Novel methods are required to improve knowledge on the activity of marine recreational fishers (which can be impactful); the stocks they prosecute, and the ecosystems the fishers and their quarry interact with. “Traditional” survey methods rely on complex randomised designs based on sound statistical sampling methods and are the “gold standard” for evidence collection. But these methods are costly and logistically complex to deliver, which partially explains the relatively poor understanding of marine recreational fisher activity in the majority of recreational fisheries, even in developed countries. This thesis examines two separate (but interlinked) approaches to enhance knowledge acquisition. Firstly, two separate methods are described which use the local ecological knowledge of fishers to describe proxies for the estimation of the spatial and temporal distribution of effort. These proxies are validated against the best available ground truth directed-survey data. One method exploits social media, which can pose unfamiliar ethical questions to ethical research boards and the peers of researchers who propose to use social media. Consequently a review of the ethical issues surrounding the use of social media as a source of scientific data for fisheries research is provided. The second approach automatically derives accurate estimates of a morphological measurement (total length) of the European sea bass (Dichentrachus labrax) under real survey conditions (i.e. with limited control of the camera and related paraphernalia). There were two aspects to the length estimation process; (i) images were corrected for distortion, and length estimates corrected for parallax effects; and (ii) machine vision (transfer learning using three pretrained regional convolutional neural networks) and a machine recognisable marker were used to detect European sea bass in images taken with different cameras and on different angling platforms. These detections, together with the methods validated in (i) provided accurate length estimates with a percent mean bias error of -0.1%.