Developing Precision Oyster Farming Methods Using Environmental Data

Progress report for FNE23-037

Project Type: Farmer
Funds awarded in 2023: $21,268.00
Projected End Date: 12/31/2024
Grant Recipient: Ferda Farms LLC
Region: Northeast
State: Maine
Project Leader:
Max Burtis
Ferda Farms LLC
Expand All

Project Information

Project Objectives:

This project seeks to…

  1. Collect environmental data using monitoring buoys at three growout sites on the New Meadows River. We anticipate an uninterrupted data set for each location with high temporal resolution throughout the growing season. 
  2. Collect oyster weight and mortality data from bags located adjacent to the data buoys. We anticipate collecting this data every two weeks at each site. Additionally, market oysters will be sampled once to calculate market volume and weight. 
  3. Model of oyster growth rate and mortality. We anticipate the results to be a multiple regression model with the response variable being oyster growth rate and the input parameters being temperature, dissolved oxygen, salinity, and current speed. 
  4. Use this model to make farming recommendations. We anticipate these recommendations to address density splitting, sales planning, and methods to limit mortality risk. 
  5. Package these results into a generalizable protocol. The anticipated result of this step will be both open source code and a web application that takes in a location of interest, environmental data from the closest data buoys then outputs predicted growth rates, and farm management recommendations such as a sorting or harvest schedule.

Most oyster farmers don’t use environmental data to manage their farm’s operations. This leads to a whole host of problems including inaccurate growth and sale projections, unexpected mortality events, biofouling, improper site selections, and untimely harvest closures. These problems hurt oyster farmers’ profitability, income, and relationships with buyers. All of these issues can be addressed by using environmental data to model the oyster farm. 

Using buoys equipped with sensors, we will be able to predict conditions that may be harmful to oysters, conditions that may result in harvest closures, and the conditions that will impact oyster growth rate. We sought input from oyster farmers, kelp farmers and aquaculture researchers in order to choose the strongest indicators of site viability, crop health, and value to researchers. As a result, the monitoring buoys that we will use are designed to generate data including current and tide profiles, temperature, water depth, conductivity, chlorophyll, and dissolved oxygen. With growth rate models we can predict stocking density. This will allow for efficient scheduling of sorting, grading, defouling, and density splitting.  Maximizing efficiencies in processing steps will result in a higher market value for the oysters because they will be handled more (resulting in better oyster shape), while simultaneously reducing labor per oyster, and therefore operational expense. Additionally, environmental data provided by these buoys can predict environmental conditions harmful to oysters and other potential crops. By providing real-time environmental data, farmers can take mitigating actions and protect their oysters from often deadly conditions. Algal blooms are a perfect example of this. When all of the algae die and begin to decompose a large amount of oxygen is consumed which can lead to anoxic conditions [1]. These conditions may cause mortality events on the farm. Understanding environmental trends that often take place before mortality events could be very beneficial as it would allow farmers to harvest ahead of these events to protect their shellfish from loss. 

Addressing this problem will improve the sustainability of oyster farms because it will give them a tool to help manage the effects that a changing climate has on their farms. In the Gulf of Maine, we are seeing a period of rapid temperature increase and ocean acidification. There are more frequent extreme weather events. Optimal management of an oyster farm in a rapidly changing environment is impossible without an understanding of that environment. This project aims to model environmental changes using data buoys. The Gulf of Maine is expected to continue changing quickly, so environmental data will be even more valuable to farmers in the future.

Time can be used as a predictor for oyster growth, but oysters don't grow based on time. The amount of food, the flow rate of the water, and the temperature of the water are more direct predictors of oyster growth rate. However, these metrics change from location to location and from year to year. A growth curve from last season may not be the same as this season and growth rates from week to week may vary greatly based on environmental conditions. This makes it difficult for oyster farmers to predict how their farm will perform.  This research will explore a novel method for data driven, predictive oyster farming. In summary, this solution will result in improved productivity, reduction of costs, and increase of net farm income by developing a protocol that will allow farmers to train a model of environmental data to the growth rate of their oysters, and then use that model to better manage their farm. 


