lable at ScienceDirect

Environmental Modelling & Software 80 (2016) 185e200

Contents lists avai

Environmental Modelling & Software

journal homepage: www.elsevier .com/locate/envsoft

Model of pathogen transmission between livestock and white-taileddeer in fragmented agricultural and forest landscapes

Andrey K. Guber*, David M. Williams, Amy C. Dechen Quinn, Sushil B. Tamrakar,William F. Porter, Joan B. RoseMichigan State University, East Lansing, MI 48824, USA

a r t i c l e i n f o

Article history:Received 17 May 2015Received in revised form18 February 2016Accepted 18 February 2016Available online 10 March 2016

Keywords:WildlifeLivestockPathogen transmissionWatershed scaleModelingSWAT

* Corresponding author.E-mail addresses: andrei.guber@gmail.com (A.K

(D.M. Williams), QuinnAC@cobleskill.edu (A.C. Degmail.com (S.B. Tamrakar), porterw@anr.msu.edu (W(J.B. Rose).

http://dx.doi.org/10.1016/j.envsoft.2016.02.0241364-8152/© 2016 Elsevier Ltd. All rights reserved.

a b s t r a c t

The study summarizes the current knowledge on infection and recovery of white-tailed deer and cattle,and integrates this knowledge into the Soil and Water Assessment Tool (SWAT) via a new add-on moduleSIR (Susceptible - Infected - Recovered) for predicting pathogen transmission between livestock and deer.New processes modeled by the SWAT-SIR model include: (a) seasonal changes in deer population andhabitat; (b) resource selection and seasonal changes in foliage consumption by deer; (c) ingestion ofpathogens with water, foliage and via grooming soiled hide by deer and grazing cattle; (d) infection andrecovery of deer and co-grazing cattle; (e) pathogen shedding by infected animals; (f) survival ofpathogens in manure; (g) kinetic release of pathogens from applied manure and fecal material. Themodel output is linked with ARC-GIS to allow spatial and temporal analysis of pathogen distributionacross the watershed for specific land use, weather and management scenarios.

© 2016 Elsevier Ltd. All rights reserved.

Software and/or data availability

The SWAT-SIR module with example of SWAT ARC GIS projectcan be obtained from the first author upon request.

1. Introduction

Manure-borne pathogens such as Escherichia coli O157:H7, Sal-monella, Campylobacter and Cryptosporidium have become a subjectof growing concerns due to continuous water body impairmentcausing increased number of waterborne disease outbreaks in theUnited States and Canada (Besser et al., 1993; Cieslak et al., 1993;Jackson et al., 1998). In 2014, the U.S. Environmental ProtectionAgency reported 3451 impaired water bodies in the United Statesbased on E. coli monitoring (U.S. EPA. 2014). Livestock is commonlyconsidered among the major sources of fecal contamination (Jones,1999; Chapman, 2000; Gallagher et al., 2012) with cattle as aprincipal reservoir of E. coli O157:H7 (Borczyk et al., 1987; Ørskov

. Guber), dmwill@msu.educhen Quinn), sushil.tamra@.F. Porter), rosejo@msu.edu

et al., 1987; Zhao et al., 1995). However, in fragmented agricul-tural and forest landscapes, wildlife can also serve as a reservoir forpathogens, thus contributing a considerable portion of the fecalpollution (Daszal et al., 2000; Ishii et al., 2007; Parajuli, 2007;Harmel et al., 2010). Several recent E. coli O157:H7 outbreakswere associated with deer. Specifically, 15 illness cases, includingtwo deaths in Oregon in JulyeAugust 2011 were caused bystrawberry-transmitted infection of E. coli O157:H7 produced byblack-tailed deer (Laidler et al., 2013). Consumption of unpasteur-ized apple juice caused infection of at least seventy people by E. coliO157:H7 in the western United States and British Columbia, Can-ada, in October 1996. This outbreak of E. coli O157:H7 infection wassuspected to be associated with apples coming from orchards fre-quented by deer (Cody at al., 1999). Since E. coli O157:H7 is spreadvia a fecal-oral route and both cattle and deer may harbor thispathogen, there is a possibility for the pathogen transmission be-tween the two groups of animals through exposure to contami-nated water and foliage (Branham et al., 2005). This possibility wassupported by the results of Rice et al. (1995), Sargeant et al. (1999),and Renter et al. (2001) who isolated E. coli O157:H7 from feces ofwhite-tailed deer co-grazing with cattle.

Multiple outbreaks associated with deer have promptedenhanced research of possible interaction and pathogen trans-mission between co-grazing domestic and wild animals (Rice et al.,

Applied manure

Pathogens in bo om sediment

Resuspension

Stream water

Inges on of pathogens with water, foliage and via grooming

Wash off pathogens Surface and subsurface transport of pathogens

Infec on of co-grazing wild and domes c animals

Shedding pathogens

Foliage Soil

Release of pathogens from manure and fecal deposits

Fig. 1. Mechanism of pathogen transmission in the Susceptible - Infected - Recovered(SIR) add-on module.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200186

1995; Sargeant et al., 1999). LeJeune et al. (2001) showed that calvescould become colonized with E. coli O157 after drinking fromwatersources that were fecally contaminated 183 days earlier. Branhamet al. (2005) concluded that white-tailed deer could potentiallybecome infected by E. coli O157:H7 and Salmonella spp. via con-sumption of water from cattle troughs, and conversely spread thepathogens to livestock and other wildlife.

Modeling is commonly used to evaluate risk of surface watercontamination by fecal bacteria. Several models have been recentlyused to predict fate and transport of manure-borne bacteria pro-duced by livestock and wildlife at the watershed scale. For example,Hydrological Simulation Programe FORTRAN (HSPF) (Bicknell et al.,1997) was used to simulate total and source-specific contributionsof fecal coliform bacteria to instream load from domestic and wildanimals by Yagow et al. (2001), Moyer and Hyer (2003), Benhamet al. (2006), Chin et al. (2009), and other studies. Soil and WaterAssessment Tool (SWAT) (Sadeghi and Arnold, 2002; Neitsch et al.,2005) was used for bacteria source tracking by Baffaut and Benson(2003), Parajuli (2007), Parajuli et al. (2009), Coffey et al. (2010),Frey et al. (2013), and for predicting E. coli and fecal coliform con-centrations in stream water by Baffaut and Sadeghi (2010), Kimet al. (2010), Cho et al. (2012), Iudicello and Chin (2013), andJayakody et al. (2013). A relatively simple tool for bacteria sourcecharacterization, BSLC (Bacteria Source Load Calculator), wasdeveloped by Zeckoski et al. (2005) to characterize the bacteriasources and loads for development of the TMDL (Total MaximumDaily Load) allocation scenarios. TMDL is a program established byU.S. Environment Protection Agency (EPA) in 1992 in response toClean Water Act of 1972 describing the maximum amount of pol-lutants that a body of water can receive while still meeting waterquality standards. Dorner et al. (2006) coupled a microbial fate andtransport model with the WATFLOOD/SPL9, a watershed hydrologymodeling system, to determine the primary sources of pathogeniccontamination in a watershed in Southwestern Ontario, Canada.Ferguson et al. (2007) developed a process-based mathematicalmodel PCB (pathogen catchment budgets) to predict Cryptospo-ridium, Giardia and E. coli loads generatedwithin and exported fromdrinking water catchments and applied this model for the Wing-ecarribee catchment located approximately 200 km south west ofSydney, Australia. Park et al. (2014) recently extended the Agri-cultural Policy/Environmental eXtender (APEX) model to predictmicrobial fate and transport at farm and small watershed scales.

