Visualization Technique Research On The Time-varying Weather Volume Dataset

Weather data visualization is an important direction in the development of visualization technique. In this paper, we carry out some research work which focuses on the time-varying hurricane dataset. The hurricane itself involves lots of factors and it can evolve as the time continues. Thus the hurricane datasets that generated from measure or simulation are usually kinds of time-varying and multi-variables datasets. Some familiar variables include temperature, pressure, wind, moisture, cloud, precipitation and so on. Weather data visualization technique can greatly help the meteorologists to analyze the datasets and make decisions quickly. However, the time-varying and multi-variables characters of the weather datasets present some new challenges to the visualization technique. First, the system must have the capabilities to analyze the dynamic process in the datasets. Second, it must be able to capture all the variables' information as well as the correlation between them.Our contribution in this paper is the construction of a visualization system which can be used to render and analyze the hurricane datasets. The first problem we faced is how to deal with the huge memory cost that introduced by the hurricane datasets. In this paper, we introduce an effective compression algorithm which is suitable for the time-varying and multi-variable volumes. And our algorithm is specially designed for the floating-point data which can preserve higher precision compared with the fixing-point ones. There are two stages to process the temporal and spatial coherence of the volume respectively in the algorithm. These two stages can be divided into more sub-stages, which include data format conversion, spatial octree construction, spatial prediction, encoding and the "LZO" compression. In the decompression side, we can decompress the data quickly through the use of GPU processing, so that the rendering efficiency can achieve the interactive or even real-time level.Based on the system mentioned above, we implement some visual analysis on the hurricane datasets using an importance-driven method. In this method, first, we calculate each block's "importance" in the volume based on the "entropy" which is a concept in the information theory. Then, we apply the k-means clustering algorithm to these "importance"s and realize the classification visualization of the hurricane datasets through the users'interaction. Some experiments are executed on the Hurricane Isabel datasets, and the results show our methods are effective.