Search

We describe RAPID: a Real-time Automated Plankton Identification Dashboard, deployed on the Plankton Imager, a high-speed line-scan camera that is connected to a ship water supply and captures images of particles in a flow-through system. This end-to-end pipeline for zooplankton data uses Edge AI equipped with a classification (ResNet) model that separates the images into three broad classes: Copepods, Non-Copepods zooplankton and Detritus. The results are transmitted and visualised on a terrestrial system in near real time. Over a 7-days survey, the Plankton Imager successfully imaged and saved 128 million particles of the mesozooplankton size range, 17 million of which were successfully processed in real-time via Edge AI. Data loss occurred along the real-time pipeline, mostly due to the processing limitation of the Edge AI system. Nevertheless, we found similar variability in the counts of the three classes in the output of the dashboard (after data loss) with that of the post-survey processing of the entire dataset. This concept offers a rapid and cost-effective method for the monitoring of trends and events at fine temporal and spatial scales, thus making the most of the continuous data collection in real time and allowing for adaptive sampling to be deployed. Given the rapid pace of improvement in AI tools, it is anticipated that it will soon be possible to deploy expanded classifiers on more performant computer processors. The use of imaging and AI tools is still in its infancy, with industrial and scientific applications of the concept presented therein being open-ended. Early results suggest that technological advances in this field have the potential to revolutionise how we monitor our seas. [ABSTRACT FROM AUTHOR]

We describe RAPID: a Real-time Automated Plankton Identification Dashboard, deployed on the Plankton Imager, a high-speed line-scan camera that is connected to a ship water supply and captures images of particles in a flow-through system. This end-to-end pipeline for zooplankton data uses Edge AI equipped with a classification (ResNet) model that separates the images into three broad classes: Copepods, Non-Copepods zooplankton and Detritus. The results are transmitted and visualised on a terrestrial system in near real time. Over a 7-days survey, the Plankton Imager successfully imaged and saved 128 million particles of the mesozooplankton size range, 17 million of which were successfully processed in real-time via Edge AI. Data loss occurred along the real-time pipeline, mostly due to the processing limitation of the Edge AI system. Nevertheless, we found similar variability in the counts of the three classes in the output of the dashboard (after data loss) with that of the post-survey processing of the entire dataset. This concept offers a rapid and cost-effective method for the monitoring of trends and events at fine temporal and spatial scales, thus making the most of the continuous data collection in real time and allowing for adaptive sampling to be deployed. Given the rapid pace of improvement in AI tools, it is anticipated that it will soon be possible to deploy expanded classifiers on more performant computer processors. The use of imaging and AI tools is still in its infancy, with industrial and scientific applications of the concept presented therein being open-ended. Early results suggest that technological advances in this field have the potential to revolutionise how we monitor our seas.

The small sizes of most marine plankton necessitate that plankton sampling occur on fine spatial scales, yet our questions often span large spatial areas. Underwater imaging can provide a solution to this sampling conundrum but collects large quantities of data that require an automated approach to image analysis. Machine learning for plankton classification, and high-performance computing (HPC) infrastructure, are critical to rapid image processing; however, these assets, especially HPC infrastructure, are only available post-cruise leading to an ‘after-the-fact’ view of plankton community structure. To be responsive to the often-ephemeral nature of oceanographic features and species assemblages in highly dynamic current systems, real-time data are key for adaptive oceanographic sampling. Here we used the new In-situ Ichthyoplankton Imaging System-3 (ISIIS-3) in the Northern California Current (NCC) in conjunction with an edge server to classify imaged plankton in real-time into 170 classes. This capability together with data visualization in a heavy.ai dashboard makes adaptive real-time decision-making and sampling at sea possible. Dual ISIIS-Deep-focus Particle Imager (DPI) cameras sample 180 L s-1, leading to >10 GB of video per min. Imaged organisms are in the size range of 250 µm to 15 cm and include abundant crustaceans, fragile taxa (e.g., hydromedusae, salps), faster swimmers (e.g., krill), and rarer taxa (e.g., larval fishes). A deep learning pipeline deployed on the edge server used multithreaded CPU-based segmentation and GPU-based classification to process the imagery. AVI videos contain 50 sec of data and can contain between 23,000 - 225,000 particle and plankton segments. Processing one AVI through segmentation and classification takes on average 3.75 mins, depending on biological productivity. A heavyDB database monitors for newly processed data and is linked to a heavy.ai dashboard for interactive data visualization. We describe several examples where imaging, AI, and data visualization enable adaptive sampling that can have a transformative effect on oceanography. We envision AI-enabled adaptive sampling to have a high impact on our ability to resolve biological responses to important oceanographic features in the NCC, such as oxygen minimum zones, or harmful algal bloom thin layers, which affect the health of the ecosystem, fisheries, and local communities.

Phytoplankton induce defensive traits in response to chemical alarm signals from grazing zooplankton. However, these signals are potentially vulnerable to changes in pH and it is not yet known how predator recognition may be affected by ocean acidification. We exposed four species of diatoms and one toxic dinoflagellate to future pCO2 levels, projected by the turn of the century, in factorial combinations with predatory cues from copepods (copepodamides). We measured the change in growth, chain length, silica content, and toxin content. Effects of increased pCO2 were highly species specific. The induction of defensive traits was accompanied by a significant reduction in growth rate in three out of five species. The reduction averaged 39% and we interpret this as an allocation cost associated with defensive traits. Copepodamides induced significant chain length reduction in three of the four diatom species. Under elevated pCO2Skeletonema marinoi reduced silica content by 30% and in Alexandrium minutum the toxin content was reduced by 30%. Using copepodamides to induce defensive traits in the absence of direct grazing provides a straightforward methodology to assess costs of defense in microplankton. We conclude that copepodamide signalling system is likely robust to ocean acidification. Moreover, the variable responses of different taxa to ocean acidification suggest that there will be winners and losers in a high pCO2 world, and that ocean acidification may have structuring effects on phytoplankton communities.

The oceans’ uptake of anthropogenic carbon dioxide (CO2) decreases seawater pH and alters the inorganic carbon speciation – summarized in the term ocean acidification (OA). Already today, coastal regions experience episodic pH events during which surface layer pH drops below values projected for the surface ocean at the end of the century. Future OA is expected to further enhance the intensity of these coastal extreme pH events. To evaluate the influence of such episodic OA events in coastal regions, we deployed eight pelagic mesocosms for 53 days in Raunefjord, Norway, and enclosed 56–61 m3 of local seawater containing a natural plankton community under nutrient limited post-bloom conditions. Four mesocosms were enriched with CO2 to simulate extreme pCO2 levels of 1978 – 2069 μatm while the other four served as untreated controls. Here, we present results from multivariate analyses on OA-induced changes in the phyto-, micro-, and mesozooplankton community structure. Pronounced differences in the plankton community emerged early in the experiment, and were amplified by enhanced top-down control throughout the study period. The plankton groups responding most profoundly to high CO2 conditions were cyanobacteria (negative), chlorophyceae (negative), auto- and heterotrophic microzooplankton (negative), and a variety of mesozooplanktonic taxa, including copepoda (mixed), appendicularia (positive), hydrozoa (positive), fish larvae (positive), and gastropoda (negative). The restructuring of the community coincided with significant changes in the concentration and elemental stoichiometry of particulate organic matter. Results imply that extreme CO2 events can lead to a substantial reorganization of the planktonic food web, affecting multiple trophic levels from phytoplankton to primary and secondary consumers.