• Open Access

A Model to Approximate Lake Temperature from Gridded Daily Air Temperature Records and Its Application in Risk Assessment for the Establishment of Fish Diseases in the UK

Authors


M. A. Thrush, Centre for Environment, Fisheries and Aquaculture Science, Weymouth, DT4 8UB, UK. Tel.: 00 44 (0)1305 206723; Fax: 00 44 (0)1305 206601; E-mail: mark.thrush@cefas.co.uk

Summary

Ambient water temperature is a key factor controlling the distribution and impact of disease in fish populations, and optimum temperature ranges have been characterised for the establishment of a number important aquatic diseases exotic to the UK. This study presents a simple regression method to approximate daily average surface water temperature in lakes of 0.5–15 ha in size across the UK using 5 km2 gridded daily average air temperatures provided by the UK Meteorological Office. A Geographic information system (GIS) is used to present thematic maps of relative risk scores established for each grid cell based on the mean number of days per year that water temperature satisfied optimal criteria for the establishment of two economically important pathogens of cyprinid fish (koi herpesvirus (KHV) and spring viraemia of carp virus (SVCV)) and the distribution and density of fish populations susceptible to these viruses. High-density susceptible populations broadly overlap the areas where the temperature profiles are optimal for KHV (central and south-east England); however, few fish populations occur in areas where temperature profiles are most likely to result in the establishment of spring viremia of carp (SVC) (namely northern England and Scotland). The highest grid-cell risk scores for KHV and SVC were 7 and 6, respectively, out of a maximum score of 14. The proportion of grid cells containing susceptible populations with risk scores of 5 or more was 37% and 5% for KHV and SVC, respectively. This work demonstrates a risk-based approach to inform surveillance for exotic pathogens in aquatic animal health management, allowing efficient use of resources directed towards higher risk animals and geographic areas for early disease detection. The methodology could be used to examine the change in distribution of high-risk areas for both exotic and endemic fish diseases under different climate change scenarios.

Introduction

Ambient water temperature is a key factor controlling the distribution and impact of disease in fish populations. Firstly, as fish are poikilothermic, temperature directly influences all aspects of their physiology, including immune response (Bowden et al., 2007), and it is generally accepted that lower water temperatures adversely affect both specific immune responses mediated by T helper cells and non-specific defences including phagocytosis and cytotoxicity (see Le Morvan et al. (1998) for review). Generally, the fish immune system is optimal at that species’ normal summer temperatures (Manning and Nakanishi, 1996). Secondly, pathogens have optimal temperature ranges for replication. The generation time of bacteria, fungi and parasites with direct lifecycles decreases with increasing water temperature (Gubbins, 2006) thus increasing the size of pathogen populations (Harvell et al., 2002), challenge levels, disease and mortality. Different viruses have their own optimal temperature range [for example, infectious haematopoietic necrosis virus grows optimally in vitro at 15°C (Mulcahy et al., 1984) and spring viraemia of carp virus (SVCV) grows optimally in vitro at 20–22°C (Fijan, 1998)]. These factors result in specific host–pathogen interactions that determine characteristic temperature thresholds for the expression of clinical disease and critical ranges for maximum pathogen virulence and host mortality in a susceptible population (Marcos-López et al., 2010).

Many salmonid diseases endemic in the UK show clinical expression above a threshold temperature including enteric redmouth (>8°C) (Horne and Barnes, 1999), furunculosis (>10°C) (Malnar et al., 1988); bacterial kidney disease (>13°C) (Jones et al., 2007) and proliferative kidney disease (PKD) (>15°C) (Sterud et al., 2007), while other pathogens cause disease below a threshold temperature including sleeping disease (<11°C) and rainbow trout fry syndrome (<10°C) (Faruk et al., 2002). For species that inhabit higher thermal ranges, disease expression may be constrained by lower and upper bounds, for example koi herpesvirus (KHV) does not cause morbidity below 13°C or above 30°C (Hedrick et al., 2000). Clear optimum temperature ranges have been characterised for the establishment of a number of important exotic disease threats to the UK, for example the most severe outbreaks of viral haemorrhagic septicaemia and infectious haematopoietic necrosis (IHN) occur between 9–12 and 10–12°C, respectively (Smail, 1999; Amend, 1975); spring viremia of carp (SVC) has an optimal range of 10–17°C (Ahne et al., 2002) and epizootic haematopoietic necrosis occurs at temperatures in excess of 11°C (Whittington and Reddacliff, 1995). Seasonal temperature characteristics are therefore a key factor for disease emergence in the aquatic environment, influencing the likelihood that an exotic pathogen will become established and spread following introduction to a new location. Spatial and temporal analyses of air temperatures have provided predictions for the likely spread of vector human livestock diseases (Cook, 1992; Ogden et al., 2005; Patz et al., 2005). A similar analysis of ambient water temperatures will help identify geographic areas at highest risk of establishment of different exotic pathogens following introduction and thus inform where surveillance activities should be targeted.

