1. Extracting Dense Regions From Hurricane Trajectory Data
Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Department of Computer Science and Engineering
University of Texas at Arlington
First International ACM Workshop on  Managing and Mining Enriched Geo-spatial Data
GeoRich 2014
Praveen Kumar Tripathi
Madhuri Debnath
Ramez Elmasri

2. Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Contents:
Spatio temporal data:
Examples
Applications
Data mining tasks
Extracting Dense Regions from Hurricane Trajectory data:
Proposed concept
DBSCAN: core concepts
Experiments and Analysis: spatial, and spatial with non spatial.
Temporal extension for the analysis:
Temporal DBSCAN
Proposed relative time attribute
Experimental analysis of spatio temporal clustering
11/13/2018

3. Spatio temporal data: Hurricane JIG 1950
Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Spatio temporal data: Hurricane JIG 1950
Has spatial and temporal attributes as key attributes. Example :
'ADV'
'Year'
'Name'
'LAT'
'LON'
'TIME'
'WIND'
'PR'
'STAT' 1
1950
'HurricaneJIG'
24.3
-47.2
'10/11/12Z' 40
'-'
'TROPICAL' 2
24.2
-48.2
'10/11/18Z' 3
-49.2
'10/12/00Z' 45 4
-50.2
'10/12/06Z' 5
-51.2
'10/12/12Z' 50 6
24.4
-52.4
'10/12/18Z' 55 7
24.7
-53.7
'10/13/00Z' 8 25
-54.6
'10/13/06Z' 60 9
25.4
-55.5
'10/13/12Z' 65
'HURRICANE-1' 10
26.1
-56.7
'10/13/18Z' 70 11
26.9
-57.8
'10/14/00Z' 75 12
27.6
-58.6
'10/14/06Z' 13
28.4
-59.3
'10/14/12Z' 80 14
29.5
-60
'10/14/18Z' 15
30.8
-60.5
'10/15/00Z' 85
'HURRICANE-2' 16 32
-60.2
'10/15/06Z' 90 17
33.2
-59.2
'10/15/12Z'
100
'HURRICANE-3' 18
34.2
-58.2
'10/15/18Z'
105 19
35.1
-57.2
'10/16/00Z' 20
35.9
-56.2
'10/16/06Z' 95 21
36.8
-55
'10/16/12Z' 22
38.8
-51.5
'10/16/18Z' 23
40.8
-47.1
'10/17/00Z' 24
41.9
-44.5
'10/17/06Z' 43
-42
'10/17/12Z'
'EXTRATROPICAL' 26
44.1
-39.9
'10/17/18Z'
11/13/2018

4. Hurricane data Extracting Dense Regions From Hurricane Trajectory Data
11/13/2018

5. Dataset* Attributes : ADV : Tracking ID LAT : Position in latitude
Extracting Dense Regions From Hurricane Trajectory Data
Dataset*
Attributes :
ADV : Tracking ID
LAT : Position in latitude
LON : Position in longitude
Time : Track time point
WIND : Winds in knots ( 1 knot = 1.15 mph)
PR : Pressure in millibards
STAT : State category ( Tropical Storm, Hurricane..)
Temporal Coverage : 1950 – 1999 (50 years) (15319 data points, 496 trajectories)
Spatial Coverage : North Atlantic
Interval : 6 hourly (0000,0600,1200,2400)
Data format : txt file *
11/13/2018

6. Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Spatio temporal data
Example
Global positioning data (GPS).
Cell phone tracking, vehicle traffic data, emergency vehicle data, transportation data
Hurricane and storm tracking data.
Animal movement data.
Environmental data:
Land use data.
Demographic : population data.
Soil moisture data.
Temperature data.
11/13/2018

7. Spatio Temporal data: contd..
Extracting Dense Regions From Hurricane Trajectory Data
Spatio Temporal data: contd..
Applications:
Managing vehicle traffic patterns.
Monitoring and predicting weather conditions.
Examining wild animal behavior.
Analyzing spread of diseases.
Monitoring land use, climate change.
Data Mining Tasks:
Clustering.
Flocking behavior mining.
Convoy detection.
Classification.
11/13/2018

