Lobster Imagery meeting minutes
Meeting Minutes, Sept 9 2019 4pm-6pm Harbor Classroom.
In attendance: James Voss, Jason Robinson, Julia Blanton, Gary Burke, Shane Robinson, John Colgate, Melissa Mahan, Joe Garrigan, Mike McCorkle, Jono Wilson, Frank Hurd, Chris Voss, Mike Nelson, Kim Selkoe, Adrian Stimson
CFSB has been collaborating with TNC to help them beta test a new tool based on taking images of product at the docks with a set up that can be used to size the animal. See the end of this post for more information about the context for this meeting.
All images were collected at the docks. All data belongs to the fisherman whose catch was imaged, and CFSB, and cannot be used by TNC or any other entity with out express permission of CFSB.The meeting opened with Frank Hurd of The Nature Conservancy giving a powerpoint presentation about the tool development. The minutes were re-organized somewhat by topic to make them easier to follow.
The motivation for this tool is to more easily collect monitoring data on size-frequency distribution based on length of catch. Length based on some past measure of spawning potential ratio (SPR) is what most FMPs are based on. So using length distribution now is better than average weight, in that it is consistent with the metrics in the FMP.
The tool includes a web-based app for uploading and analyzing length-frequency data. This system collects and stores photos and estimates a standardized length from each image. The distribution of lengths from all the photos collected gives the length frequency distribution of the catch.
This collection and storage system speeds up the analysis process. Once a bunch of data has been inputted, there are options for immediately viewing different graphs and even comparing to a harvest control rule that can be inputted.
TNC has been beta testing this system, called Poseidon, by collecting several thousand images for north coast abalone, pinshell scallop in Baja Mexico, recreational finfish in So. California, and spiny lobster in Santa Barbara Channel.
The tool is being developed for all fisheries. In other parts of the world, fishermen are coming to TNC and asking for access to tools like this. The motivation to do this project was based on other fisheries wanting/needing this type of data. N. Coast Ab and Pin Scallop asked for this tool.
Lobster was then added in because a lobster carapace is a difficult shape with unclear boundaries which makes it very challenging fro the computer to size, and so it will test the limits of the tools utility. Jono and Frank also thought it would not be controversial because the fishery is in such good shape there is low risk that the new data would inadvertently lead to a new management decision that would be detrimental to the fishermen. Santa Barbara is a convenient place to do the testing because of the relationships established.
All the images are taken dockside.
At this point, the computer is struggling to accurately find the boundaries of the carapace from the images and someone needs to go in and correct the line drawn from tip to tip by hand.
Also, because of the different size distribution of males and females, the lobster in the images should be identified as male or female. This wasn’t done during image collection, but it is now being done by looking at the leg morphology on the images.
Once that is completed, TNC will share the size distribution results from the data collection so far. This should be in the next few weeks.
The tool is not done and still needs to be proven.
Comments and Questions (Audience in italics, TNC reps not)
We need to see the data before we can continue to support.
There’s the coast fishery, the MPA line fishery, different trap designs catch different sizes, and different regions and times of year. You need to consider all this in generating an accurate SPR.
Could be just as efficient to size by hand given the problems with the tool. The technology exists (Facebook, etc.) to read an image really detailed and accurately. Can you access better technology?
TNC has already sunk a lot of money into this project. They are still able to put additional budget into acquiring better technology, but haven’t yet found something they can afford. Hoping to work out a deal soon.
Helping this effort is consistent with the CFSB mission statement to support collaborative research, better fisheries management and tools for co-management.
The folks who participate in helping develop the tool want fisheries to managed with good information not bad information.
Lobster is currently monitored as average size of a load. Data is coming from the landing receipt based on number and weight. You can’t extrapolate anything about the variation in lobster size from average weight. Switch to size frequency is good.
At some point the State will want to get a new estimate of lobster SPR. By having the size structure you know if a lot of little or big lobster are being caught, not the single average size. If you have more accurate high resolution data, you can make better management decisions.