This study presents a simple regression method to approximate daily average surface water temperature in lakes across the UK using gridded daily data sets of average air temperature at a 5 km by 5 km resolution provided by the UK Meteorological Office. A GIS is used to create thematic maps to display the estimated average number of days per year that water temperature satisfies key criteria for disease establishment. Example risk maps are presented and discussed in relation to two diseases: KHV, which is endemic to the UK (Taylor et al., 2010a), and SVC, which is exotic. Both are economically important and may impact wild and stocked populations of cyprinid fish in lakes and fisheries.

Materials and Methods

Water temperature data

Data sets of daily average water temperature for comparison with air temperature records in specific locations were derived from information generated by Tinytag automatic loggers (Gemini Data Loggers (UK) Ltd, Chichester, UK) deployed in the surface waters (top 0.5 m) of four lakes and set to record temperature at 3-hourly intervals from May 2002 to June 2003 (see Taylor et al. (2009) for further information on temperature data collection, Table 1 for lake characteristics). Average daily temperatures were calculated as the mean of 8 temperature records per day (raw data were provided by Nick Taylor, Cefas).

Table 1. Location, size and data collection information for four lakes providing surface water temperatures for the predictive model
Lake ID and countyUK grid cellSurface area (ha)Average depth (m)Estimated volume (m3/ 1000)Start and end dates of water temperature records (days)
Lake 1, Dorset33705 0.62.0 1231/5/2002 – 8/6/2003 (373)
Lake 2, Gloucestershire35751 3.62.2 7929/5/2002 – 10/6/2003 (377)
Lake 3, Gloucestershire35751 9.03.5315 3/6/2002 – 3/6/2003 (365)
Lake 4, Staffordshire3403315.06.090023/5/2002 – 5/6/2003 (378)

An additional long-term data set (1970–2011) of daily surface water temperatures recorded at the same time each day (9:00 am) for Lake Windermere (Cumbria, UK) was used to independently estimate the most likely date of peak annual temperature (using the mean of 3–36 records (median = 32) for each of 366 days of the year) (Fig. 1). These data were made available by the Freshwater Biological Association.

Figure 1.

 Average 9:00 am surface water temperature of Lake Windermere (mean for each day of year [Doy], 1970–2011; Freshwater Biological Association data) and corresponding offset for maximum Cos(Doy).

Air temperature data

Gridded daily average air temperatures were provided by the Meteorological Office. These data are formatted in text files corresponding to 180 × 290 (52200) 5 km2 squares covering the UK (one file for each day). The data are compiled from raw meteorological station records extracted from the climate data archive using National Climate Information Centre climate data analysis software according to well-defined rules and quality control (see Perry et al. (2009), Perry and Hollis (2005) for further information). A VBA procedure (MS-Excel) was developed to manage the grids, and daily air temperatures were extracted from specific geographical grids over timeframes corresponding to the water temperature data sets of the four lakes detailed in Table 1.

Statistical analysis

A linear regression model based on the method described by Matuszek and Shuter (1996) was developed to approximate daily average still-water surface temperatures using average air temperatures of preceding days combined with a time function:

image(1)

where

image

Optimal air temperature averaging period

The optimal air temperature averaging period (x) for predicting water temperature was estimated based on maximum r2 and minimum root mean square error (RMSE) of regressions of mean water temperature on mean average air temperature over preceding days for each of the four test lakes using STATA 10 (StataCorp, 2007) (see results section and Table 2).

