Statement |
Acoustic data acquisition was carried out from 9th to 19th February 2013 from the survey vessel MV Seabeam. Bathymetry and backscatter data were acquired using a Kongsberg EM3002D multibeam echosounder (MBES) system. A Kongsberg Seafloor Information System (SIS) version 3.7.5 (Build 93) was used for online data logging. No towed sonar was deployed. The bathymetric data were collected and processed in accordance with the International Hydrographic Organisation Standards for Hydrographic Surveys - Order 1 (Special Publication 44, Edition 4) (IHO, 2008). Post-survey data processing was performed using a combination of CARIS HIPS version 7.1.1 (SP1, Hot Fix 1) and IVS Fledermaus version 7.3.3c (Build 481 Professional) to produce a backscatter mosaic for export as a geotiff file.
The Runswick Bay rMCZ ground truth survey was carried out between 15th September 2013 and 3th May 2014 from the survey vessel Humber Guardian. Ninety-five target sampling stations were identified for the collection of ground truth data within the rMCZ. This selection of stations was deemed to give the best possible representation of the rMCZ area and potential BSHs, based on interpreted MBES bathymetry and backscatter data and UKHO Admiralty charts. A drop camera survey was carried out first, and stations were selected for grab sampling activity if suitable sublittoral sediment was observed in at least 50% of the images collected. Drop video camera equipment was used at all 95 stations to collect video and still images of the seabed, and benthic grabs were then used at 51 stations to collect sediments and infauna.
All new maps and their derivatives have been based on a WGS84 datum. A new habitat map for the site was produced by analysing and interpreting the acoustic data and ground truth data collected by the dedicated surveys of this site, using an object based image analysis process.
Habitat classes were created for the five BSHs for which ground truth data had been captured by the 2013-14 groundtruthing survey. Moderate energy infralittoral rock and high energy infralittoral rock were not included in the classification as they were not recorded during the ground truth survey. As a result, there is a probable overestimation of moderate energy circalittoral rock in the updated habitat map. Infralittoral rock is likely to occur in the inshore area which was not surveyed, and the map could be amended in future if evidence can be collected to support this. Classification was carried out in four steps:
1. Rule-based classification of moderate energy circalittoral rock based on thresholds of mean backscatter and slope for three different depth ranges (>33 m, -33 m to -43 m and <-43 m).
2. Nearest neighbour classification of shallow sediments (>-33 m) using the following object features: mean backscatter, standard deviation backscatter, standard deviation bathymetry, maximum pixel value slope and grey level cooccurrence matrix (GLCM) correlation backscatter.
3. Nearest neighbour classification of deep sediments (<-33 m) using the following object features: mean backscatter, standard deviation backscatter, standard deviation bathymetry, GLCM correlation backscatter and GLCM correlation bathymetry.
4. Rule-based classification of artefacts based on thresholds of size, compactness, direction and ratio of length to width. These objects were then re-classified as the habitat by which they were surrounded, or with which they shared a significant border.
Separate classification processes were required for different depth ranges, as increasing depth resulted in reduced backscatter intensity for the same seabed type. This is was evident when comparing the acoustic data with the ground truth data. Classification step 1 eliminated this issue. Samples to train the nearest neighbour classifications were created by selecting objects that overlapped PSA ground truth points. The Feature Space Optimization Tool in eCognition® was used to identify the optimum combination of features for separating the classes using the nearest neighbour classification. The classified objects were merged and the resulting habitat map was exported as a shapefile containing 953 classified polygons. |