Methodology:

Image Acquisition

WorldView-03 images (Standard 2 A product type and level) were purchased with a panchromatic resolution of 0.31m and a multispectral resolution of 1.24m from Maxar Technologies. Images were acquired over three seasons: (i) 2018 to 2019; (ii) 2019 to 2020; and (iii) 2021 to 2022. Within each season, images were acquired from the earliest weather window until acquisitions were not possible due to low light levels. This gave a range of 114, 108, and 122 days from first to last image acquisition for each of the three seasons respectively. Images were acquired close to noon each day. VHR image metadata is provided in the table 'vhr_image_metadata.csv'. The average area scanned was calculated from each observer's scanned areas. This varied due to individuals' determination of the edge of Wilhelmina Bay. For the 2018-2019 season, the average area scanned was 130.17km2 (±19.63). It was 143.28km2 (±1.73) in the 2019-2020 season, and 104.86km2 (±27.1) in the 2021-2022 season. This gave a multi season average of 115.5km2 (±37.92).

Detection and labelling of whales

Pansharpened (using the ESRI algorithm), 4-band images were analysed in ESRI ArcMap v10.6 for images from 2018 to 2020, and in ArcPro v2.8.0 for the 2021 to 2022 season (www.esri.com). Entire images were systematically scanned by multiple observers using a 0.0125 km2 grid and identified FOIs recorded as ESRI point shapefiles, with the point being placed on the centre of the FOI after (Cubaynes et al., 2019). Seven observers participated, analysing a total of 17 VHR images of Wilhelmina Bay. Due to logistical limitations, different observers analysed the images from different seasons. Two observers analysed both the 2018 to 2019 and 2019 to 2020 seasons, and four observers analysed the 2021 to 2022 season only. Of these four, two did not analyse the last of the seven images from that season. For consistency, one observer analysed all 17 images across the three seasons. Identified features were reviewed prior to classification, with obvious non-whale objects removed. Candidate features were initially assigned to either 'yes' or 'no' whale category based on an instantaneous classification. This is akin to the decision process that occurs during traditional line-transect ship-borne surveys, where observers rely on their judgement and undefined cues to determine the validity of a sighting. Only those initially classified as likely ('yes') to be a whale were then classified into either "Definite", "Probable" or "Unclassified" following the criteria set out in Cubaynes et al. (2019) and equation and weighting factors from Bamford et al. (2020). Environmental conditions in the immediate vicinity of the FOI were recorded for sea state, sea ice levels, and cloud cover following the scales provided in the document 'localised_environmental_condition_scales.pdf'.

Data collection:

ESRI ArcMap v10.6 for images from 2018 to 2020, and in ArcPro v2.8.0 for the 2021 to 2022 season.

Data quality:

NA's present in the attached 'identified_features.csv' file - these occur when a particular record was not completed by an observer during the classification process (i.e., cloud cover was omitted from the record). Whilst these correspond to missing information on the classification process the locational information is intact.