Date of Award

Fall 2025

Project Type

Dissertation

Program or Major

Biological Sciences

Degree Name

Doctor of Philosophy

First Advisor

Jennifer A. Dijkstra

Second Advisor

Christopher Neefus

Third Advisor

David Plachetzki

Abstract

Kelp habitats in the Gulf of Maine are threatened by rising ocean temperatures and marine invasions. The replacement of kelps with red turf macroalgae fundamentally alters these habitats and disrupts the ecosystem services they provide. Thorough assessment and monitoring of kelp habitats is needed for proper conservation of these habitats, but most long-term monitoring is performed using traditional survey methods that are restricted in the amount of space they can cover. Habitat modelling approaches are required to expand our knowledge of Gulf of Maine kelp habitats, and to better understand the interactions between kelp and red turf habitats across the region. The goal of this dissertation is to develop and assess habitat modelling methods that will improve our monitoring capabilities. In Chapter II, we developed a Random Forest (RF) method that uses two sources of remotely sensed data, sea surface temperature (SST) variables derived from Landsat 8 imagery and variables derived from high-resolution bathymetry, to classify benthic habitats at the Isles of Shoals, an island group in the Gulf of Maine. The models classify four types of benthic habitat: bare substrate, kelp, mixed macroalgae, and red turf habitats, with 71% accuracy. Bathymetric variables including depth, rugosity (vector ruggedness measure), and slope, and median and standard deviation of SST variables, are all highly important in modelling these habitats. In Chapter III, we explored the RF model predictions further. We find that kelp and red turf habitats are strongly separated by depth; in particular, predictions of red turf habitat essentially cease below 16 m. We find that vector ruggedness measure and median SST are highly important in identifying kelp habitats, slope is highly important in identifying bare substrate, and standard deviation of SST is highly important in identifying mixed macroalgae habitats. We also identify a pattern of depth zonation between two kelp species, shallower Saccharina latissima and deeper Agarum clathratum, the latter of which may be better sheltered against rising temperatures and marine invasions. We highlight the importance of depth-stratified sampling and consideration of all kelp species when assessing kelp habitat health in a region. In Chapter IV, we developed species distribution models for six macroalgae using Maxent approaches trained on existing species occurrence datasets. Despite extensive effort to clean and correct erroneous records in these datasets, and to correct spatial bias among the records, the Maxent models are highly inconsistent. While the models are highly accurate in terms of area-under-curve statistics, changes to the extent of the study area and the resolution of the variables greatly affect the predicted overlap between species and the overall spatial distribution of habitat suitability predictions; correction of spatial bias was also largely ineffective as most models seem to recapitulate sampling effort. The predictions of relative habitat suitability made by the Maxent models around the Isles of Shoals do not align with the predictions of benthic habitat made by the RF models. Species occurrence datasets cannot replace proper sampling in the context of monitoring Gulf of Maine macroalgae habitats.

Download

Share

COinS