Howdy

Satellite imagery now permits dynamic wetland management, yet mapping requires significant fieldwork to run classification algorithms and estuarine environments can be difficult to access. In this study we used drone imagery to create training data to classify WorldView-3 and RapidEye satellite imagery of the Rachel Carson Reserve in North Carolina, USA. We examine change between 2004 and 2017 and conclude that drones can be highly effective in training and validating satellite imagery.

Drones are a potentially promising tool for assessing marine animal populations, but a central challenge lies in analyzing the data generated. Neural networks are emerging as a method for automating detection across domains and can be applied to imagery to generate new population‐level insights. We used a neural network to count turtles in drone imagery and our model detected 8% more turtles than manual counts while effectively reducing the manual validation burden from 2,971,554 to 44,822 image windows.

Photogrammetry is contributing to the study of marine mammal individual health, population structure, and understanding changes within communities. This study combines drone surveys with convolutional neural networks for species identification and morphometric assessment of large marine vertebrates. We validate the accuracy of these workflows with traditional approaches. This study presents substantial new morphometric data on humpback, minke, and blue whales and describes an open source CNN for automated species identification and photogrammetric measurements from drone imagery.

Monitoring great whales across their broad spatial scales is prohibitively time consuming using manual analysis methods. We describe a framework using convolutional neural networks to detect great whales in high res satellite imagery. We outline challenges both in data management and model development and propose future research directions for overcoming these obstacles.