In the Spring of 2020, a 24’ Carolina Skiff, generously donated by Aaron Hassan and Fran and Gail Collins greatly improved our field work efficiency. Inspired to apply techniques from Laura Humphrey’s statistics course at Cohasset High School, Clay and Hoadley wrote a grant proposal to conduct a survey of the Cohasset Harbor, correlating eelgrass health with sediment type. Upon winning the grant from the Marjot Foundation, Hoadley and Clay added Beck LaBash (now at Northeastern University) and long-time CSCR student Brendan Burke (now at George Washington University with a scholarship resulting from his CSCR work) to the team and evaluated the underwater situation in 88 different sites. Burke continued contributing good questions that brought together observations in new and unique ways. Using LaBash’s python coding skills, learned through an online Harvard Extention course, they also invented a new technique that geocoded and video-documented the entire area of the eelgrass in the outer harbor. Below is their mid-project report (written) and final presentation (at CSCR’s State of the Harbor). They also presented at Zosterapalooza and New England Estuarine Research Society in the Spring of 2021.
2020 Eelgrass Project Report to Marjot Foundation
Lucy Clay and Andrew Hoadley
- Introduction:
We are very lucky to have been given this opportunity by the Marjot Foundation to conduct eelgrass research this summer. Our team consists of our leaders Lucy Clay and Andrew Hoadley along with two additional varsity teammates, Beck LaBash and Brendan Burke. Additionally, other junior teammates have significantly helped us with our workload. We are thankful that our research was not greatly affected by the Covid guidelines and are pleased to report that no one at CSCR has contracted Covid. Despite wearing masks, gloves, and staying six feet apart, we had an extremely productive, research-heavy summer!
Our eelgrass team started outdoor expeditions in June. With a newly donated 24’ Carolina skiff and approval from the Board of Health, we were able to proceed with our original plan of having 3 expeditions a week at low tide (about 2-4 hours each) while staying 6’ apart. We finished our sample collection for our Marjot grant in early August, with a total of 88 sites. We are now beginning to work on data management and analysis as the snorkeling and reliable boating season has drawn to a close. [See the 2021 State of the Harbor video above for the results].
- Methods
Because we found better, more efficient ways of collecting and classifying data, which saved time and improved the quality of our data, our expedition procedures evolved from the original procedure stated in our proposal.
In terms of sediment type, we changed our sediment classifications: [sand, shells, solid granite rock, mud, gravel, stoney cobble, clay, or “other”] to adhere to the USDA’s: [loam, sand, sandy gravel, gravel, gravelly cobble, and cobble]. This allowed us to focus on observed particle size, make interpolation maps more feasible, and organize our data more efficiently in the future.
We decided to define eelgrass health solely in terms of coverage density instead of defining health also in terms of leaf length, as stated in our proposal. We concluded that defining eelgrass health in terms of leaf length would be very hard to analyze and correlate, considering that the leaves grow significantly throughout the summer. If we had defined health in terms of leaf length, a possible confounding variable would have been that the sun and other external factors were making the eelgrass grow longer, and not necessarily the sediment type it was growing in.
We refined our procedures to determine the coverage density at each site, as unbiased as possible, by deploying what we’ve named “our democratic quadrat”, which is quite large at 1.5 x 1.5 meters. Once at an eelgrass site/area, this quadrat is thrown into the water randomly and each person on the research vessel classifies the coverage density of eelgrass inside the quadrat as none, 1% (to represent one solo colonizing plant), <25%, <50%, <75%, or <100%. The majority opinion is then entered into both Survey 123 (our ARCGIS app) and our field notebook.
Our initial sampling approach was superimposing a grid of equal-sized sectors over the study area and using our existing data to categorize the soil type such that we could, upon numbering the grid squares, randomly choose 15 in each soil area and take two random samples in each sector. Instead, we developed procedures to survey the whole harbor by recording tracks of underwater video footage along with the geolocation of the track. To do this, we took a gopro camera and clamped it onto a 2.2 meter stick so that the gopro is faced downward towards the sediment as the boat is moving. As GoPro’s can’t record their location underwater, Beck created an app built on ArcGIS’s Quickcapture to record streaming GPS points. By doing these tracks, we were able to create an extensive grid of our harbor with our gopro/GPS tracks. Once we have organized the video and gps files, every point on the GPS track can be matched to a frame of the gopro footage using a script that our teammate, Beck, wrote in Python. Since we had covered the whole harbor, we were able to successfully sample our whole area of study.
In addition to making moving tracks, we studied 88 stationary sites. Another ArcGIS app, Survey 123 was used to input our data at each site. With this app we recorded coverage density, location, date/time, sediment type, the organisms we see, the Democratic Quadrat, and more. At most sites, we usually sent Lucy, our “sediment sampler” to collect soil and eelgrass samples using various equipment like snorkel gear, jars, and shovels. We may find that these samples can’t be concatenated with our larger geotagged picture dataset.
Inspired by Forest Schenke a graduate student we’d met at Zosterapalooza who was using image-J to analyze eelgrass leaves, we again tapped the expertise of our colleague, Beck LaBash to experiment with using google’s teachable machine programs to train the computer to identify eelgrass in underwater photos. Our initial tests were successful, indicating fewer human hours of watching and categorizing the undulating underwater video footage. We will have to confirm that it can differentiate Ulva (green sea lettuce) seaweed from eelgrass.
In addition, Beck collected over-the-water drone imagery at different cloud coverages and tides to try to create photogrammetry that provides a high quality aerial view of the eelgrass in the harbor to corroborate our underwater footage.
Findings so far
Initial analysis of Survey 123 site data for 2020 revealed a -0.09 correlation (a weak correlation) between coverage density and sediment type. However, this data set is only from our specific anchored sites, so is significantly smaller than the dataset that is yet to be generated from our moving video footage tracks. [See the 2021 State of the Harbor video above for updates.]
In early August, we made a discovery. Lucy had noticed that despite what the surface looked like, some areas under the eelgrass were springier to walk on than others. She and Brendan wondered whether that might be more indicative of eelgrass health than the observable particle size differences in the surface sediments she was collecting. Brendan found that the sediment she collected was different from the soil that the anchor brought up. The soil that the anchor brought up was usually thicker, and varied in color and texture from site to site. We began investigating this curiosity more towards the end of the summer, and would like to continue. We hope that we will be able to find a partner with a furnace to assess the organic matter of the many soil samples we currently have in the CSCR refrigerator.
Beck’s preliminary drone footage, the updated basemap he found, and our Survey 123 findings all appear to line up well in regards to presence of eelgrass.
- Next steps
We have our handwritten notebooks, Survey 123 data, Quickcapture tracks, and hours of underwater GoPro footage, and documentary videos to organize. We are beginning to sort through this footage and data with the main focus of finding effective and efficient methods of data management. From there, we will create geotagged images with Beck’s script then classify these images by coverage density and average particle size to calculate the correlation coefficient between eelgrass health and sediment type.
Thank you again for this amazing opportunity!
Lucy, Andrew, Brendan, Beck
Within this team with its seamless collaborative spirit, we exceed expectations, we learn by doing, and even inspire other researchers. Over the last two months, CSCR’s eelgrass team presented updates to a local group of Seagrass researchers (from DMF, EPA, WHOI, etc.), to the Marjot Foundation, and to the New England Estuarine Research Society for which Beck LaBash won an Honorable Mention. They also presented their grant deliverables and their enhancements to the initial plan, at the regional Massachusetts Science and Engineering Fair, to the EPA-hosted eelgrass conference Zosterapalooza 2021, and to local stakeholders at CSCR’s State of the Harbor of 2021.