Original Article Risk assessment of the environmental impact of Norwegian Atlantic salmon farming

(iii) Viral disease outbreaks (pancreas disease, infectious pancreatic necrosis, heart and skeletal muscle inﬂammation, and cardiomyopathy syndrome) in Norwegian salmon farming suggest extensive release of viruses in many areas. However, screening of wild salmonids revealed low to very low prevalence of the causal viruses. (iv) From (cid:2) 500 yearly investigations of local organic loading under ﬁsh farms, only 2% of them displayed unacceptable conditions in 2013. The risk of eutrophication and organic load beyond the production area of the farm is considered low. Despite severallimitations,especiallylimitedmonitoringdata,thisworkrepresentsoneoftheworld’sﬁrstriskassessmentofaquaculture.ThishasprovidedtheNorwegiangovernmentwiththebasisuponwhichtotakedecisionsforfurtherdevelopmentoftheNorwegianaquacultureindustry.


Background
The Atlantic salmon (Salmo salar L.) farming industry was first started in Norway in the early 1970s and has now grown to become one of the country's largest export industries by economic value. In addition to Atlantic salmon, which is by far the most significant species farmed in Norway, there are also commercial farming of rainbow trout [Oncorhynchus mykiss (Walbaum)] and other marine species such as Atlantic cod (Gadus morhua L.) and halibut [Hippoglossus hippoglossus (L.)]. In 2012, the production of Atlantic salmon and rainbow trout in Norway was 1 232 095 and 74 583 tons, respectively, and a total of 1006 marine farms was licensed. These farms are distributed along much of Norway's coastline.
Aquaculture of salmonids in Norway, and other countries where these species are farmed in significant numbers, is primarily based around the production of eggs and juveniles in freshwater facilities on land, combined with grow out of fish in open marine cages. During the last decades, technical standards for the production of aquaculture infrastructure has improved dramatically. However, the primary methods for cultivation of finfish have remained similar, with the size of sea cages (up to 160 m in circumference) and the number of stocked smolt (up to 200 000 individuals per cage) increasing.
The rapid expansion of the aquaculture industry, both in Norway and other regions where this form of open-cage production has increased, has not occurred without environmental challenges. However, although a significant body of evidence suggests various environmental impacts of aquaculture, the rapid expansion of this industry means that management guidelines and targets to address potential negative effects have generally not developed in association with the rapid expansion of the industry. Therefore, there is a need for more coordinated efforts to identify hazards related to open sea cage farming and evaluate environmental risks.

Risk analysis
Several approaches have been suggested and discussed for risk analysis of marine ecosystems and marine aquaculture activities (Anon., 2006(Anon., , 2010Nash, 2007;GESAMP, 2008;Samuel-Fitwi et al., 2012), and similar approaches of risk analysis and assessment have been adapted to animal welfare including welfare of farmed fish (e.g. EFSA, 2012). According to GESAMP (2008), a risk analysis should first involve hazard identification, then risk assessment of these hazards including the assessment of release, exposure, and consequences, followed by risk estimation/evaluation. The latter preferably related to politically defined thresholds of acceptability or level of protection. Subsequently, this can be followed up by appropriate risk management and appropriate risk communication.
A full risk analysis is based on the ability to quantify both the probability of a certain event and its consequences, but in biological systems it is normally very difficult to quantify these factors precisely. Hence, risk analyses in biological systems are often conducted using broad qualitative categories, by scoring the probability and consequences from low to high (e.g. GESAMP, 2008). This can in turn be based on some semi-quantitative assessment or on expert opinion as suggested by Anon. (2006).
In 2009, the Norwegian government established a set of environmental goals for sustainability in the "Strategy for an Environmentally Sustainable Norwegian Aquaculture Industry" (Anon., 2009b; Table 1). In response to this, the Institute of Marine Research, Norway, initiated a risk assessment of Norwegian salmon farming in 2010, and yearly since (Taranger et al., 2011a(Taranger et al., , b, 2012a(Taranger et al., , 2014. These risk assessments were based on identified hazards and specific endpoints or proxies related to environmental impacts of salmon farming ( Table 2). The endpoints/proxies were in turn derived from the governmental goals for environmental sustainability mentioned above. Moreover, evaluation thresholds for some of these endpoints/proxies (acceptance levels of impact or level of protection) were proposed (Taranger et al., 2012b), and subsequently used in the risk assessments in 2013 and 2014. Here, we describe the way in which these assessments have been conducted, the methodological limitations and challenges, as well as future needs to data and analytical tools.

Hazard identification
The first step in a risk assessment is to identify the most important hazards. A range of criteria for hazard identification was proposed by GESAMP (2008). This includes an analysis on how potential hazards relates to undesirable changes in the environment/ecosystem. To this end, potential hazards are evaluated for their possible severity, extent and duration of the change, either based on past experiences, analogue situations, or models. Some of the environmental challenges (i.e. hazards) identified include ecosystem and benthic community effects of organic loading and nutrients (Buschmann et al., 2006;Kutti et al., 2008;Bannister et al., 2014), transfer of parasites to native populations (Krkošek et al., 2005(Krkošek et al., , 2013aJackson et al., 2013;Skilbrei et al., 2013;Torrissen et al., 2013;Serra-Llinares et al., 2014), disease interactions Madhun et al., 2014a), and ecological (Jonsson and Jonsson, 2006) and genetic interactions with wild populations (Crozier, 1993;Clifford et al., 1998b;Skaala et al., 2006;Glover et al., 2012Glover et al., , 2013a. Based on the accumulating evidence of the severity, geographical extent and duration and/or reversibility of the various impacts related to open sea cage salmon farming in Norwegian coastal waters, we have based the current risk assessment on the following hazards: (i) genetic introgression of escaped farmed salmon into wild populations, (ii) impact of salmon lice (Lepeophtheirus salmonis) on wild salmonid populations, (iii) potential disease transfer from farmed salmon to wild salmonid populations, and (iv) local and regional impacts of organic load and nutrients from marine salmon farms.

Risk assessment
In the following chapter, we have considered the following elements of risk assessment; release, exposure, and consequences, in the following manner. Release assessment is defined as the number of farmed salmon escaping into the natural environment, both as reported and unreported numbers of escapees. Exposure assessment is defined as the physical mixing of farmed escaped salmon on the spawning grounds of wild populations, and the subsequent level of genetic introgression resulting from successful spawning. Finally, we have defined consequence assessment as the consequence of genetic introgression for both the short fitness consequences and the long evolutionary consequences on the native populations.
One of the challenges to the continued development of a sustainable aquaculture industry is containment, and each year, thousands or hundreds of thousands of farmed salmon escape into the natural environment in Norway ( Figure 1). Furthermore, the official statistics for numbers of escapees reported to the Norwegian Directorate of Fisheries underestimate the real number of escapees. This has been documented through extensive simulated release experiments and statistical modelling (Skilbrei et al., 2015), and is clearly supported by the fact that the legal authorities in Norway have implemented DNA tracing methods to identify the farm of origin for escapees where they have not been reported Glover, 2010;Zhang et al., 2013). While the majority of escapees disappear post-escape (Hansen, 2006;Skilbrei, 2010aSkilbrei, , b, 2013, each year, significant numbers of farmed salmon are nevertheless observed in rivers inhabited by wild populations (Fiske et al., 2006, Fiske, 2013. It is therefore considerable potential for genetic interaction between these escapees and native populations. The Atlantic salmon displays considerable population genetic structure throughout its native range. This variation is partitioned in a hierarchical manner, with the largest genetic differences being observed between populations located in different continents or countries, and the smallest differences being observed among neighbouring populations within regions (Ståhl, 1987;Taggart et al., 1995;Verspoor et al., 2005). This structure reflects various processes, for example recolonization patterns, genetic isolation by distance , and landscape features which modify population connectivity within regions (Dillane et al., 2008). In addition to differences in allele frequencies of molecular genetic markers, Atlantic salmon populations display different life history characteristics. While much of this phenotypic variation is environmentally caused, some of these differences are influenced by underlying genetic variation, and it is generally accepted that these differences potentially reflect adaptations to their native rivers (Taylor, 1991;Garcia de Leaniz et al., 2007;Fraser et al., 2011).
Norwegian farmed Atlantic salmon dominates global production, originates from over 40 Norwegian rivers, and has been subject to approximately ten or more generations of domestication selection (Gjedrem, 2010). Breeding programmes have successfully selected for fish that outgrow their wild counterparts multiple times under farming conditions (Glover et al., 2009;Solberg et al., 2013a, b). In addition to traits that have been directly selected for in the breeding programmes, genetic changes in non-targeted traits have also been observed, for example in predator awareness (Einum and Fleming, 1997), stress tolerance (Solberg et al., 2013a), and gene transcription (Roberge et al., 2006). In addition, decreased genetic variation, as revealed by molecular genetic markers (Norris et al., 1999;Skaala et al., 2004), and lower estimates of heritability for growth (Solberg et al., 2013a), has been observed in farmed populations. Reduced genetic variation in molecular genetic markers reflects founder effects and genetic drift, driven by limited farmed population sizes, while reduction in heritability for growth is likely to be a result of successful directional selection for this trait over multiple generations.
An early study in Ireland estimated introgression of farmed escaped salmon in a native population based upon escapement  Table 2. Identified hazards, process of concern, and endpoint of concern for goals 1 -3 for the future development of the Norwegian aquaculture industry as established by the Norwegian government in 2009.

