titel der präsentation - sosto.net filefederal department of home affairs fdha federal office of...
TRANSCRIPT
Federal Department of Home Affairs FDHAFederal Office of Meteorology and Climatology MeteoSwiss
Cometa
Fribourg, 17th April 2008
Meteorological Metadata Combination Framework
DIVA Group
University of Fribourg, Switzerland
Master thesis of Lorenzo Clementi
Second presentation
2 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
3 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
4 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Architecture (1/2)
• Goals of the framework• Support for combining meteorological data• Help in developing and validating nowcasting
algorithms
• Use of metadata to describe data types and processing information
• “Nowcast validation and comparison with extrapolation is incomplete” [J.W.Wilson, National Centre for Atmospheric Research, Boulder]
5 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Configuration +Metadatacombination
Architecture (2/2)
IdeaTesting, validation
Exp
erim
ent
Procedure file
Data combination
6 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Key features
• Products metadata hierarchy / inheritance
• Metadata: XML to Java mapping thanks to JAXB
• Parameter perturbation
Experiment1Experiment2 …ExperimentN
Metadata 1
Metadata 2
Combinationmethod
Param
eter P
erturbation
7 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
8 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Use case: overview
• Precipitation field extrapolation based on satellite displacement vectors in the Alpine region
• Goals• Illustrate the contribution of the Cometa framework• First assessment of forecast reliability• First optimization experiments
9 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Use case: some details
• Atmospheric motion field (vector field) is computed on satellite images (external software) and then applied onto radar images to generate a forecasted image
• Satellite channel: IR 10.8 μm (cloud top temperature)
• Radar products• OMC (reflectivity)• PJC (precipitation estimate)
• Note that other satellite and radar products over different resolutions could be used (metadata extensibility)
10 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Evaluation: visual inspection
• Residual vectors: AMF between forecasted and extrapolated images
• Visual inspection• Systematic shift → synchronization, parallax (limb view)
• Speed and direction analysis (RMS)
11 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Evaluation: quality flag
Difference between the extrapolated image and the measured one.
q = Ngood / (Ngood + Nbad)
• Simple approach
• Improvements:• CRA• Root mean square error
tolerance
q=Mean f x ,y f x , y
2
12 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
13 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
14 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
OMC versus PJC
Tolerance ± 1Mean difference: 0.08
July 18, 2005
00.10.20.30.40.50.60.70.80.9
1
11.3
0 h
11.4
5 h
12.0
0 h
12.1
5 h
12.3
0 h
12.4
5 h
13.0
0 h
13.1
5 h
13.3
0 h
13.4
5 h
14.0
0 h
14.1
5 h
14.3
0 h
time
qu
alit
y fl
ag
OMC
PJC
15 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Pattern size perturbation
Pattern = window used for the pattern matching algorithm
• Pattern of size 60 to 80 pixels, best result
OMC extrapolation, July 12, 20065 minutes step, tolerance +/- 1
0.7
0.75
0.8
0.85
0.9
0 20 40 60 80 100 120 140 160
pattern size (pixels)
ex
tra
po
lati
on
qu
alit
y
16 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Pattern size perturbation
• Pattern size: cost - benefit
• The AMF algorithm uses a pattern matching technique
• Computational complexity:
O(m2 ∙ n)
m: pattern size n: number of grid points
• Outside optimum: • Big effort (computational resources)
but small benefit (quality increase)
17 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Prediction reliability over time
Strong relation: R2 > 0.9
March 8, 2006 (11:00 - 14:45 UTC)
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
5 25 45 65 85 105
extrapolation time [minutes]
qu
alit
y quality
Logarithm
18 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Presentation outline
• Framework architecture: reminder
• Use case description• Overview• Forecast evaluation
• Demo
• Result discussion
• Conclusion
19 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Outlook
Limitations• Quality flag needs to be improved• The creation of metadata files has to be done manually
Future works• Simultaneous perturbation of several parameters
• Assess correlations• Combinatorial problem
• Machine learning approach
20 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Timeline
November December January February March April
November 5
Project start
December 14
1st milestone- Literature study- Implementation choices- First prototype
February 8
2nd milestone- End to end use case
April 4
Final deadline- Project delivery (source code +thesis)
April 17
Final presentation- in Fribourg
21 MEMcaf | MEteorological Metadata CombinAtion FrameworkLorenzo Clementi
Conclusion
• The use case proved the validity of Cometa• We have found some interesting result• There is place for improvement
Thank you!
Questions?