National Electrofishing Programme for Scotland (NEPS) 2023: status of juvenile Atlantic salmon and brown trout populations

Methods

NEPS 2023 survey design including sample frame, strata and site allocations

NEPS surveys were undertaken in 2018, 2019, 2021 (Malcolm et al., 2019b, 2020, 2023) and 2023. The same survey design was used in 2018 and 2019; however a larger than anticipated number of oversamples necessitated a new survey design in 2021. To address several logistical challenges it was decided that the NEPS 2021 survey design would operate for a single year, whilst further work was undertaken to design a longer-term multi-year survey. The NEPS 2023 survey design was the result of this work, adding new strata to provide increased flexibility to address national (Site Condition Monitoring) and local reporting requirements, while also improving the representation of the target population (i.e. accessible wadable reaches in rivers with salmon fisheries) through revisions to the sample frame (i.e. the digital representation of rivers where samples are allocated).

The sample frame for the NEPS 2023 survey was updated from previous years with input from local fisheries managers (Figure 1). Edits included the addition of larger rivers where sampling was possible, and removal of rivers above newly identified migration barriers.

Figure 1. NEPS Survey sample frames. Blue lines denote where the sample frame was unchanged between 2018/2019 and subsequent surveys; green lines indicate where river segments were added and purple lines show where river segments were removed.

The spatial definition of strata was also refined from earlier surveys to simplify the monitoring landscape, streamline data analysis and reporting, and to provide flexibility in the allocation of sample numbers among years (Figure 2, Appendix 1). Specifically, relative to 2018/2019 (Malcolm et al., 2019b) new strata were created where 1. More than one fisheries management organisation collected data within a NEPS 2018/2019 region (e.g. Brora and Helmsdale) 2. An SAC was a small component of a larger NEPS 2018/2019 region (<75% by length) and there was a need to create flexibility in sample size to improve the accuracy of reporting (e.g. Caithness_Thurso_SAC) 3. There was an opportunity to incorporate detailed data from intensively studied index monitoring sites (Girnock, Baddoch, Gala, Laxford), or 4. There was a desire to support additional local monitoring or reporting requirements, and additional resources were available (e.g. to assess the efficacy of barrier removal on the Forth or the effects of genetic introgression on the Shin). The NEPS 2023 survey design had 54 separate strata that could be aggregated to match the 27 NEPS 2018/2019 geographic reporting regions.

Figure 2. NEPS Survey strata. Blue polygons denote strata that remain unchanged from 2018/2019 survey, purple polygons denote strata that were added to improve sample management where more than a single organisation operated within a NEPS region or where there were additional local reporting requirements and resource. Green polygons indicate where new strata were added for smaller SAC rivers within a NEPS region.

In common with previous surveys, the NEPS 2023 survey design is an unequal probability Generalised Random Tessellation Stratified (GRTS) survey design. Sample site selection was weighted towards rivers where higher juvenile salmon densities were expected based on the salmon benchmark model predictions from Malcolm et al (2019a). This increases the probability of sites falling in areas with higher salmon densities and is expected to improve the precision of salmon abundance estimates. A point density (minimum distance between sites) of 75m was specified to minimise the risk of overlapping electrofishing sites, which should be 100m2 (or 50m length for narrow streams) and could be micro-sited by moving upstream or downstream by 50m.

It was intended that the NEPS 2023 survey should be capable of operating over a cycle of at least nine surveys, with panel designs allowing for repeat sampling at the same sites (to improve trend detection) and new sites (to increase spatial coverage) each year. Once a survey has been designed it is not possible to add further sites so it is important to ensure that the total number of sites in the design is likely to meet requirements over the intended monitoring period. There is thus a need to ensure that defined sample requirements are supplemented by a proportionate number of replacement sites (oversamples) to address circumstances where sites could not be sampled (e.g. due to access or health and safety considerations). Additional sites can also be incorporated within the survey design to allow flexibility in sample numbers depending on funding and reporting priorities between years. The number of sites in the NEPS 2023 national survey (and individual strata) thus balanced scientific requirements, available resources, an expectation of oversample requirements based on previous surveys and practical considerations (e.g. the river length within each stratum).