Hazard
Process of concern Endpoint of concern Genetic interaction (Goal 2) Farmed escaped salmon successfully interbreed with wild salmon populations Changes observed in the genetic characteristics of wild salmon populations Salmon lice (Goal 1) Salmon lice from fish farming affects wild fish Salmon lice from fish farming significantly increase the mortality of wild salmonids Viral diseases (Goal 1) Disease transmission from fish farming affects wild fish Viral transmission from fish farming significantly increase the mortality of wild salmonids Discharges of organic material: (i) local effects (ii) regional effects (Goal 3) Emissions of organic materials to the surrounding environment (i) Unacceptable change in sediment chemistry and faunal communities in the production zone (ii) Significant change in bottom communities beyond the production zone-regional impact Discharges of nutrients: (i) local effects (ii) regional effects (Goal 3) Emissions of nutrients to the surrounding environment (i) Nutrients from fish farms results in local eutrophication (ii) Nutrients from fish farms results in regional eutrophication Risk assessment of the environmental impact 999 from a nearby farm (Clifford et al., 1998b). However, despite the fact the genetic changes in native wild populations have been observed in molecular genetic markers as a result of farmed salmon introgressing in Canadian (Bourret et al., 2011), Irish (Crozier, 1993(Crozier, , 2000Clifford et al., 1998a, b), and Norwegian rivers (Skaala et al., 2006;Glover et al., 2012), until a recent breakthrough in a study of 20 Norwegian rivers , the cumulative level of introgression of farmed salmon has not been calculated for any native Atlantic salmon population. This is due to the fact that estimation of cumulative introgression of farmed salmon is statistically challenging (Besnier et al., 2011).
In the Norwegian study of 20 rivers , a combination of Approximate Bayesian Computation, and genetic data for wild-historical, wild-contemporary and a diverse pool of farm samples that were genotyped for a set of collectively informative single-nucleotide polymorphic markers (Karlsson et al., 2011) was used to estimate introgression for a period of 2-4 decades. Overall, the study by Glover et al. (2013a) revealed less introgression of farmed Atlantic salmon in many Norwegian populations ( Figure 2) than may be expected based upon the reported numbers of escapees in these populations, and estimations from introgression models (Hindar et al., 2006). The authors concluded that spawning success of farmed escaped salmon has been generally low in many Norwegian rivers, a suggestion consistent with earlier estimates of spawning success in controlled experiments (Fleming et al., 1996(Fleming et al., , 2000. Nevertheless, results from the study demonstrated high levels of admixture in some native populations, and together with an earlier study using microsatellites, reported decreased genetic differentiation among populations over time . The latter of which is consistent with suggestions that widespread introgression of farmed salmon will lead to erosion of population genetic structure among native populations (Mork, 1991).
Estimating the genetic consequences of farmed salmon introgression on life history traits, population fitness, and long-term evolutionary capacity of wild populations is more challenging than estimating introgression. This is in part due to the fact that wild populations display large natural variation in, for example, marine survival, and at the same time are influenced by a wide range of anthropogenic factors (Parrish et al., 1998), which may potentially mask biological changes caused by introgression of farmed salmon. Nevertheless, comparative studies in Ireland and Norway have demonstrated additive genetic variation for fitness in the wild, with offspring of farmed salmon displaying lower survival than fish of native origin (McGinnity et al., 1997(McGinnity et al., , 2003Fleming et al., 2000;Skaala et al., 2012). Similar studies conducted on other salmonid species in response to releases of hatchery fish have also arrived at similar conclusions (Araki et al., 2008;Araki and Schmid, 2010).
In summary, the presence of farmed escaped salmon on the spawning grounds of native populations, and the potential for genetic interactions between escapees and wild conspecifics, is of concern. This is because farmed escapees may be genetically different from the recipient wild population for several reasons. (i) Farmed salmon usually do not originate from the same wild population into which they migrate post-escape and will therefore display population genetic differences to the native population. (ii) Farmed salmon have been subject to directional selection and thus differ to all wild salmon for those traits. (iii) Through relaxation of natural selection and inadvertent adaption to the domestic environment, farmed salmon have undergone domestication selection and will also differ to wild salmon.  Table 3 in Glover et al. (2013a). The computed median level of introgression is 9.1%.

Risk estimation
As part of a national strategy for an environmentally sustainable aquaculture industry (Anon., 2009b), the Norwegian government established the following management goal to prevent genetic interactions of farmed escapees with wild salmon populations: "Aquaculture will not contribute to permanent genetic changes in the genetic characteristics of wild fish stocks" (Table 1). This political target, which forms the basis for the risk estimation, is clearly open for scientific interpretation. However, it was interpreted in a conservative sense for the estimation of risk (Table 2). Thus, any observed genetic change in allele frequencies of molecular genetic markers caused by introgression of farmed salmon would be regarded as permanent genetic change, and therefore in violation of the management goal for sustainability established by the government. The rationale behind this interpretation was first and foremost because molecular genetic markers would be required to directly measure genetic changes in the wild populations. Furthermore, while natural selection will influence the genetic composition of any population, including those where farmed salmon have successfully introgressed, it is unlikely that natural selection will revert the population back to its exact genetic composition before introgression of farmed salmon. This is despite the possibility that natural selection may potentially restore fitness in the natural population.
The documentation of genetic change in a wild population is most directly achieved through the analysis of molecular genetic markers. However, while introgression of farmed Atlantic salmon has been estimated for 20 wild salmon populations in Norway in a 3-4 decade period from 1970 onwards , genetic data to estimate introgression of farmed salmon does not exist for the great majority of Norwegian populations. Furthermore, the analysis was being used to address risk of continued and future genetic changes in relation to today's aquaculture industry rather than changes that have already occurred through historical introgression. Therefore, the frequency of farmed escaped salmon observed in wild populations, which is correlated with genetic introgression of farmed escapees over time  was chosen as the indicator to estimate risk of further genetic changes in each wild population for the estimation of risk (Taranger et al., 2012a).
From 2012 onwards, risk was estimated against the below categories for probability of further genetic changes in wild populations caused by introgression of farmed salmon: No or low risk of genetic change: ,4% incidence of farmed salmon Moderate risk of genetic change: 4 -10% incidence of farmed salmon High risk of genetic change: .10% incidence of farmed salmon The threshold values were set according to knowledge of natural straying (reviewed by Stabell, 1984) and knowledge about the present correlation between frequency of farmed fish and corresponding genetic introgression . Stabell (1984) showed that most fish returned to their natal river, although in two of the experiences referred to, straying rates were as high as 10% and nearly 20%. We have chosen a threshold value for no or low risk for genetic change at a frequency of farmed fish in the river corresponding to the lower part of the natural straying estimates (4%), while 10% as high risk of genetic change from the upper part of the distribution. Especially, the threshold value for the upper limit is uncertain and might be modified according to new knowledge about the corresponding correlation between frequency of farmed fish and actual introgression Taranger et al., 2014).
The frequency of farmed salmon in each river surveyed was based upon autumn data where the frequency has been reported for a series of Norwegian rivers (Fiske, 2013). To estimate risk, the frequency of farmed salmon observed in autumn survey was recomputed into an "incidence of farmed salmon" per population using a formula for normalizing data from summer angling catches and autumn surveys (Diserud et al., 2010). This was done because the percentage of farmed salmon in autumn samples is usually higher than in summer angling catches (Fiske et al., 2006), which is in part because farmed salmon enter rivers later than wild salmon. Thus, without normalization of data, the frequency of farmed salmon in summer and autumn surveys are not directly comparable. While many rivers have both summer and autumn estimates, some only have one or the other estimate and therefore require transforming into what has been defined as the "incidence of farmed salmon". The formula for normalizing data from summer and autumn surveys to create the "incidence of farmed salmon" were obtained from Diserud et al. (2010) and are presented below: For the risk assessment, the mean incidence of farmed salmon was estimated in 34 rivers distributed along the Norwegian coast using autumn survey data collected in the period 2010-2012. Only rivers having autumn survey data from a minimum of 2 of the 3 years in this survey period were included in the risk assessment. Based on these data, the risk for genetic changes as a result of farmed salmon introgression was low, moderate, or high for 13 (38%), 11 (32%), and 10 (29%) surveyed rivers, respectively ( Figure 3).