Table 2. Results of linear regression analyses used to determine the optimal air temperature averaging period for four lakes (maximum r2 and minimum root mean square error (RMSE) for groups of uni- and multi-variate models are marked with asterisks, lakes are ordered by increasing size)
Regression variablesCoefficientsStatistics
a1 a2 b r2 F df RMSE
  1. P = 0.000 for all regressions.

Lake 1
 1 day mean air temp1.009 1.5570.77312661, 3712.561
 5 day mean air temp1.127 0.3600.87124661, 3671.934
 6 day mean air temp1.142 0.1980.87926581, 3661.869
 7 day mean air temp1.556 0.0610.88528061, 3650.823
 8 day mean air temp1.167 −0.0580.88929111, 3641.793
 9 day mean air temp1.177 −0.1570.89129621, 3631.778
 10 day mean air temp1.186 −0.2390.891*29631, 3621.777*
 11 day mean air temp1.193 −0.3090.89029261, 3611.786
 12 day mean air temp1.199 −0.3690.88828611, 3601.803
 10 day mean air temp; Cos(Doy150)0.5220.593−0.5520.938*27202, 3611.345*
 11 day mean air temp; Cos(Doy150)0.5120.601−0.5470.93525872, 3601.377
Lake 2
 1 day mean air temp1.050 1.6420.81816901, 3762.501
 5 day mean air temp1.158 0.4940.89833011, 3721.871
 6 day mean air temp1.172 0.3440.90635561, 3711.808
 7 day mean air temp1.183 0.2200.91037501, 3701.764
 8 day mean air temp1.193 0.1110.91439061, 3697.730
 9 day mean air temp1.202 0.0180.91640181, 3681.708
 10 day mean air temp1.211 −0.0660.91740741, 3671.696
 11 day mean air temp1.218 −0.1390.918*40921, 3661.692*
 12 day mean air temp1.224 −0.2010.91740531, 3651.699
 10 day mean air temp; Cos(Doy150)0.5470.633−0.4170.951*35272, 3661.312*
 11 day mean air temp; Cos(Doy150)0.5450.636−0.4220.94933852, 3651.338
Lake 3
 1 day mean air temp1.081 0.8800.81415861, 3632.625
 5 day mean air temp1.201 −0.3980.91036281, 3591.828
 6 day mean air temp1.219 −0.5830.92041161, 3581.725
 7 day mean air temp1.234 −0.7340.92745221, 3571.650
 8 day mean air temp1.247 −0.8640.93248571, 3561.595
 9 day mean air temp1.257 −0.9740.93551171, 3551.556
 10 day mean air temp1.266 −1.0640.93752481, 3541.536
 11 day mean air temp1.273 −1.1380.937*52731, 3531.532*
 12 day mean air temp1.279 −1.1980.93751991, 3521.540
 10 day mean air temp; Cos(Doy150)0.6610.576−1.3660.963*45722, 3531.180*
 11 day mean air temp; Cos(Doy150)0.6670.571−1.3740.96143122, 3521.213
Lake 4
 1 day mean air temp1.109 0.1170.80515421, 3742.647
 5 day mean air temp1.248 −1.2710.90836481, 3701.825
 6 day mean air temp1.259 −1.4520.91841111, 3691.727
 7 day mean air temp1.274 −1.6080.92444791, 3701.660
 8 day mean air temp1.287 −1.7380.92847551, 3671.614
 9 day mean air temp1.298 −1.8480.93149481, 3661.584
 10 day mean air temp1.307 −1.9440.933*50541, 3651.568
 11 day mean air temp1.315 −2.0250.93350471, 3641.565*
 12 day mean air temp1.321 −2.0910.93249991, 3631.571
 10 day mean air temp; Cos(Doy150)0.7380.522−2.1700.956*39452, 3641.270*
 11 day mean air temp; Cos(Doy150)0.7490.514−2.1790.95337212, 3631.304
All data
 10 day mean air temp; Cos(Doy150)0.6550.522−1.1630.947129002, 14551.359

Time function

The cosine of day of year, Cos(Doy [radians]), was used to provide the time function for the predictive model. Doy was derived using the Days360 Excel worksheet function, which returns the number of days between two dates from a year based on 12 equal months of 30 days. For day n this is

image(2)

where

image

An offset (y) was applied to match the maximum time function value (1) to the most likely date of peak annual temperature for application of a general (multi-year) model. This was estimated, using the Lake Windermere data set, to be 30 July (Doy = 210). The offset required for the time function was therefore 360 (maximum Doy) −210 = 150 (see Fig. 1).