Click linked name(s) to expand/collapse or show everyone's info


Materials and methods:
  1. We will deploy the three monitoring buoys on the south end of three different limited purpose aquaculture (LPA) sites along the New Meadows that are operated by Ferda Farms. Installation of the buoys will occur as the aquaculture gear is floated in the spring. The buoys will be installed by tying them into the existing mooring lines between the LPA marker buoy and the first surface cage of oysters. The first site (license number MBUR520) is located just east of Lower Coombs Island at 43.86140N, -69.90440W; the second (license number SSHA218) is located in an adjacent channel 0.81 km to the east of MBUR520 at 43.86064N, -69.89453W; the third and final site (license number CBUR420) is located 2.2 km to the north of SSHA218. All sites use the same gear, start with the same seed, and are farmed using the same husbandry techniques. Once deployed, each of the three monitoring buoys will be continuously measuring, logging and reporting data back to the cloud. The measurement frequency is user configurable and can be done without removing the buoy from service. Through prior research, and advice, we will initially configure them to take measurements every hour. We plan to evaluate the data frequency every month with our technical advisor and adjust based on analysis with them. All the buoys are connected to the cloud with a satellite modem. The frequency at which data is transmitted is user configurable, and we initially expect to transmit new data as it is collected, every hour. In addition, we can poll the buoy remotely and get new data points on demand if necessary. Early prototypes of the data buoys will employ anti-fouling films where possible but will be checked every two weeks to verify their integrity, and clean any organic buildup that may occur.
  2. An initial random sample of 200 one year old oysters (Crassostrea virginica) from each of the LPA sites will be taken. These oysters will be bagged separately and marked while they are in grow out, and they will be stored in the cage nearest the data buoy. Measurements will begin as soon as oysters are raised from the bottom in the spring and will occur every two weeks until they are sunk in the winter. The bag of oysters will be brought back to the processing float and the oysters will be removed from the bag and any fouling or pseudofeces will be washed from the oysters using a high pressure washdown hose. Mortality will be assessed by counting the number of oysters in each bag that are dead. Any oyster with an open shell will be considered dead. Dead oysters will be discarded, and the sample size will be adjusted. The grow out bag will then be used to tare a scale that will hang from the superstructure on the processing float. The oysters will then be placed into the grow out bag and weighed using the scale. An average weight will be calculated by dividing the weight of the oysters by the number of oysters in the bag. After this, the oysters will be returned to grow out for another two weeks. Additionally, a metric for weight and volume of market ready oysters will need to be found for future modeling. A random sample of 200 market ready oysters will be taken and the weight will be measured using the same methods as above. The volume of the oysters will be calculated using the water displacement method with a graduated cup. Measuring both mass and volume in this step will allow us to calculate oyster density which can be used to map weight to volume throughout the rest of the analysis. Collection of the samples, processing of the oysters, and associated measurements are expected to take one person three hours every two weeks throughout the growing season. 
  3. After one season of growing, we will have three separate datasets, one for each site, that contain environmental data, oyster weights, and mortalities. The weight and mortality will be analyzed in separate multiple linear regression models as response variables. The environmental data will be used as inputs to this model using similar methods to [12], except with different input parameters. The Python library, Scikit-learn will be used to implement the model. Scikit-learn is freely available and well documented, and will ensure easier reproducibility of the results by others. The three test sites will provide more statistical power and will result in a more general model that is less likely to be overfitted or confounded by environmental factors that are not measured in this experiment. The result will be a set of models that map the environmental data to the response variables across the growing season. 
  4. The focus areas for the farming recommendations will be mortality, density splitting, and sales planning. We will examine patterns in environmental data that lead to mortality events. The coefficients of the regression model will demonstrate the weighted impact that each of the environmental factors has on oyster mortality. These weights will be used to warn farmers of risks. For example, if dissolved oxygen is found to strongly correlate to oyster mortality in the regression model, an oyster farmer should prioritize harvesting oysters if dissolved oxygen drops below a certain level, as detected by environmental monitoring buoys. Farmers need to reduce the number of oysters in each bag as the size of the oysters increase so that the oysters don’t become too crowded. Knowing when to thin out bags will help plan out space, gear and labor requirements. Scheduling bag splits will follow a similar method to [11]. Weight will be multiplied by the density to get volume.  When the volume is predicted to double, the amount of gear and space to house the oysters will also double. Sales can be predicted by finding the time along the growth curve that the oysters will reach market weight. 
  5. A primary product of this research is to develop an easy protocol that can be performed by farmers, so weight is a preferred measurement of regular sampling. Hanging a bag of oysters on a scale is easier than using the water displacement method in a graduated cup. Volume is more relevant to oyster farmers as it relates to how much gear is needed to hold the oysters, but by dividing mass by density, a farmer can obtain volume. This protocol will include a guide for taking measurements of oysters and collecting the environmental data. The ultimate goal of this grant is a publicly available web application that any farmer can access to evaluate their operation. As farmers use the portal, the first thing they see is our protocol for measuring, weighing and estimating their oysters. This is presented as an instructional video and interactive guide. The interactive guide helps farmers input their data into the web app where it is stored in perpetuity. Throughout a growing season, farmers can continue to sign-in to their account, input new data and access their old data. As they enter data throughout the season, the models are continuously improved and their predictions and management suggestions become tailored to their own farm. 
Participation Summary

Project Outcomes

Project outcomes:

1/15/23 Report: 

Unfortunately, there were significant delays in the production of the environmental monitoring buoy, produced by Fathom Fishing. These delays prevented us from deploying the data buoys during the summer 2023 growing season. Instead, we plan to deploy these buoys during the summer of 2024, which will push back the grant timeline by one year. During the summer of 2023, we made traction of the web app development and hosted farm tours with the software developer. A mobile version of the web app prototype can be viewed here, and a diagram of the information processing workflow for the back end of the app can be seen here

Any opinions, findings, conclusions, or recommendations expressed in this publication are those of the author(s) and do not necessarily reflect the view of the U.S. Department of Agriculture or SARE.