In NEPS 2018/2019, 30 sites were allocated to each reporting region (which were also individual strata). Where additional strata were added in subsequent NEPS surveys, the regional sample numbers remained consistent, but sample numbers were spread between new strata in proportion to river length. Consistent with previous years, the NEPS 2023 survey included 810 sites that were centrally funded. However, additional resource from the Forth Rivers Trust, Kyle of Sutherland Fisheries Trust (above Shin Dam), Tweed Foundation (Gala), “Project Laxford”, and Marine Directorate (Girnock and Baddoch) increased the overall number of sites to 921. Although there was an intention to increase sample densities in the Tay region using local resources to provide improved individual assessments of the Tay, Earn and Eden this was not possible in 2023. Consequently the 30 centrally funded sites available for the Tay 2018/19 NEPS reporting region was allocated across the three new strata; Tay (20 samples), Tay_Eden (5 samples) and Tay_Earn (5 samples).

The survey design was generated using R version 4.2.0 (R Core Team, 2022) and spsurvey version 5.3.0 (Dumelle et al., 2023).

Electrofishing data and oversamples

Full details of the electrofishing sampling protocols for NEPS are provided elsewhere (Malcolm et al., 2019b). In brief, electrofishing was undertaken by local fisheries managers following standard operating procedures developed for NEPS. All electrofishing data were area delimited (the surveyed area was measured for each site) with a target date for sampling between 01 July and 30 September. Approximately a third of sites were fished using three passes, with the rest fished with a single pass. The effort expended on the first pass of the multi-pass electrofishing and the single pass electrofishing should be the same. Basic habitat information was recorded at each site, and water quality samples were collected for analysis at MD-FFL (Marine Directorate, Freshwater Fisheries Laboratory). Genetic samples were obtained from up to 30 salmon parr at all sites to assess levels of genetic introgression from farmed fish (Gilbey et al., 2021). Fish data were stored in the Marine Directorate Fish Observation (FishObs) database, making use of the FishObs Data Processing Utility (DPU) for data entry. Chemistry data were stored in the Marine Directorate (MD) Science, Evidence, Data and Digital (SEDD) Laboratory Information Management System (LIMS).

Oversamples are requested when a site cannot be sampled. Where this occurs, this is recorded alongside the cause (e.g. site too deep, no access permission). The next site on the survey site list is then provided as a replacement, maintaining the statistical integrity of the overall survey design. In 2023 Marine Directorate launched an online ‘NEPS Oversample Portal’, which allowed NEPS collaborators to independently request and download oversamples (without needing to contact Marine Directorate).

Generation of covariates for electrofishing sites

All locations require landscape covariates to fit and estimate capture probability and to determine benchmark densities (i.e. the densities expected for healthy salmonid populations). Landscape covariates are proxies for habitat that can be derived from spatial data and include; gradient, altitude, river distance to sea (RDS), upstream catchment area (UCA) and percentage of different land uses in the riparian zone. All covariates were calculated using an in-house R Package (FFLGIS) and associated R scripts. Detailed methods can be found in Jackson et al. (2017) and Malcolm et al. (2019a).

Estimating capture probability

Capture probability can vary between teams, regions and habitats, and over time. It is essential to account for this variability when estimating abundance and performance (population health). Capture probability was modelled for juvenile salmon and trout following the methods described by Millar et al. (2016) and Malcolm et al. (2019a), and was consistent with previous NEPS surveys (Malcolm et al., 2019b, 2020, 2023). In brief, NEPS 2023 multi-pass electrofishing data were combined with previously analysed multi-pass electrofishing data collected across Scotland between 1997- 2015 (Malcolm et al. 2019a), ad-hoc data collected between 2016 and 2017 and new data collected under NEPS in 2018, 2019 and 2021 (Malcolm et al., 2019, 2020, 2023). Capture probability was modelled as a logistic function of covariates representative of staff and equipment (Organisation-Team), fish size and behaviour (Lifestage and Electrofishing pass), time (Year and Day of the Year, DoY), habitat (e.g. Altitude, Upstream Catchment Area, UCA; River Distance to Sea, RDS; and Gradient), land use (Conifer, Deciduous and Mixed trees, Urban area), and geographical region (Hydrometric Area, HA). The term Organisation (as an indicator of staff and equipment) was divided into broad time periods (Organisation-Team) to reflect any major organisational changes. Where there were small sample sizes for a given Organisation-Team, these were grouped with teams in an adjacent region or time period.

