\r\n The Marine Geospatial Ecology Lab (MGEL) of Duke University worked with NOAA’s Fisheries Science Centers, the Cetacean & Sound Mapping Working Group, partners at universities and non-governmental research organizations, and the Navy to create comprehensive cetacean habitat-based density surface models for the US east coast. Models were created for all species sighted at least once during NOAA broad-scale marine mammal abundance surveys of the US east coast conducted since 1992. Depending on the data available for a species and what is known about it, the species was modeled either individually or as part of a species guild.\r\n
\r\n\r\n Models were created by applying distance sampling and density surface modeling methods to visual line transect surveys with sighting data for 30 cetacean species or species guilds, and linking physiographic and oceanographic covariates via Generalized Additive Models (GAMs). The database of line-transect data sources consists of data from multiple organizations, platforms (aerial and ship-based), and time periods (1992 – 2016) spanning the entire US East Coast and into Canadian waters. Oceanographic covariates may be climatological (e.g. mean sea surface temperature at the location of the sighting for an 8-day period averaged over 30 years) or contemporaneous (daily sea surface temperature on the date of the sighting). Models were created using both types of covariates, and the better performing model was selected. Model performance was assessed with diagnostic tools and plots such as the Q-Q plot and explained deviance. A density surface was then predicted from the model at a monthly or yearly temporal resolution. When possible, fitted seasonal models used species-specific season definitions, based on known ecology.\r\n
\r\n\r\n Many trade-offs and decisions were made by MDAT in the creation of the cetacean density models. Density models are complex, involving variables that can be difficult to determine unambiguously, and must account for many factors, including the probability of detecting an animal according to how far it is from the observer, the speed and viewing characteristics of the observation platform, the size of the animal group, the sea state, the presence of sun glare, the availability of the animal at the ocean surface for detection, cryptic behaviors of the species being observed, and, ideally, the biases of individual observers, etc. During many expert review processes, Duke MGEL considered and decided upon these options. A few specific caveats and considerations are highlighted below, as being most relevant to the ocean planning processes and efforts that they are likely to be used in.\r\n
\r\n\r\n Several measures of model uncertainty are provided with each habitat-based density model. For this tool, we decided to visualize the Coefficient of variation (CV). The CV is the ratio of the standard error to the estimated density, and helps inform users about the magnitude of variation in model predictions from one place to another. Values greater than 1, i.e. where the standard error is greater than the density estimate, indicate substantial uncertainty. When high CVs occur where the density estimate is very low, as is often the case, there is little cause for concern. But when high CVs occur where the density estimate is high, it suggests the model cannot predict density well there. We combined all CV layers for all species in a single layer for each season to provide a general view of model uncertainty. However, we recommend referring to the species CV layers available in both the NE Data Portal and the MARCO data portal before making any decisions using these datasets.\r\n
\r\n\r\n The NE and Mid-Atlantic data portals also provide other measures of model uncertainty:\r\n
\r\n\r\n NCCOS employed a statistical modeling framework that relates avian density to environmental predictor variables Spatial predictive modeling was applied to the survey data to account for spatial and temporal heterogeneity in survey effort, platform, and protocol. An ensemble machine-learning technique, component-wise boosting of hierarchical zero-inflated count models, was used to relate the relative density of each species to multiple spatial and temporal predictor variables while accounting for survey heterogeneity and the aggregated nature of sightings. Seasonal climatologies of dynamic spatial environmental predictors were used (i.e., a climatological habitat modeling approach).\r\n
\r\n\r\n These models incorporate virtually all known science-quality at-sea seabird surveys from 1978- 2016, including all AMAPPS and USFWS aerial and boat surveys, BRI’s Mid-Atlantic Baseline surveys, and recent surveys conducted by states, BOEM, and wind energy companies to inform energy siting off Rhode Island, Massachusetts, Maine, and elsewhere in the Northeast and Mid-Atlantic\r\n
\r\n\r\n It is important to recognize that the model predictions do not represent absolute density, rather they are indices of density. This is because during visual surveys individual birds may be missed and animal movement can bias estimates of abundance, and probabilities of detection are unknown. Avian relative abundance predictive maps may inform users in answering the question “relative to other areas, how many more of species X are there likely to be in this area?”\r\n
\r\n\r\n Masks showing areas with no survey effort are provided to aid the user in understanding where caution should be used when interpreting model results. Model predictions in areas with no survey effort should be interpreted cautiously. Individual model performance statistics are included, and should be referenced when individual layers are used in agency decisions (see below).\r\n
\r\nFor each season, NCCOS created survey coverage masks that indicate hatched areas, areas without survey effort at a 10 x 10 km spatial resolution. While model predictions were created for these areas, values should be interpreted cautiously as there were no survey data to support them. The tool shows whether the area selected is included as part of the hatch area (i.e. low quality) or not (high quality).
\r\nNCCOS also provides two extra measures of model uncertainty:
\r\nWe recommend to review these measures of uncertainty for each species by going to either the Northeast data portal or the Mid-Atlantic data portal.
\r\nPlease refer to the information available on the following report:
\r\nWinship, A.J., B.P. Kinlan, T.P. White, J.B. Leirness, and J. Christensen. 2018. Modeling At-Sea Density of Marine Birds to Support Atlantic Marine Renewable Energy Planning: Final Report. OCS Study BOEM 2018-010. Sterling, VA. 67 pp. https://coastalscience.noaa.gov/data_reports/modeling-at-sea-density-of-marine-birds-to-support-atlantic-marine-renewable-energy-planning-final-report/
\r\nThe team at NOAA used statistical modeling techniques to predict areas of the seafloor that are capable of supporting deep-sea corals. Using a statistical machine-learning algorithm called maximum entropy (MaxEnt), they combined databases of known deep-sea coral locations provided by the NOAA Deep-Sea Coral Research and Technology Program (DSCRTP) and other contributors with environmental and oceanographic data to generate predictive models of deep-sea coral distribution. These models are often used to produce regional maps of deep-sea coral habitat.
\r\nThis project was a cross-NOAA collaboration involving funding and collaborators at NCCOS, the NOAA National Marine Fisheries Service (DSCRTP, NEFSC, SEFSC), and the NOAA Office of Ocean Exploration and Research (OER). The original data used to support this modeling effort were collected by a large number of Federal, State, academic, NGO, and industry efforts over many years.
\r\n\r\n Please refer to NOAA’s website about this project.\r\n
\r\n\r\n Following this link, you can download the original datasets used for this analysis. Also, full descriptions of data contributors to each regional modeling effort can be found in the metadata for each component in the Digital Data Packages available for download.\r\n
\r\nExporting fact sheet\r\n
Uploading & Converting SHP\r\n
Missing features? Contact us
\r\n\r\n By entering this site, you agree to the terms outlined in the Legal Disclosure and Privacy Statement.\r\n
\r\n\r\n