Creating and displaying estimated average surface water temperatures

Gridded data files of predicted average daily water temperature across the UK were generated for 1997–2006. Coefficients provided by regression of combined data for all test lakes were used with a cosine offset of 150 days and air temperatures averaged over the preceding 10 days (see results section for detailed analysis) for each grid cell (using Met Office air temperature files starting 23 December, 1996). The parameters for final predictive model (Eqns 1 and 2) were therefore:

  • = 10

  • = 150

  • a1 = 0.655

  • a2 = 0.522

  • = −1.163

  • DateRef = 31 December, 1995

The water temperature files were then analysed to calculate the mean number of days per year (over 10 years) that average lake temperatures were predicted to be (i) above, (ii) below single threshold values or (iii) between sets of two threshold values. Various threshold temperatures were chosen based on the establishment criteria of different fish diseases of interest. Outputs (mean numbers of days satisfying input criteria) were converted to ten 36.5-day time interval categories. These categories were used as a relative risk scoring scheme for ranking 5 km squares for the establishment of disease (see Table 3). Results were displayed thematically on maps using a 180 × 290 grid projected onto the British National Grid coordinate system (origin x, y = −200 k, −200 k) (ArcMap V9.3; ESRI Corp. Redlands, CA, USA).

Table 3. Methodology for calculating risk categories for individual UK 5 km2 grid cells
Temperature profilePopulation densityCombined risk
Days per year satisfying risk condition (mean of 10 years)Score 1 Score 2*Score 3**
  1. *maximum value for Score 2 = 4; **Score 3 was set to zero if either Score 1 = 0 or Score 2 = 0.

329–36510(i) Total farms and fisheriesScore 1 + Score 2
293–3289
256–2928 9–123
220–2557 5–82
183–2196 1–41
147–1825 None0
110–1464+
74–1093
34–732(ii) Wild populations
1–361 Present1
00 Absent0

Distribution of susceptible species

Fish species were determined to be susceptible to specific diseases as listed by the European fish health directive 2006/88/EC (Anon., 2006) (Part II Annex IV, amended by 2008/53/EC). For KHV, these are common carp and koi carp (Cyprinus carpio). Species susceptible to SVC include, in addition, bighead carp (Aristichthys nobilis), goldfish (Carassius auratus), crucian carp (Carassius carassius), grass carp (Ctenopharyngodon idellus), silver carp (Hypophthalmichthys molitrix), sheatfish (Silurus glanis), tench (Tinca tinca) and orfe (ide) (Leuciscus idus).

The location of fish farm and fishery sites holding managed stocks of susceptible fish in England and Wales was extracted from the Cefas live fish movements database (LFMD). A relative risk scoring scheme for site density was applied based on the number of sites contained in geographically referenced 5 km2 grid squares (1–4 sites scored 1, 5–8 sites = 2, 9–12 sites = 3). The distribution of susceptible wild fish in England, Wales and Scotland was determined using the Database and Atlas of Freshwater Fish (DAFF) (Davies et al., 2004). Point data were matched to 5 km2 grids squares by GIS SQL select query (MapInfo v10; Pitney Bowes, New York, NY, USA.) Sampling bias in the DAFF data prevented accurate assessment of stock density, hence a score (=1) for wild fish presence only was assigned. Results for managed and wild stocks were combined to provide scores of overall population density for each grid square (maximum = 4, see Table 3), which were displayed thematically on maps. No fish population data were available for Northern Ireland.

Overall risk of establishment of disease

An additive model was used to combine scores to achieve a relative approximation of overall risk of disease establishment. Each grid square was scored according to the sum of (i) the risk score for number of days per year that average water temperature satisfied the disease risk condition and (ii) the score for susceptible population density for that disease. Overall risk was set to zero if either the temperature or density components were zero (maximum score = 14, Table 3). The results were again presented thematically on maps (categories of high-to-low risk).