8. Extracting Dense Regions from Hurricane Trajectory data
The data set has been considered as point data set.
Density Based Spatial Clustering of Application with Noise (DBSCAN) [1] has been used.
DBSCAN has been used to identify the dense regions of hurricane activity viz., hot spots.
Clustering has been done using only the spatial attributes (latitude and longitude), as well as spatial along with non spatial attribute (latitude, longitude and wind speed) for comparative analysis.
Clustering results have been evaluated using clustering evaluation measures.
[1] Martin Ester , Hans-peter Kriegel , Jörg S , Xiaowei Xu, A density-based algorithm for discovering clusters in large spatial databases with noise , Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96), pp. 226
11/13/2018

9. Extracting Dense Regions From Hurricane Trajectory Data
Contd..
The post processing of the obtained clustering results have been done to obtain:
Regions from where the storms are most likely to originate.
Regions where the storms are more likely to land.
Key regions through which the storms are more likely to traverse.
11/13/2018

10. Extracting Dense Regions From Hurricane Trajectory Data
DBSCAN: Core Concepts
Cluster Definition: A cluster 𝐶 is a subset of objects satisfying the following:
Connected: ∀𝑝, 𝑞∈𝐶, 𝑝 𝑎𝑛𝑑 𝑞 are density connected
Maximal: ∀𝑝, 𝑞, 𝑖𝑓 𝑝∈𝐶, 𝑎𝑛𝑑 𝑞 is density reachable from 𝑝, 𝑡ℎ𝑒𝑛 𝑞∈𝐶.
11/13/2018

11. Experiments and Analysis : Spatial analysis
Extracting Dense Regions From Hurricane Trajectory Data
Experiments and Analysis : Spatial analysis
11/13/2018

12. Experiments and Analysis(1): Spatial and non spatial analysis
Extracting Dense Regions From Hurricane Trajectory Data
Experiments and Analysis(1): Spatial and non spatial analysis
11/13/2018

13. [2] Quality measure = Two distances used for two quality measures :
Extracting Dense Regions From Hurricane Trajectory Data
[2] Quality measure =
Two distances used for two quality measures :
Spatial distance (latitude and longitude)
Non-spatial distance (wind speed).
11/13/2018

14. Extracting Dense Regions From Hurricane Trajectory Data
11/13/2018

15. Temporal extension to DBSCAN
Extracting Dense Regions From Hurricane Trajectory Data
Temporal extension to DBSCAN
The spatio temporal data has time as another key attribute.
DBSCAN needs to be modified to incorporate the time domain.
In the temporal analysis possibilities:
Absolute temporal analysis (vehicle traffic analysis)
Periodic temporal analysis (repeating everyday)
Relative temporal analysis (hurricane data analysis)
Present analysis uses the relative temporal analysis:
Hurricanes do not occur at the same time. (absolute temporal analysis not possible).
11/13/2018

16. Relative temporal attribute
Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Relative temporal attribute
Let there be two trajectories of length 𝑚 𝑎𝑛𝑑 𝑙 where, 𝑙 𝑎𝑛𝑑 𝑚 may be different:
Absolute relative temporal attribute.
𝑇𝑟 1 =[< 𝑥 11 , 𝑦 11 , 1>, < 𝑥 12 , 𝑦 12 , 2>… 𝑥 1𝑙 , 𝑦 1𝑙 , 𝑙]
𝑇𝑟 2 =[< 𝑥 21 , 𝑦 21 , 1>, < 𝑥 22 , 𝑦 22 , 2>… 𝑥 2𝑚 , 𝑦 2𝑚 , 𝑚]
Length normalized relative temporal attribute.
𝑇𝑟 1 =[< 𝑥 11 , 𝑦 11 , 0>, < 𝑥 12 , 𝑦 12 , 1 𝑙−1 >… 𝑥 1𝑙 , 𝑦 1𝑙 , 1 ]
𝑇𝑟 2 =[< 𝑥 21 , 𝑦 21 , 0>, < 𝑥 22 , 𝑦 22 , 1 𝑚−1 > … 𝑥 2𝑚 , 𝑦 2𝑚 , 1]
11/13/2018