Different approaches to modeling pathogen transport and riskassessment were further developed in an Integrated EnvironmentalModeling Framework (IEM) (Whelan et al., 2014).

Current microbial transport models were developed for indi-cator organisms of fecal contamination and bacterial impairment ofwatersheds. The most commonly tested fecal bacteria indicatorssuch as total coliforms, fecal coliforms, E. coli, fecal streptococci, andenterococci are relatively harmless. They indicate possibility ofpresence of pathogenicmicroorganisms, though the concentrationsof the pathogens themselves are generally unknown. To extend theexisting models to predict pathogen transport and assessment ofpathogen water contamination, the models must include themechanisms and processes of pathogen transmission within thesame groups of animals as well as between different groups. Thesemechanisms are not well studied and information about pathogentransmission between animals is scarce. Moreover, infection andpathogen shedding by domestic and wild animals depend ondifferent environmental factors that are difficult to account for dueto their high spatial and temporal variability.

In this study we (i) summarized the current knowledge oninfection and recovery of white-tailed deer and cattle, and (ii) in-tegrated this knowledge into the SWAT model via a new add-onmodule SIR (Susceptible - Infected - Recovered) which can predict

pathogen transmission between livestock and deer. We demon-strated the SWAT-SIR features that can be helpful for analyzingpathogen sources and for development of better managementpractices for reducing pathogen loads on fragmented agriculturaland forest landscapes.

2. Materials and methods

2.1. Theory

2.1.1. Modeled processes and structure of the SIR moduleAs the basis for the development of our new SIR module, we

used SWAT model. The SWAT software is frequently used to modelbacteria fate and transport at a watershed scale, has a well-developed graphical user interface linked to ArcGIS (ESRI® Arc-GIS™), and supports commonly used soil (SSURGO, NRCS) andland-use (NLCD2006) databases. Modified version of theSWAT2012 software (Kim et al., 2010) includes following bacterialprocesses:

� bacteria deposition on soil and foliage with applied manure orwith fecal material of grazing animals;

� bacteria die-off/re-growth in soil, water and on foliage; bacteriawash-off from soil and foliage;

� bacteria leaching from soil; bacteria subsurface, overland andinstream transport;

� bacteria deposition to and resuspension from streambedsediment.

This version of SWAT software was developed further to predicttransmission of E. coli O157:H7 between livestock and grazingwhite-tailed deer. New add-on module SIR includes the processesshown in Fig. 1. A brief description of the modeled processes andSIR structure can be found in Guber et al. (2014). Here we present acomplete description of the SIR module that includes governingequations, model parameters and simulation results.

Pathogens can be ingested by co-grazing cattle and deer withwater, foliage and via grooming soiled hide and can cause animalinfection (Fig. 1). The SIR module implements a doseeresponseapproach to compute the probability of animal infection based onthe ingested daily dose of pathogens. It is assumed that the infectedanimals shed pathogens at grazing areas in the amounts propor-tional to fecal material produced daily until their full recovery. Theshed pathogens can grow, die-off in fecal material, and/or be

Fig. 2. Structure of the deer population, infection and recovery components of theSusceptible e Infected e Recovered (SIR) add-on module.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 187

released into the soil or on the foliage with rainfall water. Thepathogens can also be washed off soil, manure and fecal deposits,and be transported further toward streams and rivers with runoffwater. Microorganisms that survive in the bottom sediment can beresuspend during high flow events in streams and transported totributaries of higher order. The resuspended pathogens and thosethat entered the streams with runoff water can be ingested bygrazing animals in pathogen free areas and cause infection ofhealthy animals.

New processes that were integrated into the SWAT-SIR model toaccount for pathogen transmission between infected and healthyanimals include:

� seasonal changes in deer population and habitat;� resource selection and seasonal changes in foliage consumptionby deer;

� ingestion of pathogens with water, foliage and via groomingsoiled hide by deer and grazing cattle;

� infection and recovery of deer and co-grazing cattle;� pathogen shedding by infected animals;� survival of pathogens in manure;� kinetic release of pathogens from applied manure and fecalmaterial;

The processes listed above are modeled in the SIR module andthen coupled with the processes simulated in SWAT2012 tocompute the daily values of the following characteristics:

� number of infected deer and cattle;� number of pathogenic and nonpathogenic cells ingested andshed by deer and co-grazing cattle;

� number of non-pathogenic and pathogenic cells in soil, on fo-liage and in runoff water;

� total number of non-pathogenic and pathogenic cells trans-ported with overland and subsurface flow from each HRU tostreams;

� instream concentrations of non-pathogenic and pathogeniccells.

The values of these characteristics are computed per unit area ofhydrologic response units (HRUs) and sub-basins. An HRU isdefined as a combined land area within the sub-basin representedby a unique combination of land cover, soil, and managementfeatures. While a sub-basin is defined as a group of HRUs borderingthe same stream.

Stand-alone postprocessors transfer results of SWAT-SIR simu-lations to ASCII files for import to SWAT database for further spatialdata analysis in ArcGIS. ASCII files generated by the postprocessorscan also be used by any standard software supporting graphicalanalysis (e.g. MS Excel (Microsoft Co.), Sigma Plot (Systat Software,Inc.), or Grapher (Golden Software, Inc.).

2.1.2. Deer infection, recovery and population dynamicsDeer population is constantly changing across the year due to

birth and mortality associated with accidents, predation, hunting,and diseases. These changes are modeled in the add-on SIR moduleseparately for each sub-basin using forcing functions that accountsfor deer birth (b), recruitments (r), and mortality (d) (Fig. 2). Thepopulation model coupled with a three-compartment SIR module(Keeling and Rohani, 2008) predicts deer infection and recovery.Infection of deer in the susceptible group (S) occurs throughpathogen ingestionwithwater and foliage. Number of infected deer(I) is defined by the probability of the infection computed using adose response-model. The infected deer recover at a daily rate a

and constitute the recovered group (R), which is transitioned

further to the group S at a daily rate w. The infected deer shedpathogens during the whole infection period at rates, which arespecific for deer gender, age, season, and grazing conditions. Thesame approach is used to model infection and recovery of grazingcattle with parameters specific for this species.

The SIR module was applied for grazing cattle in a form:

dScdt

¼ �bcSc þwcRc

dIcdt

¼ bcSc � Icac

dRcdt

¼ acIc � Rcwc

(1)

where index c denotes cattle; S, I and R are the abundance of sus-ceptible, infected, and recovered animals, respectively, head ha�1;1/a is recovery period, days; b is infection rate, day�1; and 1/w isimmune period, day.