Results

Model development

Daily air temperature was found to be a significant predictor of surface water temperature in the four lakes examined. Increasing the averaging period for air temperature in preceding days in single parameter regressions improved model fit (explained variance indicated by r2) from 77% (1-day average air temperature, Lake 1) to 94% (10 days, Lake 3) (see Table 2). The coefficient for average air temperature in the resulting relationships (a1) increased with lake size (Table 2). The optimal air temperature averaging period was relatively consistent for all lakes (either 10 or 11 days) resulting in RMSE values of between 1.5 and 1.8°C. Inclusion of the time function (Doy150) in regression models increased the r2 for all lakes (to 94–96%) and reduced RMSEs to 1.2–1.3°C. The optimal air temperature averaging period (x) for the multiple regressions was 10 days for all lakes. The regression of daily average water temperature on 10-day mean air temperature and Doy150 for all lake data combined also provided a good fit (r2 = 95%, RMSE = 1.36°C), and the coefficients obtained from this were used to provide a general predictive model for average water temperature. Graphical comparisons of resulting outputs for the four test lakes (Fig. 2) did not reveal any substantial or consistent departure of observed and predicted water temperatures.

Figure 2.

 Observed average daily water temperatures (black lines) and water temperatures estimated by the predictive model using air temperature data (blue lines) for four lakes.

Figure 3 demonstrates how analysis of the gridded data files of predicted average daily water temperature generated by the model can be used to provide a general spatial representation of specific temperature threshold characteristics for lake surface waters across the UK. The mean number of days per year above a range of threshold values illustrates the north-south trend in average temperatures and the influence of altitude (for example, Peak district and Scottish Highlands).

Figure 3.

 Number of days per year that estimated daily average lake surface temperatures exceed different thresholds (means of 10 years, 1997–2006).

Risk mapping

The average number of days per year that mean daily lake water temperatures were estimated to be optimum for the establishment and expression of KHV (i.e. ≥16°C) across the UK ranged from 3.6 to 140. The results fall into four time interval categories (1–4, Fig. 4i, see Table 3 for explanation of risk categories), with 4958 grid cells (48% of the UK) in category 4 (110–146 days) (Table 3) and covering areas predominantly in south and central England and coastal margins of Wales. All other areas are category 3 (74–109 days) with the exception of high altitude ground in the Scottish Highlands [category 2 (7.6%), category 1 (1.7%)]. Populations of fish susceptible to KHV (common and koi carp) occur in 3480 grid cells and broadly overlap the areas with the highest number of days per year where the temperature is optimal for the establishment of this disease (Fig. 4ii) (distribution data were not available for Northern Ireland). The highest density populations (13 cells with a density score of 4) occur in the west midlands and south-east England. Combining the temperature and population distribution information established that the highest overall risk score for KHV was 7 (Fig. 4iii). Twelve grid cells received the highest score and 116 cells scored 6 (a combined total of 3.7% of grid cells containing susceptible fish). A further 33% of grid cells scored 5. The mean risk score for KHV was 4.31. High-score cells clustered predominantly in northwest and south-east England.

Figure 4.

 (i) Number of days per year that estimated daily average lake surface temperatures satisfy optimum criteria for the establishment koi herpesvirus (KHV) (≥16°C, mean of 10 years, 1997–2006), (ii) the distribution of KHV susceptible species (population density) and (iii) combined risk categories for the establishment of KHV.

The average number of days per year that mean daily water temperatures were optimum for the establishment of SVC (i.e. 10–17°C) across the UK ranged from 90.3 to 158.6. Results for individual grid cells fell into three time interval categories (Fig. 5i): category 3 (74–109 days) covering most of England, with the exception of the extreme southwest, south-west moorlands, northern high ground and south Wales; category 4 (110-146 days), including all remaining areas with the exception of all small minority (95 cells = 0.92%) in category 5 (147–182 days) located in the west Scottish mainland, the Hebrides and eastern Shetland. The inclusion of additional species susceptible to SVC increased the distribution of populations at risk by 557 grid cells and the number of cells with high-density susceptible populations to 17 (Fig. 5ii). In contrast to the situation with KHV, we find that very few populations susceptible to SVC in the UK occur in areas with temperature profiles most likely to result in the establishment of disease (Fig. 5i, ii). The highest score for overall risk of disease establishment was 6 (19 cells = 0.5% of those containing susceptible fish) and 184 cells had a risk score of 5 (4.6%). Again, areas of highest risk are in the west midlands and south-east England (Fig. 5iii). The majority of cells (48%) score 3 for overall risk of establishment of SVC (mean risk score for SVC, 3.6).

