Spatial Analysis Infrastructure for Tropical Cyclones Observation from Ground-Based Doppler Radar Towards Big Data and Cloud Computing

During tropical cyclones' (TC) landfall, complicated interactions occur among TCs, atmospheric environments and the surfaces in a fast evolving pattern that might lead to complicated rainbands structure in TCs. As NEXRAD radar network is able to observe and measure TCs' process in a real-time way, timely processing and distribution is necessary for rainfall forecasting as well as decision making in emergency situation. Radar data are voluminous and geospatial analyzing algorithm requires large amount of computation due to high spatio-temporal resolution in radar data. However, most Geographical Information System (GIS) software in desktop environments (e.g. ArcGIS) is not able to handle them in a timely manner. To support fast geospatial analysis on tropical cyclones based on NEXRAD observations, in this research, we propose the Infrastructure-as-a-Service (IaaS) computational model, a high-efficient cloud computing model. We implement this computational model as an integrated infrastructure with following data-intensive technology: MapReduce (MR) combined with Hadoop File System (HDFS); Large scale spatial analysis tools using R language, an advanced statistic language, combined with R-Hadoop connector, and a geodatabase backend powered by MySQL spatial extension and MySQL Hadoop Applier. There are three parts in this infrastructure: 1. a high-performance radar merging tool to compose three-dimensional reflectivity maps; 2. a set of geospatial and geostatistic models programmed by R language for developing the geodatabase holds spatio-temporal properties of rain bands in TCs; and 3. a geodatabse backend supporting geospatial query required during reflectivity map composing, models running and other external GIS software. Case study of Hurricane Charley (2004) from Sep. 13 to 14 is carried out to evaluate both system performance and geospatial models on analyzing hurricane rainbands. In further research, we plan to migrate this infrastructure to run Amazon Web Services (AWS) cloud with Amazon Elastic Map Reduce (EMR) to make it available to public domains.