150 likes | 282 Views
Meteorological data, evaluation results, defaults and advantages of the different motion estimators. André Szantai, P. Lopes, M. Desbois, A. Lahellec, G. Sèze Laboratoire de Météorologie Dynamique Palaiseau, France. Plan. 1) New data : AMVs from EUMETSAT, Trajectories (LMD method).
E N D
Meteorological data, evaluation results, defaults and advantages of the different motion estimators André Szantai, P. Lopes, M. Desbois, A. Lahellec, G. Sèze Laboratoire de Météorologie Dynamique Palaiseau, France FLUID Meeting Las Palmas (8-9 Dec. 2005)
Plan • 1) New data : • AMVs from EUMETSAT, • Trajectories (LMD method). • 2) Comparison of different motion estimators : • Qualitative comparisons • 3) Prospects • Quantitative results for comparisons - statistics. • Cloud classification (LMD method) • ECMWF analysed winds. FLUID Meeting Las Palmas (8-9 Dec. 2005)
Atmospheric motion vectors (AMV) • EUMETSAT product, available every 3 hours (2:45, 5:45, 8:45UTC…) • 1 file per instant with all channels (currently 3) : • IR 10.8, • VIS 0.8 during daytime, • WV 6.2 cloud + WV 6.2 non cloudy areas. • Parameters : • Position (longitude, latitude), • Wind speed and direction, • Pressure and corrected temperature. FLUID Meeting Las Palmas (8-9 Dec. 2005)
Main differences between LMD motion vectors and EUMETSAT AMVs • AMVs repositioned on clouds (from original regular grid) ==> irregular grid. • Calculation of a quality coefficient (QI). • Pressure level derived mainly from satellite data (corrected brightness temperatures). • In some cases, 2 vectors can be found on the same grid point and channel (with different quality coefficients). FLUID Meeting Las Palmas (8-9 Dec. 2005)
LMD motion vectors vs. EUMETSAT AMVs (VIS 0.8) • (low) 1000 - 925 - 850 - 700 - 500 - 400 - 300 - 250 - 200 hPa (high) • Best fit level (LMD) EUMETSAT pressure level (QI > 0.5) FLUID Meeting Las Palmas (8-9 Dec. 2005)
High-level cloud structures Quality indicator QI > 0.5 (both images) Pure water vapour structures. Better tracking than in LMD fields (esp. dry areas) Motion at 2 levels----. AMVs in the WV 6.2 channel (mid- and high-level structures) FLUID Meeting Las Palmas (8-9 Dec. 2005)
QI = 0.5 More inconsistent vectors QI = 0.8 Less dense vector groups Influence of the quality indicator (QI) of AMVsIR 10.8 image FLUID Meeting Las Palmas (8-9 Dec. 2005)
Cloud and atmospheric structure trajectories • Calculation method developed at LMD : • Extension of cloud motion vector calculation method (block matching, minimization of the sum-of-squared differences). • Similar quality tests applied along trajectories • Consistency tests detect the end of trajectories (dissipation…) Variable number of vectors. FLUID Meeting Las Palmas (8-9 Dec. 2005)
IR 10.8 : Long living low level cloud trajectories, Medium-level trajectories (red, orange), High-level cloud trajectories (green, blue) Maximal duration : 13 h 45. IR 10.8 cloud trajectories FLUID Meeting Las Palmas (8-9 Dec. 2005)
WV 6.2 Numerous high-level cloud trajectories, Westward motion in bright (moist) pure WV areas Few trajectories in very dry areas (bottom) WV 7.3 More trajectories at medium-level, Different motion (medium-level) at the edge of the pure WV structure (lower part). WV 6.2 and WV 7.3 trajectories FLUID Meeting Las Palmas (8-9 Dec. 2005)
Qualitative comparisons Examined cases : vector fields from ULPGC Comparison of different motion estimators FLUID Meeting Las Palmas (8-9 Dec. 2005)
LMD motion vectors More vectors in areas with complex motion ULPGC optical flow - correlation method Vector field comparisons (VIS 0.8, NAtl, image 48) FLUID Meeting Las Palmas (8-9 Dec. 2005)
Closest fields with LMD motion vector fields (visually): 1) OF correlation vector field, 2) OF PDE vector field (smoother), 3) OF Structure Tensor vector field. But statistics indicate : Larger differences for OF correlation vector fields than for PDE fields. Planned : Best-fit level for all fields Vectors expected at lower altitude : bias > 0 ==> (stronger winds at higher altitude) Remarks about the ULPGC vector fields(NAtl, VIS 0.8) FLUID Meeting Las Palmas (8-9 Dec. 2005)
Conclusions and prospects • Systematic comparisons (all motion vector fields) : • Qualitative comparisons • Quantitative results - statistical parameters : • RMS vector difference, bias, NRMSVD, angular error. • With / without use of a cloud classification-based selection. • Use of complementary data : • New cloud classification (LMD dynamical clustering-based) method. • ECMWF analysed winds. FLUID Meeting Las Palmas (8-9 Dec. 2005)
Thank you ! FLUID Meeting Las Palmas (8-9 Dec. 2005)