In common with previously published capture probability models, model selection followed a step-up-down procedure starting from a large model that included a 3-way interaction between Species, Lifestage and Pass, a continuous smoother for Year, two level factor for Urban area (presence / absence) and smoothed main effects for the remaining continuous variables. The model scope (i.e. most complex model possible) allowed 4-way interactions between Species, Lifestage, Pass and the other covariates.

Estimating site-wise (observed) salmon and trout densities

Fish densities were estimated for each species, lifestage and electrofishing site visit following the methods described by Glover et al. (2018) and Malcolm et al. (2019a). In brief:

where the total fish count for each species / lifestage combination across all passes and

is the cumulative capture probability across all passes (in the example above 3 passes) and Pn denotes the fitted capture probability for pass n (where n can be pass 1,2,3). To ensure consistency with the benchmark models, wetted area (m2) measured at the time of electrofishing was used to represent the site area.

Calculating site-wise benchmark densities for salmon and trout

Salmon (Malcolm et al., 2019a) and trout benchmark densities (Malcolm et al., in prep) were calculated separately for fry and parr life stages and represent the densities that would be expected on average for a particular habitat. Benchmarks can be considered a target for healthy populations, which is similar in concept to meeting the “intrinsic habitat potential” (Burnett et al., 2007).

Benchmark densities for both species were calculated using GIS derived habitat proxies and habitat – abundance models. The national juvenile salmon density benchmark model reported by Malcolm et al. (2019a) was used to predict salmon densities. A new trout benchmark model, derived following similar procedures was used to predict trout benchmark densities (Malcolm et al., in prep.)

The salmon benchmark model predictors are Upstream Catchment Area, River Distance to Sea, Altitude, Percentage Conifer and a residual regional spatial effect (Hydrometric Area). Upstream catchment area has the greatest effect with a positive relationship on both fry and parr, whereby densities increase with upstream catchment area. However, it is necessary to cap upstream catchment areas at 250km2 in the prediction model to prevent unrealistically high estimates of benchmark abundance in 5th order rivers, which were not well represented in the original benchmark dataset. Salmon densities also increased with distance to sea. Salmon densities declined with increasing altitude and the percentage of conifer trees on the river banks. Reductions with altitude were greater for fry than parr. Finally, there was a regional spatial effect which resulted in lower salmon densities in the south than in the north. To ensure that the benchmark reflects healthy natural conditions, unimpacted by anthropogenic pressures, the purely spatial effects (which can reduce expectations in the urbanised central belt of Scotland) and negative effects of conifer woodland (related to commercial conifer plantations) were removed from benchmark predictions.

The trout benchmark was derived using the same modelling approaches outlined previously for salmon (Malcolm et al., 2019a) but were based on an extended multi-pass electrofishing dataset collected across Scotland between 1997 and 2017 by a wide range of organisations including Fisheries Boards, Trusts, Marine Directorate and Scottish Environment Protection Agency - SEPA (6245 visits to 2924 sites). Full details of the trout benchmark model are in the process of being published (Malcolm et al., in prep.).

The trout benchmark model predictors were Day of the Year, River Distance to Sea, Altitude, Upstream Catchment Area and a residual regional spatial effect. Benchmark densities were higher for trout fry than parr. Parr densities declined with distance to sea, increased with altitude, and exhibited a modal response with upstream catchment area, with highest densities in relatively small catchments. Fry densities showed a strong negative effect of altitude and upstream catchment area. Both fry and parr densities declined over time (day of the year). Having accounted for habitat covariates, trout densities were generally lower in the west and north and higher on the east of the country.

Scaling benchmark salmon and trout densities to larger spatial scales

Consistent with previous NEPS reports benchmarks at different spatial scales were calculated by obtaining a benchmark estimate of expected densities and river length for each of the digitised river segments in the sample frame. Benchmark densities were predicted for upstream and downstream river nodes. Where nodes had the same river order, the segment benchmark was the geometric mean of the two. Where the downstream node had a higher river order to the upstream node (e.g. a tributary entering a larger river) then upstream benchmark predictions were assigned to the edge to avoid over inflating benchmark estimates for the segment.

