Integration of Remote Sensing Data in Machine Learning to Predict Marine Microbial Community Composition

Abstract

Microbes are the most abundant and diversified living things on Earth. They play important roles in their ecosystems such as marine, soil, and human body. However, little is known on the relationship between microbial communities composition and their surrounding environment. Robust species distribution models are necessary to provide a comprehensive understanding of the patterns of microbial communities and their interactions with the environment. Similar models have only been applied to macro-organisms for the prediction of species distribution across space and time based on different environmental conditions. However, these cannot be used directly to microbial communities as microbes have more complex structure and interactions with their surroundings. We here propose ensemble-based nonlinear models to predict the relative abundance of marine microbes across time by combining compositional data of the communities and environmental data. Both in-situ(ground) and remote-sensing data were used in the thesis work. We have obtained a set of predictive models with an aggregated average performance of 23.93% normalized root mean square error(NRMSE).

Date of Award	May 2015
Original language	American English
Supervisor	Andreas Henschel (Supervisor)