For grazing deer, the SIR module was formulated for adults (a)and fawns (f) as:

dSadt

¼ farFA�SaðdaþbaÞþwaRadSfdt

bAF�Sf�df þbf þ r

�þwf Rf

dIadt

¼baSa� IaðdaþaaÞdIfdt

¼ bf Sf � If�df þaf þ r

�dRadt

¼aaIa�RaðdaþwaÞdRfdt

¼af If �Rf�df þwf þ r

�(2)

where index a denotes male (AM) or female (AF) adult deer,therefore the system of equations in the left part of Eq. (2) was usedfor bothmale and female adult deer; index f denotes fawns (FA); r isthe recruitment rate, day�1; b is the birth rate, day�1; and d is themortality rate, day�1.

The system of the first order differential Equation (2) can bereduced to a single equation to compute the changes in abundanceof deer groups (AM, AF, FA) over the period of time dt as:

dðAM þ AF þ FAÞ=dt ¼ bAF ��dAMAM þ dAFAF þ df FA

�(3)

where parameters b, dAM, dAF and df are defined using forcingfunctions that account for seasonal changes in deer birth andmortality (e.g., Altizer et al., 2006; Hosseini et al., 2004):

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200188

b ¼ B.h�

1þ expðcðSD� tÞÞ��

1þ 1=expðcðSDþ d1 � tÞÞ�i

da ¼ Mb þMw exp�� 0:5ððt � mwÞ=DwÞ2

�þMha exp

�� 0:5ððt � mhÞ=DhÞ2

�dfa ¼ Mb þMw exp

�� 0:5ððt � mwÞ=DwÞ2

�þMhf exp

��� 0:5ððt � mhÞ=DhÞ2

�þMf

.h�1þ expðcðSD� tÞÞ

���1þ 1=expðcðSDþ d2 � tÞÞ

�i(4)

where B denotes the height of birth function, day�1; c is a shapeparameter controlling the steepness of the double-logistic birthfunction, day�1; SD is the start date for births, day; d1 is the dura-tion of parturition/recruitment, day; d2 is the end date for post-parturition mortality of fawns, day; Mb is the background mortal-ity level (base height of mortality functions), day�1; Mf denotes theheight of the fawnmortality function in spring-summer, day�1;Mw

is the amplitude of winter mortality, day�1; Dw is the variance ofwinter mortality, day; Dh is the variance of harvest mortality, day;mw is the mean day of winter mortality, day; mh is the mean day ofharvest mortality, day; MhAM, MhAF, and Mhf are the amplitude ofharvest mortality for adult males, females, and fawns respectively,day�1.

Recruitment of white-tailed deer occurs in April, just prior toparturition, within a short period of time at a rate and partitioningbetween male and female computed as:

r ¼ 1=½30ð1þ expðcðSD� d1 � tÞÞÞð1þ 1=expðcðSD� tÞÞÞ�fAM ¼ 1=ð1þ ratioÞ fAF ¼ 1� fAM ratio ¼ bAM=bAF

(5)

where fAM and fAF are fractions of the male and female recruitment;and bAM and bAF are birth rates for male and female deer,respectively.

2.1.3. Dose-responseThe doseeresponse component of SIR module computes daily

rates of cattle and deer infection by E. coli O157:H7 based on dailyvalues of pathogen ingestion by cattle and deer with water, foliage,and via grooming soiled hide. Two doseeresponse modelscommonly used for wildlife were implemented in this study (Haaset al., 2000):exponential model:

b ¼ 1� expð � dose=kÞ (6)

and beta-Poisson model:

b ¼ 1�h1þ dose

�21=ad � 1

�.D50

i�ad(7)

where b is the daily infection rate, day�1; D50 is the daily dose ofE. coli O157:H7 that infect 50% of the exposed population, CFU day�1

head�1; ad is the slope parameter; dose is the number of pathogencells ingested daily by cattle or deer, CFU day�1 head�1; and k is thenumber of organisms that are ingested for one to survive and causeinfection, CFU day�1 head�1.

Parameter k in Eq. (6) was related to the parameter D50 in Eq. (7)to simplify the data input for SIR module as:

k ¼ �D50=lnð0:5Þ (8)

Daily doses of E. coli O157:H7cells ingested by grazing cattle anddeer were estimated based on the daily foliage and water con-sumptions, and on grooming soiled hide:

dose ¼ CwatVwat þ CfolMfol þ CgshMgsh (9)

where Cwat, Cfol and Cgsh are E. coli O157:H7 concentrations in water,on foliage, and on animal hide, respectively, CFU g�1; Vwat, Mfol andMgsh are mass of ingested water, dry forage and fecal material (i.e.grooming) daily by single animal, respectively, g day�1 head�1.

Daily water and foliage consumption for cattle is assumed to betime-invariant during the grazing period, but varied for deer withrespect to age, sex, and season. To account for seasonal variation inwater consumption bywhite-tailed deer fawns, does, and bucks weused the equations derived by Moen (1978) that relate deer weightto their age and day of the year:

MFAf ¼ 2:6þ 0:229AFaf 0<AFAf � 100

MFAf ¼ exp�1:617þ 0:381 ln

�AFAf

��n0:9þ 0:1 sin

�0:9863Jday þ 90

�oAFAf >100

MFAm ¼ 3:3þ 0:25AFAm 0<AFAm � 100MFAm ¼ 3:31A0:4589

FAm 100<AFAm � 260

MFAm ¼ 9:415A0:3415FAm

n0:83þ 0:17 sin�

0:9863Jday þ 180�o

AFAm >260

(10)

where MAF, MAM, MFAf and MFAm are the weights of adult females,adult males, fawn females and fawn males, respectively, kg; AFAf

and AFAm are the age of female and male fawns in days; Jday is theday of the year. Since modeling the age of an individual deer at awatershed scale was impractical, the Moen's equation was used fortwo deer gender groups in a simplified form:

MAF ¼ MAF

n0:9þ 0:1 sin

�0:9863Jday þ 90

�oMAM ¼ MAM

n0:83þ 0:17 sin

�0:9863Jday þ 180

�o (11)

where MAF and MAM are the average weights of adult females andmales at Jday ¼ 365. Then daily water consumption amount by in-dividual deer was computed from deer weights Mdeer as:

Vwat ¼ KwatM3=4deer (12)

where Kwat is deer daily water consumption rate, g water per kg3/4.Daily foliage consumption by deer relates to the seasonal deer

metabolism and availability of the forage (Moen, 1978) as:

Mfc ¼mblmð70$0:9$MdeerÞ3=4

mefo

Mfol ¼ Mfc Mfc <Mavl

Mfol ¼ Mavl Mfc >Mavl

(13)

where Mfc and Mavl are the potential and available daily foliageconsumption values by deer, g day�1 head�1; mefo is the metabo-lize energy in the foliage, kcal g�1; and mblm is the baselinemetabolism (kcal kg�3/4 day�1) computed as:

mblm ¼ 1:0285$sin�0:9863$Jday þ 239

�þ 2:6981 (14)

Grooming soiled hide has been shown to cause animal in-fections by smaller doses of E. coli O157:H7 compared to the bac-teria ingestion with water and foliage (McGee et al., 2004). Socialgrooming is common among the same and different deer sexgroups (Miller, 1971; Forand and Marchinton, 1989), and thereforecan cause the pathogen transmission from one individual toanother. Ingestion of E. coli O157:H7 via grooming was assumed

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 189

proportional to the daily mass of fecal material produced by ani-mals in the SIR module:

Mgsh ¼ fgshMfclAHRUI (15)

where Mfcl is the dry weight of fecal material produced daily byinfected animals at HRU, g day�1 head�1; I is the abundance ofinfected animals, head ha�1; AHRU is area of HRU, ha; fgsh is thefraction of this material on soiled animal hide that was licked viagrooming, head�1.