17. Extracting Dense Regions From Hurricane Trajectory Data
Extending DBSCAN
In [2], DBSCAN extended to incorporate additional neighborhood criteria:
Let 𝐷= 𝑑 1 , 𝑑 2 ,… 𝑑 𝑛 be the data set where 𝑑 𝑖 = 𝑥 𝑖 , 𝑦 𝑖 , 𝑎 𝑖 , 𝑏 𝑖 .
𝑑𝑖𝑠 𝑡 𝑠 = (𝑥 1 − 𝑥 2 ) 2 + (𝑦 1 − 𝑦 2 ) spatial attributes ( 𝑥 𝑖 , 𝑦 𝑖 )
𝑑𝑖𝑠 𝑡 𝑛𝑠 = (𝑎 1 − 𝑎 2 ) 2 + (𝑏 1 − 𝑏 2 ) non spatial attributes ( 𝑎 𝑖 , 𝑏 𝑖 )
Let there are two thresholds 𝜖 1 and 𝜖 2 given by the user for the spatial and non-spatial neighborhood…………………………(1)
Let there be now two neighborhood (spatial and non-spatial):
𝑁 𝜖 1 𝑑 𝑘 = 𝑑 𝑗 ∈𝐷 𝑑𝑖𝑠 𝑡 𝑠 𝑑 𝑘 , 𝑑 𝑗 ≤ 𝜖 1 }
𝑁 𝜖 2 𝑑 𝑘 = 𝑑 𝑗 ∈𝐷 𝑑𝑖𝑠 𝑡 𝑛𝑠 𝑑 𝑘 , 𝑑 𝑗 ≤ 𝜖 2 } then the final common neighborhood will be:
𝑁=𝑁 𝜖 1 𝑑 𝑘 ∩ 𝑁 𝜖 2 𝑑 𝑘 …………………………………(2)
[2] Birant, D., and Kut, A. ST-DBSCAN: An algorithm for clustering spatial-temporal data. Data Knowledge Engineering , 2007 pp
11/13/2018

18. Extracting Dense Regions From Hurricane Trajectory Data
Temporal extension
We use the formulation in (2), where the thresholds 𝜖 1 is used for spatial neighborhood and 𝜖 2 for the temporal neighborhood.
By incorporating the spatial as well as temporal neighborhood in the basic DBSCAN algorithm we now have the spatio-temporal version of the DBSCAN algorithm.
11/13/2018

19. Spatio temporal clustering results
Extrating Dense Regions From Hurricae Trajectory Data
11/13/2018
Extracting Dense Regions From Hurricane Trajectory Data
Spatio temporal clustering results
11/13/2018

20. Extracting Dense Regions From Hurricane Trajectory Data
References:
Martin Ester , Hans-peter Kriegel , Jörg S , Xiaowei Xu, A density-based algorithm for discovering clusters in large spatial databases with noise , Proceedings of the Second International Conference on Knowledge Discovery and Data Mining (KDD-96), pp. 226–231
Birant, D., and Kut, A. ST-DBSCAN: An algorithm for clustering spatial-temporal data. Data Knowledge Engineering , 2007 pp
Gil Lee and Han, J. Trajectory Clustering: A partition-and-group framework. In SIGMOD 2007, pp
Gudmundsson, J, Thom A, Vahrenhold J, Of motifs and goals : mining trajectory data. In Proceedings of the 20th International Conference on Advances in Geographic Information Systems, SIGSPATIAL’
Gil Lee, J Han , X Li and H Gonzalez, Traclass : Trajectory Classification using hierarchical region- based and trajectory-based clustering VLDB 2008.
Vieira M., Bakalov P., and Tsotras V., On-Line Discovery of Flock Patterns in Spatio-Temporal data, Proceedings of 17th ACM SIGSPATIAL International conference, GIS’09, pp
Jeung H., Yiu M., Zhou X. Jensen C., and Shen H., Discovery of convoy in Trajectory databases. PVLDB, 2008.
Kalnis P., Mamoulis, and Bakiras S., On Discovering Moving Clusters in Spatio-Temporal data, In SSTD 2005, pp

21. Thanks ! Extracting Dense Regions From Hurricane Trajectory Data
11/13/2018