Toward wind farm monitoring optimization: assessment of ecological zones from marine landscapes using machine learning algorithms
Document Type
Article
Date of Original Version
12-12-2015
Abstract
Within the perspective of siting wind farms offshore of Rhode Island, USA, the State and National Environmental Agencies had requested a local marine ecological assessment, which led to an ecological zoning of the area. In view of expanding this zoning outside its limit of the test area and filling gaps in ecological zones, an effort to model those ecological zones using marine landscape or abiotic features was carried out. This study tests the accuracy of selected machine learning algorithmic models, decision tree, and random forest, for relating marine landscapes features to ecological sub-regions. Both models show to be good predictive tools with accuracy after cross validation of the order of 5–3%. Key abiotic variables to provide an accurate model were investigated. The study demonstrates the importance of the distance to coast, the sediment characteristics (fraction of clay, median size of the sediments), the hydrodynamic features, in particular not only tidal current/drag force, but also wave drag force, and finally the oceanographic characteristics such as stratification and sea surface temperature to built a good predictive model. Those findings provide some insight on the pre-monitoring effort optimization.
Publication Title, e.g., Journal
Hydrobiologia
Volume
756
Issue
1
Citation/Publisher Attribution
Grilli, Annette R., and Emily J. Shumchenia. "Toward wind farm monitoring optimization: assessment of ecological zones from marine landscapes using machine learning algorithms." Hydrobiologia 756, 1 (2015): 117-137. doi: 10.1007/s10750-014-2139-3.