modelling the impact of climate change on spatial patterns of disease risk: sheep blowfly strike by...

8
Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain Hannah Rose , Richard Wall Veterinary Parasitology and Ecology Group, School of Biological Sciences, University of Bristol, Bristol BS8 1UG, UK article info Article history: Received 5 November 2010 Received in revised form 19 January 2011 Accepted 25 January 2011 Available online 11 March 2011 Keywords: Blowfly strike Lucilia sericata Ovine cutaneous myiasis Maxent Species distribution model Climate change Disease abstract Understanding the spatial scale and temporal pattern of disease incidence is a fundamental prerequisite for the development of appropriate management and intervention strategies. It is particularly critical, given the need to understand the elevated risks linked to climate change, to allow the most likely changes in the distribution of parasites and disease vectors to be predicted under a range of climate change sce- narios. Using statistical models, the spatial distribution and climatic correlates of a range of parasites and diseases have been mapped previously, but their development into dynamic, predictive tools is less com- mon. The aim of the work described here, was to use a species distribution model to characterise the environmental determinants of the monthly occurrence of ovine cutaneous myiasis (blowfly strike) by Lucilia sericata, the most frequent primary agent of northern European myiasis, and to then use this model to describe the potential spatial changes that might be expected in response to predicted climate change in Great Britain. The model predicts that the range of elevated temperatures predicted by current climate change scenarios will result in an increase in the risk of strike and an elongated blowfly season. However, even for the most rapid warming scenario predictions over the next 70 years, strike is not predicted to occur throughout the winter. Nevertheless, in this latter case, parts of central and southern England are likely to become too hot and dry for strike by L. sericata, to persist in mid-summer. Under these con- ditions, it is possible that other, more pathogenic Mediterranean agents of myiasis, such as Wolfhartia magnifica, could potentially replace L. sericata. Where the phenology of strike is altered by climate change, as predicted here, significant changes to the timing and frequency of parasite treatments and husbandry practices, such as shearing, will be required to manage the problem. The results suggest that the model- ling approach adopted here could be usefully applied to a range of disease systems. Ó 2011 Australian Society for Parasitology Inc. Published by Elsevier Ltd. All rights reserved. 1. Introduction Understanding the spatial scale and temporal pattern of disease incidence is a fundamental prerequisite for the development of appropriate management and intervention strategies (Ward and Carpenter, 2000; Koopmans et al., 2007; Rose et al., 2009). It is par- ticularly important given the need to understand the elevated risks linked to climate change. During the past three decades many geographically explicit spatial modelling approaches have been employed to map the distribution of parasites and vectors of dis- ease, with the aim of explaining their distribution with respect to environmental variables (Hay et al., 2000; Randolph, 2000). Many of these studies use statistical tests of association such as logistic regression (Randolph, 2002; Rossi et al., 2007; Medlock et al., 2008), while others develop and adapt mathematical models (Kleinschmidt et al., 2000; Robinson, 2000; Rogers et al., 2002; Big- geri et al., 2006; Brooker and Clements, 2009). However, while con- siderable success has been achieved with models that describe and explain the spatial distributions of disease, attempts to use such models dynamically to predict future changes have been more limited. Species distribution models (SDMs) are empirical tools that de- scribe the geographic distribution of a population, based on the environmental conditions where it is known to occur (Elith and Leathwick, 2009). These models are widely used in conservation and ecology (Zaniewski et al., 2002; Guisan and Thuiller, 2005; Elith et al., 2006; Hernandez et al., 2006; Puschendorf et al., 2009). Using SDMs it is possible not only to characterise the eco- logical requirements of the population in question, but also to pre- dict its distribution in space and time in unsampled study areas and in future climates (Peterson et al., 2005; Peterson, 2007). This provides potential for the development of disease forecasting sys- tems (Rogers et al., 2002; Peterson et al., 2005). However, their application in spatial epidemiology has been limited largely to the development of low resolution climate matching models, prob- ably as a result of the lack of reliable data in areas where disease is absent in many epidemiological datasets (Perry et al., 1991; 0020-7519/$36.00 Ó 2011 Australian Society for Parasitology Inc. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.ijpara.2011.01.012 Corresponding author. Tel.: +44 1179 287 489; fax: +44 1173 317 985. E-mail address: [email protected] (H. Rose). International Journal for Parasitology 41 (2011) 739–746 Contents lists available at ScienceDirect International Journal for Parasitology journal homepage: www.elsevier.com/locate/ijpara

Upload: hannah-rose

Post on 05-Sep-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

International Journal for Parasitology 41 (2011) 739–746

Contents lists available at ScienceDirect

International Journal for Parasitology

journal homepage: www.elsevier .com/locate / i jpara

Modelling the impact of climate change on spatial patterns of disease risk:Sheep blowfly strike by Lucilia sericata in Great Britain

Hannah Rose ⇑, Richard WallVeterinary Parasitology and Ecology Group, School of Biological Sciences, University of Bristol, Bristol BS8 1UG, UK

a r t i c l e i n f o

Article history:Received 5 November 2010Received in revised form 19 January 2011Accepted 25 January 2011Available online 11 March 2011

Keywords:Blowfly strikeLucilia sericataOvine cutaneous myiasisMaxentSpecies distribution modelClimate changeDisease

0020-7519/$36.00 � 2011 Australian Society for Paradoi:10.1016/j.ijpara.2011.01.012

⇑ Corresponding author. Tel.: +44 1179 287 489; faE-mail address: [email protected] (H. Rose).