Limitations
There are a number of challenges and limitations to the conducted risk assessment and estimation, its approach, and calibration against threshold values and potential impacts. One limitation is the fact that the observed frequency of farmed escapees in rivers has been used as the proxy for potential genetic changes in wild populations caused by introgression of the farmed salmon. This is a limitation because the correlation between the frequency of escapees observed in a river, and documented genetic introgression is only modest (R 2 ¼ 0.47, p ¼ 0.0007) . As a result, some rivers display higher and lower levels of genetic introgression from farmed than would be estimated by analysis of the frequency of escapees on the spawning grounds. The consequence of this is that the observed frequency of farmed escapees in each population will not accurately reflect the true risk of genetic changes for all populations, and only by using genetic methods directly will the risk be able to be quantified accurately. The underlying causes of the lack of a strong relationship between the observed frequency of escapees and genetic introgression are important to identify to help improve the accuracy of the risk assessment in the future. From their genetic study of introgression in 20 Norwegian rivers, Glover et al. (2013a) identified both technical and biological elements that are likely to influence the Risk assessment of the environmental impact  strength of the relationship between the observed frequency of escapees and genetic introgression. Two of the primary components suggested were potential inaccuracies in the frequency of escapees reported for each river (e.g. limited, biased, or non-standardized sampling or reporting), and the fact that the density of the native population, especially on the spawning grounds, may also influence relative success of farmed escapees through spawning competition (Fleming et al., 1996(Fleming et al., , 2000. In the future, models may be used which include covariables in addition to just the incidence of farmed salmon to predict genetic changes. A further limitation of the present risk assessment is that it was only conducted for 34 rivers for the 2013 risk assessment. These rivers were chosen as they had autumn survey data published (Fiske, 2013) and thus readily available for assessment of risk for the period 2010-2012. In Norway, there are over 400 salmon rivers and for 220 of them the status of the stocks are assessed (Anon., 2013b). Thus, the rivers investigated in the current risk assessment only represent a small proportion of those in Norway. Therefore, it is not possible to make clear regional inferences regarding introgression of farmed salmon, only for a small number of specific rivers. Data for the frequency of farmed salmon exist for a larger number of rivers than are currently included in this risk assessment. However, the quality of some of these data, the reporting and availability of the data are highly variable. It was for this reason that the risk assessment was only conducted for the 34 rivers.
Clearly, there is a significant need to increase efforts to expand and improve the monitoring of escaped salmon in a larger number of Norwegian rivers using data gathered and reported in a standardized manner. This will initially be able to improve estimates of the proportion of escaped salmon and will also provide a better foundation for the collection of representative samples for subsequent use in genetic analysis to validate introgression in rivers. An effort to coordinate data collection of escapees has been initiated within Norway in 2014, and in the future it is predicted that the risk assessment will be conducted in a much larger number of rivers.
In addition to technical and data availability challenges linked with the risk assessment and its implementation, there are gaps in current knowledge which limit the ability to identify threshold tolerance limits for introgression of escapees and the level of potentially detrimental effects on the wild populations. Current knowledge points toward a potential negative effect of introgression of farmed Atlantic salmon on the fitness and future evolutionary capacity of recipient wild populations. This is when taking into consideration data available from experimental comparisons of farmed and wild salmon especially in the natural environment (McGinnity et al., 1997(McGinnity et al., , 2003Fleming et al., 2000;Skaala et al., 2012), background knowledge of salmon biology, life history and ecology, and extensive information from hatchery-fish supplementation for both Atlantic salmon as well as other salmonid species in both the Atlantic and Pacific (Araki et al., 2008;Araki and Schmid, 2010). Nevertheless, significant gaps in understanding of the biological consequences of introgression of farmed salmon remain. These need to be quantified in the future to make a full assessment of risk of biological consequences following introgression of escapes. The major points are summarized in Table 3.

Salmon lice impact on wild salmonids
Salmon lice (Lepeophtheirus salmonis) from salmon farms are recognized as an important hazard to wild anadromous salmonids in Norwegian coastal waters (Serra-Llinares et al., 2014). Salmon lice on farmed salmon produce large amounts of planktonic larvae stages that spread via the water currents and can infect migrating Atlantic salmon post-smolts, as well as sea trout (Salmo trutta) and Arctic charr (Salvelinus alpinus) that stay in coastal waters . Hydrodynamic models coupled with biological data show that salmon lice can be transported up to 200 km over a 10-d period, although most dispersed 20 -30 km (Asplin et al., 2011;Serra-Llinares et al., 2014). The number of salmon lice allowed on farmed salmon is tightly controlled by Norwegian legislation (www.mattilsynet.no). However, the large number of farmed salmon, with 300 million smolts put into sea cages every year along the Norwegian coast, results in worse case releases in the order of more than a billion salmon lice larvae daily from salmon farms in Norway (Taranger et al., 2014).
New analyses reveal strong correlation between salmon farms and lice infections on wild salmonids in Norwegian coastal waters (Helland et al., 2012;Serra-Llinares et al., 2014). The Norwegian salmon lice monitoring programme on wild salmonids demonstrate annual lice epidemics, most likely connected to the density of salmon farms in the surrounding areas as well as the seasonal dynamics of salmon lice infections on farmed salmon (Jansen et al., 2012;Serra-Llinares et al., 2014;Taranger et al., 2014). A series of experiments has shown that salmon lice may affect anadromous salmonids (reviewed in Finstad and Bjørn, 2011;Anon., 2012;Torrissen et al., 2013).
To assess the risk of salmon lice infection on wild populations, we have considered the following elements of risk assessment: release, exposure, and consequences in the following manner. The release Table 3. Some identified gaps in current knowledge with respect to understanding the potential negative consequences of introgression of farmed salmon in native populations.

Question
Hypothesis to be tested What is the fitness differential between the offspring of wild, hybrid, and farmed salmon, including multiple generation back-crossed individuals, in different rivers and environments?
To what degree can the results from the few comparative experiments of survival in the wild be generalized for all types of rivers and populations, and is it possible to generalize results? How strong is natural selection, and will the offspring of farmed salmon readapt to the natural environment?
Lower survival of the offspring of farmed salmon and hybrids in the wild also implies that natural selection will purge poorly adapted individuals from the recipient population, but how strong is natural selection and what is the time-scale of this potential re-adaptation process? What are the threshold limits of introgression?
What degree of genetic introgression will be tolerated in wild populations before biological, life history and ecological characteristics of the population, and population productivity are compromised? What is the underlying genomic architecture of domestication and local adaptation in salmonids?
What genetic changes have occurred during domestication, is it possible to identify genetic markers linked with these changes, and fitness in the wild to provide more accurate measurements of functional genetic changes in native populations?
Risk assessment of the environmental impact assessment is based on estimating the production and distribution of infectious salmon lice. The exposure assessment is based on estimating the lice infection on wild salmonid populations using different methods for direct measurements of salmon lice infections on salmon and sea trout. The consequence assessment is the effect salmon lice have on salmonid populations in terms of estimated likelihood of increased marine mortality and/or reduced reproduction based on the exposure assessment.

Risk assessment
All salmon farming sites in Norway report the numbers of salmon lice on the fish weekly when the temperature is .48C. This is reported together with fish biomass and number of individual salmon per cage (reported each month). Based on the number of sexually mature female salmon lice on the fish on each farming site, the number of infectious salmon lice larvae produced from the different salmon farms are calculated (Jansen et al., 2012). However, at present we do not have enough information and validated models to accurately estimate the impact of the salmon lice infections on wild populations based on reported data from the fish farms (Taranger et al., , 2014. The lice infection on wild salmonid populations is estimated using different methods as part of a national monitoring programme (Serra-Llinares et al., 2014;Taranger et al., 2014). These methods encompass catch of sea trout and Arctic charr in traps or nets (Bjørn et al., 2011a), and salmon post-smolts caught in special surface trawls in fjord systems (Holm et al., 2000;Bjørn et al., 2007b;Holst et al., 2007). In addition, groups of small sentinel cages containing on average 30 farmed salmon post-smolts are placed in the fjords to monitor the salmon lice infection rate. The fish are kept in the cages for 3 weeks before lice are counted on all the fish, and the procedure repeated three times during spring and summer (Bjørn et al., 2011a. The sampling programme is focused on areas with high salmon farming activity (Figure 4), as well as some fjords that are protected against salmon farming. The assessment in the period 2010-2013 is based on data from 1 to 5 sites per fjord in 13-16 fjord systems annually, and with increasing numbers of fish sampled at each site in the later years (Table 4). More details about the national salmon lice monitoring programme are provided elsewhere (Helland et al., 2012;Bjørn et al., 2013;Serra-Llinares et al., 2014).
To conduct the risk assessment on the potential impact on salmonid smolts that migrate from the rivers in spring and early summer on the one hand, and the risk of sea trout and Arctic charr that stay in fjords and coastal waters during summer, the national monitoring programme covers two different periods. These periods are adjusted for different timing of smolt migration and seawater residence along the Norwegian coast (Anon., 2011), with an earlier sampling window in the southern part of Norway and later further north. This corresponds to two assumed "critical periods", the first during spring when the salmonid smolts leave the rivers and enters the estuaries and fjords (Period 1), and the second period to estimate the accumulated infection rate on sea trout and Arctic charr that remain in fjords and on the coast during summer (Period 2). See also Anon. (2011) for further information on median migration dates and migration speeds for Atlantic salmon post-smolts in Norwegian fjords.
A range of laboratory studies demonstrate the impact of salmon lice on salmon post-smolts (Grimnes and Jakobsen, 1996;Finstad et al., 2000Finstad et al., , 2010Heuch et al., 2005;Wagner et al., 2008). It has been shown that 0.04 -0.15 lice per g fish weight can increase stress levels, reduce swimming ability and create disturbances in water and salt balance in Atlantic salmon (Nolan et al., 1999;Wagner et al., 2003Wagner et al., , 2004Tveiten et al., 2010). In sea trout, 50 mobile lice are likely to give direct mortality (Bjørn and Finstad, 1997), and only 13 mobile lice, or 0.35 lice per g fish weight might cause physiological stress in sea trout (weight range of 19-70 g; Wells et al., 2006Wells et al., , 2007. Moreover, 0.05-0.15 lice per g fish weight were found to negatively affect sexually maturing Arctic charr (Tveiten et al., 2010).
According to a review by Wagner et al. (2008), infections of 0.75 lice per gram fish weight, or 11 salmon lice per fish, can kill a recently emigrated wild smolt of 15 g if all the salmon lice develop into preadult and adult stages. Studies of naturally infected wild  salmon post-smolts indicate that only those with ,10 lice survived the infection (Holst et al., 2007). This is consistent with field studies on salmon lice infections in salmon post-smolts in the Norwegian Sea. Over a decade of surveys, no post-smolts was found with .10 salmon lice, and fish with up to 10 mobile lice were observed to be in poor condition with low blood count and poor growth (Holst et al., 2007). New studies of naturally infected wild salmon post-smolts also show that sea lice are fatal at high infections (Berglund Andreassen, 2013). More work in this field is in progress.
Population-wise effects of salmon lice have been demonstrated on wild salmonids in Ireland and Norway. This was studied by protecting individually tagged Atlantic salmon smolts against salmon lice (using Substance EX or Slice) before they were released into the sea near their respective home rivers (Finstad and Jonsson, 2001;Hazon et al., 2006;Skilbrei and Wennevik, 2006;Hvidsten et al., 2007;Skilbrei et al., 2008Skilbrei et al., , 2013Jackson et al., 2011Jackson et al., , 2013Anon., 2012;Gargan et al., 2012;Krkošek et al., 2013a, b;Vollset et al., 2014). These studies suggest that salmon lice infections increase the marine mortality in areas with intensive salmon farming activity.