Federal groundfish has voluntary fish measurement. This could be similar.
It would take a lot of motivation to choose to implement that in this fishery.
What are chances of a barely legal lobster erroneously being categorized as a short due to an error in the sizing on the computer? Who would be on the hook for getting the data right?
Those are good points that should be addressed in the tool development. The original image will always be kept so that even if the computer inaccurately estimates a lobster as shorter than it is, you can return to the image and see that it was not. Some level of acceptable error in the lengths of images would need to be established from seeing how a barely legal sized lobster could be misconstrued as short in the photo.
The FMP was $4 million and only resulted in the trap limit. New spending on this fishery isn’t worth it.
What did Chris do pre-season last year? Was this part of it?
Chris was involved in a short pre-season lobster study unrelated to this that did provide some weight-length conversions. The study was focused on MPA production. Lobster in MPAs may have a large fraction of the spawning potential of the whole stock (or just region?).
What is the Data ownership agreement?
CFSB owns the data and TNC cannot use it outside the agreed upon internal analyses without prior agreement. We have this in a contract.
How did they sort out data ownership in the other fisheries?
For rec abalone and rec groundfish fisheries they dove themselves and worked with a small number of divers/anglers that agreed to the terms. So they are in close contact with the individuals involved in data ownership.
Using this tool in lobster fishery would help us get a look at what a healthy resource looks like. To demonstrate to other fisheries.
How accurately does length represent spawning potential? Shortlived species might have more variable growth rates.
We should consider the contexts in which CFSB fishermen could get some benefit from this since we have put time and effort into helping develop it.
When the tool is better developed, maybe the rock crab fishermen would be interested since the fishery may need some new strategies.
Rock crab fishery could get really bad news from this technology.
In the case of black cod, the trawl stock assessments are not reflecting the fishermen’s catch.
How could this tool benefit us? Use it inside and outside the MPAs and show how much spawning biomass is in the MPAs and see that the harvest is small.
Using it with cowcod and abalone would help reopen these fisheries. Seems like a better focus than fisheries that are already well managed.
There could be value in having our SPR calculation in hand and in case the managers try to paint a negative picture and we could counter them.
SPR is critical info in fishery re-opening - Rockfish are not considered overfished anymore because of the SPR and biomass calculations.
Ridgeback – could be a useful tool to bring high quality, validated data to the managers about change in size over time and in different areas.
Meeting was adjourned at 6pm.
Lobster Imagery Project Summary
The Nature Conservancy (TNC) is working to improve the management of California fisheries by facilitating opportunities for fishermen to play active roles in data collection and decision-making processes. One such area of investment is the development of digital data collection tools that can provide more accurate information in a timely manner. The Conservancy has recently developed an image recognition technology (Poseidon) that can be used on board vessels or at the dock to take photos of animals and automatically record the size/length. The Conservancy is looking to pilot this technology with fishermen to explore its utility and functionality.
Size data is one of the indicators for management specified in the FMP. The managers are supposed to use size data, but currently they are using average weight as a proxy for size data, which is flawed and inaccurate. With size, they can get an estimate of reproductive potential of the overall population, which feeds into estimating the maximum sustainable yield. With more accurate size data, managers can be more confident in the calculations of maximum sustainable yield, and this allows them to be less ‘conservative’ and risk-averse when it comes to managing fishing effort for maximum sustainable yield.
The scope of this collaboration is to demonstrate that this technology has the potential to generate an estimate of the stock’s Spawning Potential Ratio.
In Phase 1, conducted in Spring of 2018, CFSB tested out image collection equipment and protocols, built a library of 3000 lobster images. These images had many redundancies, and were useful only for training a computer algorithm for detecting the carapace length of a lobster from the images.
In Phase 2, conducted in Winter and Spring of 2019, CFSB was asked to collect several thousand new, unique images demonstrate that image recognition technology can be used to generate a an estimate of spawning potential ratio in the California Spiny lobster fishery.