a b s t r a c t

Understanding the spatial scale and temporal pattern of disease incidence is a fundamental prerequisitefor the development of appropriate management and intervention strategies. It is particularly critical,given the need to understand the elevated risks linked to climate change, to allow the most likely changesin the distribution of parasites and disease vectors to be predicted under a range of climate change sce-narios. Using statistical models, the spatial distribution and climatic correlates of a range of parasites anddiseases have been mapped previously, but their development into dynamic, predictive tools is less com-mon. The aim of the work described here, was to use a species distribution model to characterise theenvironmental determinants of the monthly occurrence of ovine cutaneous myiasis (blowfly strike) byLucilia sericata, the most frequent primary agent of northern European myiasis, and to then use this modelto describe the potential spatial changes that might be expected in response to predicted climate changein Great Britain. The model predicts that the range of elevated temperatures predicted by current climatechange scenarios will result in an increase in the risk of strike and an elongated blowfly season. However,even for the most rapid warming scenario predictions over the next 70 years, strike is not predicted tooccur throughout the winter. Nevertheless, in this latter case, parts of central and southern Englandare likely to become too hot and dry for strike by L. sericata, to persist in mid-summer. Under these con-ditions, it is possible that other, more pathogenic Mediterranean agents of myiasis, such as Wolfhartiamagnifica, could potentially replace L. sericata. Where the phenology of strike is altered by climate change,as predicted here, significant changes to the timing and frequency of parasite treatments and husbandrypractices, such as shearing, will be required to manage the problem. The results suggest that the model-ling approach adopted here could be usefully applied to a range of disease systems.

� 2011 Australian Society for Parasitology Inc. Published by Elsevier Ltd. All rights reserved.

1. Introduction

Understanding the spatial scale and temporal pattern of diseaseincidence is a fundamental prerequisite for the development ofappropriate management and intervention strategies (Ward andCarpenter, 2000; Koopmans et al., 2007; Rose et al., 2009). It is par-ticularly important given the need to understand the elevated riskslinked to climate change. During the past three decades manygeographically explicit spatial modelling approaches have beenemployed to map the distribution of parasites and vectors of dis-ease, with the aim of explaining their distribution with respect toenvironmental variables (Hay et al., 2000; Randolph, 2000). Manyof these studies use statistical tests of association such as logisticregression (Randolph, 2002; Rossi et al., 2007; Medlock et al.,2008), while others develop and adapt mathematical models(Kleinschmidt et al., 2000; Robinson, 2000; Rogers et al., 2002; Big-geri et al., 2006; Brooker and Clements, 2009). However, while con-

sitology Inc. Published by Elsevier

x: +44 1173 317 985.

siderable success has been achieved with models that describe andexplain the spatial distributions of disease, attempts to use suchmodels dynamically to predict future changes have been morelimited.

Species distribution models (SDMs) are empirical tools that de-scribe the geographic distribution of a population, based on theenvironmental conditions where it is known to occur (Elith andLeathwick, 2009). These models are widely used in conservationand ecology (Zaniewski et al., 2002; Guisan and Thuiller, 2005;Elith et al., 2006; Hernandez et al., 2006; Puschendorf et al.,2009). Using SDMs it is possible not only to characterise the eco-logical requirements of the population in question, but also to pre-dict its distribution in space and time in unsampled study areasand in future climates (Peterson et al., 2005; Peterson, 2007). Thisprovides potential for the development of disease forecasting sys-tems (Rogers et al., 2002; Peterson et al., 2005). However, theirapplication in spatial epidemiology has been limited largely tothe development of low resolution climate matching models, prob-ably as a result of the lack of reliable data in areas where disease isabsent in many epidemiological datasets (Perry et al., 1991;

Ltd. All rights reserved.

Page 2: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

740 H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746

Morgan et al., 2009). The recent development and refinement ofmethods requiring presence data only (Stockwell and Peters,1999; Phillips et al., 2004) has, therefore, enabled wider applica-tion of higher performing SDMs to disease mapping.

One such SDM is Maxent (Elith et al., 2006). Maxent estimatesthe probability distribution with the maximum entropy (the mostuniform distribution) for the species in question, subject to theconstraints that the distribution matches the observed averagesfrom the occurrence data (Phillips and Dudík, 2008). The Maxentalgorithm is able to fit complex responses to the data and inbuiltfeatures prevent the model from overfitting (matching the datatoo closely) (Phillips and Dudík, 2008). As a result, Maxent is par-ticularly valuable for modelling the distribution of rare species orwhere there is limited observational data. Its predictions have beenshown to be robust to changes in sample size, and useful modelsmay be developed using as few as five to 10 sample locations (Her-nandez et al., 2006, 2008); the Maxent algorithm was found to beamongst the least sensitive to sample size out of 12 species distri-bution modelling algorithms tested (including General LinearModels), outperforming all other methods at the smallest samplesize tested (n = 10) (Wisz et al., 2008). In addition, several studiescomparing the performance of this modelling approach with otherSDM methods have identified Maxent as one of the highest per-forming (Elith et al., 2006; Hernandez et al., 2006, 2008).

Species distribution models such as Maxent have been used forover a decade to predict range shifts, changes in species diversityand potential extinctions in response to climate change (e.g. Iver-son and Prasad, 1998; Peterson et al., 2002; Thomas et al., 2004;Franklin, 2009; Rebelo et al., 2010). Such models are increasinglyused to estimate the distribution and ecological requirements ofparasites and vectors for disease, including mosquitoes (Rogerset al., 2002; Levine et al., 2004; Peterson et al., 2005; Barroset al., 2007; Masouka et al., 2009), malaria (Kleinschmidt et al.,2000), fleas (Adjemian et al., 2006), canine leishmaniasis (Chama-illé et al., 2010) and psoroptic mange in sheep (Rose et al., 2009).Relatively few studies however, have extended these models toexplicitly include predictions about future changes in the distribu-tion and risk of disease under conditions of dynamic environmen-tal conditions such as climate change (Peterson and Shaw, 2003;González et al., 2010). Nevertheless, such applications have consid-erable potential (Peterson and Shaw, 2003; Peterson, 2007; Gon-zález et al., 2010).

