Zooplankton, the smallest animals in the food chain, are critical to the ecology of lakes like Lake George. But studying them in the wild – by measuring biomass, species composition, behavior, and diet – is a challenge. How do you track a borderline microscopic animal in the vast volume of water (an estimated 550 billion gallons) that fills the Lake George basin?

Typically, a researcher like Alex Pezzuoli relies on data that he collects by hand – visiting five different sites on Lake George twice a week, and using a fine net to collect samples of zooplankton and water for analysis. For each sampling event, he gets about 20 data points across the whole lake.

But through the Jefferson Project at Lake George, a collaboration between Rensselaer Polytechnic Institute, IBM, and The FUND for Lake George to make Lake George a global model for ecosystem understanding and protection, Pezzuoli may soon have access to hundreds of thousands of sensor network data points daily. And with the unprecedented capabilities of the network, he will also be able to correlate zooplankton data to indicators of water quality and weather, which the network collects and autonomously monitors for anomalies and previously undetected relationships.

Pezzuoli, a doctoral candidate in biological sciences, is working with Jefferson Project researchers to establish whether the project’s Acoustic Doppler Current Profilers (ADCP) – which use the reflection of sound waves off particulate in the water column to track water circulation – might also be useful in tracking zooplankton.

To test the concept, Jefferson Project researchers headed by Mike Kelly, an IBM senior research engineer, suspended a pair of ADCPs – one facing downward and one facing up toward the surface  – in the water alongside one of the project’s vertical profilers, a floating sensor platform that includes a communication station. Kelly connected the ADCP to the communications station, making it possible to see the ADCP data in real time, and to correlate it with indicators of water quality and weather information the vertical profiler captures. Kelly is also developing customized on-device machine learning techniques called edge analytics, which would transform the sensor platform into a cognitive Internet of Things (IoT) network with the ability to seek, understand, and sense the relationships between different data streams, data collection, measurement, and analysis.

Edge analytics are central to the utility of the sensor network. In Kelly’s mind, a sensor is something that provides a single type of measurement, but a sensor network is an integrated semi-autonomous system. Here’s how he put it:

Cognitive technologies such as edge analytics are fundamental to the nature of the Internet of Things. Not only do you want to collect data near real-time, you also want to be able to feed back control to actuators and motors in real time. We want to take the observations that are collected in real-time, analyze the data, and potentially change how we’re taking the measurements in the physical system.

The zooplankton tracking effort is an example of how the Jefferson Project is building a powerful tool for environmental research by combining cognitive computing with the IoT sensors.

The idea for using the sensor network to track zooplankton sprang from a pattern Rensselaer biologists Jeremy Farrell and Sandra Nierzwicki-Bauer noticed in data that an ADCP collected while perched on the lakebed in 2015. The pattern coincides with the diurnal vertical migration of zooplankton, which are known to sink into the lake to evade predators by day, and rise to the surface at night to eat phytoplankton.

Pezzouli searched research literature and found accounts of marine researchers using a sonar signal – like backscatter or the density of a sonar return – at about 120 kilohertz to measure fish and larger invertebrates in the ocean. Freshwater plankton are smaller, and can only be detected with a higher frequency signal, and Pezzuoli found fewer accounts of research that uses ADCP data to track zooplankton in lakes. On Lake George, the Jefferson Project is using devices that range from 300 to 600 kilohertz, which should be high enough to do the job.

A data stream from the ADCPs alone would be an enormous improvement, Pezzuoli said:

When we do this by hand, we pull a 64-micron mesh net from the bottom to the surface and we get a discrete sample of zooplankton in the water column. You can’t get any information on how density changes with depth, and you only get a sample when you physically go out and drop the net, so you only have a specific point in time. If something happens when you’re not there, you don’t know about it.

Add that continuous data stream to other information from the vertical profiler – like temperature throughout the water column, or weather information on solar radiance or wind speed – and researchers can be begin to look for relationships between the physical, chemical, and biological. The sensor network offers researchers a more intimate view of how physical functions, like weather, drive biology.

That intimate view doesn’t just appear from a flood of raw numbers. It comes from the story hidden in the numbers. To discern those stories, researchers in the life sciences work directly with the engineers who are developing the sensor network and data portal. In adding sensors, and connections and machine learning capabilities to the network, Kelly is always looking to make the result more useful to scientists.

If the approach works, it will be transformative. Here’s what Pezzuoli has to say about the potential benefits of the new data stream:

We’re bringing all of this data together – physical, chemical, biological. It’s not only a much higher resolution, it’s also more informative. You can see in real-time how conditions in the lake are changing, how temperature and lake chemistry are changing, and how the zooplankton respond to that. It’s a huge improvement.

The project recently announced that it has received a $917,000 grant from the National Science Foundation to complete the sensor network within the next year. The below graphic depicts the location of current and planned sensors on Lake George.