Efficient and Scalable Parallel Zonal Statistics on Large-Scale Species Occurrence Data on GPUs

Analyzing how species are distributed on the Earth has been one of t he fundamental questions in the intersections of environmental sciences, g eosciences and biological sciences. With world-wide data contributions, more than 375 mil lion species occurrence records for nearly 1.5 million species have been deposite d to the Global Biodiversity Information Facility (GBIF) data portal. The she er amounts of point and polygon data and the computation-intensive point-in-polygon tests for zonal st atistics for biodiversity studies have imposed significant technical challenges . In this study, we have significantly extended our previous work on parallel primitives bas ed patial joins on commodity Graphics Processing Units (GPUs) and have developed new ef ficient and scalable techniques to enable parallel zonal statistics on the G BIF data completely on GPUs with limited memory capacity. Experiment results have shown that an impressive end-to-end response time under 100 seconds can be achieved for zonal stat istics on the 375+ million species records over 15+ thousand global eco-regions with 4+ million vertices on a single Nvidia Quadro 6000 GPU device. The achieved high perfor mance, which is several orders of magnitude faster than reference s erial implementations using traditional open source geospatial techniques, not only demonstrates the pot ential f GPU computing for large scale geospatial processing, but also makes i nteractive query driven visual exploration of global biodiversity data possible.