Post on 30-Dec-2015
description
transcript
Automatic Detection, Automatic Detection, Segmentation and Segmentation and Visualization of Mesoscale Visualization of Mesoscale Eddies in MICOMEddies in MICOM
Bin J. JohnBin J. John
CIS Dept, UMass DartmouthCIS Dept, UMass Dartmouth
Outline of talkOutline of talk Mesoscale EddiesMesoscale Eddies MICOM DatasetMICOM Dataset Detection of EddiesDetection of Eddies Segmentation of EddiesSegmentation of Eddies Visualization of extracted eddiesVisualization of extracted eddies Conclusion & Future WorkConclusion & Future Work
Mesoscale EddiesMesoscale Eddies What are mesoscale Eddies?What are mesoscale Eddies? Why are they important?Why are they important? Why an automatic detection of Why an automatic detection of
eddies?eddies?
What are mesoscale What are mesoscale eddieseddies
Mesoscale: Size scale of about Mesoscale: Size scale of about 100km100km
Eddies are energetic swirling Eddies are energetic swirling currents found all over the oceancurrents found all over the ocean
Mesoscale eddies have deep roots Mesoscale eddies have deep roots of about 5km of about 5km
They can transport momentum to They can transport momentum to the sea floorthe sea floor
Why are mesoscale eddies Why are mesoscale eddies important?important?
Poleward Heat transportPoleward Heat transport• Heat moved by the from the equator Heat moved by the from the equator
towards polestowards poles• An important quantity in Climate An important quantity in Climate
StudiesStudies• Oceans are partly responsible for Oceans are partly responsible for
poleward heat transportpoleward heat transport• Contribution of mesoscale eddies to Contribution of mesoscale eddies to
heat transportheat transport
Why are mesoscale eddies Why are mesoscale eddies important?important?
Biological communities of the oceanBiological communities of the ocean• Strongly affected by ocean circulation Strongly affected by ocean circulation
and mesoscale eddiesand mesoscale eddies• Nutrient rich cold water at the bottom Nutrient rich cold water at the bottom • Cold cyclonic eddies can lift this nutrient Cold cyclonic eddies can lift this nutrient
waters towards surfacewaters towards surface• Sunlight on surface helps nutrient rich Sunlight on surface helps nutrient rich
water promote chlorophyll rich plankton, water promote chlorophyll rich plankton, and thereby the food chain from and thereby the food chain from plankton to whalesplankton to whales
Why an automatic Why an automatic detection of eddies?detection of eddies?
A large number of eddies present in A large number of eddies present in the oceanthe ocean
Computation of contribution of Computation of contribution of mesoscale eddies to poleward heat mesoscale eddies to poleward heat transporttransport
Enable motion tracking of eddies to Enable motion tracking of eddies to study it’s trajectorystudy it’s trajectory
Visualize the different kinds of Visualize the different kinds of eddies present all over the oceaneddies present all over the ocean
MICOM DatasetMICOM Dataset Miami Isopycnic Coordinate Ocean Miami Isopycnic Coordinate Ocean
ModelModel Available ParametersAvailable Parameters Size of this DatasetSize of this Dataset Data ProfileData Profile
Miami Isopycnic Miami Isopycnic Coordinate Ocean ModelCoordinate Ocean Model
Is the dataset used in this projectIs the dataset used in this project Is a numeric 3D model of the ocean Is a numeric 3D model of the ocean
where vertical coordinate is densitywhere vertical coordinate is density Is where the ocean is divided into 11 Is where the ocean is divided into 11
layers, each of which maintains its layers, each of which maintains its own density -- hence the term own density -- hence the term isopycnic, meaning constant densityisopycnic, meaning constant density
Available parametersAvailable parameters
MICOM provides Temperature, MICOM provides Temperature, Velocity, Salinity information of the Velocity, Salinity information of the oceanocean
Information available in a spatial Information available in a spatial resolution of 1/12resolution of 1/12thth degree degree
Temporal resolution of every 3 Temporal resolution of every 3 days, for a whole yeardays, for a whole year
Entire Atlantic ocean in the northern Entire Atlantic ocean in the northern hemisphere consideredhemisphere considered
Size of this datasetSize of this dataset
Snapshots of the ocean, is available for Snapshots of the ocean, is available for 120 different days at 11 different layers120 different days at 11 different layers
