Species Distribution Modeling with Scalability: The Case Study of P-GARP, a Parallel Genetic Algorithm for Rule-set Production

Author(s)

Santana, Fabiana ; Bravo Pariente, Cesar Alberto ; Saraiva, Antonio Mauro ; Zhang, C ; Palanisamy, B ; Khan, L ; Sarvestani, SS

• Document type: Journal article
• Source: 2017 IEEE 18TH INTERNATIONAL CONFERENCE ON INFORMATION REUSE AND INTEGRATION (IEEE IRI 2017); v. N/A, p. 9-pg., 2017-01-01.

Abstract

Species distribution modeling (SDM) calculates a species' probabilistic distribution by combining Environmental raster layers with species datasets. Such models can help to answer complex questions in Ecology/Biology/Health, e.g., by calculating impacts of climate changes in Biodiversity, or the potential for a disease spread (vectors' modeling). Machine learning is largely applied in SDM, being the Genetic Algorithm for Rule-set Production (GARP) one of the most reliable solutions. However, GARP's convergence needs to speedup under certain conditions (high resolution or number of layers), for which this paper proposes P-GARP, a parallel, scalable implementation of GARP. P-GARP was implemented onto a SGI Altix XE 1300 cluster with 2 quad-core processors/node. Preliminary results show an expressive 3.2/node speedup. Premature convergence is not observed in PGARP and its accuracy is very similar to GARP ' s. Effective solutions to improve this speedup in even larger scale are proposed, along with a discussion about P-GARP correctness and efficiency. (AU)

• FAPESP's process: 04/11012-0 - Computer environment for modeling of spice distribution
• Grantee: Vanderlei Perez Canhos
• Support Opportunities: Research Projects - Thematic Grants