Date of Award
Doctor of Philosophy
School of Science
Associate Professor Ute Mueller
Dr Stephen M Taylor
Digital camera monitoring has evolved as an active application-oriented scheme to help address questions in areas such as fisheries, ecology, computer vision, artificial intelligence, and criminology. In recreational fisheries research, digital camera monitoring has become a viable option for probability-based survey methods, and is also used for corroborative and validation purposes. In comparison to onsite surveys (e.g. boat ramp surveys), digital cameras provide a cost-effective method of monitoring boating activity and fishing effort, including night-time fishing activities. However, there are challenges in the use of digital camera monitoring that need to be resolved. Notably, missing data problems and the cost of data interpretation are among the most pertinent. This study provides relevant statistical support to address these challenges of digital camera monitoring of boating effort, to improve its utility to enhance recreational fisheries management in Western Australia and elsewhere, with capacity to extend to other areas of application.
Digital cameras can provide continuous recordings of boating and other recreational fishing activities; however, interruptions of camera operations can lead to significant gaps within the data. To fill these gaps, some climatic and other temporal classification variables were considered as predictors of boating effort (defined as number of powerboat launches and retrievals). A generalized linear mixed effect model built on fully-conditional specification multiple imputation framework was considered to fill in the gaps in the camera dataset. Specifically, the zero-inflated Poisson model was found to satisfactorily impute plausible values for missing observations for varied durations of outages in the digital camera monitoring data of recreational boating effort.
Additional modelling options were explored to guide both short- and long-term forecasting of boating activity and to support management decisions in monitoring recreational fisheries. Autoregressive conditional Poisson (ACP) and integer-valued autoregressive (INAR) models were identified as useful time series models for predicting short-term behaviour of such data. In Western Australia, digital camera monitoring data that coincide with 12-month state-wide boat-based surveys (now conducted on a triennial basis) have been read but the periods between the surveys have not been read. A Bayesian regression framework was applied to describe the temporal distribution of recreational boating effort using climatic and temporally classified variables to help construct data for such missing periods. This can potentially provide a useful cost-saving alternative of obtaining continuous time series data on boating effort.
Finally, data from digital camera monitoring are often manually interpreted and the associated cost can be substantial, especially if multiple sites are involved. Empirical support for low-level monitoring schemes for digital camera has been provided. It was found that manual interpretation of camera footage for 40% of the days within a year can be deemed as an adequate level of sampling effort to obtain unbiased, precise and accurate estimates to meet broad management objectives. A well-balanced low-level monitoring scheme will ultimately reduce the cost of manual interpretation and produce unbiased estimates of recreational fishing indexes from digital camera surveys.
Access to Chapters 2, 4, 5, 6 and 7 of this thesis is not available.
Afrifa-Yamoah, E. (2021). Imputation, modelling and optimal sampling design for digital camera data in recreational fisheries monitoring. https://ro.ecu.edu.au/theses/2387