The daily E. coli O157:H7 ingestion dose (dose) was furtheradjusted for time duration that deer spend at each HRU:

dose ¼Xni¼1

fHRUidosei (16)

where dosei is the dose E. coli O157:H7 ingested at i-HRU, CFU; andfHRUi is the fraction of day that deer spend at i-HRU. The parameterfHRUi is computed in the resource selection component of SIRmodule described in Section 2.1.4. The values of daily doses ob-tained in Eq. (9) and corrected for deer grazing durations inparticular HRUs using Eq. (16) were used to calculate the values ofdaily infection rates b for grazing cattle and deer.

2.1.4. Resource selectionSub-basin is considered as the smallest spatial unit for deer

habitat in the resource selection component of the SIR module.Each sub-basin includes one tributary and several HRUs delineatedby the SWAT model based on a soil map, land use, and digitalelevation model (DEM). Deer abundance at HRUs typically exhibitsseasonal trends following the changes in forage availability anddeer preferences, type of land use, and distances to water sourcesand shelters. These trends are expressed as temporal variations ofodds ratios, which indicate the prevalence of deer at specific landuse relative to the deciduous forest. To define fractions of day thatdeer spend at each HRUwith specific land use for eachmonth of theyear fHRUi, the odds ratios ORjk were converted into probabilityvalues by solving the system of equations:

8>>>>><>>>>>:

ORjk ¼

Pjk1� Pjk

!, PjFRSD

1� PjFRSD

!

PjFRSD þXnk¼1

Pjk ¼ 1

(17)

where Pjk is the probability that in a given j-month deer choose aHRU with k-land use for grazing; and PFRSD is the probability that ina given j-month deer choose a deciduous forest.

The sub-basins may consist of several HRUs with the same landuse, but different soil types or terrain slopes. To account for frac-tions of the day that deer spend on each HRU of the same sub-basin,the probability values computed for each land use via Eq. (17) werenormalized by the total area of HRUs with the same land use withinone sub-basin as:

f jHRUi ¼ PjkAkHRUi

.Ak (18)

where AkHRUi is the area of the i-HRU with k-land use in the sub-

basin, ha; and Ak is the total area with k-land use in the subbasin,ha.

2.1.5. Bacteria sheddingThe shedding component of the SIR module computes daily

rates of pathogenic and nonpathogenic strains of E. coli shed by

cattle and deer. Several studies have reported that the duration andrates of E. coli O157:H7 shedding differed for animal species anddepended on animal age, gender, and diet as well as season (Rogers,1987; Sanderson et al., 1999; Fischer et al., 2001; Khaitsa et al.,2003). SWAT model uses constant in time rates of bacteria shed-ding by domestic and wild animals. These rates are averaged for allanimals belonging to the same group or species. It is impractical toassign different or changing in time shedding rates to individualwildlife animals, because their routes even within a small water-shed are generally unknown. Therefore averaging the pathogenshedding rates is a necessary procedure. However, it was possible toaccount for the daily variation in E. coli shedding within each sea-son that partly compensated for averaging of the shedding ratesand introduced temporal variability into the pathogen shedding. Todo that, we assumed that the deer defecation rates Ddoe are nor-mally distributed within the season with mean values, Ddoe, re-ported by Rogers (1987) for free-ranging female white-tailed deerand the standard deviation, s, computed as:

s ¼ ðDmax � DminÞ=4Ddoe ¼ ðDmax þ DminÞ=2

fdoe ¼1

sffiffiffiffiffiffi2p

p exp

��Ddoe � Ddoe

�22s2

(19)

where Dmin and Dmax are the minimum and maximum daily defe-cation rates for does, g head�1 day�1; and fdoe is the frequency ofdoe defecation rate Ddoe, unitless.

The rates for fawns and bucks were computed based on theratios derived from Van Etten (1959) publication as:

Dfawn

.Ddoe ¼ 1:14

Dbuck=Ddoe ¼ 1:52(20)

where Dfawn and Dbuck are the daily defecation rates for fawns andbucks, respectively, g head�1 day�1.

To account for the seasonal variability in the deer defecationrates the parameters s and Ddoe for does were set individually foreach season of the year. Finally, computed daily defecation rateswere converted into the mass of fecal materialMfcl and the numberof pathogenic and nonpathogenic E. coli cells Mbac (CFU) depositeddaily at HRUs based on deer grazing durations and dry mass ofdeposited feces:

Mfcl ¼ fHRUiDfclAHRUiIMbac ¼ MfclCbac

(21)

where Dfcl is the deer daily defecation rate, g head�1 day�1; and Cbacis the concentration of pathogenic and nonpathogenic E. coli cells indeer feces, CFU g�1.

Equation (21) was implemented in the SIR module separately tothe susceptible, infected, and recovered groups of animals. Notethat the susceptible and the recovered animals produced solelynonpathogenic E. coli, while the infected animals produced bothpathogenic and nonpathogenic E. coli at concentrations defined bythe parameter Cbac.

2.1.6. Bacteria survival in and release from fecal material andmanure

Applied manure and fecal deposits are considered as materialswith specific bacterial survival and release properties in the SWAT-SIR model. For the bacterial die-off/re-growth in the manure andfecal material we used a modified Chick's first order decay equationin a form used by the SWAT model:

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200190

Ctbac ¼ Ct�1

bac expð � kTdtÞ � Cminbac (22)

where Ctbac is the amount of bacteria present in the material (i.e.

manure, soil, foliage, water) on a current day, CFU m�2; Ct�1bac is the

amount of bacteria present on previous day, CFU m�2; Cminbac is the

minimum daily loss of bacteria for die-off, CFU m�2; dt is the timeinterval in SWAT simulations, 1 day by default; and kT is the first-order rate coefficient for bacteria growth/die-off corrected fortemperature as:

kT ¼ k20qT�20 (23)

where T is the mean daily temperature (�C); q is the temperaturecorrection coefficient; and k20 is the bacteria growth/die-off ratemeasured at temperature of 20 �C (day�1).

To describe bacteria release from applied manure and fecalmaterial deposited by the grazing cattle and deer we used theBradford-Schijven release model (Bradford and Schijven, 2002) in aform:

Mtbac ¼ ErMt�1

bac

�1�

�1þ afclbfclPrec

��1=bfcl�

(24)

where Mtbac is the number of bacteria cells released from fecal de-

posits into the aqueous phase on a current day, CFU; Mt�1bac the

number of bacteria cells in fecal deposits on previous day, CFU; Er isthe bacteria release efficiency, unitless; Prec is the daily precipita-tion, mm; afcl (mm�1) and bfcl (unitless) are the empirical param-eters defining the shape of the release curve. The released bacteriaare distributed between soil surface and vegetation proportionallyto the vegetation density computed by the SWAT model.