The aim of the work described here was to use a presence onlyspecies distribution model to characterise the environmentaldeterminants of ovine cutaneous myiasis (blowfly strike), and touse this model to describe the potential spatial changes in inci-dence that might be expected in response to predicted climatechange. Blowfly strike is a common disease, particularly affectingsheep; it results from the infestation of living animals by fly larvae.The risk of strike is known to be highly sensitive to even smallchanges in climate (Pitts and Wall, 2004; Wardhaugh et al.,2007). Although the aetiology of strike is complex, ambient tem-perature has a particularly strong influence on strike incidence(French et al., 1995; Broughan and Wall, 2007) and higher temper-atures would be expected to increase strike incidence directlythrough higher blowfly development rates, increased numbers ofgenerations per year and prolonged periods during which condi-tions are favourable for fly survival. Warmer climates are alsolikely to change strike incidence indirectly through changes tothe seasonal pattern of sheep susceptibility and in the timing ofseasonal farm management practices (Morgan and Wall, 2009).

In sheep farming areas of northern Europe and parts of thesouthern hemisphere such as New Zealand, strike is largely initi-ated by larvae of the blowfly Lucilia sericata. In north western Eur-ope, the disease is widespread between April/May and October/November with, for example, at least one case of myiasis reported

per year in 52% of sheep farms in the Netherlands (Snoep et al.,2002) and over 80% of sheep farms in Great Britain (French et al.,1995; Bisdorff et al., 2006). Higher strike incidence resulting froman altered climate could significantly increase the risks to animalwelfare and threaten the economic sustainability of sheep farmingin some areas.

2. Materials and methods

2.1. Species distribution model

The strike incidence data that were used in the current study in-cluded both farms which had treated their sheep with insecticideto prevent strike, as well as farms that had not. As a result, the ab-sence of strike in any particular month at any one farm could notbe taken to indicate that environmental conditions were necessar-ily unsuitable. Hence, presence–absence models such as the Gener-alised Linear Model were not appropriate and Maxent, a presence-only species distribution model, was therefore preferred.

Maxent estimates a given species’ most uniform distributionacross the study area (the distribution with maximum entropy),subject to the environmental constraints at the locations of theincidence data. The model is then extrapolated to other unsampledareas within the study area to give the probability distribution ofthe species (the probability that the species will occur at any givenpixel in the study area) (Phillips et al., 2004; Phillips and Dudík,2008). Unlike presence–absence models which attempt to differen-tiate the environmental conditions where the species is observed(presence points) from the conditions where the species is absent(absence points), presence-only models use pseudo-absences (arandom sample of locations from the study area) in place of trueabsences and attempt to differentiate the conditions at the pres-ence points from the random distribution of pseudo-absences(Phillips, 2007). This method has been shown to achieve high levelsof predictive accuracy (Elith et al., 2006; Hernandez et al., 2006).Here, Maxent software (http://www.cs.princeton.edu/~schapire/maxent) was used to characterise the environmental correlates ofblowfly strike and build an initial descriptive model. After valida-tion, this model was then projected onto present and future cli-mate data to model the distribution and risk of fly strike insheep in Great Britain.

2.2. Model parameters and validation

Monthly presence data for blowfly strike in Great Britain wereobtained from two surveys of sheep farms. The 1991 data were col-lected as part of a longitudinal study of 453 farms in England andWales, during which farmers were asked to record each case ofstrike they saw throughout the 1991 blowfly season (Frenchet al., 1995). The 2003/2004 data were collected as part of postalsurvey of 1067 farmers in Great Britain, where farmers were askedto report retrospectively the number of cases of strike that theyhad experienced each month of the previous season (Bisdorffet al., 2006). These two studies provided a total of 938 and 1031monthly strike records, respectively. The two studies gave a good,well dispersed distribution of strike records over Great Britain.

The geographic position of each reporting farm was obtainedfrom its postcode. Monthly precipitation and monthly mean tem-perature data between January and December 1991, and March2003 and February 2004 were obtained from the UK Climate Im-pacts Programme (UKCIP) as 25 km2 (5 � 5 km) grid datasets,available from the UK Met Office (Perry and Hollis, 2005).

The strike occurrence data were input into Maxent in ‘SamplesWith Data’ (SWD) format, where the farm coordinates were pro-vided together with the corresponding monthly value for temper-

Page 3: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746 741

ature and precipitation for the month in which the strike was re-corded. To match Maxent’s default of 10,000 background (pseu-do-absence) points, a random sample of 417 pseudo-absencelocations in SWD format were generated in ArcMap 9.3 (ESRI,New York, USA 2008) for each study month. Background pointswith missing environmental data were removed prior to modeldevelopment. To avoid sample-selection bias, background pointsfor January to December 1991 were restricted to England andWales only (Phillips, 2007).

Ten replicates of the model were run using 75% of the occur-rence data as a training dataset and the remaining 25% as a testdataset; training and test data were chosen at random using the‘‘random seed’’ function in the Maxent software. Setting aside a dif-ferent random test sample for each replicate not only served to val-idate each model but also tested whether using different subsets ofdata as parameters for the models significantly affected their per-formance. Model performance was tested using the AUC value (de-fined as the area under the receiver operating characteristic (ROC)curve), which is the probability that a random presence data pointwill be ranked above a random absence data point (or pseudo-ab-sence point) (Hanley and McNeil, 1982). The ROC curve shows sen-sitivity over 1-specificity at a range of threshold probability values,where sensitivity is the proportion of observed presences correctlypredicted by the model, and specificity is the proportion of ob-served absences (or pseudo-absences) correctly predicted by themodel (Raes and ter Steege, 2007). An AUC value above 0.5 indi-cates that the model performs better than random (Elith et al.,2006; Phillips and Dudík, 2008). A potentially useful model willhave an AUC above 0.75 (Elith et al., 2002).

Maxent provides a number of threshold probabilities, deter-mined from the ROC curve, which can be used to indicate pres-ence/absence, and tests their statistical significance against thetest dataset. The mean maximum training sensitivity plus specific-ity threshold of the training replicates was used as a binary thresh-old for presence/absence of blowfly strike, above which strike isconsidered to occur. This is the point where the proportion of cor-rectly predicted presences and pseudo-absences are maximised(Liu et al., 2005).

Heuristic estimates of the relative contribution of each variableto the model, reported as a percentage (±SD), are also given. Thisindicates the contribution of each variable to the model’s gain,which is the ability to distinguish sample localities from randombackground points.

