In-Situ Microscope Provides Continuous Time-Series Of Plankton And Particulates

Plankton images generated by CPICS (Continuous Particle Imaging System).

Plankton images generated by CPICS (Continuous Particle Imaging System).

The Continuous Plankton Imaging and Classification System (CPICS) is providing automated measurements of plankton biodiversity on long-term observing systems

Marine and freshwater ecosystems are changing on surprisingly rapid time scales as a function of a diverse suite of forcing functions, both natural and anthropogenic. Plankton are at the base of virtually all aquatic food chains supporting ecosystem function and are particularly relevant to commercially important fisheries.

 

CPICS

The CPICS instrument.

Click here for the 3D model of CPICS

Plankton and their resulting breakdown products called marine snow directly support the biogeochemistry of aquatic communities by providing more than half of the oxygen we breathe and the removal of more than half of the carbon produced by burning fossil fuels to the deep sea. Understanding the balance between plankton, their community structure, and the production of marine snow is essential to understanding ecosystem function and the survival of our species. To this end, establishing a continuous plankton and marine snow time series at key locations throughout marine and freshwater systems consisting of sampling scales from rapid (seconds) to longterm (decades) would provide a sentinel for ecosystem change. The key is to measure plankton abundance and establish indices of biodiversity at sufficiently fast time scales that allow disentanglement of physical (transport) and biological (growth) properties of an ecosystem.

Traditionally, plankton and particle studies are carried out using nets towed through the water that screen out material between 50 microns and several millimeters, followed by laborious hours in the laboratory under a microscope sorting and identifying plankton by hand. This manual sampling approach precludes the kind of rapid sampling necessary to build indices of biodiversity that will allow us to better understand ecosystem dynamics. Conversely, the CPICS instrument puts the microscope in the water and, along with high-resolution optics and embedded processing, provides a continuous stream of data consisting of plankton and particle classifications, size, shape, volume, and other data types necessary for calculating the contribution of plankton to carbon flux to the deep ocean and lakes.

Figure 1. Workflow for embedded, automated classification of plankton on CPICS instrument.

Figure 1. Workflow for embedded, automated classification of plankton on CPICS instrument.

Figure 2. Example screen shot of website showing cross-validation confusion matrix with classification accuracies ranging from 92% to 100% and percentage composition of each class in a given sample of 16 classes.

Figure 2. Example screen shot of website showing cross-validation confusion matrix with classification accuracies ranging from 92% to 100% and percentage composition of each class in a given sample of 16 classes.

Figure 3. Time series of Shannon-Weaver Biodiversity Index (H’) for 21 plankton classes classified automatically for the autumn months in 2014 and 2015 at a long-term observatory site in the Kuroshio Current south of Tokyo, Japan. Blue dots are hourly calculations of H’, red lines are 6-hour running means, and green lines are linear regressions. Notes: 1) very high variability in H’ hour to hour; and 2) the negative slope in 2014 and not in 2015, suggesting that biodiversity did not decrease in 2015 as it normally should due to lack of decreasing fall temperature.

Figure 3. Time series of Shannon-Weaver Biodiversity Index (H’) for 21 plankton classes classified automatically for the autumn months in 2014 and 2015 at a long-term observatory site in the Kuroshio Current south of Tokyo, Japan. Blue dots are hourly calculations of H’, red lines are 6-hour running means, and green lines are linear regressions. Notes: 1) very high variability in H’ hour to hour; and 2) the negative slope in 2014 and not in 2015, suggesting that biodiversity did not decrease in 2015 as it normally should due to lack of decreasing fall temperature.

 

Continuous automated plankton classification has been accomplished recently on fixed observing systems allowing long-term, high-frequency biological measurements that are anti-aliased for physical processes. The OceanCubes program (oceancubes.whoi.edu) was established specifically to provide measurements of biological, chemical, and physical properties of the coastal ocean to capture the response of the plankton community to ecosystem change and to expose and quantify the drivers causing such change. This observing capability will be reported in the September issue of ON&T.

The CPICS plankton camera (Figure 1) has been installed on OceanCubes observatories off Okinawa, on Oshima island off Tokyo, on the Pacific and Caribbean coasts of Panama, and in Lake George, New York. In addition, it is proposed for several new sites where plankton communities are impacted by physical and geochemical features such as upwelling, intense horizontal mixing, and terrestrial run off.

CPICS is manufactured by CoastalOceanVision, Inc. in North Falmouth, Massachusetts (www.coastaloceanvision.com) and produces high-resolution dark-field images in vivid color six times per second and can process hundreds of particles per image. The short exposure (100 μs) from a custom LED ring eliminates motion blur, while the open flow design is non-invasive and non-restrictive, providing images of very fragile plankton in their natural orientation. Several magnifications are available from 0.5 to 20 x, forming a Field of View (FOV) of ~5 centimeters to ~100 microns, respectively. Images shown in this article were taken at a magnification of 1x and FOV of 12x11 millimeter. Image processing algorithms running on an Nvidia Jetson TX1 embedded processor extract and store images locally for later processing. The next step is to extract feature sets for texture, color pattern, morphology, shape, and volume to train a Random Forest machine learning classifier resulting in classifications that are cross-validated using confusion matrices (Figure 2). High classification accuracies (80% to 100%) are possible depending on the number of training categories and target complexity. A web-based utility running on the CPICS instrument will allow access to raw images, training sets, classifiers, and classification results over the Internet. CPICS may be stand-alone such as on a CTD, towed vehicle, AUV, or connected by Ethernet or serial to ship or shore. An ecologically meaningful plankton index of biodiversity and its variance is developed using a combination of species and taxon groups, which provides a novel approach for understanding ecosystem change (Figure 3).

Subscription

Corporate Headquarters

Ocean News & Technology is a publication of TSC

7897 SW Jack James Drive
Suite A, Stuart, FL 34997
(772)-221-7720

Go to top