2.2. Watershed description and simulation scenarios

2.2.1. Watershed descriptionThe add-on SIR module for the SWAT software was used to es-

timate the impact of E. coliO157:H7 transmission between livestockand white-tailed deer on the pathogen spread at a small watershedwith fragmented agricultural and forest land uses. The land usewithin the 75.2 km2 watershed included forest (74%), hay (15.6),pasture (6.5%) and urban lands (3.8%). Data for the SWAT projectwere obtained from publically available sources. Specifically, DEM(the National Elevation Dataset (NED)) and land use (National LandCover Database (NLCD)) were downloaded from the USGS website(http://nationalmap.gov/viewer.html), the soil map layer wascreated using the SWAT US SSURGO soils database incorporatedinto the SWAT software, and three monitoring sites located at thewatershed provided the input weather data for the SWAT model(Fig. 3). The SWAT software delineates hydrological response units(HRUs) that have similar soils, land uses and slopes. To reduce thenumber of computational HRUs, soil types were grouped accordingto the MUID key corresponding to the State Soil Geographic Data-base (STATSGO, U.S. Department of Agriculture (1994)). Total of1388 HRUs were delineated within the simulated watershed. Threesites located along the watershed with contribution areas of18.8 km2, 36.8 km2, and 68.0 km2, respectively (Fig. 3) were used torecord water hydrograph, sample water for E. coli analysis andcalibrate the SWAT model (Kim et al., 2010).

2.2.2. Parameters of the deer population componentThe values of the parameters in Eqs. (3)e(4) were selected so as

to provide the steady-state annual dynamics of the deer popula-tion. The seasonal variation in buck, doe and fawn quantities isaccounted for, but the annual deer population is kept constant

(Fig. 4). Fractions of different gender and age groups correspondingto these parameter values were 0.221, 0.428 and 0.351 for deerbucks, does and fawns, respectively. In the simulations we set theanimal densities for grazing cattle to 4 head ha�1. Deer populationdynamics is generally unknown. Bureau of Wildlife Management ofPennsylvania Game Commission use sex-age-kill model PASAK(Rosenberry et al., 2011) to monitor deer population trends. Themodel provides estimates of white-tailed deer population for 23wildlife management units (WMU) for 2003e2009 years. To esti-mate deer density for the simulated watershed the deer densitieswere normalized by forest area forWMU 5A and thenmultiplied byforest area of the simulated watershed. Then deer abundance forJanuary 1 2005 (0.14 head ha�1) was estimated by fitting Eq. (2) andEq. (3) to the PASAK-estimated values (Fig. 4).

Parameters of the forcing functions that account for the seasonalchanges in deer birth and mortality (Eqs. (4) and (5)) were esti-mated from the literature. Birth rates for adult white-tailed deervary from 0.67 to 2.27 fawns per female (DeYoung, 2011). Parturi-tion of white-tailed deer in similar environments occurs during ashort period of time (~3 weeks), with captive deer studies reportingbirths as early as May 4 (Dechen Quinn, unpublished data) to midJune (Pekins et al., 1998). We fit a birth function that results in onefawn per female over the course of 20 days beginning 20 May:B¼ 0.05 day�1, c¼ 0.8 day�1, SD¼ 140 day, and d1 ¼ 20 day. White-tailed deer mortality is highly variable across their range andamong sex and age classes (DeYoung, 2011).

We fit mortality functions with 7.5% winter-caused mortalitiesfor all deer classes, 50% fawn mortality post-parturition, and class-specific mortalities due to fall harvest (fawns: 15%, adult females:13%, and adult males: 38%): d2 ¼ 63 day, Mb ¼ 10�6 day�1,Mf ¼ 0.008 day�1,Mw ¼ 0.002 day�1, Dw ¼ 15 day�1, Dh ¼ 15 day�1,mw ¼ 75 day, mh ¼ 320 day,Mham ¼ 0.01 day�1,Mhaf ¼ 0.0035 day�1,and Mhfa ¼ 0.0004 day�1. The annual dynamics of deer birth,recruitment and mortality rates calculated using these parametervalues are shown in Fig. 5. Fawn birth occurs in May and June,followed by fawn recruitment, while fawn death occurs, in Junethrough August due to illness and predation, in October throughDecember during hunting season and in March due to starvation.Hunting also causes high death rates for bucks and does in Octoberthrough December.

The value of the ratio parameter for males and females born intothe deer population in Eq. (5) was estimated based on Burke andBirch (1995) study. The authors analyzed offspring sex ratio ofwhite-tailed deer during three sequential years in Michigan withrespect to thematernal condition and age in total of 705 individualsand found that male to female ratio in newborn deer varied be-tween 1.04 and 1.24 with average of 1.14. The infection rates forcattle and deer were computed using the doseeresponse models(description of parameter value estimations is provided below).

2.2.3. Parameters of the doseeresponse and infection componentsValues of parameters k, ad and D50 for deer and cattle infection

by E. coli O157:H7 are generally unknown and are found to varygreatly for other ruminants. For example, for pigs, Cornick andHelgerson (2004) estimated the k values for the enter-ohemorrhagic (EHEC) E. coli to be 4.59$103 CFU. For rabbits, Haaset al. (2000) reported values of parameters in beta-Poisson modelas D50 of 5.9$105 CFU and ad of 0.49. Value of k parameter computedfor rabbits using Eq. (9) was 8.52$105 CFU, that is two orders ofmagnitude higher than the value determined for pigs. The infectivedose of E. coli O157:H7 for cattle varied from 2 to 10 log CFU instudies of Cray and Moon (1995), Cray et al. (1998), Shere et al.(2002), McGee et al. (2004) and likely was influenced by diet, mi-crobial flora, animal age, and immune status (Cray andMoon, 1995;Zhao et al., 1995; Johnson et al., 1996; Hovde et al., 1999; McGee

0 1 20.5

Kilometers

SoilMD020MD024PA022PA033PA059

0 1 20.5

Kilometers

Land UseWATRURLDURMDURHDFRSDFRSEFRSTHAYPASTWETF

0 1 20.5

Kilometers

Surface slope0-55-20> 20

monitoring locations

Fig. 3. Land use, soil, and surface slope maps of the watershed. NLCD land use classes in the legend denote: openwater (WATR), developed low intensity (URLD), developed mediumintensity (URLD), developed high intensity (URHD), deciduous forest (FRSD), evergreen forest (FRSE), mixed forest (FRST); hay (HAY), pasture (PAST), and woody wetlands (WETF).The soil legend includes MUID soil mapping units according to the State Soil Geographic Data Base (STATSGO, 1994).

Fig. 4. Annual dynamics of buck, doe, and fawn abundances calculated using param-eters shown in Table 1 of Appendix B and deer density estimated using PASAK software(Rosenberry et al., 2011).

Fig. 5. Annual dynamics of deer birth, recruitment and mortality rates.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 191

et al., 2004). For white-tailed deer the infective dose of 8 log CFU analidixic-acid resistant strain 86-24 Nal-R of E. coli O157:H7 hasbeen reported by Fischer et al. (2001). Scarcity of data on the cattleand deer infection dose indicates the need in further researchaimed at reducing uncertainty in estimation of these parameters.For the software demonstration purpose of this study, the values ofthe parameter D50 were set equal to 5∙105 CFU day�1 head�1 and105 CFU day�1 head�1, while the values of the parameters ad in thebeta-Poissonmodel were set to be equal to 0.5 and 0.2 for cattle anddeer, respectively.

