Predicting fish weight using photographic image analysis: a case study of broad whitefish in the lower Mackenzie River watershed

Abstract

Many small-scale fisheries are remote in nature, making data collection logistically difficult. Thus, there is a need for accessible solutions that address the data gaps present in these fisheries. One possible solution is to incorporate photography into community- or harvest-based monitoring frameworks and employ these images to estimate biological data. Here, we test this approach using łuk dagaii, or broad whitefish, Coregonus nasus (Pallus, 1776) in the Gwich'in Settlement Area, a remote region in the Mackenzie River system in Canada's Northwest Territories. We used photographs taken by Gwich'in collaborators using a simple, standardized set-up to ask the question: how accurately can weight be estimated from a photo? Using random forest models based on morphometric photograph measurements as well as season and location of harvest, we predicted broad whitefish weight to within 13% of true weight (257 g, for fish weighing an average of 2036 g). The model predictions were well distributed in their residuals for most fish, though we discuss biases at low and high weights. Image analysis is a simple, low cost, and accessible method that may contribute to ongoing, community/harvest-based fishery data collection where fish length (measured) and weight (predicted) can be tracked through time.

Keywords

• image analysis
• community-based monitoring
• broad whitefish
• random forest analysis
• fisheries monitoring
• analyse d’image