Name
Abstract | ||
|---|---|---|
Weather data is a classic example of spatio-temporal data, with time and space as two of its key attributes. Clustering has been one of the key techniques used for analyzing the storm trajectories. Trajectory based clustering algorithms consider whole trajectories as clustering units, or in some cases the segments of the trajectory, i.e., sub-trajectories, are considered in order to capture local similarities among long trajectories. Our work takes a different approach, by considering a trajectory as a set of points, then focusing on the point data for finding the regions that are hot spots for the storms. We use DBSCAN algorithm, and consider spatial (longitude, latitude) as well as non-spatial attributes (viz., wind speed and time) for the similarity measure. The results show the impact of the respective non-spatial attributes on the spatial attributes during clustering and hence the identified dense regions. For the temporal analysis, we used a relative temporal framework by normalizing relative time stamp order in the trajectory by the length of the trajectory to consider storms of different lengths. We use quality measures to validate our clusters. Post processing on the obtained clusters identifies the regions from where the storms are more likely to originate, and the regions where the storms are most likely to land. Another useful result is the key regions that the storms are most likely to traverse. |
Year | DOI | Venue |
|---|---|---|
2014 | 10.1145/2619112.2619117 | GeoRich@SIGMOD |
Keywords | Field | DocType |
interface | Longitude,Data mining,Wind speed,Similarity measure,Hotspot (geology),Computer science,Cluster analysis,Trajectory,DBSCAN,Traverse | Conference |
Citations | PageRank | References |
1 | 0.36 | 11 |
Authors | ||
3 | ||
Name | Order | Citations | PageRank |
|---|---|---|---|
| Praveen Kumar Tripathi | 1 | 179 | 11.83 |
| Madhuri Debnath | 2 | 11 | 2.74 |
| Ramez Elmasri | 3 | 1 | 0.36 |