The other parameters related to the doseeresponse model wereset for cattle as: Vwat ¼ 60,000 ml day�1 head�1,

Mfol ¼ 22.7 kg day�1 head�1 (Redfearn and Bidwell, PSS-2871); andfor the deer: MAM ¼ 118 kg, MAF ¼ 77 kg, 70/(pdec·gefo·meco)¼0.0301, Kwat ¼ 118 ml water per kg3/4 (French et al., 1956; Ullreyet al., 1968, 1970; Moen, 1978; Lautier et al., 1988). Annual dy-namics of deer weight, consumption of foliage and water calculatedusing Eqs.11e13 and these parameter values are shown in Fig. 6.

2.2.4. Parameters of the resource selection componentThe odds ratios were estimated based on monitoring data ob-

tained using GPS collared white-tailed deer in central NY (DechenQuinn et al., 2009). GPS collars acquired a location every 5 h andcollars were remotely detached and retrieved after approximately 1year. All deer were captured and handled in accordance with pro-tocols approved by the Institutional Animal Care and Use

0 1 2 3 4 5

Wat

er c

onsu

mpt

ion

(L h

ead-1

day

-1)

0123456

Dee

r wei

ght (

kg)

0

20

40

60

80

100

120

Female fawn Male fawn DoesBucks

Folia

ge c

onsu

mpt

ion

(kg

head

-1 d

ay-1

)

0

1

2

3

Year

Fig. 6. Annual dynamics of deer weight, consumption of foliage and water.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200192

Committee at the State University of New York College of Envi-ronmental Science and Forestry (protocol no: 2005-1). We incor-porated a use-versus-availability design structure to determinerelative probability of selection of various resources bywhite-taileddeer (Boyce et al., 2002; Manly et al., 2002) and defined the domainof available habitat using a step selection function similar to thatemployed in Fortin et al. (2005) using R 2.15.2 (R Development CoreTeam 2012). For each GPS location observed at time (t), we sampledrandomly from an empirically derived exponential distribution ofstep lengths (distance between two GPS locations) and a von Misesdistribution of turn angles (direction of travel between two GPSlocations). Each step length/turn angle pair originated at theobserved point (t) and generated a random point on the landscape.Then five random points for each GPS location were produced andeach random point generated from (t) was compared to the usedlocation at time (tþ1). In this way, we were able to compare step-specific availability at (t) to what the animal actually selected at(tþ1). Available points were considered random, but conditional onthe location of the observed point fromwhich theywere generated.Landscape attributes were extracted from 30 � 30 m resolutionraster layers to each corresponding observed or available locationin ArcGIS 10.0 (ESRI 1999e2010). Land cover typewas recorded as acategorical variable; distance to water as a continuous variable.Land cover identification was determined from the 2006 NationalLand Cover Database (Fry et al., 2011) that we reclassified from 21land cover categories to 9 categories of the Anderson classificationsystem: open water, urban land, coniferous forest, mixed forest,hay, row crops, rangeland, and wetland (Anderson et al., 1972). Thereclassified land cover categories were used to elucidate the func-tional differences between forage and cover that were presumed tobe important in driving space use by white-tailed deer while pre-senting a parsimonious model that reflected reasonable statisticalpower.

A surface that calculated continuous, Euclidian distance fromhydrological features to the deer location was generated for thewatershed used in this study. This variable evaluated their distanceto water features, such as streams and drainages. It is distinguishedfrom the land-cover type ‘open water’, which is typified by largerbodies of water. We developed candidate models that reflectedthese variables hypothesized to serve as important resources towhite-tailed deer and used a conditional logistic regression withinthe Cox Proportional Hazards (coxph) function from the survivalpackage (Therneau and Lumley, 2009) in R 2.15.2 (R DevelopmentCore Team 2012). To obtain odds ratios we divided the dataset intomonthly subsets and fit the same set of models to each monthlysubset:

logit�Pjk

�¼ b

j0 þ

Xnk¼1

bjkX

jk; n ¼ 8 j ¼ 1� 12 (25)

where Xjk is categorical variable for k-land use; k is the land cover

index; j stands for month of the year; bjk are log(ORjk). Average

monthly values of odds ratios ORjk for different land uses are shown

in Table 1 of Appendix B.

2.2.5. Parameters of the recovery and pathogen sheddingcomponents

The recovery periods 1/a for cattle and deer were estimatedfrom Sanderson et al. (1999), Khaitsa et al. (2003) and Fischer et al.(2001). Sanderson et al. (1999) studied fecal shedding of E. coliO157:H7 by one-week-old calves orally inoculated with 5∙108 CFUof E. coli O157:H7 that was a nalidixic-acid resistant strain 86-24Nal-R. The authors showed that following the first inoculation, thecalves shed E. coli O157:H7 in their feces for 20e43 days, and theE. coli O157:H7 shedding period decreased to 3e8 days after thesecond and third inoculations. Khaitsa et al. (2003) found for

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 193

feedlot cattle that the duration of E. coli O157:H7 shedding variedfrom 2 to 4 days during pre-epidemic period, from 21 to 30 daysduring epidemic, and from 4 to 14 days in post-epidemic period.Fischer et al. (2001) investigated fecal shedding of E. coli O157:H7by deer that received 108 CFU inoculum of nalidixic acid-resistantstrain. The authors found that deer can shed E. coli O157:H7 up to26 days post inoculation, (DPI) with maximum E. coli O157:H7concentrations in deer feces from 3.5 to 5.1 log10 CFU during thefirst 10 DPI followed by a decrease to less than 10 CFU of E. coliO157:H7 per gram of feces during 12e17 DPI.

Simulation of infection and recovery of individual animals isimpractical at a watershed scale since their exact grazing area isgenerally unknown. In this study we assumed, therefore, that allinfected animals shed the same number of E. coli O157:H7 cell pergram of feces during the whole recovering period of time. The re-covery periods and shedding E. coli O157:H7 concentrations wereestimated from linear parts of the cumulative curves of E. coliO157:H7 cells shed by cattle and deer as a function of time. Thesecurves were generated for cattle based on the results of Sandersonet al. (1999) and for deer based on the results of Fischer et al. (2001)(Fig. 7). We first computed the slopes of the cumulative curves, thatare time averaged E. coliO157:H7 concentrations in feces of infectedanimals, and then calculated recovery periods dividing the totalnumber of E. coli O157:H7 cells shed during the infection period bythe slope of the cumulative curves. Themean values of the recoveryperiods are shown for deer and cattle in Table 2 of Appendix B.Concentrations of nonpathogenic E. coli in cattle and deer feces inthis table were set based on Soupir et al. (2006) and Guber et al.(2015) studies.

Seasonal values of deer defecation rates were estimated fromRogers (1987) publication, who reported seasonal variation in thenumber of fecal pellet groups deposited daily by seven free-rangingfemale white-tailed deer in mixed coniferous-deciduous forests innortheastern Minnesota (Table 3 of Appendix B). The same studyreported that the number of pellets in groups ranged from 30 to 96with an average of 68.7 pellets group�1. Dry weight of a deer pelletwas set equal to 0.214 ± 0.056 g pellet�1 based on our directmeasurements. The seasonal variability in cattle defecation rateswas ignored in simulations and the daily rates were taken fromASAE D384.2 publication.