Finally, to increase its power, the model was calibrated usingthe entire incidence dataset and projected onto precipitation andtemperature data for March 2003 to February 2004 to producebaseline probability distribution maps, considered to be represen-tative of the present distribution of blowfly strike, against whichfuture projections could be compared.

2.3. Projection onto future climate scenarios

To consider potential changes in the distribution of strike in re-sponse to anticipated climate change, the parameterised strikemodel was projected onto monthly 50% probabilistic climate pro-jections for temperature and precipitation for the years 2069–2099, referred to here as the 2080s (Fig. 1). These were obtainedas grid datasets, with a resolution of 25 � 25 km, from the UKClimate Impacts Programme 09 (Murphy et al., 2009). These pro-jections were generated from a perturbed physics ensemble ofdownscaled global climate model projections, with predicted cli-mate change added onto the baseline projections for 1961–1990.The 50% probability level refers to the probability that the truevalue is as likely as not to exceed that predicted and provides acentral estimate of future climate (Jenkins et al., 2009). High andlow emission scenarios, which correspond to the Intergovernmen-

tal Panel on Climate Change (IPCC) ‘A1fi’ and ‘B1’ emissions scenar-ios, were considered (Nakicenovíc and Swart 2000). The ‘A1fi’scenario describes a high rate of economic growth and fossil fuelconsumption whereas the ‘B1’ scenario assumes the introductionof clean and resource-efficient technologies (Murphy et al.,2009). Measured against the 1960–1999 baseline projections, bothhigh and low emission scenarios predict warmer, drier summersand warmer, wetter winters in Great Britain, with some regionalvariation. By the 2080s, mean monthly temperature is predictedto increase 1.23–10.8 �C (±4.6 SD) under the low emission scenarioand 2.55–12.1 �C (±5 SD) under the high emission scenario. A sim-ilar increase in total monthly precipitation of 13–14% is predictedunder both emission scenarios. However, this increase is not con-sistent throughout the year; significantly less rainfall is predictedduring the spring/early summer (Fig. 1).

For each model projection, probability values were extracted atthe location of 10,000 random points in Great Britain, generated inArcMap 9.3, to compare with the 2003/2004 projections. Fromthese data, the mean change in risk over a 12 month period underboth climate change scenarios, compared with the 2003/2004baseline projections, were calculated and plotted as probabilitydensity histograms in R (R Development Core Team, 2009, Vienna,Austria). Tests of kurtosis were carried out on the data in SPSS 14.0(IBM Corporation, New York, USA). To quantify the increase in riskfor all projections, the percentage of the random sample with prob-ability values exceeding the threshold for presence/absence wascalculated for each monthly projection. This was considered to berepresentative of the area in Great Britain at risk of strike duringeach month and was used to calculate the average percentage ofGreat Britain at risk per month.

3. Results

3.1. Model parameters and validation

A total of 1969 strike presence points and 9876 backgroundpoints were used to derive parameters for the model. Initially,492 of these presence points were set aside at random to test eachmodel replicate. The 10 replicates achieved a mean training AUCvalue of 0.832 (±0.002 SD) and mean test AUC value of 0.830(±0.005 SD). The final model, with parameters derived by usingall 1969 strike presence points, achieved an AUC value of 0.832.The mean maximum training sensitivity plus specificity thresholdwas estimated at 0.349 and was able to accurately classify a meanof 92% of the test dataset (mean test and training omission = 0.08,P < 0.001).

Temperature and precipitation were estimated to contribute amean of 99.2% (±0.224 SD) and 0.84% (±0.224 SD), respectively, tothe model across all 10 training runs. Removing precipitation fromthe model did not significantly reduce the model performance(mean training AUC = 0.827 ± 0.005 SD) whereas removing tem-perature from the model reduced the performance to little betterthan random (mean training AUC = 0.593 ± 0.01 SD). Hence, themodel suggests that temperature is clearly the key driver in theblowfly strike system in Great Britain. Amongst the recorded cases,the probability of strike was observed to increase rapidly at tem-peratures of between 8 and 16 �C and decrease above 17 �C. Theprobability of strike varied relatively little with changes inprecipitation.

3.2. Projection onto future climate scenarios

A probability density distribution allows the overall spatial im-pact of the climate change model projections for the 2080s to beassessed in comparison with the present risk (Fig. 2). The models

Page 4: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

Fig. 1. Observed and predicted annual patterns of temperature and rainfall. (A) Mean monthly mean temperature for Great Britain during 2003/2004 and the 2080s under thelow and high emission scenarios. (B) Mean monthly total precipitation for 2003/2004 and the 2080s under the low and high emission scenarios.

742 H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746

predict an overall mean increase in the probability of strike occur-rence throughout the year of 0.07 (±0.03 SD) and 0.05 (±0.02 SD)under the high and low emissions scenarios, respectively (Fig. 2).Under the low emissions scenario the risk distribution was rela-tively leptokurtic (g2 = 2.86 ± 0.05 SD), suggesting that there waslikely to be relatively little spatial variation in the impact of climatechange on increased strike risk. However, under the high emissionscenario, the distribution of strike risk was more platykurtic(g2 = 1.1 ± 0.05 SD), indicating a higher level of variation in the im-pact of climate change on strike risk in different areas (Fig. 2).

Under current conditions (2003/2004) the model indicates thatblowfly strike will occur between April and November. In spring, asmean temperatures increase and the conditions become suitablefor blowfly emergence, the risk of strike gradually increases andspreads further north. The majority of Great Britain is predicted

to be at relatively high and generally equal risk of strike betweenJuly and August. As conditions become less favourable, the areapredicted to be at risk of strike contracts from North to Southand by October and November, only the southern English andWelsh coast are at risk (Fig. 3A–D). This matches closely observedpatterns seen on sheep farms (French et al., 1995).