Risk estimation
The risk assessment on salmon lice is based on Goal 1 in the policy document on sustainable aquaculture by the Norwegian government from 2009 (Anon., 2009b, Table 1) stating that "Disease in fish farming will not have regulating effect on populations of wild fish. . .". This target has in this risk estimation been interpreted as "Salmon lice from fish farming significantly increase the mortality of wild salmonids" (Table 2).
A salmon lice risk index, attempting to estimate the increased mortality due to salmon lice infections, was proposed by Taranger et al. (2012a) and are based on the assumption that small salmonid post-smolts (,150 g body weight) will suffer 100% lice-related marine mortality, or return prematurely to freshwater for sea trout, in the wild if they are infected with .0.3 lice g 21 fish weight. Furthermore, the lice-related marine mortality is estimated to 50%, if the infection is between 0.2 and 0.3 lice g 21 fish weight, 20% if the infection rate is between 0.1 and 0.2 lice g 21 fish weight, and finally 0% lice-related mortality if the salmon lice infection is ,0.1 lice g 21 fish weight.
For larger salmonids (over 150 g), we assume that lice-related mortality or compromised reproduction will be 100% in the group if they have .0.15 lice g 21 fish weight, 75% for lice infections between 0.10 and 0.15 lice g 21 fish weight, 50% for lice infections between 0.05 and 0.10 lice g 21 fish weight, 20% for lice infections between 0.05 and 0.01 lice g 21 group, and finally 0% if the salmon lice infection is ,0.01 lice g 21 fish weight.
For both indices, increased mortality risk or compromised seawater growth or reproduction at population level are calculated as the sum of the increased mortalities/compromised reproduction for the different "infection classes" in the sample, reflecting the distribution of the intensity of salmon lice infections of the different individuals sampled. This assumes that individuals caught with traps, gillnets or trawls are representative for the various salmonid populations in that fjord area. The risk was further scored according to the system proposed by Taranger et al. (2012a); as low (,10% estimated increase in mortality; green colour), moderate (yellow) for those with between 10 and 30% increase, and high (red) if the increase is calculated as .30%.
The current assessment is based on these scorings, and we use data from the national monitoring programme on salmon infections in wild salmonids (Bjørn et al., , 2011b. Separate result tables are presented for the risk for Atlantic salmon smolts (Period 1; Table 5) and for sea trout and the Arctic charr populations (Period 2; Table 6). The results are sorted by county from south to north and by fjord system.
The results indicate considerable variation in risk between years and sampling locations. Moreover, these data strongly indicate a much higher risk for sea trout (and also Arctic charr in the Northern regions) compared with Atlantic salmon post-smolts and reveal moderate-to-high risk of population-reducing effects on sea trout in most counties with high salmon farming activity. The risk of population-reducing effects for Atlantic salmon varies much more between years and sites, and was low at most sites in 2010 and 2013, but moderate and high at several sites 2011 and 2012.

Limitations
The assessment in the period 2010-2013 is based on data from 13 to 16 fjord systems annually. Despite large field effort (Bjørn et al., 2011b, the geographical coverage is insufficient in terms of the distribution of salmon farms and wild salmonid populations along the Norwegian coast. There are also problems considering how well the different sampling methods are representative for the different anadromous populations in that area. We have limited data on salmon lice infections in migrating Atlantic salmon smolts, so the risk assessment for salmon is mainly conducted by the use data on salmon lice infections on sea trout caught in traps and gillnets in Period 1 as proxy for the risk to Atlantic salmon post-smolts. Lice infections on trout may not be directly proportional to lice infections on migrating salmon smolt. It is likely that differences in, for example, migratory behaviour and marine ecology exposes salmon and sea trout smolts for different sea lice infection risk (Anon., 2011), even within the same fjord system (Bjørn et al., 2007a(Bjørn et al., , 2011b(Bjørn et al., , c, 2013Serra-Llinares et al., 2014). The link between individual lice infections and population effects is also very uncertain. There is therefore uncertainty of the current risk assessment both for Atlantic salmon, and for sea trout and Arctic charr. Moreover, the current data are presented without any estimates of uncertainty, which must be included in future analyses.

Future
The "Strategy for an environmentally sustainable aquaculture industry" (Anon., 2009b) states that no disease, including lice, should have a regulatory effect on wild fish. The monitoring of salmon lice infection of wild salmonids is an important verification of whether this goal is achieved, and whether the measures taken are appropriate and sufficient. An indicator system that allows detection of possible problems needed therefore to be established. Hitherto, this assessment has been based on data from the national monitoring of sea lice. An analysis of the historical data in this monitoring series (2004)(2005)(2006)(2007)(2008)(2009)(2010) shows that both the extent and nature of the data have had some weaknesses that limit the ability to analyse and understand the observed variation in infections on wild fish (Helland et al., 2012;Serra-Llinares et al., 2014). Therefore, monitoring and risk assessment based only on lice counts on wild salmonids is not considered sufficient.
Consequently, a rather radical change in the monitoring, advisory and management system for lice has therefore been proposed (Taranger et al., 2012b(Taranger et al., , 2014Bjørn et al., 2013;Serra-Llinares et al., 2014). This system is based on (i) detection and forecasting of increased production of infectious salmon lice using models, (ii) verification of infection pressure through risk-based Risk assessment of the environmental impact Table 5. Estimated risk for lice-related mortality (%) of Atlantic salmon post-smolts based on lice levels on sea trout caught in traps or gillnets and with weight ,150 g in Period 1 at the sites from south to north in Norway in 2010 -2013.
The colour code refers to the assessment of potential population-reducing effect (red ¼ high, yellow ¼ moderate, and green ¼ low). a Small number of fish caught, all fish were used in the assessment irrespective of body size. 1006 G. L. Taranger et al. Table 6. Estimated risk for salmon lice-related mortality or compromised reproduction (%) of sea trout based on observations in Period 2 at the various sites from south to north in Norway in 2010 -2013.
Mortality estimates are based on all fish sampled in the period, with different thresholds for small (,150 g) and large (.150 g) fish. The colour code refers to the assessment of potential population-reducing effect (red ¼ high, yellow ¼ moderate, and green ¼ low).
Risk assessment of the environmental impact 1007 and adapted surveillance on wild salmonids and (c) extended risk assessment based on a considerably larger dataset and fine scale hydrodynamic lice dispersal modelling to assess the effect on wild populations, which then allows adjustment of measures taken by management and industry to reduce this effect to levels within the objective of the strategy. This is now possible due to better knowledge about the relationship between intensive salmon farming activity and infection pressure (Helland et al., 2012;Jansen et al., 2012;Serra-Llinares et al., 2014;Taranger et al., 2014), and better and more accessible farming and environmental data (Jansen et al., 2012;Taranger et al., 2014).
A preliminary analysis indicates that under realistic conditions of lice infections, water currents, temperature and salinity, and relative lice infections may be predicted using a coupled hydrodynamicbiological lice dispersion model. With further calibrations and validation, such a system can probably be developed at least for specific areas along the Norwegian coast. As validation, calibration, and implementation of such a risk-based monitoring system is done, more of the monitoring could be based on the model and less on catch of wild salmonids. The preliminary model results are encouraging in terms of validation and calibration the model predictions against observed infection of wild salmonids. However, considerable research and development remains, where the main challenges are: † The system for detecting problem areas based on farming data and sea lice infections of notification of problem areas need to be further developed and operationalized, and systems for risk-based and adapted surveillance on wild salmonids must be developed. † Coupled hydrodynamic-biological lice dispersion models must be validated and calibrated against observed infection levels on wild salmonids in the field. † Knowledge about the ecological effects of a given infection pressure on stocks of wild salmonids (population-reducing effects) must be increased so that more precise predictions can be developed.

