Modelling the bioenergetics of salmon migration: Defra Project Code: SF0237, CEH Project Number: C02029

Executive Summary

Populations of Atlantic salmon have declined throughout the species’ range. The primary causes may vary among populations, but it appears that there has been a widespread decrease in survival during the marine phase, associated with reduced growth rates. It is therefore important for efforts to conserve salmon populations to understand how changes in oceanic conditions might affect growth and survival.

The aim of this report is to record the methodology, results and findings of a DEFRA funded project called “modelling the bioenergetics of salmon migration”. This project used a physiologically and physically-based numerical modelling approach to investigate the effects of changing oceanographic conditions on the growth and survival of migrating Atlantic salmon. A bioenergetic approach was taken, which seeks to assess survival and growth by quantifying the balance between energy gained from feeding and energy lost through maintenance, activity, digestion, food capture, growth, nitrogenous excretion and faeces.

Results from published work on salmon physiology were used to design and parameterise an individual-based bioenergetic model to predict growth and survival for post-smolt Atlantic salmon. A spatially and temporally explicit virtual marine world was then created. This allowed assessment of the effects of different patterns of ocean surface currents, sea surface temperature, and salmon prey availability on salmon growth and survival using the bioenergetic model. Data were used from the National Oceanography Centre’s ocean circulation model OCCAM to represent surface currents. Data from the Meteorological Office’s ocean-ecosystem model HadOCC were used to calculate a representation of prey availability. Observed historical data from the Hadley Centre’s HadSST data set and modelled predictions from the Hadley Centre’s couple ocean-atmosphere model HadCM3 were used to represent patterns of sea surface temperature. The particular future scenario that we used employed observed anthropogenic emissions to 1990 and the IS92a scenario thereafter.

We compared OCCAM simulated currents with data from the WOCE drifting buoy data set. A comparison of the simulated currents with data derived from drifters showed that the calculated currents reproduced many detailed circulation features of the surface circulation of the North Atlantic ocean. In order to reproduce the observed drifter trajectories the momentum in the model was mixed over a climatological mixed layer depth. Comparison of Eulerian velocities showed generally good correspondence between the direction of the drifters and the direction of the model currents. The speed of the model currents was generally underestimated. Data describing the concentrations of zooplankton and zooplankton lost to higher predators from the HadOCC model were used to create a representation of salmon prey availability. Comparisons between these representations and data on zooplankton availability from continuous zooplankton records showed good correspondence.

The sensitivity of length predicted by the bioenergetic model to parameters of the bioenergetic non-spatial functions was tested by varying parameters individually with combinations of water temperature and swimming speed. Modelled growth varied with water temperature and swimming speed in the manner expected from empirical studies, but was less than that achieved by wild post-smolt salmon. Growth was very sensitive to certain parameters of the main bioenergetic functions, i.e. those for maximum daily consumption, respiration and allocation of assimilated energy. Other parameters had moderate effects, whereas parameters associated with prey encounter rate and ingestion had little influence on growth, given the defined prey availability. Some of the sensitive parameters have a reliable empirical provenance, whereas appropriate values for others are less certain, owing to the species or circumstances from which they were derived. The studies needed to obtain more-reliable estimates are considered.

Spatially explicit sensitivity tests were used to investigate the spatial aspects of the full bioenergetic model. The model was used to assess how changes in smolt length, smolt year, date of entry to the ocean, location of feeding ground and home-river can affect growth and survival of Atlantic salmon. Our results indicated that salmon that left their home-rivers earlier in the year and as large smolts were likely to return as larger adults. Results indicated that salmon travelling towards more northerly feeding grounds did not grow as much as those travelling on shorter migrations. Results also showed that several of the model parameters did not significantly affect the model results. Overall, these sensitivity experiments suggested that our ability to model the affects of changing oceanographic conditions on salmon growth and survival are limited more by our lack of knowledge of salmon behaviour (i.e. feeding behaviour, swimming speeds and swimming directions) in the open ocean than a lack of knowledge of the physiological processes controlling growth.

The representations of surface currents were used to investigate the impact of ocean surface current patterns on the speed and timing of salmon migration routes during the first two months at sea. Trajectories of tagged and recaptured salmon that left the west coast of Ireland in 1996 and 1997 were simulated using rheotaxis as a direction-finding mechanism and a constant swimming speed of 0.2 m s-1. Simulated trajectories were significantly affected by the way in which surface currents were represented. When the mixed layer depth was used in the calculation of currents, 78% of the simulated trajectories passed within 40 km of the observed recapture location. This work indicated that rheotaxis is a possible direction-finding mechanism for migrating Atlantic salmon.

