Abstract
CREEM were commissioned to review statistical modelling methods currently used in the marine renewables industry. We also compared the performance of these methods and appropriate alternatives not presently in use.
An extensive review of the available literature was undertaken and three modelling methods (GAMs, GAMMs and CReSS) were identified for the methodological comparison. This comparison was carried out using simulated scenarios based on off-shore and near-shore data collected from existing marine renewable developments. In particular, off-shore and near-shore data were each generated with: no-change post-impact, 30 % post-impact decrease and post- impact redistribution scenarios.
The ability of CReSS, GAMs and GAMMs to recover genuine impact-related changes was examined. In addition to evaluation based around pre/post-impact changes, the relative performance of the methods at returning both accurate predictions (either pre or post impact) and realistic measures of precision about these predictions was also measured. Further, the ability of each method to correctly identify post-impact effects (if any) was also quantified (e.g. no-change post-impact, post-impact decreases and post-impact redistribution).
CReSS performed better than GAMs and GAMMs at successfully locating spatially explicit impact-related change. CReSS is also recommended for site characterisation because spatial predictions from this method showed fidelity to the simulated animal distributions. Uncertainty in the predicted spatial distribution of animals was close to its nominal (95% coverage) level.
This document contains a discussion about the issues involved with the data collection process and in particular, the differences in survey methods across platforms (e.g. boat, plane, vantage point). Related platform-based issues about the observation process (and associated imperfect detection) for the data collection, and the associated need to correct observed counts prior to input for analysis are also outlined. A description of the methods comparison process and the associated results then follow, along with recommendations based on the results contained therein. Two worked examples (based on the recommended approach) provide not only a suggested approach to analysis, but also offer signposts that consumers of analyses can use to assess reliability of the results.
In addition to this guidance document we have produced a literature review related to statistical modelling of animal distribution in the UK marine renewable industry. Based on the evaluation presented in the guidance document, we have also produced software for the assessment of animal distribution.