Disease transfer from farmed salmon to wild fish
Background Infectious diseases represent a major problem in Norwegian fish farming, despite successful development and application of vaccines against a range of pathogens. In addition to lice (considered separately above), viral diseases currently represent the largest disease problems in Norwegian aquaculture (Johansen, 2013). In the period 2005-2012, the four most frequent viral diseases [infectious pancreatic necrosis (IPN), pancreas disease (PD), heart and skeletal muscle inflammation (HSMI), and cardiomyopathy syndrome (CMS)] had 400-500 outbreaks annually (Johansen, 2013). The main reason for the dominance of viral diseases is the lack of effective vaccines. Bacterial diseases, on the other hand, cause only 20 outbreaks annually, reflecting that the currently used bacterial vaccines provide good protection (Austin and Austin, 2007). Among the parasites, parvicapsulosis due to the myxosporean Parvicapsula pseudobranchicola is a problem mainly in northern Norway, whereas heavy gill infections with the microsporidian Paranucleospora theridion and the amoeba Paramoeba perurans occur mostly in the southern parts of the country. The significance of infections with the former two parasites is unclear, while amoebic gill disease (AGD) has so far been detected only during fall in 3 years (2006,(2012)(2013).
For most pathogens, clear evidence for transmission from farmed to wild fish is limited (Raynard et al., 2007). Most of the diseases that currently cause problems in fish farms are likely enzootic, originating from wild fish. This implies that these infections occur or occurred in the past at some "background" level in wild stocks. Such considerations complicate an estimation of the impact of aquaculture, since the "normal" prevalence range of many important disease agents is unknown. However, in two cases exotic pathogens have been introduced in association with farming activities. These have clearly affected wild Atlantic salmon populations.
The ectoparasite Gyrodactylus salaris (Monogenea) was first detected in Norway in 1975 (Johnsen et al., 1999). There have been several introductions of G. salaris to Norway (Hansen et al., 2003) linked to the import of salmonids from Sweden. Later, the parasite has spread (or has been spread) to many rivers (Johnsen et al., 1999). By 2005, G. salaris had been detected in 45 rivers and 39 freshwater farms (Mørk and Hellberg, 2005). Norwegian Atlantic salmon stocks are very susceptible to G. salaris, and gyrodactylosis in farmed salmon may lead to 100% mortality if not treated (Bakke, 1991;Bakke et al., 1992;Bakke and MacKenzie, 1993). Mortality in rivers is high, with the density of Atlantic salmon parr being reduced by 50-99% (Johnsen et al., 1999).
Aeromonas salmonicida, the causative agent of furunculosis, was introduced to Norway in 1964, when furunculosis was detected in a single farm that received rainbow trout from Denmark. The disease then spread to other farms and wild fish within a limited area, being detectable there until 1979. A second introduction occurred in 1985, in connection with an import of Atlantic salmon smolts from Scotland. The disease then spread rapidly to farms and wild fish, and in 1992 a total of 550 salmon farms and 74 river systems were affected (Johnsen and Jensen, 1994). This rapid spread of the disease was likely facilitated by frequent escapement events involving infected fish (Johnsen and Jensen, 1994). Mortality due to furunculosis was registered in many rivers among escaped salmon, wild salmon, and trout. Mortality in farmed fish was high, reaching 50%, but the disease was first controlled by antibiotics and subsequently effectively with oil-based vaccines (Sommerset et al., 2005;Johansen, 2013).
These two examples show the devastating effects that introductions of exotic pathogens can have. Even when disregarding agents only known from non-salmonids, there is a large number of potential pathogens (i.e. hazards) infecting salmonids elsewhere that could have significant impact on both salmon farming and wild fish populations in Norway if introduced (Raynard et al., 2007;Brun and Lillehaug, 2010). Import of live fish represents the major threat to both fish farming and wild stocks, since this may lead to the introduction of exotic pathogens. However, G. salaris infections have not been detected in Atlantic salmon hatcheries in recent years (Hytterød et al., 2014 and references therein), and the parasite does not survive in seawater. Furunculosis outbreaks in farms are rare, since most farmed salmon is protected through vaccination. Regarding disease transfer from farmed salmon to wild salmonids, these diseases are currently considered to be under control.
The detection of disease in wild fish and estimating disease impact on wild populations is difficult. Clinically affected fish usually disappear quickly in nature (e.g. predated). Epizootics with mass mortality of fish are rare, but have occurred in Norway (Bakke and Harris, 1998;Sterud et al., 2007) and elsewhere (Hyatt et al., 1997;Gaughan et al., 2000). Such episodes are usually caused either by an exotic pathogen introduced to naïve host populations (Bakke and Harris, 1998)  conditions such as high temperature (Sterud et al., 2007). However, infection with native (enzootic) agents under normal environmental conditions can cause disease in individuals and affect an individual's survival or investment in reproduction. Hence, all pathogens may contribute to the regulation of wild populations at some level (May and Anderson, 1979;May, 1983), although the impact may vary and is often the result of a complex interaction between hosts, pathogens, environment, and predators (Dobson and Hudson, 1986;Combes, 2001).

Risk assessment
There is relatively little data available on the infection status of Norwegian wild salmonid stocks with respect to the most important pathogens that affect farmed salmon (e.g. viral agents). The available data mainly concerns returning adult salmon and some local sea trout populations screened with molecular methods (Kileng et al., 2011;Garseth et al., 2012Garseth et al., , 2013aBiering et al., 2013;Madhun et al., 2014a, b). Studies on the occurrence of viral infections in early life stages of salmonids are only fragmentary (e.g. Plarre et al., 2005).
Due to the limited data available, the disease status (outbreak statistics) in Norwegian fish farming is used as a proxy of the infection pressure from farmed salmon to wild salmonids. Information regarding disease outbreaks on Norwegian fish farms is gathered by the Norwegian Veterinary Institute (NVI) and published annually in their Fish Health Reports (e.g. Johansen, 2013). These data record official diagnoses from NVI, as well as information from the local fish health services. This information is likely to be biased towards the more serious diseases, particularly those that are required by law to be reported to government authorities. Subclinical infections may be common and may also contribute to the spread of pathogens. However, these infections are usually not detected. Despite shortcomings, these data are the best currently available information and give a reasonably good indication of the disease status of the majority of farmed fish in Norway.
Most diseases in Norwegian salmon and rainbow trout farms are represented by only a few outbreaks, often representing geographically separate cases (Johansen, 2013). However, some diseases have a large number of outbreaks/diagnoses, and are those most likely to cause elevated infection pressures that may affect wild populations. At present the most common diseases in Norwegian salmon farming are the viral diseases PD, IPN, CMS and HSMI (Table 7; Johansen, 2013). In addition, AGD due to Paramoeba perurans is an emerging problem (Hjeltnes, 2014). The listed viral diseases have caused some 400 or more outbreaks each year since 2005. Outbreaks are often more frequent in certain regions and at certain times of the year, leading to a consideration also of spatial and temporal variation in the potential infection pressure (Table 8).
We have considered the following elements of risk assessment regarding the viral agents salmonid alphavirus (SAV), infectious pancreatic necrosis virus (IPNV), piscine myocarditis virus (PMCV), and piscine reovirus (PRV); release, exposure, and consequences as follow. The release assessment is the assumed infection pressure as proxied by the outbreak statistics. The exposure assessment is a consideration of the spatial and temporal concurrence of wild salmonids with release. A consequence assessment should consider two aspects: (i) evidence for virus transmission and (ii) impact of viral infections. However, the impact is in all the considered cases are unknown, and only evidence for virus transmission can be discussed.
Pancreas disease: salmonid alphavirus PD in Atlantic salmon and rainbow trout is caused by SAV. In Norway, there are currently two regionalized PD epidemics caused by SAV3 (south of Hustadvika, 638 N) and SAV2 (north of 638 N). Experimental studies show transmission of SAV via water, and epidemiological studies provide evidence for horizontal farm to farm spread (Nelson et al., 1995;McLoughlin et al., 1996;Kristoffersen et al., 2009;Stene, 2013;Stene et al., 2014). The virus has been shown to survive for several weeks in the environment (Graham et al., 2007) and thus may be carried long distances with currents (Stene, 2013;Stene et al., 2014). SAV2 may have a different outbreak pattern than SAV3, since outbreaks tend to occur later in the year (Johansen, 2013). For the southern region (SAV3), the period in which smolts migrate and adult salmon return coincides with many SAV3 outbreaks (Table 8; Johansen, 2013). In the northern region (SAV2), most outbreaks occur later in the year. This may signify that most of the smolt migration precedes peak virus spread in the SAV2 region. On the other hand, returning salmon and sea trout are likely more exposed, but screening indicates that very few wild fish are infected (Biering et al., 2013). Infected escaped salmon can enter rivers in fall, possibly exposing wild fish including naïve juveniles to the virus (Madhun et al., 2014a).
Screening of sea trout (Biering et al., 2013;Madhun et al., 2014b) indicates that sea trout in areas with high frequency of PD outbreaks are not infected with SAV. This is in accordance with injection  Risk assessment of the environmental impact experiments, which suggest that sea trout is more resistant to SAV than salmon (Boucher et al., 1995). SAV infections have been detected in wild salmonids and wild flatfish (Nylund, 2007;Snow et al., 2010;Biering et al., 2013), but PD has not been observed in wild fish. Altogether, there are yet no data that confirm SAV transmission from farmed salmon to wild fish, but transmission of virus to wild fish is considered likely due to the large number of outbreaks and the documented efficient horizontal transmission of SAV. The probability of transmission of SAV to wild salmon is considered to be moderate for migrating smolts in the southern PD-region due to the temporal overlap between outbreaks and migration, whereas it is considered to be low in the northern PD-region as most of the migration is finished before the major outbreak period. For returning salmon, the probability of infection is considered to be low in both PD-regions based on the available screening results. The probability of SAV transmission to wild salmon is considered to be low in areas with no or few outbreaks. The probability of infection of sea trout during the marine phase is also considered to be low.

