Unsupervised phytoplankton community detection and analysis of environmental and satellite parameters on each community

Marine dynamics largely affects the phytoplankton community com- position. The distribution characteristics of phytoplankton can reflect spatio-temporal variability in the marine ecosystem, on the other way around. In this work, we study the relation between remote sensing satellite observations, en...

Descripción completa

Detalles Bibliográficos
Autor: Fathollahi, Mohana|||0009-0009-5015-743X
Tipo de recurso: tesis de maestría
Fecha de publicación:2023
País:España
Institución:Universitat Politècnica de Catalunya (UPC)
Repositorio:UPCommons. Portal del coneixement obert de la UPC
Idioma:inglés
OAI Identifier:oai:upcommons.upc.edu:2117/394183
Acceso en línea:https://hdl.handle.net/2117/394183
Access Level:acceso abierto
Palabra clave:Cluster analysis
Remote sensing
Phytoplankton
clustering
community detection
remote sensing
phytoplankton
Anàlisi de conglomerats
Teledetecció
Fitoplàncton
Àrees temàtiques de la UPC::Enginyeria de la telecomunicació::Radiocomunicació i exploració electromagnètica::Teledetecció
Descripción
Sumario:Marine dynamics largely affects the phytoplankton community com- position. The distribution characteristics of phytoplankton can reflect spatio-temporal variability in the marine ecosystem, on the other way around. In this work, we study the relation between remote sensing satellite observations, environmental factors and phytoplankton com- munities. First, we employ network-based unsupervised clustering approaches to identify representative communities using metabarcoding data that has been collected both across depth and surface. Next, we investigate the relation between the detected phytoplankton communities and the environmental parameters (e.g., temperature, salinity, nutrients and so on). Our results show that phytoplankton communities are segregated based on the depth and basin. Additionally, for communities where the majority of samples are gathered from the Atlantic ocean, the nutrient levels are much higher than other communities. To extend this analysis to other years, a scientific ship should collect water samples in different years. This would be very costly and even infeasible for many applications such as analyzing the seasonal changes in plankton communities. Therefore, in the second part of our work, we utilize the fact that the reflected light from the ocean's surface that is captured by a remote sensing satellite has a specific relationship with the plankton composition. To this end, we first cluster the samples that are collected at the surface of the ocean. Next, we apply several machine learning algorithms to classify these representative communities from satellite data. Our top performing classifier reached 0.94 accuracy in leave-one-out cross vali- dation setting. The results show three top important features in predict- ing communities are surface temperature, chlorophyll and particulate organic carbon.