Projection onto the low emission climate scenario predicts ageneral increase in risk throughout the year by the 2080s. Overall,the length of the strike season in Great Britain is unchanged. How-ever, strike is predicted to progress north earlier and contract laterin the season, hence the risk period for strike is expected to be ex-tended in northern regions (Fig. 3B-D). Similarly, projections ontothe high emission scenario predict that by the 2080s the distribu-tion of strike will peak earlier and contract later. This will extendthe risk period for strike in northern England by as much as

Page 5: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

Fig. 2. The predicted change in mean probability of occurrence of blowfly strike (risk) in Great Britain under low and high emission scenarios for the 2080s (2069–2099),compared with the present distribution of strike (2003/2004) as indicated by zero on the abscissa.

H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746 743

2 months; it will extend the ‘at risk’ period in some areas intoDecember, although less than 1% of the total area is predicted tobe at risk at this time (Fig. 3D). The model predicts that the mostsignificant change in the distribution of strike may occur duringJuly and August under the high emission scenario, where a sharpdecrease in risk is seen in the south and southeast (Fig. 3C). Thisis due to the high temperatures and low precipitation predictedfor the region which create conditions which are unsuitable for L.sericata.

Hence, under the high emission scenario, the model predicts anincrease in risk and an extended strike season for the northernareas, but two discrete early and late periods of strike risk forsouthern regions separated by a period of lower risk. When consid-ering the average percentage of area at risk per month, by the2080s an increase is predicted compared with 2003/2004, butthere is little difference between the two emission scenarios(2003/2004 = 37.5%, 2080s low emissions = 46.9%, 2080s highemissions = 47.9%).

4. Discussion

The model developed here, using parameters from strike datafrom surveys undertaken in 1991 and 2003–2004, accurately de-scribed the present seasonal pattern of strike in Great Britain.The model predicted that strike would begin to occur at tempera-tures above 9–10 �C and this is consistent with the published rangeof threshold temperatures for development and oviposition inadult L. sericata (Pitts and Wall, 2004; Broughan and Wall, 2007).This temperature threshold explains the progressive movementnorth of strike risk over spring and summer months and its retreatsouth in autumn and winter. Precipitation appeared to contributelittle to the model predictions. This latter finding may be due sim-ply to the correlation between precipitation and temperature,however it may also be that the monthly rainfall, as used here, istoo imprecise a measure. The use of rainfall events immediatelypreceding a strike may be more informative (these data were notavailable to the current study), and this is an area where furtherdevelopment of the model may prove useful.

Model projections onto the IPCC’s ‘B1’ emission scenario (lowemission, with an emphasis on green technology), predicted an

overall increase in risk of strike by the 2080s, but with very littlechange to the length of the strike season. In contrast, an overall in-crease in risk was also predicted for much of the year under theIPCC’s ‘A1fi’ emission scenario (high emissions, fossil fuel inten-sive), but strike was expected to persist into December and the sea-son was extended in some northern regions by up to 2 months. It isnotable, however, that even under the high emission scenario,strike was not expected to occur throughout the winter. In addi-tion, the model clearly suggests that under the high emission sce-nario, the high temperature and low rainfall conditions which arelikely to occur in mid-summer in southern and central Englandare likely to restrict the incidence of strike by L. sericata. As a result,two discrete periods of high risk would be anticipated, separatedby a period of lower risk in July and August in these areas. How-ever, in hotter, drier parts of Europe, throughout the Mediterra-nean basin and in eastern Europe, the sarcophagid Wohlfahrtiamagnifica is commonly a much more prevalent and clinicallyimportant agent of myiasis in livestock than L. sericata, particularlyin mid-summer, although L. sericata may be sympatric (Hall, 1997;Hall and Farkas, 2000). The larvae of W. magnifica cause a more se-vere, traumatic myiasis than larvae of L. sericata, often invadingwounds, the anus or urinogenital system. It is possible, therefore,that under the higher emissions scenario, W. magnifica couldpotentially become established in southern England, leading to sig-nificantly greater problems in sheep husbandry. However, thepresence of the English Channel would limit the potential for dis-persal unless aided by human carriage.

It is estimated that there has been an average global decadal in-crease of 0.13 �C between 1956 and 2005; a further increase ofapproximately 0.4 �C is expected within the next two decades(IPCC, 2007). Changes in the phenology and distribution of speciesin response to climate change have been observed worldwide; ameta-analysis of published data for over 1700 species from a rangeof taxa estimated that the range limits of these species have shiftedon average 6.1 km (±2.4 km) per decade northwards (Parmesanand Yohe, 2003). Other species have shifted their range to higherelevations and the timing of spring events such as frog breedingand tree budburst has advanced, on average, 2.3 days per decade(Parmesan and Yohe, 2003). In livestock husbandry, a change inthe seasonal incidence of parasitic gastroenteritis and fascioliasisin response to climate change has also been observed (Pritchard

Page 6: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

Fig. 3. The risk of blowfly strike by Lucilia sericata in Great Britain. A–D) The predicted monthly probability distribution (risk) of blowfly strike for 2003/2004, and the 2080s(2069–2099) under both low and high emission scenarios. Parameters for the Maxent model were derived using monthly reports of blowfly strike, mean monthlytemperature and total monthly precipitation for 1991 and March 2003–February 2004, then projected onto central estimates of climate for the 2080s. Probability of strikeranges between zero (green (dark grey)) and one (red (white)). The threshold for presence/absence is estimated to be 0.349 (P < 0.001; yellow–green (light grey)).

744 H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746

Page 7: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746 745

et al., 2005; van Dijk et al., 2008) and a similar response is expectedfor a range of other livestock parasites (Gale et al., 2009; Morganand Wall, 2009). To allow these anticipated increased disease risksto be managed appropriately, it is essential that tools are devel-oped that allow future patterns of parasitic disease to be predictedin space and time (Morgan and Wall, 2009).

The modelling approach used in any particular study needs tobe selected carefully so that it is appropriate to the questionsasked. The model developed here is an example of a ‘top down’ ap-proach which attempts to predict changes in overall disease inci-dence, without incorporating a detailed understanding of themyriad of factors which drive the complex fly–sheep–farm system.Comprehensive simulation models of this system have been devel-oped (Wall and Ellse, 2010), but such models inevitably under-rep-resent the stochasticity of a real-world environment and rapidlybecome cumbersome. All modelling approaches have their limita-tions and it is important to be aware of the strengths and con-straints of each approach when interpreting the results.