Infectious pancreatic necrosis: infectious pancreatic necrosis virus
IPNV is a robust, long-lived birnavirus that infect many different fish species in both fresh water and seawater (e.g. Reno, 1999). The virus is enzootic in Norway. IPN cause significant losses in fish farming in most areas in Norway (Johansen, 2013). However, there are indications of a downwards trend in outbreaks and losses, which might be caused by the increased use of IPN resistant fish (Johansen, 2013). The virus is shed into the water by infected fish, and is spread to other farms by water currents (see, e.g. Mortensen, 1993;Wallace et al., 2005;Raynard et al., 2007;Johansen et al., 2011). A higher prevalence of IPNV has been found in wild fish near salmon farms with clinical outbreaks of IPN, compared with fish at distant sites (Wallace et al., 2005(Wallace et al., , 2008. Fish surviving an IPNV infection often become persistent carriers of the virus, but viral shedding from carriers has not been demonstrated . The prevalence of IPNV in wild fish is low (Brun, 2003;Wallace et al., 2005), and farmed fish are probably a major source of virus in the marine environment. Disease outbreaks in wild salmon have not been described, but mortality in wild marine fish due to IPN has been reported elsewhere (Stephens et al., 1980;Mcallister et al., 1984).
Due to the large number of annual outbreaks, the demonstrated robustness and infectivity of the virus, as well as the wide range of hosts, the probability of IPNV transmission from farmed to wild fish is considered as moderate. The extent and consequences of such transmission are unknown.
Heart and skeletal muscle inflammation: piscine reovirus HMSI affects farmed salmon along the entire coast of Norway. HSMI outbreaks mainly occur 5 -9 months after sea-transfer. The causative agent is an Orthoreovirus, PRV (Palacios et al., 2010;Løvoll et al., 2012). The disease can be produced experimentally using infected tissue, infected cell culture, or by cohabitation (Kongtorp et al., 2004;Martinez-Rubio et al., 2012. PRV is present in high densities in salmon with HSMI, but high infection intensity can be found also in clinically healthy salmon. PRV infections have been detected in wild salmon along the entire coast of Norway and have been detected in sea trout (Biering et al., 2013;Garseth et al., 2013b). Analyses of PRV genotypes in wild Atlantic salmon, farmed salmon, and sea trout have suggested an extensive spread of the virus along the coast, and establishment in wild populations. This spread is probably due to extensive transportation of fish between areas over a long period (Garseth et al., 2013c). However, there are no reports of HSMI in PRV infected wild salmonids (e.g. Garseth et al., 2013b). In Norway sea trout are only rarely (1.4-3%) infected (Garseth et al., 2012;Biering et al., 2013) with PRV. PRV infections have been detected in some marine fish species, but the virus genotype is unknown (Wiik-Nielsen et al., 2012). It is not known how long and in what quantities PRV is shed from infected fish nor viral survival in seawater. However, modelling suggests that the virus can be transported over longer distances than SAV (Aldrin et al., 2010;Kristoffersen et al., 2013). The latter findings suggest that the virus is relatively stable and may spread over large areas.
Since PRV infections are widespread in farmed salmon, may readily be transmitted, and is detectable in .10% of the wild salmon examined, it is considered likely that PRV is transmitted from farmed to wild salmon. HSMI occurrence in wild salmon and other wild fish are unknown (Garseth et al., 2013b, Madhun et al., 2014b.

Cardiomyopathy syndrome: Piscine myocarditis virus
CMS is a serious disease in salmon, and is caused by a Totivirus, PMCV (Løvoll et al., 2010;Haugland et al., 2011). CMS can be transmitted experimentally by injecting heart tissue homogenates from diseased fish, PMCV from cell culture, and by cohabitation (Haugland et al., 2011). Infections are long-lasting, with a gradual development of cardiac pathology.
PMCV has been detected in farmed Atlantic salmon along the entire Norwegian coast, but is not as widespread as PRV. PMCV infection has also been detected in a few wild salmon in Norway (Garseth et al., 2012), and CMS-like lesions have been observed in the hearts of wild salmon before the discovery of the virus (Poppe and Seierstad, 2003). Large-sized wild salmon represent the only known natural reservoir for PMCV. There are no studies on shedding of PMCV from diseased fish or carriers, or on virus survival in water.
Due to the large number of hosts, prolonged infections, and the documented virus spread in cohabitation experiments, we assume that the virus is present in the environment and that the infection pressure around farms harbouring the virus is elevated. Examination of wild returning salmon detected only very few (3/ 1350) infected with PMCV (Garseth et al., 2012;Biering et al., 2013). Such a low prevalence, which may represent natural rather than fish farming-related infections, suggests that PMCV transmission from farmed to wild salmon is infrequent. Therefore, the probability of infection in wild salmon due to virus released from farms is considered low. Due to a general lack of data, particularly regarding young fish, it is not possible to assess the impact of PMCV infection in wild salmon.

Limitations of the analysis
There is a scarcity of data on infections and lack of evidence for disease in wild salmonids for the four viral agents considered. A large number of outbreaks suggest extensive spread of virus, and consideration of timing also often substantiates an exposure of wild salmonids to the viral agents. A serious limitation in the risk assessment is a lack of information on infections due to these agents in wild fish, particularly in salmon smolts. These may be less susceptible than their farmed peers, but may also suffer mortality due to the infections. Such mortality would likely occur through predation (i.e. virus induced). In either case, the returning 1 or 2 sea-winter salmon could be found to be virus free, as is indeed generally the 1010 G. L. Taranger et al.
case regarding SAV. Juvenile fish is often particularly susceptible to viral infections. Escaped virus infected farmed salmon can enter rivers in fall (e.g. Madhun et al., 2014a), where juvenile salmonids may be exposed to released virus. The impact of SAV, PRV, and PMCV infections on juvenile salmonids is unknown. The abundance of escaped salmon in Norwegian rivers (see above) indeed suggests a high potential for interaction at this stage. Hence, due to lack of data, consequence assessments cannot be done for the four viral agents considered.
Here, we have focused on the four major diseases currently prevalent in Norwegian salmon farming. All are viral diseases, and a large number of outbreaks are reported annually. However, low prevalence does not necessarily mean low impact, and diseases that at present are under control might surge and become a threat to wild populations. Subclinical or apparently benign infections may also have unforeseen ecological effects in nature by affecting survival (i.e. predator avoidance) or recruitment.
We have not considered the possible impact of exotic pathogens on Norwegian salmon farming. There are a large number of potential pathogens (i.e. hazards) infecting salmonids elsewhere that could affect farmed and wild salmonids in Norway if introduced (Raynard et al., 2007;Brun and Lillehaug, 2010). Such introductions may be irreversible, and difficult or impossible to contain, and the main risk factor is the movement of live fish or fertilized eggs.

Future work
To evaluate the effects of salmon and rainbow trout farming on the infection status of wild salmonids, there is a need to increase our knowledge about the complete pathogen repertoire (viruses, bacteria, fungi, and parasites) present in wild fish in areas of highand low-intensity fish farming. Long-term monitoring of selected wild populations would allow detection of changes in the infection status in the population. Experimental challenge experiments with SAV, PRV, and PMCV must be performed on juvenile Atlantic salmon and brown trout in freshwater. This would better allow a consideration of the threat posed by escaped salmon ascending rivers during fall. A more extensive genotyping of virus from wild and farmed fish would improve our understanding of both virus spread and genetic changes in pathogens that may occur in the high host density of farming areas.

Organic load and nutrients from salmon farms
The salmonid aquaculture industry has continuously restructured since 1999, with reductions in the number of farms, increased farm size and relocation of farms to deeper fjord (50-300 m) and current rich coastal aquaculture sites. During this period, the production has doubled (Gullestad et al., 2011), with typical salmon farm produces between 3000 and 5000 tons in a 18 months period in sheltered coastal waters and as much as 14 000 tons at more dynamic coastal sites. This rapid development has led to increased concerns about the environmental impacts both at present and future predicted finfish production levels.
Increased awareness of elevated discharges of nutrients, excess feed, and faeces to the marine environment has resulted in greater scrutiny of the aquaculture industry (Mente et al., 2006;Taranger et al., 2012a). To this end, the Ministry of Fisheries and Coastal Affairs have stated that "the environmental impact of aquaculture must be kept at an acceptable level and be within the assimilative capacity of the area" (Anon., 2009b).
Assessing the risk of organic enrichment and nutrient overloading of Norwegian finfish aquaculture at local and regional scales focus both on benthic and pelagic systems and is based on a combination of scientific knowledge and industry monitoring data, coupled with best professional judgment, and the precautionary principle.

Risk assessment Release assessment
Intensive farming of finfish in open sea cages results in the release of organic and inorganic effluents (i.e. carbon, nitrogen, and phosphorus) in the form of waste feed, faeces, and metabolic by-products to the surrounding aquatic environments (Holmer et al., 2005;Strain and Hargrave, 2005). At current production levels in Norway, salmonid farming (1.3 M tons cultivated fish in 2012) releases 34 000 tons of N, 60 000 tons of C, and 9750 tons of P annually .

