Extracting storm-centric characteristics from raw rainfall data for storm analysis and mining

Abstract

Most rainfall data is stored in formats that are not easy to analyze and mine. In these formats, the amount of data is enormous. In this paper, we propose techniques to summarize the raw rainfall data into a model that facilitates storm analysis and mining, and reduces the data size. The result is to convert raw rainfall data into meaningful storm-centric data, which is then stored in a relational database for easy analysis and mining. The size of the storm data is less than 1% of the size of the raw data. We can determine the spatio-temporal characteristics of a storm, such as how big a storm is, how many sites are covered, and what is its overall depth (precipitation) and duration. We present formal definitions for the storm-related concepts that are needed in our data conversion. Then we describe storm identification algorithms based on these concepts. Our storm identification algorithms analyze precipitation values of adjacent sites within the period of time that covers the whole storm and combines them together to identify the overall storm characteristics.