Regional benchmark estimates were thus calculated as follows:

where Edge Benchmark is the density estimate for each river segment (edge) in the DRN and Edge Length is the length of each line feature (m). Consistent with site-wise estimates of the benchmark, upstream catchment area was capped at 250km2 when making predictions.

Scaling site-wise observed densities to larger spatial scales

The R package "spsurvey" was used to analyse the NEPS data. Sample weights were adjusted to reflect the final list of sampled locations (i.e. removing sites that could not be sampled and including replacement sites). Analysis was conducted using the "cont_analysis" function for continuous data for each survey year and for different spatial scales. The response variable was the site-wise observed wetted area densities (n m-2 wetted area). The "cont_analysis" function estimates the mean density (per unit length of sample frame) in each strata, together with associated two-sided 90% confidence intervals (i.e. one-sided 95% confidence intervals).

It is also possible to combine estimates from multiple strata (e.g. for national or regional estimates) and to post-stratify to smaller spatial scales of interest (e.g. individual catchments). This allows reporting at a range of spatial scales beyond the original survey strata. This approach was used to provide assessments of SAC rivers where these were not included as individual strata. It was also used to combine multiple strata to provide regional assessments (e.g. Dee_Girnock, Dee_Baddoch, Dee).

Importantly, such approaches also allowed integration of data from different surveys (e.g. NEPS 2018/2019, NEPS 2021 and NEPS 2023) which is necessary for inter-annual comparisons. The challenges of integrating different surveys are discussed in detail in Malcolm et al., (2023). In brief, wherever comparisons are made across NEPS surveys, sites within 5th order rivers and above barriers that were absent from earlier surveys were removed, and strata aggregated to the original NEPS Regions as necessary. With the exception of the NEPS 2023 strata reporting, all data from the Forth_DLCAAETA_Above_Barriers_2021 strata were excluded from analyses.

Differences in juvenile density across different river orders and within sub-catchments were also explored.

Assessments of status for salmon and trout (grades)

Since 2016, Scottish salmon rivers have received one of three conservation grades associated with an adult assessment method (Marine Scotland, 2020c). These grades are based on the probability of meeting a spatially varying egg deposition target indicative of maximum sustainable yield (Conservation Limit). Results are averaged over a 5-year period to prevent any single poor year from bringing down the status of the river (Marine Scotland, 2020c). The grades are associated with particular management advice (below). Importantly Grade 3 rivers (the poorest grading) are associated with compulsory catch and release and preclude the killing of salmon.

• Grade 1: Exploitation is sustainable therefore no additional management action is currently required. This recognises the effectiveness of existing non-statutory local management interventions.
• Grade 2: Management action is necessary to reduce exploitation: catch and release should be promoted strongly in the first instance. The need for mandatory catch and release will be reviewed annually.
• Grade 3: Exploitation is unsustainable therefore management actions required to reduce exploitation for 1 year i.e. mandatory catch and release (all methods).

It is possible to obtain similar status assessments for fry and parr (for both salmon and trout) by comparing the regional estimates of mean salmon density, obtained from the GRTS sampling, with the benchmark regional densities (Malcolm et al., 2019a; Malcolm et al., in prep) to assess how likely it is that a particular area meets its benchmark. The classification method for NEPS 2023 remains consistent with the method reported in NEPS 2021 and is summarised below.

• Grade 1: The mean density estimate exceeds the benchmark and the lower one-sided 95% confidence interval does not include zero.
• Grade 2: The mean density estimate exceeds the benchmark but the lower one-sided 95% confidence interval includes zero, or the mean density estimate exceeds 50% of the benchmark and the upper one-sided 95% confidence limit includes the benchmark.
• Grade 3: The mean density estimate exceeds 50% of the benchmark but the upper 1-sided 95% confidence limit is below the benchmark, or the mean density estimate is less than 50% of the benchmark.

Figure 3. Theoretical scenarios under which an area would be classified as Grade 1, 2 or 3. Green squares denote the benchmark and black circles the mean observed density with associated confidence intervals.