Expose assessment
Accumulation of these effluents into the marine system can negatively impact the ecosystem by contributing to eutrophication of pelagic systems, fertilization of benthic macrophytes in the euphotic zone, and organic enrichment of benthic systems (Strain and Hargrave, 2005). However, the area of influence (local or regional locations) and degree of enrichment of the environment depends on a number of factors including the size of the farm (i.e. the biomass of fish), the ambient environmental conditions (i.e. hydrodynamics, water depth, wave exposure, topography and substrate type) and the husbandry practices at the individual fish farms (Holmer et al., 2005). In Norway, detailed knowledge about the environmental effects of organic and nutrient enrichment from finfish aquaculture is mainly based on studies around sheltered and coastal fjord aquaculture sites.
At deep aquaculture sites, fish farming effluents can be traced into the wider environment and into benthic foodwebs up to at least 1 km from the farming site (Kutti et al., 2007b;Olsen et al., 2012). At low deposition levels, organic enrichment of benthic sediments (up to 500 m from the farming location) stimulates secondary production in soft bottom communities, resulting in shifts in benthic faunal community structure (Kutti et al., 2007a;Kutti, 2008;Bannister et al., 2014). In addition, excessive loading of organic effluents to sediments often leads to dramatic changes in biogeochemical processes leading to grossly anoxic conditions (Valdemarsen et al., 2012). The emissions of dissolved nutrients from finfish farms are quickly diluted in the water column at dynamic sites and elevated nutrient levels are hardly detected 200 m away from the farm (H. Jansen, IMR, unpublished data;Sanderson et al., 2008).

Consequence assessment
Many studies have investigated benthic impacts of fish farming on soft sediment benthic systems, demonstrating that intensive fish farming modifies biogeochemical processes (Holmer and Kristensen, 1992;Holmer and Frederiksen, 2007;Norði et al., 2011). Remineraliztion of the highly labile organic waste (i.e. fish feed and faeces) results in increased sediment oxygen demand and altered metabolic pathways, and a shift from aerobic (i.e. heterotrophic respiration) to anaerobic (i.e. sulphate reduction and methanogenesis) microbial degradation (Holmer and Kristensen, 1992;Holmer et al., 2003;Valdemarsen et al., 2009). Excessive organic enrichment can thus lead to highly modified sediment conditions (Valdemarsen et al., 2012), impacting the structure and biomass of faunal communities (Kutti et al., 2007b;Hargrave et al., 2008;Valdemarsen et al., 2010;Bannister et al., 2014). Increased release Risk assessment of the environmental impact of dissolved nutrients from fish farming activities may stimulate phytoplankton growth and plankton blooms (Gowen and Ezzi, 1994) and may change the composition of seaweed communities in the littoral zone (Rueness and Fredriksen, 1991;Bokn et al., 1992;Munda, 1996;Pihl et al., 1999;Worm and Sommer, 2000;Krause-Jensen et al., 2007).

Risk estimation Organic loading on a local scale
On a local scale the endpoint: unacceptable change in faunal communities and sediment chemistry in the production zone is estimated. The criterion of unacceptable change is determined by Norwegian authorities and all salmon farms in Norway are monitored through mandatory investigations (MOM system;Hansen et al., 2001). The MOM-B investigations are performed regularly under and in the closest vicinity of the fish cages and are based on qualitatively determined indicators such as chemical parameters (pH and redox potential), sensory parameters, and presence and/or absence of macro-infauna. The performance of these indicators against predefined thresholds categorizes the farming locations into different environmental conditions (1. low-, 2. medium-, 3. high-organic loading, and 4. organic overloading). The environmental condition 4 represents an unacceptable state when production cannot continue before the farming location has recovered. Data from the monitoring of Norwegian salmon farms are obtained from the Norwegian Directorate of Fisheries, including 2761 electronically reported MOM-B investigations undertaken beneath Norwegian fish farms between 2009 and 2013. The percentage of farms in an unacceptable ecological state (4) has been stable and ,3% the last few years ( Figure 5). This is probably a result of better localization of farms. The risk of unacceptable change in sediment chemistry and fauna communities in the production zone is low. However, according to national set thresholds for management of the production zone, low impact on this scale does not reflect pristine conditions, but merely that the farm is managed within acceptable conditions in regard to its local impact.
Several rigorous scientific examinations of benthic impacts have been conducted near salmon farms in western Norway (Hordaland) where benthic carbon loadings, benthic fauna responses, and sediment biogeochemical processes were studied (Kutti et al., 2007a, b;Kutti, 2008;Valdemarsen et al., 2012;Bannister et al., 2014). In addition, there are generic scientific knowledge in respect to the flow of organic waste into benthic foodwebs along the Norwegian coast (Olsen et al., 2012).

Organic loading on a regional scale
The MOM-C system is an extended investigation of several sites (1-5) in the extended influence zone around farms and consists of quantitative measurements of the organic enrichment and the impact on biodiversity in infauna communities. The MOM-C investigations of fauna communities are following the Norwegian Standard (NS 9410) and farming sites are categorized into different environmental states (i.e. very good, good, moderate, poor, and very poor) according to nationally set thresholds (Molvaer et al., 1997). Based on the hydrographical conditions around the farm a distant point should be indentified in the most likely accumulation area beyond the production zone. The ecological condition at this distant site could be used as a proxy to estimate the risk of the endpoint; significant change in bottom communities beyond the production zone (regional impact).
Data from MOM-C investigations on 122 salmon farms, which represents 10% of the farms currently operating in Norwegian waters, are compiled from the Norwegian Directorate of Fisheries. The data show that distant sites at 95% of the farms had a high or very high ecological classification according to national set thresholds for Shannon -Wiener diversity index (H ′ ) (Molvaer et al., 1997) while 5% was classified in moderate conditions.
To provide an estimate of risk for impact on a regional scale, data have been retained from case studies in the Hardangerfjord (Husa et al., 2014a) and regional monitoring in some other areas in Norwegian coastal waters (Vassdal et al., 2012) according to parameters and thresholds defined in the Norwegian implementation of the European Water Framework Directive (Molvaer et al., 1997;WFD, 2000/60/EC;Anon., 2009a). These data show that the ecological conditions in fauna communities and oxygen values in deep regional basins are high to very high in fjords with high salmon farming activity. These findings were also supported by analysis of the relative importance of the extra contribution of organic farm waste to decomposing communities in the deep basins in the Hardangerfjord, estimating that current farming production increased oxygen consumption by 10% and decreased oxygen levels in bottom water with 0.09% (Aure, 2013). However, we do not have sufficient data from the entire Norwegian coastline to make a full risk estimation of the impact of organic loading on a regional scale.

Nutrient emissions on a local scale
Local impact from nutrients and fine particulate material in the euphotic zone are currently not monitored around Norwegian fish farms, and we therefore have no data to estimate the endpoint; nutrients from fish farms results in local eutrophication.

Nutrient emissions on a regional scale
To estimate the endpoint, nutrients from fish farms results in regional eutrophication, we do not have sufficient data from Norwegian coastal waters to fulfil a complete risk estimation. However, three years monitoring of nutrient values and chlorophyll a in the Hardangerfjord area (Husa et al., 2014b) and in Rogaland County, a sensitive area for fish farming due to lower water exchange (Vassdal et al., 2012), show that ecological conditions for these parameters are within national acceptances thresholds (Molvaer et al., 1997) suggesting high or very high water quality.
These data coupled with modelling estimations on potential increase in phytoplankton production ( Figure 6; Skogen et al., 2009) suggest low risk of regional impacts from aquaculture in Norway. The potential increase in phytoplankton production is based on knowledge about the water transport mechanisms, coupled with typical natural values of nitrogen and phosphorous in the Norwegian Coastal Current and the calculated extra contribution to nutrient concentrations from fish farms in each Norwegian county. Assuming that theoretically all the nitrogen released from fish farms is assimilated in phytoplankton growth, an increase in the natural phytoplankton biomass were calculated and compared with the threshold of a 50% increase in phytoplankton biomass that is defined as eutrophication by OSPAR (Anon., 2010).

Limitations
The risk assessment in the period 2010-2013 is based on a limited dataset, restricted scientific knowledge, and national monitoring methods that require upgrading/revising. On a local scale, the use of the MOM-B monitoring dataset for assessing local impacts of organic enrichment should be used cautiously. The MOM-B system is built on a qualitative assessment limited in its efficacy outside of soft sediment habitats. Using the "MOM system" to monitor the environmental effects to other benthic habitats such as hard, mixed and sandy bottom habitats, seaweed and kelp habitats, or other sensitive habitats including sponge aggregations and cold water corals reefs will lead to uncertain monitoring results. Therefore, reported benthic conditions underneath fish farms should be used cautiously in the risk assessment approach. The number of MOM-C investigations available for this risk assessment was limited to 122 fish farming sites.
Considering there are more or less 1000 fish farming locations along the Norwegian Coastline, the use of these investigations to provide an overview of the impact of organic enrichment from aquaculture within the influence area, should be approached cautiously, given they only represent 10% of current production sites. Scientific data are restricted to fjord habitats in western Norway, there is a dearth of scientific studies representing the different geographical settings (i.e. southern, mid, and northern Norway) and the different benthic habitat types (coastal sandy sediments, hard bottom habitats, and sensitive and vulnerable habitats).
Furthermore, at regional scales national monitoring programmes have only started to begin (from 2008); therefore, existing knowledge and data of regional impacts from fish farming are mainly restricted to two counties and limited time. These limitations result in a precautionary approach used in the above risk assessment.