Fig. 8 illustrates seasonal variation in deer defecations rates

Deer

DPI

0 2 4 6 8 10 12 14Cum

ulat

ive

E.c

oli O

157:

H7

shed

by

anim

als

(CFU

/g)

0

105

2x105

3x105

4x105

5x105

Fig. 7. Cumulative amounts of E. coli O157:H7 shed in cattle and deer feces calculated bassymbols denote values measured for individual animals; lines denote regression models fit

calculated for does, fawns, and bucks using Eq. (19) and parametervalues adjusted for bucks and fawns using Eq. (20). It can be seen inFig. 8, that minimum and maximum defecation rates for all deergroups corresponded to Winter and Fall, respectively, while theintermediate defecation rates occurred in Spring and Summermonths.

There are a few studies that reported an immune response ofdeer and cattle to infection by E. coli O157:H7. No antibodies againstE. coli O157:H7 were detected in exposed deer in Fischer et al.(2001); that lack of detection was attributed to minimal interac-tion between the bacteria and the host. The calves reinoculated by10 log CFU of E. coli O157:H7 13e14 weeks after the last positivefecal sample were all reinfected in Cray and Moon (1995) study.However, after the first inoculation the calves shed considerablylonger and higher E. coli O157:H7 concentrations than after sub-sequent reinoculation. The calves which were positive for 14 and 20weeks after primary inoculation were positive for 4 and 3 weeksafter reinoculation. No or little change in O157 antibody titers incalves administered 106 and 107 CFU were found in Johnson et al.(1996) and Shere et al. (2002) studies. The first study alsorevealed changes in the antibody titers at E. coli O157:H7 inocula-tion dose of 1010 CFU, however the antibodies did not influence thereinfection of the animals with the same E. coli strain. Based on theresults of these studies, the value of the immune period parameter1/w was set equal to 1 day meaning that all recovered animals (R)transferred to the susceptible group (S) on the first day after theirrecovery.

2.2.6. Bacteria survival and release parametersParameters of the Chick's first order decay Equation (22) and

Arrhenius temperature correction Equation (23) were estimatedbased on the E. coli die-off kinetics or the kinetics parametersmeasured by different authors in soil, bottom sediment, animalfeces, dry manure, and manure slurry for different temperatureranges (Fig. 9). Part of this dataset was obtained in E. coli O157:H7studies and therefore should describe more accurately the path-ogen survival compared to the other E. coli strains. Due to limitedpublished data on the pathogens, the parameters summarized inTable 4 of Appendix B were applied for both pathogenic andnonpathogenic E. coli strains in SWAT-SIR simulations of this study.

Parameters of the Bradford-Schijven bacteria release model (Eq.

Cattle

DPI

0 5 10 15 20 25 30 35 400

2x105

4x105

6x105

8x105

106

1x106

1x106

2x106

ed on Sanderson et al. (1999) and Fischer et al. (2001) studies, respectively. Differentto the linear part of the curves.

Does

0.0 0.2 0.4

Freq

uenc

y

0.00

0.02

0.04

0.06

0.08

0.10

0.12

0.14Fawns

Defecation rates (kg day-1 head-1)

0.0 0.2 0.4

Bucks

0.0 0.2 0.4 0.6

WinterSpring Summer Fall

Fig. 8. Seasonal changes in defecation rates for three deer groups.

Temperature oC

0 10 20 30 40

κ T (d

ay-1

)

0.2

0.4

0.6

0.8

1.0

1.2

1.4Bovine feces Old dairy slurryFresh dairy slurry Fresh dairy manure Sheep fecesDeer pellet Sandy loam soil Loamy sand sediment Sandy loam sedimentStream water

Fig. 9. Relationships between E. coli die-off rates and temperature estimated fordifferent environments based on the parameters of the E. coli inactivation model ob-tained from published studies listed in Table 5 of Appendix B. Symbols denotemeasured data, while lines denote the values computed using Eqs. (22) and (23).

Precipitation, mm

0 2 4 6 8 10

Frac

tion

of in

itial

con

tent

0.0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

Dairy fresh manureSolid dairy manureLiquid dairy manureDeer feces

Fig. 10. Release kinetics of E. coli from manure and deer feces.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200194

(24)) adopted from Guber et al. (2015) publication are shown inTable 5 of Appendix B. Fig. 10 illustrates the release kinetics ofbacteria from four different fecal materials as a function of pre-cipitation computed based of the parameter values shown inTable 5 (Appendix B). The release is the fastest from liquid manure,followed by fresh manure and deer feces with solid manure beingthe lowest.

2.2.7. Simulation scenariosPathogen transmission between co-grazing cattle and deer was

evaluated in three simulation scenarios. The same managementpractices, weather conditions, duration and conditions of grazing,animal population dynamics, and resource selection parameterswere used in all three scenarios.

The scenarios differed in E. coli O157:H7 content in the appliedmanure and in the initial number of infected cattle, and the specificsettings used in the three scenarios are shown in Table 1. Specif-ically, in scenario 1, 4% of cattle were infected by E. coli O157:H7 inthe beginning of grazing season while the applied manure con-tained no E. coli O157:H7. In scenario 2, the cattle was not infectedby E. coli O157:H7 in the beginning of grazing season, however the

applied manure contained E. coli O157:H7 in amount of1.11∙103 CFU (g dry mass)�1. In scenario 3, both cattle infection byE. coliO157:H7 and application of manurewith E. coliO157:H7werepresent. Whole deer population was not infected in the beginningof the simulations in either of the studied scenarios (January 1st).

We ran the SWAT-SIR model with values of the hydrologicalparameters obtained by Kim et al. (2010) using SWATcalibration onwater hydrograph data and concentrations of E. coli measured insteam at monitoring location shown in Fig. 3. Manure productionby cattle was set equal to 51, and 68 wet kg day�1head�1 for beefand lactating cows (American Society of Agricultural Engineers(ASAE) D384.2), respectively. Average water content of the manurewas estimated as 86% (ASAE D384.2). E. coli concentrations in theliquidmanure applied from an aerobic digester were reduced by 0.5log factor compared to fresh cattle deposits (CH2M Gore & StorrieLimited (CG&S), 2000; Daigger and Bailey, 2000). The manure wasapplied on pastures and hay fields onMay 8th. The application rates

Table 1Percentage of initially infected animals, total E.coli and E.coli O157:H7 contents in the solid manure applied at pastures used in the three scenarios.

Scenario Infected cattle Infected deer E.coli E.coli O157:H7

% % CFU (g dry mass)�1 CFU (g dry mass)�1

1 4 0 7.11∙106 02 0 0 7.11∙106 1.11∙103

3 4 0 7.11∙106 1.11∙103

Fig. 11. Precipitation and daily air temperature used in simulations.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 195

were 47.8 ton ha�1, 3.0 ton ha�1and 14 ton ha�1for the solidmanureat hay, liquid manure at hay and solid manure at pastures,respectively. Cattle grazing season at pastures started May 24 andlasted 150 days. The simulations were conducted for one year withprecipitation and temperature shown in Fig. 11.