Figure 5.

 (i) Number of days per year that estimated daily average lake surface temperatures satisfy optimum criteria for the establishment of spring viremia of carp (SVC) (10–17°C, mean of 10 years, 1997–2006), (ii) the distribution of SVC susceptible species (population density) and (iii) combined risk categories for the establishment of SVC.

Discussion and Conclusions

It is well documented that air temperature is a reliable predictor of lake surface water temperature, and models of varying complexity have been published (Matuszek and Shuter, 1996; Honzo and Stefan, 1993; Robertson and Ragotzkie, 1990; McCombie, 1959; Livingstone and Padisák, 2007). Sharma et al. (2008) reported that multiple regression provided the best approach to modelling lake water temperatures, based on model performance and computational complexity, in a comparison of techniques including regression tree, artificial neural network and Bayesian methodology. The methodology developed in this study is based the approach taken by Matuszek and Shuter (1996), who demonstrated that a robust and practical air–water temperature regression model is achievable using relatively modest data requirements (average air temperatures over preceding days and a seasonal time function). Essentially, a parabolic seasonal trend is described, around which daily water temperatures fluctuate in response to changes in mean air temperature.

The predictive ability of the model we have developed has been constrained by the availability of long-term data from which reliable daily average water temperature can be derived for matching with air temperature records, and more validation work is required using independent data sets. However, the strong regression statistics obtained from analysis of the combined data suggest that this methodology provides a good platform for refinement. Model fit for the linear regression of water temperature on air temperature is likely to be reduced at extremes of temperature: that is, for water temperatures >18°C, evaporation will exert an increasing cooling effect on surface waters and for water temperatures <4°C, latent heat effects will delay the response to changing air temperature. These effects may be addressed using logistic regression functions; however, estimation of the total numbers of days per year above or below the key threshold temperatures for most aquatic diseases should be relatively unaffected as these do not fall in extreme ranges. The results are similarly likely to be minimally compromised by minor temporal phase-shift errors in water temperature predictions and provide sufficient resolution to inform surveillance activities for early detection of disease.

The estimated date of maximum water temperature used to position the time function was based on data from multiple years collected at a single location (Lake Windermere). This lake is much larger than those used for comparisons of air and water temperature, and it is possible that surface temperatures peak in Windermere later than in smaller water bodies as a result of its increased heat capacity. Furthermore, considerable between-year variation in maximum water temperature was evident (M. A. Thrush and E. J. Peeler, unpublished data this study) suggesting that the approximation of year-specific offsets using additional independent data sets from different locations and investigation of the sensitivity of the model to changes in time of maximum water temperature may improve predictive accuracy. There was also a significant difference in the volumes, and therefore the heat capacities, of the four lakes used to develop the model. Lake 4 is 75 times larger in volume but only 25 times larger in surface area than lake 1 (Table 1) and therefore requires three times the energy per unit surface area to heat by 1°C than lake 1. Incorporating a relatively long averaging time for air temperature (10 days) is likely to reduce the impact of these differences; however, one may expect the regression coefficient for temperature to increase with lake size. This was not the case in this study, an increasing gradient in the relationship between air and water temperature with lake size was observed and more work is required to investigate the reason for this. This result indicates that the application of the outputs from this study is likely to be limited to water bodies of equivalent size. However, for the proposed application, this would include the majority of fisheries in England and Wales (70% of registered fisheries in England and Wales are 0.5–15 ha, LFMD data). Further study of this relationship with data sets from additional lakes would provide a more accurate indication of the general applicability of the model.