A particular strength of empirical species distribution models istheir ability to capture these complex interactions by modellingboth direct and indirect species responses to spatial variables(Franklin, 2009). In this way, many processes determining speciesoccurrence can be modelled using relatively few variables. The riskof blowfly strike is dependent on a complex interaction betweenclimate, sheep availability, sheep susceptibility, fly abundanceand activity, and farmer behaviour (Wall et al., 2002; Morganand Wall, 2009). Temperature, for example, may directly affectstrike incidence through changes to the development rate of blow-flies, but also indirectly by affecting sheep susceptibility throughthe timing of shearing and gastrointestinal nematode infection(Morgan and Wall, 2009). However, limited by the availability ofsimilar historic and future climate datasets, the model developedhere does not explicitly include the range of individual factorswhich may affect the probability of occurrence of strike and there-fore the output of the model gives an overall estimate of potentialblowfly strike risk. In addition, blowfly population dynamics arenot explicitly modelled here and therefore the observed risk forany month may be different to that predicted if conditions in thepreceding months were particularly favourable or unfavourablefor blowfly development. For example, the decrease in temperatureand precipitation predicted during the summer under the highemission scenario may result in a decrease in the size of the blow-fly population and the risk over the subsequent months may belower than is predicted. The model presented here also makes noallowance for the wide range of changes in farm husbandry prac-tices that might occur in response to perceived changes in climateand which might mitigate its effects, such as changes in sheepstocking rates, treatment intervention thresholds or sheep breeds(Wall and Ellse, 2010).

Despite their limitations, species distribution models such asMaxent are powerful tools, enabling the forecast and mitigationof probable changes in the seasonality and spatial distributionof disease under future climate scenarios. Changes in the timingof farm husbandry procedures which reduce the susceptibility ofsheep to strike could mitigate the effects of these predicted in-creases in strike risk; early shearing and strategic prophylactictreatment of ewes and lambs may serve to reduce incidence signif-icantly (Morgan and Wall, 2009). In particular, where the phenol-ogy of strike is changed, as predicted here, significant changes tothe timing and frequency of parasite treatments and shearing willbe required to manage the problem. Nevertheless, as shown here,species distribution models are a promising approach to predictingchanges in the distribution and risk of disease. Novel presence-onlymethods such as Maxent are particularly suited to risk mappingwhere reliable data on species absences are not readily available,or where limited incidence data exists. Such models could readily

be applied to a range of parasite problems to forecast changes inboth space and time, facilitating better risk management.

Acknowledgements

We are grateful to the Natural Environment Research Council(NERC), UK for funding, Dr. Jon Flanders (University of Bristol,UK) for his advice with modelling and Dr. Betty Bisdorff (Klinikumder Universistät München, Germany) and Professor Nigel French(Massey University, New Zealand) for access to their original data.

References

Adjemian, J.C.Z., Girvetz, E.H., Beckett, L., Foley, J.E., 2006. Analysis of GeneticAlgorithm for Rule-Set Prediction (GARP) modelling approach for predictingdistributions of fleas implicated as vectors of plague, Yersinia pestis, inCalifornia. J. Med. Ent. 43, 93–103.

Barros, F.S.M., De Aguiar, D.B., Rosa-Freita, M.G., Luitgards-Moura, J.F., Gurgel, H.C.,Honório, N.A., Arruda, M.C., Tsouris, P., Vasconcelos, S.D., 2007. Distributionsummaries of malaria vectors in the northern Brazilian Amazon. J. Vector Ecol.32, 161–167.

Biggeri, A., Dreassi, E., Catelan, D., Rinaldi, L., Lagazio, C., Cringoli, G., 2006. Diseasemapping in veterinary epidemiology: a Bayesian geostatistical approach. Stat.Methods Med. Res. 15, 337–352.

Bisdorff, B., Wall, R., Milnes, A., 2006. Prevalence and regional distribution of scab,lice and blowfly strike in Great Britain. Vet. Rec. 158, 749–752.

Brooker, S., Clements, A.C.A., 2009. Spatial heterogeneity of parasite co-infection:determinants of geostatistical prediction at regional scales. Int. J. Parasitol. 39,591–597.

Broughan, J.M., Wall, R., 2007. Fly abundance and climate as determinants of sheepblowfly strike incidence. Med. Vet. Ent. 21, 231–238.

Chamaillé, L., Tran, A., Meunier, A., Bourdoiseau, G., Ready, P., Dedet, J., 2010.Environmental risk mapping of canine leishmaniasis in France. Parasite. Vector.3, 31.

Elith, J., Leathwick, R., 2009. Species distribution models: ecological explanation andprediction across space and time. Annu. Rev. Ecol. Evol. S. 40, 677–697.

Elith, J., Burgman, M.A., Regan, H.M., 2002. Mapping epistemic uncertainties andvague concepts in predictions of species distribution. Ecol. Model. 157, 313–329.

Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J.,Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A.,Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T.,Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J.,Williams, S., Wisz, M.S., Zimmermann, N.E., 2006. Novel methods improveprediction of species’ distributions from occurrence data. Ecography 29, 129–151.

Franklin, J., 2009. Mapping species distributions: spatial inference and prediction.Cambridge University Press, Cambridge, UK.

French, N.P., Wall, R., Morgan, K.L., 1995. The seasonal pattern of blowfly strike inEngland and Wales. Med. Vet. Ent. 9, 1–8.

Gale, P., Drew, T., Phipps, L.P., David, G., Wooldridge, M., 2009. The effect of climatechange on the occurrence and prevalence of livestock diseases in Great Britain:a review. J. Appl. Microbiol. 106, 1409–1423.

González, C., Wang, O., Strutz, S.E., González-Salazar, C., Sánchez-Cordero, V., Sarkar,S., 2010. Climate change and risk of leishmaniasis in North America: predictionsfrom ecological niche models of vector and reservoir species. PLOS Neglect.Trop. D 4, e585.