Future work
To achieve the goal of the Ministry of Fisheries and Coastal Affairs that the environmental impact of aquaculture must be kept at an acceptable level and be within the assimilative capacity of the area, further scientific research and coastal monitoring efforts are needed. It is crucial that further knowledge is developed on understanding the interaction of organic and nutrient waste release on different habitat types (ca. hard bottom habitats, coastal sandy sediment habitats, and benthic boreal systems). In addition, habitats of ecological significance and sensitive species (ca. coral reefs, sponge aggregations, maerl beds, seagrass meadows, and spawning areas) require detailed investigations to understand their responses to organic and nutrient loadings, thus allowing more informed decisions on ecological impacts to be made. Furthermore, given the heterogeneity of benthic substrates along the Norwegian coastline, and the existing limitations of the MOM-B monitoring standard, improved monitoring tools need to be established to enable monitoring of local impacts from fish farms on non-soft sediment substrates. MOM-C investigations should be performed more often and at a greater number of stations along a gradient from enrichment. Furthermore, sampling for therapeutants and fatty acids should be incorporated in the MOM-C investigations to detect the pressure of drugs on the environment and also to determine if the impacts detected are in fact related to fish farming activities.
Moreover, there is also a need to develop new modelling tools that can predict the dispersal of organic and nutrient wastes from fish farms, which will enable better placement of monitoring stations for both MOM-C and regional monitoring to increase the likelyhood of detecting impacts. Finally, regional monitoring programmes should incorporate greater sampling coverage (i.e. sampling locations) and frequency along the entire Norwegian coastline. Monitoring programmes should also identify possible risk areas for regional impacts and place further emphasis on monitoring in these habitats.

Discussion
The main approach in the risk assessment on the environmental impact of fish farming in Norway was to review the state-of-the-art on various hazards and potential risk factors, review national Figure 6. Estimated percentage increase in phytoplankton production due to emissions of dissolved nitrogen from finfish farms in 2012 in each Norwegian county, based on 100% exploitation of the nitrogen to carbon fixation.
Risk assessment of the environmental impact monitoring data with relevance for the risk assessment, and score qualitatively the risk of impact in broad categories; low, moderate, and high risk of impact when monitoring data were available (Taranger et al., 2011a(Taranger et al., , b, 2012b. When the first risk assessment of the environmental impact of Norwegian fish farming was initiated in 2010 (Taranger et al., 2011a), only very broad definitions and goals of aquaculture sustainability were presented in a strategy document by the Norwegian government (Anon., 2009b). The goals for sustainability put forward in this document are listed in Table 1, and covers the topics; diseases/parasites, genetic interactions with wild populations, pollution and discharges, marine site structure and zoning, as well as sustainability of feed and feed resources. These goal are very generic and do not specify sustainability indicators and related thresholds for societal/political acceptance of the environmental impact of fish farming in Norway.
The Institute of Marine Research in Norway and the Norwegian Veterinarian Institute were requested by the Norwegian Ministry of Fisheries and Coastal Affairs in 2011 to propose scientifically based sustainability indicators and related thresholds to score the severity of the potential environmental impacts of aquaculture. This included indicators for the risk of genetic introgression of escaped farmed salmon as well impact of salmon lice from fish farming on wild salmonid populations. As a result, indicators and suggested thresholds to score the environmental impact as low, moderate, or high were proposed in 2012 (Taranger et al., 2012a). These recommendations were in part approved and implemented into a governmental report on Norwegian seafood policy in 2013 (Anon., 2013c), and in a newly established Quality Norm for the management and protection of wild Atlantic salmon in Norway approved by the Norwegian government in late 2013 (Anon., 2013a).
As a result, the suggested indicators for environmental impact as the associated suggested thresholds for scoring of the impacts have been adopted in the most recent version of the Risk assessment of Norwegian fish farming (Taranger et al., 2014). They are the basis of the current assessment of risk of genetic introgression and impact on salmon lice on wild salmonid populations. The assessments are based on monitoring programmes on the numbers of escaped salmon in Norwegian rivers over the last 3 years, and the level of salmon lice infections on wild salmonids in Norwegian coastal waters during the last 4 years.
New scientific findings have recently emerged, such as the measured level of introgression of escaped farmed salmon into wild salmon populations in 20 Norwegian rivers , and new estimates on the additional marine mortality caused by salmon lice on Atlantic salmon from the rivers Dale and Vosso, in Hordaland, Norway Vollset et al., 2014). Such findings assist in developing more quantitative assessments of the impact of number of escaped salmon in rivers, and the impact of salmon lice from farming on wild salmonids, respectively.
In contrast to the situation for genetic introgression and impact of salmon lice, there are very limited monitoring data on potential transfer of other diseases and parasites from salmon farming to wild fish in Norway. Hence, the assessment on the risk of disease transfers to wild fish is mainly based on an analysis of the frequency of disease outbreaks in the 1000 sites for salmon farming along the Norwegian coast. This is supported with a review on the knowledge of risk of disease transmission for the most relevant pathogens, and some available data on prevalence of pathogens and/or any disease outbreaks in wild Atlantic salmon and sea trout populations in Norway (Taranger et al., 2014).
Regarding the local and regional impact of organic load and release of nutrient from marine salmon farming, such environmental impact indicators and associated monitoring programmes are defined and adapted by the Norwegian fish farming authorities regarding the local zone under and close to the farms, whereas the indicators and monitoring programmes for regional effects are being implemented in some counties in Norway (Vassdal et al., 2012;Husa et al., 2014a). The local zone under and close to the farm is monitored with a risk-based frequency using the relatively simple MOM-B method, while the more sensitive MOM-C method with detailed analysis of the species compositions in soft bottom samples near the farms is only applied occasionally (Hansen et al., 2001). Both these methods have limitations, e.g. they require soft bottom, and are currently under revisions. Data from regional monitoring has only become available in a few counties in the last years, but new programmes are starting up in several counties. The regional monitoring will to a large degree be based upon environmental indexes and environmental quality elements and related threshold for scoring of quality according to the Norwegian implementation of EUs Water Framework Directive (Anon., 2009a).
The current risk assessment of local of organic matter and nutrient release are based on the mandatory MOM-B monitoring from all farms and MOM-C analyses from a limited number of fish farms (Taranger et al., 2014). The regional impact is evaluated based on available models and some investigations in the counties of Hordaland and Rogaland which both have high salmon farming activity compared with the area of available coastal water. As discussed above under the section on impact of organic matter and nutrients, the analyses have limitations, and a more extensive and improved monitoring programme is needed both on local and regional scale.
Moreover, a range of other potential risk factors, such as use of various pharmaceuticals, transfers of xenobiotics with the feed, use of copper as antifouling agent on sea cages, interactions with fisheries and other ecological impacts of sea cage farming, as well as ecological impacts of catch, transport and use of wrasses as cleaner fish against salmon lice are discussed in the risk assessment of Norwegian fish farming (Taranger et al., 2014), but are not included in the current analysis.

Conclusions and summary of main findings
This represents the first risk assessment of cage-based salmonid aquaculture in Norway, which is world's largest producer of farmed Atlantic salmon. While there are several limitations in the approaches used to estimate the risks, as has been discussed in the sections above, this work has provided the Norwegian authorities with a framework upon which to evaluate the most important identified hazards against environmental goals for sustainability. The primary findings from the present risk assessment can be summarized as follows: † Based upon the observed frequency of farmed escaped salmon on the spawning grounds of wild populations in the period 2010-2012, 21 of the 34 populations included in the risk assessment were in moderate-to-high risk of experiencing genetic changes due to introgression of farmed salmon. However, a recent study of 20 Norwegian rivers has demonstrated that there is only a moderate correlation between the observed frequency of escapees and introgression of farmed salmon ; therefore, validation of the level of introgression in a higher number of native populations will be required in the future. † During the period 2010-2013, salmon lice infections mainly resulting from salmon farming were estimated at a total of 109 stations covering relevant areas of the Norwegian coastline using wild sea trout as a proxy for local infection pressure on wild salmonids. Twenty-seven of these stations indicated moderate or high likelihood of mortality for wild migrating salmon smolts. For sea trout later in the season, 67 of the stations indicated moderate or high likelihood of mortality on wild sea trout. † The high frequency of the viral disease outbreaks for PD, IPN, heart and skeletal muscle inflammation, and CMS in Norwegian salmon farming suggests extensive release of the causal pathogens for these diseases in many areas. Migrating wild salmon and local sea trout are likely to be exposed to these pathogens. However, the extent of this exposure and consequences remains largely unknown. Screening of wild salmonids has revealed low to very low prevalence of the viruses SAV, IPNV, PMCV, and low prevalence of PRV in salmon. Furthermore, these viruses have never been documented to cause disease in wild Norwegian salmonids. Thus, a general lack of data prohibits complete risk estimation for these diseases. † From a total of 500 yearly investigations of local organic loading under fish farms, 2% of them displayed unacceptable conditions in the benthic sediments and faunal composition in 2013, whereas 11% classified with a high organic loading but still within the threshold. The remaining 87% of the farms had a moderate-to-high ecological conditions. The risk of eutrophication and organic over loading in the benthic communities beyond the production area of the farm is considered low based upon case studies and monitoring data from a limited area of the Norwegian coast.
Given the rapid expansion of open sea cage farming in Norway, and internationally, and the range of ecological impacts that either are demonstrated or suspected, there is an urgent need for better knowledge about such impacts, to implement improved monitoring programmes for the most important hazards, and also to improve procedures for risk assessments including useful environmental risk indicators and to facilitate processes that involves definitions on the societal acceptance levels of the various impacts.