The NEPS categories reflect both the pointwise estimates of density and uncertainty, relative to the Benchmark. Grade 1 indicates that a region is likely to be healthy with a reasonable degree of confidence. Grade 2 indicates that a region may be meeting the benchmark, but with lower confidence. Grade 3 indicates with a reasonable level of confidence that a region is not meeting its benchmark.

The grades for the two lifestages are then combined to provide a single juvenile assessment grade for each year. The combined status favours the better of the two lifestage assessments, reflecting the ability of populations to rebound from a single year of poor recruitment. Strong evidence is required that both lifestages are failing to meet the benchmark before obtaining an overall Grade 3 categorisation (Figure 4).

Figure 4. Matrix showing the rule-based system for generating an overall juvenile status assessment (grading) from individual lifestage assessments. Fry grades run horizontally, parr grades run vertically. Blue denotes Grade 1, grey denotes Grade 2 and orange Grade 3, these colours are used throughout the report.

Assessing trends in juvenile abundance in rivers with contrasting NEPS grades

The NEPS benchmark was intended to be broadly representative of “intrinsic habitat potential” (Burnett et al., 2007) or saturated habitat. This should be broadly consistent with the maximum production of juveniles from a given habitat, an important target for recreational fisheries or conservation ecology (Thorstad et al., 2020) where commercial harvest is not the primary management objective. Where ova deposition is adequate to ensure freshwater habitats are saturated, juvenile densities should be relatively constant over time, varying only due to density independent processes (e.g. effects of hydroclimatic extremes). Similarly, juvenile densities (recruits) should be independent of adult numbers (stock), as density dependent processes in freshwater would set an upper limit to production (Glover et al., 2020). These characteristics should be indicative of NEPS Grade 1 rivers. In contrast, where juvenile densities decline over time, or vary with adult abundance, this indicates that juvenile populations are below freshwater carrying capacity and are consequently more consistent with NEPS Grades 2 or 3.

Except for a few detailed long-term monitoring sites (e.g. Girnock Burn, Glover et al., 2018) there are very few high-quality long-term electrofishing datasets extending over more than a couple of decades. Even where longer-term electrofishing datasets exist, they are often patchy in terms of spatial coverage and highly variable in terms of sampling effort among years (Malcolm et al., 2019). These data limitations can make it hard to detect catchment or regional trends in juvenile salmonid abundance from historical data. The rivers Tweed, Spey, and to a lesser extent the Aberdeenshire Dee have some of the better catchment-scale electrofishing data coverage for Scotland, with multi-pass electrofishing data extending back to the late 1990s, and substantial electrofishing programmes in at least some years during the 2000’s (Appendices 5-7). These rivers have different NEPS Grade profiles. Except for 2021, the Spey had a Grade 1 classification across NEPS surveys. In contrast, the Tweed and Aberdeenshire Dee have lower Grades in the recent surveys.

Two sets of generalised additive mixed models (GAMMs) were fitted to juvenile density data for each of the rivers to investigate whether there was 1. evidence of temporal trends and 2. evidence of relationships between juvenile density and rod catch. Because of the potential for ad-hoc sampling strategies to generate temporal trends from changes in the spatial pattern of sampling (e.g. a move from sampling smaller to larger rivers, or from sampling upstream to downstream locations), habitat proxies were included as potential explanatory variables in each set of models. Both sets of models assumed a Poisson distribution and a log link.

In the first set of spatio-temporal models salmon counts were modelled as a function of Lifestage (Fry or Parr), CohortYear (a continuous term lagged by a year for parr), Day of Year (DoY) and a range of spatial habitat covariates (i.e. Altitude, River Distance to Sea, Upstream Catchment Area and percentage landuse in a buffer around the site) explored in previous benchmark modelling exercises (Malcolm et al., 2019). An offset allowed for differences in fished area and capture probability. Random effects were included for Lifestage within Site Visit, Lifestage within Site and Lifestage within FCohortYear (a factor level for each Cohort Year). A step-up-down model selection was then undertaken starting from a large model where all continuous variables were fitted as smooth effects. At each stage of model selection it was possible to drop interaction terms between Lifestage and the continuous variables, replace smooth functions of the continuous variables with linear effects and drop the main effects of categorical and continuous variables (provided they were not involved in any interactions or expressed as smooth functions). It was also possible to introduce interactions between the continuous variables and Lifestage to allow for different relationships for fry and parr. Model selection was based on BIC, with the sample size taken to be the number of Site-Visits. This model selection criteria places an emphasis on identifying strongly influential spatial covariates, but over-penalises temporal covariates. Consequently, at the end of the model selection process, CohortYear was added back into the final model (if absent) and the significance of this term was tested to determine if there was evidence of temporal trends.