Guisan, A., Thuiller, W., 2005. Predicting species distribution: offering more thansimple habitat models. Ecol. Lett. 8, 993–1009.

Hall, M.J.R., 1997. Traumatic myiasis of sheep in Europe: a review. Parasitologia 39,409–413.

Hall, M.J.R., Farkas, R., 2000. Traumatic myiasis of humans and animals. In: Papp, L.,Darvas, B. (Eds.), Manual of Palearctic Diptera. Science Herald, Budapest,Hungary, pp. 751–768.

Hanley, J.A., McNeil, B.J., 1982. The meaning and use of the area under the receiveroperating characteristic (ROC) curve. Radiology 143, 29–36.

Hay, S.I., Omumbo, J.A., Craig, M.H., Snow, R.W., 2000. Earth observation, geographicinformation systems and Plasmodium falciparum malaria in Sub-Saharan Africa.Adv. Parasit. 47, 173–215.

Hernandez, P.A., Graham, C.H., Master, L.L., Albert, D.L., 2006. The effect of samplesize and species characteristics on performance of different species distributionmodeling methods. Ecography 29, 773–785.

Hernandez, P.A., Franke, I., Herzog, S.K., Pacheco, V., Paniagua, L., Quintana, H.L.,Soto, A., Swenson, J.J., Tovar, C., Valqui, T.H., Vargas, J., Young, B.E., 2008.Predicting species distributions in poorly-studied landscapes. Biodivers.Conserv. 17, 1353–1366.

Intergovernmental Panel on Climate Change, 2007. IPCC Fourth Assessment Report(AR4) – Climate Change 2007: Synthesis Report. Geneva, Switzerland.

Iverson, L.R., Prasad, A.M., 1998. Predicting abundance of 80 tree species followingclimate change in Eastern United States. Ecol. Monogr. 68, 465–485.

Jenkins, G.J., Murphy, J.M., Sexton, D.S., Lowe, J.A., Jones, P., Kilsby, C.G., 2009. UKClimate Projections: Briefing report. Met Office Hadley Centre, Exeter, UK.

Page 8: Modelling the impact of climate change on spatial patterns of disease risk: Sheep blowfly strike by Lucilia sericata in Great Britain

746 H. Rose, R. Wall / International Journal for Parasitology 41 (2011) 739–746

Kleinschmidt, I., Bagayoko, M., Clarke, G.P.Y., Craig, M., Le Sueur, D., 2000. A spatialstatistical approach to malaria mapping. Int. J. Epidemiol. 29, 355–361.

Koopmans, M.M.M., Reusken, C., van Maanen, K., 2007. West Nile virus in Europe:waiting for the start of the epidemic. In: Takken, W., Knols, B.G.J. (Eds.),Emerging Pests and Vector-Borne Diseases in Europe. Wageningen AcademicPublishers, Wageningen, Netherlands, pp. 123–151.

Levine, R.S., Peterson, A.T., Benedict, M.Q., 2004. Geographic and ecologicdistributions of the Anopheles gambiae complex predicted using a geneticalgorithm. Am. J. Trop. Med. H. 70, 105–109.

Liu, C., Berry, P.M., Dawson, T.P., Pearson, R.G., 2005. Selecting thresholds ofoccurrence in the prediction of species distributions. Ecography 28, 385–393.

Masouka, P.M., Burke, R., Colaccico, M., Razuri, H., Hill, D., Murrell, K.D., 2009.Predicted geographic ranges for North American sylvatic Trichinella species. J.Parasitol. 95, 829–837.

Medlock, J.M., Pietzsch, M.E., Rice, N.V.P., Jones, L., Kerrod, E., Avenell, D., Los, S.,Ratcliffe, N., Leach, S., Butt, T., 2008. Investigation of ecological andenvironmental determinants for the presence of questing Ixodes ricinus (Acari:Ixodidae) on Gower, South Wales. J. Med. Ent. 45, 314–325.

Morgan, E.R., Wall, R., 2009. Climate change and parasitic disease: farmermitigation? Trends Parasitol. 25, 308–313.

Morgan, E.R., Jefferies, R., Krajewski, M., Ward, P., Shaw, S.E., 2009. Caninepulmonary angiostrongylosis: the influence of climate on parasitedistribution. Parasitol. Int. 58, 406–410.

Murphy, J.M., Sexton, D.M.H., Jenkins, G.J., Booth, B.B.B., Brown, C.C., Clark, R.T.,Collins, M., Harris, G.R., Kendon, E.J., Betts, R.A., Brown, S.J., Humphrey, K.A.,McCarthy, M.P., McDonald, R.E., Stephens, A., Wallace, C., Warren, R., Wilby, R.,Wood, R.A., 2009. UK Climate Projections Science Report: Climate changeprojections. Met Office Hadley Centre, Exeter.

Nakicenovíc, N., Swart, R. (Eds.), 2000. Special Report on Emissions Scenarios. ASpecial Report of Working Group III of the Intergovernmental Panel on ClimateChange. Cambridge University Press, Cambridge, UK and New York.

Parmesan, C., Yohe, G., 2003. A globally coherent fingerprint of climate changeimpacts across natural systems. Nature 421, 37–42.

Perry, M., Hollis, D., 2005. The generation of monthly gridded datasets for a range ofclimatic variables over the United Kingdom. Int. J. Climatol. 25, 1041–1054.

Perry, B.D., Kruska, R., Lessard, P., Norval, R.A.I., Kundert, K., 1991. Estimating thedistribution and abundance of Rhipicephalus appendiculatus in Africa. Prev. Vet.Med. 11, 261–268.

Peterson, A.T., 2007. Ecological niche modelling and understanding the geographyof disease transmission. Veter. Ital. 43, 393–400.

Peterson, A.T., Shaw, J., 2003. Lutzomyia vectors for cutaneous leishmaniasis inSouthern Brazil: ecological niche models, predicted geographic distributions,and climate change effects. Int. J. Parasitol. 33, 919–931.

