Scottish Marine and Freshwater Science Vol 6 No 6: Development of a Model for Predicting Large Scale Spatio-Temporal Variability in Juvenile Fish Abundance from Electrofishing Data

Recommendations for Future Work

Models are only as good as the data that underpins them. This report identified a number of problems with the location data and the generation of covariates. The location of sites was often incorrect or too imprecise. This caused particular problems where grid references placed sites near both tributary and mainstem rivers. This could be resolved by providing maps of 'snapped' electrofishing sites back to data providers and asking them to confirm the precise location of sampling which could improve model fits. There were also problems where the SEPA river line dataset did not match with the OS MasterMap dataset due to differences in the spatial scale of data capture. This resulted in erroneous estimates of channel width. It is possible that this could be resolved by snapping sites to the forthcoming OS rivers dataset or by introducing rule based quality controls on the distance that points could be snapped to rivers and the adjacency of different rivers. These options would require further investigation. It should also be considered whether the size of the buffers could be adjusted depending on channel width, such that standard "land areas" could be characterised and whether additional landuse datasets could provide improved information on agricultural landuse.

In addition to the issues of site location and covariates, there were a number of occasions where it appears that partial and unrepresentative electrofishing surveys were performed on larger rivers e.g. sampling a single braid or sampling only the peripheral areas. In the context of the current analysis these partial samples could introduce bias. Again this could be resolved by asking data providers to identify any site visits where electrofishing did not encompass the full channel and excluding these data from the final analysis.

Despite successful model application, a number of developments could potentially improve the ability of these models to characterise and predict spatio-temporal variability in fish abundance at the national scale. At present it has only been possible to fit models for single species and life stages. It is suggested that future work considers interactions between species and life stages and further develops capacity to fit multiple river catchments with the aim of producing a single spatial model for salmonids. Furthermore, there is a need for improved model diagnostics to check model fits and identify influential outliers.

Although the existing datasets were spatially extensive, there is poor replication between Organisation and HA (or other spatial metrics). It is therefore possible that the effects of e.g., equipment or methods are confounded with spatial effects. This could be further investigated if data providers were to fish outside of their geographic areas in locations with similar GIS characteristics. There was also poor data coverage in the case of certain unusual (and broadly correlated) covariate combinations and for larger rivers with a high UCA and river width. Strategic data collection, for example, in areas of high upstream catchment area with low distance to sea (and vice versa); and low channel width with low distance to sea would improve spatial models and help to separate the effects of these variables. Potentially valuable sites could be identified using approaches similar to those established for the Scottish River Temperature Monitoring Network ( SRTMN)

( http://www.gov.scot/Topics/marine/Salmon-Trout-Coarse/Freshwater/Monitoring/temperature).

The generation of covariates in ARC GIS was a major resource constraint in the current project. It is also anticipated as a constraint for future model development and predictions where there are potentially even greater data requirements. Increasingly sophisticated spatial tools are being developed in R which allow for rapid automated generation of certain covariates. It is therefore recommended that skills and knowledge are developed in this area.

In the current MasterMap dataset, any waterbodies <1m (urban) or 2m (rural) are characterised using line data. Previous studies ( e.g., Wyatt, 2005) have attempted to generate models of channel width from landscape covariates. Although not a priority, similar work could be undertaken for Scotland, with electrofishing site data being used to calibrate these models.

The development of a capture probability model allows for the integration of single and multi-pass electrofishing data on the assumption that they were both collected in a consistent way ( i.e., same effort, equipment and staff). Were single pass data to be included in future models, this would greatly increase the size and coverage of available covariates given the quantity of one-pass data collected by fisheries trusts. To ensure that the capture probability models remain valid, it is recommended that multi-pass electrofishing data is still collected by all data providers. Where there is only one-pass data available for particular catchments it is recommended that this is supplemented with multi-pass data from which to estimate capture probability. In addition, given the effect of Organisation on catch probability models, it is likely that further improvements could be made by incorporating information on fishing team and equipment used. It is recommended that the fished area is always recorded as there is no way of incorporating timed fishings.

Given the potential of biomass models to integrate across species and life stages, it would be useful if the sizes of fish (length) are recorded in future electrofishing work in addition to numbers. Finally, reliable estimates of juvenile ages, obtained via scale reading, would allow for the development of more complex age structured models which potentially indicate additional system resilience over and above numbers alone.

A substantial amount of resource was required to collate and harmonise data formats for the electrofishing data and also in allocating recorded sites to locations on rivers. Future projects would benefit greatly if these data were collected and stored using the same data format, preferably in a single database. MSS has found that the use of a single flexible database reduces the need for multiple extractions, followed by complex collation, and can allow for validation of site locations, removing some of the potential errors associated with snapping sites to rivers.

Finally, the selection of covariates in the current report was pragmatic given the need for large scale spatial coverage and the limited time available for the project. However, future work should consider the inclusion of metrics of hydrological impact, hydrochemistry and river temperature given the importance of these variables in controlling fish distribution and abundance. Due to the discrete nature of water quality observations, these data would need to be modelled using similar approaches to those reported here.