We further test the models ability to predict the positions of salmon migration trajectories using data on the positions of tagged salmon that were recaptured after several months as sea. Results showed that it is unlikely that these salmon would have obtained their recaptured locations by swimming in random directions. Comparisons of the positions simulated using different direction finding mechanisms showed that they can result in similar trajectories, and similar areas of the ocean being occupied. It is possible that, in reality, combinations of mechanisms are being employed. For example, salmon may be heading for cooler waters and using local currents beneficially at the same time. Rheotaxis, thermoregulation and prey searching were all mechanisms that may have been responsible for creating the patterns of migration that were inferred from trawl data.

Observed lengths of tagged salmon that were caught of the north east coast of the Faroes were compared with lengths calculated by the model. Lengths of the tagged and recaptured salmon caught in December were in the range 0.40-0.47 m. Calculated lengths in mid-December were in the range 0.35-0.43 m. Calculated lengths therefore did overlap with the observed data, however, on average the calculated lengths did underestimate lengths in comparison with the observations. This test of the model was undertaken using an observed data set with a very small sample size (n = 9). The observed data show that there was considerable variation in sizes of the recaptured salmon. This was even the case for salmon caught in the same month and in the same position. Furthermore, the lengths of the two salmon that were caught in March were 0.45 and 0.43 m. Calculated lengths for the corresponding number of days at sea were in the range 0.42 to 0.47 m. It would therefore appear that calculated lengths were slightly over predicted for these observed salmon. It was therefore concluded that this model scenario reproduced observations of salmon length given uncertainty in the estimate of the length distribution of the true population. Larger data sets on the condition of salmon at sea are required in order to more fully test this type of model.

Observations from the Rivers Wear, Dee and Frome suggest that typical 1SW fish returned to their home-rivers in mid-August after having grown to sizes ranging from 0.5 to 0.8 m with an average size of 0.65 m. The model was run using different direction finding mechanisms, and calculated lengths were compared with observed lengths. When forced to swim toward a representative final destination simulated salmon leaving the rivers Wear, Coquet, Dee and Frome grew to lengths of 0.54, 0.59, 0.59 and 0.72 m respectively. When forced to follow a set route, simulated salmon leaving the rivers Wear, Coquet, Dee and Frome grew to lengths of 0.60, 0.60, 0.59 and 0.66 m respectively. Calculated lengths for both sets of simulations overlapped with the observed range of lengths. Random effects were included in the model when a thermoregulation scenario was used as a direction finding mechanism. This meant that a distribution of lengths was calculated by the model. Therefore calculated mean and distribution of lengths could be compared with the observed data. Results indicated that, for this particular model scenario, the model replicated observed patterns of growth for the river Dee. The distribution of calculated lengths was narrower than the observed data and the model slightly underestimated the mean observed lengths.

Comparisons between observed and calculated lengths were hampered by several difficulties. The model was able to predict death of salmon caused by starvation. In reality the salmon may die because of a variety of reasons including starvation, disease and predation. Predation may be size dependent (i.e. smaller salmon may be more vulnerable to being preyed upon). This may partly explain why calculated lengths were smaller than the observed data. In these simulations a single date of entry to the ocean and smolt size were used as initial conditions to the model. In reality salmon smolts enter the ocean over a range of dates and after having grown to a range of sizes. Without paired data on the conditions of individual smolts and adults it is difficult to assess the causes of variations within the observed data.

We used different sources of information in order to represent the spatial and temporal patterns of oceanographic conditions experienced by the salmon. We first investigated the relative importance of inter-annual variations in prey availability, sea surface temperature and surface currents on salmon growth using the most accurate sources of information on surface currents, prey availability and temperature available. Results indicated that, when the salmon were given final destinations that represented feeding grounds to swim towards and not provided with any mechanism to adapt to local conditions, inter-annual changes in currents were capable of advecting them long distances and therefore affecting growth rates through changes in temperature and prey availability. When salmon followed these fixed trajectories, inter-annual variations in prey availability and temperature had far less of an influence on length in comparison with the position of the home-river and the position of the final destination. Results suggested that, under this migration scenario, changes in growth between years could be dominated by changes in temperature rather than prey availability, although changes in prey availability did influence salmon growth in high latitudes. This was because most salmon could feed to reach their maximum consumption in the majority of situations, except at high latitudes were less prey were available. Salmon trajectories that might have occurred if the salmon were reacting to local conditions were also simulated. We choose thermoregulation as an example of this type of behaviour. When given this adaptation mechanism that allowed to local conditions, inter-annual changes in currents had less influence on growth. Under these types of behaviour the affects on length of inter-annual variations in temperature dominated over inter-annual variations in prey or currents.

