Advancing Escapement Estimates: Camera-Based Systems

The Yukon and Kuskokwim Rivers have seen steep declines in chum and Chinook salmon over the last decade, and their management requires timely and accurate monitoring of escapements. Weirs and aerial surveys are the primary methods used to estimate escapement in the Kuskokwim River Basin. Environmental and fluvial changes have posed numerous challenges to both methods, making them impractical and prone to substantial errors and data gaps. Compounding these limitations is the relatively low number of rivers within the Kuskokwim basin that are adequately monitored. Camera-based systems present an alternative survey method to capture escapement in tributary streams. These systems collect both imagery and video to derive estimates of fish passage. Pioneered in Alaska and used on select systems by the Alaska Department of Fish and Game, these simple and low-cost systems can provide reliable population estimates with minimal upkeep and impact on the surrounding environment. Further, coupling camera-based systems with deep learning (i.e., AI) models may mitigate known data storage and efficiency issues, making their application useful to in-season management. We hypothesize that camera-based systems provide reliable and accurate population estimates to supplement current monitoring techniques. In addition, our goal is to train community-based weir crews, which will aid in expanding unmonitored waterways while providing workforce training in a rapidly developing technological field. We will test and deploy the system over two field seasons and compare estimates from current methods. We will create a deep learning model that works in parallel to the camera-based system to test if deep learning can improve efficiency and aid in enumerations. By creating training materials (videos, documents, etc..), we posit that camera-based systems can be used across the Kuskokwim River Basin to assist in the critical need for consistent and reliable monitoring of salmon populations.

This project will design a camera-based system for Kuskokwim streams and conduct a set of feasibility studies for application in fisheries population monitoring. Our training

program will be developed with local community and governing agencies to engage and train local communities in salmon population estimation . The central question we aim to answer is: What are the optimal camera-based system designs and counting protocols to accurately estimate the escapement of multi-species salmon runs in the Kuskokwim Basin? This question requires a series of steps to design, validate, and deploy a valid and cost-efficient monitoring technique. We ask:

1) What are the technical requirements for operating camera-based systems in the Kuskokwim Basin?

2) How accurate is species identification from images and/or video from camera-based systems?

3) How do camera-based systems compare to current methods?

4) Can an artificial intelligence model increase counting efficiency, and how do estimates compare to other methods?

5) Can we create a transferable camera-based system and training materials for communities and research entities such as ADF&G and USFWS for deployment across the Kuskokwim Basin?