Spatial Similarity And Its Application In Tropical Cyclone Track Forecasting

Spatial similarity technology is build on the basis of similarity theory. Those features are considered similar at a particular scale and context according to the result of calculation and analysis using the techniques such as geo-computation and artificial intelligence.The purpose of this paper is to study the spatial similarity of tropical cyclone tracks and try to forecast them by means of spatial similarity technology. First of all, the method of measuring the spatial similarity of segments has been studied. Taking into account the direction of spatial line segments, the measure computes position distance and direction distance, and neither is indispensible. Several kinds of distance measuring methods and calculation formula are introduced.The similarity of tropical cyclone tracks can be judged by the result of segment clustering. A partition-and-group method is adopted in order to analyse the similar characteristics in direction and position of tropical cyclone track. According to the result of partition step, we can get the complexity of single tropical cyclone track and the set of segments which will be used in the second step; after the group step, we can know the spatial distribution of the sub-tracks of tropical cyclones. In second step, we adopt the clustering algorithm based on density of segments. The reachable-distance and core-distance in OPTICS clustering algorithm which is based on point density is introduced into segment clustering algorithm. In theory we can get all clusters where the cluster radius is smaller than the input value.According to the result of the clustering, we retrieve the similar tracks in spatial database, and forecast the track by average position of similar tracks. The effects of prediction has been tested by the data of 2009a. Since the clustering result is the only criterion in forecasting, we can improve the effect of prediction by adding other criteria such as tensity, wind direction, source of track, etc. In this paper, we have considered the source of tracks in prediction. After adding the criterion, the effect of prediction is improved.