Search

Seagrass form the basis for critically important marine ecosystems. Previously, we implemented a deep convolutional neural network (CNN) model to detect seagrass in multispectral satellite images of three coastal habitats in northern Florida. However, a deep CNN model trained at one location usually does not generalize to other locations due to data distribution shifts. In this paper, we developed a semi-supervised domain adaptation method to generalize a trained deep CNN model to other locations for seagrass detection. First, we utilized a generative adversarial network (GAN) loss to align marginal data distribution between source domain and target domain using unlabeled data from both data domains. Second, we used a few labelled samples from the target domain to align class specific data distributions between the two domains, based on the contrastive semantic alignment loss. We achieved the best results in 28 out of 36 scenarios as compared to other state-of-the-art domain adaptation methods.

Seagrass habitats are becoming extremely vulnerable due to human intrusion to seagrass meadows, which results in unbalanced marine ecosystems and extinction of marine animals. Traditionally, manual scarring has been used to identify and quantify seagrass propeller scars. However, this method requires site visitation and it is cost ineffective. In this paper, we propose deep learning method to automatically detect propeller seagrass scars in multispectral satellite images. Our proposed algorithm is more computationally efficient than our previous sparse coding detection model and can accurately detect seagrass scars. Additionally, we explored two pan-sharpening methods for obtaining high-resolution multispectral satellite images for scar detection. We evaluated our methods on four multispectral images collected in Florida and experimental results show that the proposed deep learning model combined with the Gram-Smith (GS) pan-sharpening approach achieved the best sensitivities in seagrass scar detection and this combination is also the most computational efficient method, requiring only 7 minutes for a testing image with a size of 1000x800 in testing phase.

We describe emerging capabilities to understand physical processes and biogeochemical cycles in coastal waters through the use of satellites, numerical models, and ship observations. Emerging capabilities provide significantly improved ability to model ecological systems and the impact of environmental management actions on them. The complex interaction of physical and biogeochemical processes responsible for hypoxic events requires an integrated approach to research, monitoring, and modeling in order to fully define the processes leading to hypoxia. Our effort characterizes the carbon cycle associated with river plumes and the export of organic matter and nutrients from coastal Louisiana wetlands and embayments in a spatially and temporally intensive manner previously not possible. Riverine nutrients clearly affect ecosystems in the northern Gulf of Mexico as evidenced in the occurrence of regional hypoxia events. Less known and largely unquantified is the export of organic matter and nutrients from the large areas of disappearing coastal wetlands and large embayments adjacent to the Louisiana Continental Shelf. This project provides new methods to track the river plume along the shelf and to estimate the rate of export of suspended inorganic and organic particulate matter and dissolved organic matter from coastal habitats of south Louisiana.