The second set of models explored whether any observed temporal variability in juvenile abundance was related to temporal variability in adult numbers using rod catch as a proxy of adult abundance. This modelling process was the same as above, but included lagged rod catch (-1 year for fry and -2 years for parr) as an additional explanatory variable. Where rod catch (Rod) or CohortYear were absent from the final model, they were added one at a time and the significance of the terms assessed.

In the case of the River Dee, the number of electrofishing observations from the Girnock and Baddoch index monitoring sites could strongly influence the final models. Consequently, the same set of models were also fitted excluding data from these sub-catchments.

All models were fitted in R version 4.3.1 (R Core Team, 2023) using gamm4 version 0.2-6 (Wood and Schiepl, 2020).

Assessing the status of Special Areas of Conservation (SACs)

In this report, the status of SACs is only reported where a minimum of five samples were obtained within each of the NEPS sampling years. The assessed SAC rivers are; Bladnoch, Dee, Endrick Water, Moriston, Naver and Mallart Water, Oykel, South Esk, Spey, Tay, Thurso and Tweed. The remaining SAC rivers that are not reported are; Berriedale and Langwell Waters, Borgie, Langavat, Little Gruinard River, North Harris, Teith. Alternative methods would be required to report on these rivers until such time as improved datasets are obtained.

The methods proposed in this report are an evolution of the methods outlined previously by Malcolm et al., (2023). In brief, SACs characterised by an overall (fry and parr) NEPS Grade of 3 in any survey year would be designated as "Unfavourable". Where no salmon were observed an SAC would be classified as “Destroyed”. Any other combinations of NEPS Grades would be designated “Favourable”.

The second component of site condition monitoring requires an assessment of trends in abundance. Prior to NEPS 2018 there were no formal (statistical) juvenile survey designs from which to obtain spatially balanced and unbiased assessment data, and there have only been four NEPS surveys to date. Furthermore, there have been changes in the sample frame between NEPS years which affect the benchmark densities. In such circumstances a straightforward comparison of abundance over time could be misleading and post-stratification (e.g. to river orders 2-4) would constrain the available data. Consequently, trends were assessed based on performance against benchmark, where the benchmark varies each year to accommodate changes in the sample frame. Given the limited number of surveys, no formal modelling was undertaken to assess trends. Instead, a simple and pragmatic rule-based system was used to report trends in status whereby the performance against benchmark (% of benchmark) results for each lifestage and year were compared to the proceeding NEPS year. To be considered an increasing trend each result must be higher than the previous year (e.g. 2023 is higher than 2021, 2021 is higher than 2019 and 2019 is higher than 2018) and the opposite for a reducing trend. Any inconsistency in the direction of results between years would be considered unchanging. These trends of increasing, reducing and unchanging salmon abundance could then be designated as “Recovering”, “Declining” and “No change” respectively for the purposes of SAC reporting. Where trends were different between fry and parr the designation of “No change” was applied.

Comparing abundance indicators: mean juvenile salmon density, NEPS grades and rod catch

Rod catch data are spatially extensive for Scotland, published annually as official statistics, and provide a useful proxy of adult salmon abundance in many circumstances (Thorley et al., 2005). However, they are affected by spatial and temporal variability in fisheries practices and exploitation rates (Gurney et al., 2015, Gregory et al., 2023). Juvenile assessment methods are catch independent but rely on wading and electrofishing, where it is not possible to sample the whole river system. Nevertheless, the assumption is that the areas surveyed for juvenile assessment are indicative of the status of the wider river system. Confidence in the reliability of abundance indicators is enhanced where spatio-temporal patterns of abundance are broadly coherent. However, comparisons between abundance indicators can also be informative in determining where juvenile recruitment data reside in a stock-recruitment context as juvenile densities would be expected to plateau at high stock levels where competition in freshwater controls juvenile production.