Having investigated the relative importance of inter-annual variations in temperature, prey availability and currents on salmon growth using shorter-term, more reliable data sets, we used longer-term, but possible less reliable, data on temperature and prey availability changes to investigate longer-term trends for the period 1970 to 2001. Results showed that, when thermoregulating migration strategies were used, inter-annual changes in temperature patterns are capable of causing changes in salmon growth and survival. Under this migration scenario, there was little variation in growth of salmon leaving British rivers during the 1970’s. Greater variations in predicted lengths were calculated for the 1980’s and 1990’s. Smolt years 1982, 1986 and 1993 were particularly noticeable for having reduced lengths. The North Atlantic Oscillation for these periods was mainly positive, indicating that changes in oceanographic conditions may have predictable impacts on salmon growth and survival. These were smolt years that coincided with predominantly positive NAO periods. The NAO is a major feature that affects oceanographic locations. The NAO is the major contributor to inter-annual and decadal variations of climate in the North Atlantic Ocean. In particular, it influences the patterns of sea surface temperature and surface currents, which in turn effect the distribution of salmon prey eg Calanus. Our results showed that negative NAO correlated with increased growth when thermoregulation was used as a direction finding mechanism.

The model was then used to predict the effects of inter-decadal variations in future SST on calculated lengths, depending on variations in prey availability and regardless of changes in local currents, when the simulated salmon followed the same fixed trajectories. The bioenergetic model, together with future predictions of ocean surface temperature, was used to simulate growth of salmon that might occur for the period 1990-2100 as a result of one scenario for future climate change (i.e. observed anthropogenic emissions to 1990 and the IS92a scenario thereafter). Results showed that, given fixed trajectories and steady swimming speeds, increases in temperature would lead to an increase in growth as the salmon experienced warmer temperatures that created more favourable conditions for growth regardless of prey availability. This was because more energy was gained because of the increases in consumed energy than was lost through increased respiration costs as a result of the increases in temperature. These results suggest that, if all other factors are equal, if salmon are following set routes they will not be adversely affected by climate change driven future increases in SST. The results suggest that, given this modelled behaviour, the only factors that could cause an adverse effect as a result of climate change is a decrease in prey availability or a change in current patterns, if these were caused by climate change.

A different result was found when thermoregulation was used to simulate the type of behaviour that might result when the salmon were attempting to adapt to the local conditions. Under this particular migration scenario increases in temperature lead to a slight decrease in growth. Many of the simulated salmon were able to adapt to the change in oceanographic conditions by seeking similar conditions elsewhere in the ocean. This suggests that, during their oceanic phase, salmon may be capable of adapting to future climate change. However, predicting the effects of climate change on salmon growth and survival is limited by a lack of detailed information on the behavioural mechanisms controlling salmon migration in the open ocean and on how freshwater and oceanic conditions might interact to affect salmon.

In order to simulate the growth of migrating salmon we used several assumes and different data sources. We used modelled data from several sources to create representations of future oceanographic conditions. These sources of information were the best available and represent approximations of the conditions that would be expected in the real world. This project concentrated on development of a bioenergetic model. However, we also conducted robust testing of the ocean current modelled data using observed patterns from drifting buoys. Modelled patterns of prey availability were compared with observed data on zooplankton. We used observed data on sea surface data for 1993-1999 and modelled data from the Hadley Centre for future prediction.

We assumed that entry to the ocean was on the 14th April for all salmon in all years and in all locations. We also assumed a uniform size on entry to the ocean. In reality size and date of entry to the ocean will be affected by climate change. For example, currently salmon from more northerly locations enter the ocean latter in the year in comparison with more southerly populations. This means that salmon populations could adapt to climate change during their freshwater phases as well as their marine phases. The model that was developed for this project only considered the marine phase of the salmon life-cycle. We compare results for two different direction-finding mechanisms. These two mechanisms represent two possible types of behaviour: one with no adaptation to local conditions and one with adaptation to local conditions. Other direction-finding mechanisms could be used in addition to those used here. In reality salmon migration mechanism may be more complicated than those used in these simulations.

Overall, results indicated that predicted changes in marine conditions resulting from future climate change are capable of causing a decline in the growth and survival of British salmon populations. However, our ability to predict the effects of climate change on salmon populations is limited by a lack of understanding of migration behaviour, and understanding of migration behaviour is restricted by the paucity and limited availability of observed data of salmon growth and movements in the open ocean.