Northern Ireland has been included in the analysis of optimum temperatures for the establishment of KHV and SVC, but as no population data were available, the overall risk categorisation could not be completed for grid cells in this territory. In addition, the LFMD only holds information on farms and fisheries in England and Wales, so the overall risk categories for some grid cells in Scotland may be underestimated as they are based on wild fish presence only. However, in comparison with England and Wales, there are relatively few managed coarse fish populations in Scotland as angling here is mainly for salmonids [<7% of total angler effort in Scotland is directed towards coarse fish species (Radford et al., 2004)].

The development of geographic information systems and spatial statistics has allowed detailed investigation of spatially distributed determinants of health and disease. Risk mapping has made a valuable contribution to the epidemiology of terrestrial animal and avian diseases (Lessard et al., 1988, 1990; Perry et al., 1990; Snow et al., 2007); however, its application to diseases in the aquatic environment has been limited (Thrush et al., 2011). This analysis has enabled the spatial comparison of the risk of establishment of two economically important diseases of cyprinid fish throughout the majority of their range in the UK and highlights previously unrecognised differences in the geographical distribution of these risks. Areas with the most suitable temperature profiles for the establishment of KHV coincide with high-density populations of susceptible fish (central and south-east England). In contrast, few susceptible populations in the UK occur in areas with temperature profiles most likely to result in the establishment of SVC (northern England and Scotland).

This study has investigated the risk of disease establishment based on two parameters (permissive temperature and host density). The methodology applied weights the temperature (10 categories) over host density (four categories), which was considered justifiable given that the wild fish population component was more variable and uncertain. The resulting estimates of risk were relative and for comparative purposes only. Clearly, this model could be extended to include other parameters (risk factors) for which geo-referenced data may be available to provide a more complete framework for assessment. Key factors include live fish movements (identified as a risk factor for KHV (Taylor et al., 2010b)) and, for parasitic diseases, the distribution of intermediate hosts (see Okamura et al. (2010)). The establishment of aquatic diseases may also be influenced by aspects of water chemistry or general water quality. Relevant data sources include the Environment Agency’s water information management and hydrometric databases, GIS layers of land cover (administered by the Centre for Ecology and Hydrology (Morton et al., 2011)) and geology (available from the British Geological Survey). Results from this study could be extended to include the effect of different climate change predictions. A differential response to temperature change across the UK could be accommodated using the grid-cell approach.

In addition to assessing the threat from exotic diseases and potential expansion in the range of some endemic diseases (e.g. KHV), this work may also be used to identify areas where diseases can be managed in farmed populations. For example, outbreaks of PKD in farmed rainbow trout (Oncorhynchus mykiss) typically occur at water temperatures above 15°C (Hedrick et al., 1993). At lower temperatures (<10°C), fish may become infected, but the disease is not clinically expressed (Gay et al., 2001; Clifton-Hadley et al., 1986). Endemically infected farms manage PKD by introducing naïve fish when the water temperature is below 10°C and the fish are sufficiently immuno-competent to mount an immune response (Ferguson, 1981; de Kinkelin and Loriot, 2001). These populations generally remain protected when the temperature subsequently increases and do not suffer mortalities.

Despite the limitations in data availability, we consider the model developed here makes a worthwhile contribution to risk-based surveillance for exotic pathogens in aquatic animal health management. It provides a good platform future refinement which will ultimately improve efficiency in the use of resources by directing them towards higher risk animal populations and geographic areas for early disease detection.

Footnotes

  • These data are freely available for research purposes (subject to registration) via UKCP09 - the UK Climate Impacts Programme (UKCIP): http://www.metoffice.gov.uk/climatechange/science/monitoring/ukcp09/.

  • Cefas developed and hosts the LMFD, a resource jointly owned by Cefas, Environment Agency, Welsh Government and Defra, to manage all data relating to Aquaculture Production Business authorisation and registration, aquatic animal imports and exports, the rearing and holding of non-native species and statutory aquatic disease testing and controls.

Acknowledgements

The authors acknowledge the UK Department for Environment, Food and Rural Affairs (Defra) for supporting this work (Defra contract F1185). We also thank the UK Met Office, Freshwater Biological Association and Nick Taylor (Cefas) for providing data.

Ancillary