Estimates of mean salmon and trout, fry and parr densities from NEPS surveys were compared to rod catches at national and regional scales. Regional rod catches were scaled by accessible wetted areas to provide a rod catch density (catch per unit area). Fry and parr data were lagged back to spawner years for comparison with catch data. To simplify analysis, parr densities were assumed to be dominated by 1+ parr (i.e. in their second year in freshwater). Ricker stock-recruitment curves, which are commonly used in fisheries assessment, were fitted to the regional data and overlain to aid visualisation. Data points were coloured by NEPS grade to determine if Grade 1 observations were generally consistent with maximum juvenile production.

Spatial variability in the demographic (age) characteristics of juvenile salmon: consequences for status assessment

Where salmon grow quickly, there is the potential for a proportion of parr to emigrate in the spring of their second year in freshwater (1+), prior to the summer electrofishing census. If the proportion of early migrating fish changes over time (i.e. the fish grow more rapidly), and is independent of competition (density), then this could affect assessments of juvenile status. Specifically, it may be increasingly challenging to meet the benchmark with fewer parr observed in the summer, and result in a lower NEPS grade than expected based on historically observed data.

To determine the demographic characteristics of salmonid populations, scale samples were taken from up to 50 salmon and trout parr in each electrofishing pass and site visit as part of the NEPS standard protocols. This typically results in scale samples being taken from all parr except at the most productive of sites. Work to read these scale samples is ongoing. However, provisional data are now available for salmon parr sampled during NEPS 2018. The potential for a large proportion of early migrating parr was inferred by plotting the percentage of salmon parr that were older than one year old (>1+) for each site, the assumption being that the presence of older parr indicates slower growth and that early migration, prior to the summer census would be less likely.

Relationships between mean abundance and prevalence (presence / absence) of different species and lifestage combinations

If juvenile salmonid densities at individual sites (and habitats) decline in proportion to the overall population size, then the number of sites where fish were absent would be expected to increase as mean abundance declines. In this context, the proportion of sites with fish, could provide a valuable additional indicator of fish population health in addition to mean abundance.

The "cat_analysis" function for categorical data from the R package "spsurvey" (Drumelle et al., 2023) was used to explore the number of sites without fish, of a particular species and lifestage, in each year. The categorical variables were site-wise presence / absence groupings, for each species and lifestage, where juvenile densities > 0 were coded as ‘Present’ and juvenile densities of 0 as ‘Absent’. The "cat_analysis" function estimates the proportion of river length where fish (of a given species and lifestage) are present in each strata (occupancy), together with associated uncertainty.

Occupancy was then plotted against the mean density estimate for the stratum, along with associated confidence intervals. To illustrate the relationships between the proportion of sites where fish were present and mean density, a generalised linear model, with binomial form (logistic regression) was fitted to the data, with prevalence as the response variable and log density as the predictor. Response relationships were allowed to vary by species and lifestage, effectively fitting a separate model for each species and lifestage combination.

Spatial variability in water quality: assessing pressures on salmonids

Water quality samples were obtained at the time of electrofishing and returned to MD-FFL for analysis. A broad suite of determinands were measured including major cations (sodium: Na, potassium: K, magnesium: Mg, calcium: Ca), anions (sulphate: SO4, Chloride: Cl, nitrate: NO3-N), pH, alkalinity, electrical conductivity (EC), dissolved organic carbon (DOC), phosphate (PO4-P), total dissolved phosphorous (TP), total dissolved nitrogen (TN), Nitrite (NO2-N), total ammonia (NH4-N), Silica (Si). The spatial variability of determinands was then mapped to help understand large scale spatial variability in water quality. There was high correlation between some determinands e.g. alkalinity was strongly correlated with pH, Conductivity, Ca and Mg. Consequently, the list of determinands that are illustrated was reduced to a subset that were relatively poorly correlated (Pearson Correlation Coefficient < 0.8), specifically: TN, NO3-N, NO2-N, NH4-N, TP, PO4-P, DOC, Si, pH, K, Cl, SO4.