That’s over a billion points in the ocean That’s over a billion points in the ocean for which Temperature, Velocity, Salinity for which Temperature, Velocity, Salinity information is availableinformation is available
Today we have massive datasets of Today we have massive datasets of information, powerful computers, and information, powerful computers, and advanced computing techniquesadvanced computing techniques
Utilizing these fully can unravel insights Utilizing these fully can unravel insights about the world in a way not possible about the world in a way not possible before before
Data ProfileData Profile
Histograms give a graphical overview Histograms give a graphical overview of data distribution, by plotting the of data distribution, by plotting the number of occurences of a range of number of occurences of a range of value value
With multiple histograms presented With multiple histograms presented here, a quick overview of the here, a quick overview of the distribution for different layers of the distribution for different layers of the oceans at different times can be oceans at different times can be visualizedvisualized
Temperature-copy2.gif
Detection of EddiesDetection of Eddies Visualizing EddiesVisualizing Eddies Pattern recognitionPattern recognition Structuring Element for Finding Structuring Element for Finding
EddiesEddies ResultsResults
Visualizing eddiesVisualizing eddies
What you see is what you could What you see is what you could possibly detectpossibly detect
Need to see the feature Need to see the feature prominently prominently
Need a view where a pattern is Need a view where a pattern is perceived before an automatic perceived before an automatic detection can be done based on detection can be done based on this patternthis pattern
Structuring element for Structuring element for automatic detection of automatic detection of eddieseddies
Results of automatic Results of automatic detectiondetection
Results Results
Data Number of eddies examined
False Positives
Accuracy Visualization Used
Day 27 Layer 4 40 0 100% Quiver Plots
Day 12 Layer 8 39 1 97.50% Quiver Plots
Day 15 Layer 8 18 1 94.44% Quiver Plots
Day 24 Layer 4 1059 187 82.44% Cone Plots, Streamline
Day 33 Layer 8 575 77 86.66% Streamline
Segmentation of Segmentation of EddiesEddies
Donut Structuring ElementDonut Structuring Element Statistical Border DetectionStatistical Border Detection Segmented EddiesSegmented Eddies Circle vs Ellipse MethodCircle vs Ellipse Method ResultsResults
Donut Structuring elementDonut Structuring element
Three consecutive Three consecutive circlescircles
More number of More number of sample points sample points than a single than a single circle/ellipse alonecircle/ellipse alone
Donut of radius 11 Donut of radius 11 would sample 240 would sample 240 pointspoints
StatisticalBorderDetection
Segmented EddiesSegmented Eddies
Segmented EddiesSegmented Eddies
Segmented EddiesSegmented Eddies
ResultsResults
Number of eddies Number of eddies verified manually: verified manually: 120120
Number of eddies Number of eddies that were mis-that were mis-segmented by an segmented by an area > 15% = 20area > 15% = 20
Accuracy of Accuracy of segmentation segmentation technique: 83.3%technique: 83.3%
Circle vs EllipseCircle vs Ellipse
Visualizations of Visualizations of eddieseddies
Enhanced Pseudocolor plotsEnhanced Pseudocolor plots Movies of eddies in motionMovies of eddies in motion Multi-Layer Streamline plotsMulti-Layer Streamline plots 3D Cone plots3D Cone plots
Pseudocolor Log velocity Pseudocolor Log velocity plotplot
Multi-layer Streamline PlotMulti-layer Streamline Plot
010
20
3040
0
20
40
601
1.5
2
2.5
3
Lattitude grid pointsLongitude grid points
Oce
an L
ayer
#
3D Cone plot3D Cone plot
Conclusions and Future Conclusions and Future WorkWork Fairly Accurate Pattern RecognitionFairly Accurate Pattern Recognition Able to estimate poleward heat Able to estimate poleward heat
transport of eddies from the results of transport of eddies from the results of this projectthis project
Build eddy tracking algorithmBuild eddy tracking algorithm Develop pattern recognition algorithm Develop pattern recognition algorithm
for temperature only databasesfor temperature only databases Geometrically Deformable Model can be Geometrically Deformable Model can be
used for more accurate segmentationused for more accurate segmentation