Peterson, A.T., Ortega-Huerta, M.A., Bartley, J., Sánchez-Cordero, V., Soberón, J.,Buddemeier, R.H., Stockwell, D.R.B., 2002. Future projections for Mexican faunasunder global climate change scenarios. Nature 416, 626–629.

Peterson, A.T., Martínez-Campos, C., Nakazawa, Y., Martínez-Meyer, E., 2005. Time-specific ecological niche modelling predicts spatial dynamics of vector insectsand human dengue cases. T. Roy. Soc. Trop. Med. H. 99, 647–655.

Phillips, S.J., Dudík, M., 2008. Modeling of species distributions with Maxent: newextensions and a comprehensive evaluation. Ecography 31, 161–175.

Phillips, S.J., Dudík, M., Schapire, R.E., 2004. A maximum entropy approach tospecies distribution modeling. In: Proceedings of the Twenty-first InternationalConference on Machine Learning. ACM Press, New York, pp. 472–486.

Phillips, S.J. et al., 2007. Transferability, sample selection bias, background data inpresence-only modelling: a response to Peterson et al. (2007). Ecography 31,272–278.

Pitts, K.M., Wall, R., 2004. Adult mortality and oviposition rates in field and captivepopulations of the blowfly Lucilia sericata. Ecol. Entomol. 29, 727–734.

Pritchard, G.C., Forbes, A.B., Williams, D.J.L., Salimi-Bejestani, M.R., Daniel, R.G.,2005. Emergence of fascioliasis in cattle in East Anglia. Vet. Rec. 157, 578–582.

Puschendorf, R., Carnaval, A.C., VanDerWal, J., Zumbado-Ulate, H., Chaves, G.,Bolaños, F., Alford, R.A., 2009. Distribution models for the amphibian chytridBatrachochytrium dendrobatidis in Costa Rica: proposing climatic refuges as aconservation tool. Divers. Distrib. 15, 401–408.

Raes, N., ter Steege, H., 2007. A null-model for significance testing of presence-onlyspecies distribution models. Ecography 30, 727–736.

Randolph, S.E., 2000. Ticks and tick-borne disease systems in space and from space.Adv. Parasit. 47, 217–243.

Randolph, S.E., 2002. Predicting the risk of tick-borne diseases. Int. J. Med.Microbiol. 291 (Supplement 33), 6–10.

Rebelo, H., Tarroso, P., Jones, G., 2010. Predicted impact of climate change onEuropean bats in relation to their biogeographic patterns. Glob. Change Biol. 16,561–576.

Robinson, T.P., 2000. Spatial statistics and geographical information systems inepidemiology and public health. Adv. Parasit. 47, 81–128.

Rogers, D.J., Randolph, S.E., Snow, R.W., Hay, S.I., 2002. Satellite imagery in the studyand forecast of malaria. Nature 415, 710–715.

Rose, H., Learmount, J., Taylor, M., Wall, R., 2009. Mapping risk foci for endemicsheep scab. Vet. Parasitol. 165, 112–118.

Rossi, E., Rinaldi, L., Musella, V., Veneziano, V., Carbone, S., Gradoni, L., Cringoli, G.,Maroli, M., 2007. Mapping the main Leishmania phlebotomine vector in theendemic focus of the Mt. Vesuvius in southern Italy. Geospatial Health 2, 191–198.

Snoep, J.J., Sol, J., Sampimon, O.C., Roeters, N., Elbers, A.R.W., Scholten, H.W.,Borgsteede, F.H.M., 2002. Myiasis in sheep in the Netherlands. Vet. Parasitol.106, 357–363.

Stockwell, D.R.B., Peters, D.P., 1999. The GARP modelling system: problems andsolutions to automated spatial prediction. Int. J. Geogr. Inf. Syst. 13, 143–158.

Thomas, C.D., Cameron, A., Green, R.E., Bakkenes, M., Beaumont, L.J., Collingham,Y.C., Erasmus, B.F.N., de Siqueira, M.F., Grainger, A., Hannah, L., Hughes, L.,Huntley, B., van Jaarsveld, A.S., Midgley, G.F., Miles, L., Ortega-Huerta, M.A.,Peterson, A.T., Phillips, O.L., Williams, S.E., 2004. Extinction risk from climatechange. Nature 427, 145–148.

van Dijk, J., David, G.P., Baird, G., Morgan, E.R., 2008. Back to the future: developinghypotheses on the effects of climate change on ovine parasitic gastroenteritisfrom historical data. Vet. Parasitol. 158, 73–84.

Wall R., Ellse, 2010. Climate change and livestock parasites: integratedmanagement of sheep blowfly strike in a warmer environment. GlobalChange Biol. doi:10.1111/j.1365-2486.2010.02361.x.

Wall, R., Cruickshank, I., Smith, K.E., French, N.P., Holme, A.S., 2002. Developmentand validation of a simulation model for blowfly strike of sheep. Med. Vet. Ent.16, 335–346.

Ward, M.P., Carpenter, T.E., 2000. Analysis of time-space clustering in veterinaryepidemiology. Prev. Vet. Med. 43, 225–237.

Wardhaugh, K., Morton, R., Bedo, D., Horton, B.J., Mahon, R.J., 2007. Estimating theincidence of fly myiases in Australian sheep flocks: development of a weather-driven regression model. Med. Vet. Ent. 21, 153–167.

Wisz, M.S., Hijmans, R.J., Li, J., Peterson, A.T., Graham, C.H., Guisan, A., Elith, J., Dudík,M., Ferrier, S., Leathwick, J.R., Lehmann, A., Lohmann, L., Loiselle, B.A., Manion,G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.McC., Phillips, S.J.,Richardson, K.S., Scachetti-Pereira, R., Schapire, R.E., Soberón, J., Williams, S.E.,Zimmerman, N.E., 2008. Effects of sample size on the performance of speciesdistribution models. Divers. Distrib. 14, 763–773.

Zaniewski, A.E., Lehmann, A., Overton, J.M., 2002. Predicting species spatialdistributions using presence-only data: a case study of native New Zealandferns. Ecol. Model. 157, 261–280.