Title: Model-based calibration of fishery resource survey data as a change-of-support problem
Abstract: For fish abundance monitoring, standard field surveys may employ specialized data collection equipments (gears) and data processing protocols. For example, a survey team may deploy the type of underwater video camera that is already in its possession, after which the footage videos are processed to derive MaxN's --- relative abundance values that are only meaningful when compared within camera type but not between. As such, a major challenge arises when absolute abundance (true animal count) across a large spatial domain is of interest, whereby the spatial expanse requires regional survey teams for data collection, each team deploying its own type of camera. This is the case for the current US$12M study aimed to estimate, for the first time ever, the absolute abundance of the Greater Amberjack across the South Atlantic and Gulf of Mexico. In this talk, I discuss a small-scale calibration experiment (conducted as part of the continental-scale study) in which data from 2 to 4 different camera types were simultaneously deployed on each of 21 boat trips. Alongside each suite of deployed cameras was also an acoustic echosounder that recorded fish signals that may be regarded as the most representative measure of absolute abundance. I discuss our Bayesian modeling framework that can be used to derive calibration formulas to translate camera-specific MaxN to the absolute abundance scale; the conversion is to be accompanied by uncertainty estimates. The inferential context here is a wide-sense change-of-support problem: instead of reconciling data at various spatial scales, here we reconcile abundance data observed at various gear-specific scales, some being relative, and others, absolute. The operational product of this work is an R Shiny widget that takes a camera-based MaxN count as input, and produces a corresponding expected absolute abundance and associated uncertainty as output. The widget facilitates the conversion of a MaxN observed from a regional survey in which only one camera type is deployed. (arXiv article to appear.)
