run for +24h and +48h forecast time) from Dec 02 to Feb 03. Cesana Pariol (45 N 6.8 E): station point 1545 m grid point 1970 m Cuneo Levaldigi (44.5 N 7.6 E): station point 386 m grid point 387 m
Rain gauges network 126 station above 1000 m Regions interested: Piemonte, Valle d Aosta, Liguria, Ticino Cesana sounding Cuneo sounding
LAMI00-LAMI12: comparison between the first and the second 24h versus thresholds ETS: results between 0.25-0.35 no significative differences between the two runs and between the two days of integration. BIAS: globally good results, always greater than 1, for high thresholds the first 24h of both runs perform better especially for 12UTC.
ROC diagram confirms previous results: small differences between the two runs, but less FAR for the first integration time thresholds(mm) 5 10 20 35 50 75 LAM00_0024 LAM00_2448 LAM12_0024 LAM12_2448 OSS OSS OSS OSS 1675 1499 1349 1333 1087 1009 895 867 586 576 497 457 303 305 261 218 193 191 158 121 95 93 86 61 FOR FOR FOR FOR 2136 1804 1930 1676 1282 1172 1110 1046 677 700 585 540 350 434 308 309 200 290 180 206 122 183 80 105
LAMI00-LAMI12: comparison of the 12h-QPF performance for 3 fixed thresholds For every thresholds there is a diurnal cycle of error. The precipitation is generally overestimated BIAS influenced by the diurnal cycle more than forecast time BIAS better in the morning (00-12UTC)
The same behaviour comes out in ETS index for low thresholds only; for high thresholds (not shown here) the signal is smoothed. LAMI00-LAMI12: BIAS for the 6h-QPF Concerning the low thresholds, the same diurnal cycle is evident: the worst results are found during the night (18-00UTC)
Cuneo sounding bad agreement with observation close to the ground bias > 1 in the first levels probably due to a wrong heat flux parameterisation that gives a colder model forecast above 700 hpa: good bias for both forecast times above 800 hpa: +24h rmse is better than +48h rmse, slight worsening of the results with time.
Cesana sounding 800 hpa - 700 hpa: model T is cooler than observation due to the elevation difference and to a systematic underestimation worsening in time: +24h bias is closer to 0 than +48 bias, +24h rmse is higher than +48h.
Conclusions Globally good skills for LAMI QPF verification: general overestimation in precipitation, worsening in time. Diurnal cycle present and quite evident with worst results during the coldest hours. Good results for the vertical temperature profile above 700 hpa. Problems close to the ground probably due to the physical parameterisations. Next step: verification of the other variables of the sounding (Rh, DwT, wind direction and velocity)