3. Uncertainty analysis

Uncertainty analysis was conducted to evaluate the relativeimpacts of input variable uncertainties on model prediction un-certainty for the three simulation scenarios shown in Table 1. Acomprehensive review of uncertainty evaluation methods fordeterministic models in environmental modeling can be found inHammonds et al. (1994), Refsgaard et al. (2007), Matott et al.(2009). Specifically for SWAT, Talebizadeh et al. (2010) evaluateduncertainty in predicting sediment loads associated with the un-certainty of soil erodibility and hydraulic properties. Uncertaintywas quantified using a 95% confidence interval of the predictedvariable, and p- and d-factors as described by Abbaspour et al.(2004). The p-factor indicated the percentage of observationsbracketed by the 95% confidence interval of the predicted variable,while the d-factor was the average width of the 95% confidenceinterval normalized by the standard deviation of observed values.The same statistics with little variation were used by Setegn et al.(2010), Shen et al. (2012), Singh et al. (2014), and Leta et al.(2015) to evaluate uncertainty in predicted streamflow associatedwith rainfall uncertainty and parameters of overland and ground-water flow. Zhang et al. (2009) used 66.7% and 90% confidenceintervals to evaluate uncertainty in predicted streamflow.

The analysis of model predictive uncertainty associated withparameter uncertainty assumes that the model is calibrated, vali-dated and probability distributions of the input model parametersare known. In this study due to absence of measurements on deer

and cattle infection rates, model calibration and validation was notpossible. Therefore, we relied on parameters of deer and cattleinfection components estimated from limited literature data. Wealso expected interactions between different parameters of the SIRmodule to affect predictive uncertainty, however these interactionswere generally unknown and therefore difficult to account for inthe uncertainty analysis. For these reasons a comprehensive anal-ysis of the effect of parameter uncertainty on model predictiveuncertainty using methods of uncertainty analysis commonly usedin hydrological modeling (e.g. Beven and Binley, 1992; Thyer et al.,2009; Vrugt et al., 2003; Leta et al., 2015) and statistics imple-mented for SWAT in the listed above publications was not possiblefor the SIR module. We instead used a simplified uncertaintyanalysis in this study. In performing such an analysis, we antici-pated that even a simplified uncertainty analysis of pathogentransmission between wildlife and domestic animals would pro-vide new insights into fate and transport of pathogens in mixedforest-agricultural settings and stimulate new research in this area.In this sense we agree with Jakeman et al. (2006) that “…

improvement of understanding of the system is almost always apurpose of modelling …” and “It is important to recognize thatsome purposes, particularly increased understanding of the systemand data, may be realised well even if the final model is poor inmany respects. An inaccuratemodel may still throw light on howanenvironmental systemworks”. We also used the interquartile range(a 50% confidence interval) instead of a 95% confidence interval tocharacterize the predictive uncertainty of the percentage of infec-ted co-grazing deer and cattle, and E. coli O157:H7 contents inapplied manure and fecal deposits of deer and cattle. Similar to the95% confidence interval in surface hydrological modeling, theinterquartile range is commonly used to quantify a spread of dis-eases in epidemiology (Merrill, 2009; US-HHS, 2006). Predictionuncertainty was also quantified using variance of the percentage ofinfected cattle and deer (Bennett et al., 2013).

Nineteen model parameters representing four components ofthe SIR module were selected for the uncertainty analysis (Table 2of Appendix B). To reduce the number of model runs we used theLatin hypercube sampling (LHS) approach with mean and standarddeviation values shown in Table 2 of Appendix B. The standarddeviation values corresponded to a coefficient of variation equal toCV ¼ 5% for all model parameters except the parameter ad in thedoseeresponse component, where CV was set to 25% due to rela-tively low model sensitivity to this parameter compared to theothers. Parameter values were assumed to be normally distributedand constrained by 90% confidence intervals of their values toexclude extreme parameter values.

Special attention was paid to the parameters of the resourceselection component. It was impractical to generate 84 parametersrepresenting all 7 land uses for all 12 months. The differences invariation of ln(ORjk) were not statistically significant over 12months within each land use but were significant for land uses.Therefore, we generated the values of ln(ORjk) using a normal dis-tributionwith zero mean andmonthly averaged s values separatelyfor each land use (Table 2 of Appendix B). Then the odds ratios werecalculated individually for each month and land use as:

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200196

ORjk ¼ exp�ln�OR

jk

�þ sk

�(26)

where ORjk is average monthly odds ratios for j-month and k-land

use, sk is the standard deviation of ln(ORjk) for k-land use.The UNIX Library/Standalone version of the Latin Hypercube

Sampling Software (Swiler and Wyss, 2004) was used to generatethe parameter dataset for the SWAT-SIR uncertainty analysis.

4. Results

The developed SWAT-SIR software allowed analyzing annualdynamics of deer and cattle infection and spatial distributions oftotal E. coli and E. coli O157:H7 loads across the watershed. Theannual dynamics predicted in the three scenarios indicated a pos-sibility of deer and cattle infection via consumption of foliagecontaminated by E. coli O157:H7, which was applied with manure

Fig. 12. Percentage (aec) and variance (d) of infected cattle and deer populations obtained infor scenarios (a), (b) and (c) as, respectively.

or shed by infected cattle. Up to 1.5% of deer were infected soonafter application of dry manure, which contained E. coli O157:H7 inMay (Fig. 12b,c), however deer infection did not occur in May in thefirst scenario, where the manure was not applied (Fig. 12a). Themanure application did not affect the grazing cattle. By the begin-ning of the cattle grazing period, the E. coli O157:H7 content in themanure decreased to such levels that caused infection only in 0.1%of the cattle population (Fig. 12b). The model predicted deerinfection in July through November in the scenarios where 4% ofthe cattle was infected by E. coli O157:H7 in the beginning of thegrazing season (Fig. 12ac). The percentage of the infected deerfollowed the increase in the number of co-grazing infected cattle,the increases typically occurred after intensive rainfalls. Themaximum percentage of the deer infected by co-grazing cattlereached 0.5% that was less compared with the number of infecteddeer caused by the manure application. The relatively small num-ber of infected deer in AugusteOctober compared with May in

the three simulation scenarios. Solid, dotted and dash lines (d) denote variance values

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 197

scenarios 1 and 3 (Fig. 12a,c) can be explained by overall smallerloads of E. coli O157:H7 with cattle fecal material compare to thesolid manure.

The infection periods and the numbers of infected animalsdiffered in the three studied scenarios and depended on the sourceof the infection. The first infection affected approximately 1.5% ofthe deer population and the infection period lasted approximately2 weeks, that was twice as long as the deer recovery period 1/a(Table 2 of Appendix B). The long infection period and the shapes ofthe curves in Fig. 12b and c implied that E. coli O157:H7 survived inthe applied manure and was released from it on foliage at thepasture during multiple rainfalls in MayeJune, causing several deerinfection events.

We have not seen a noticeable E. coli O157:H7 transmission fromdeer to cattle in the simulations. The maximum E. coli O157:H7content in deer feces deposited on pasture did not exceed10 CFU m�2 in the simulations. This can be partly attributed tospatial averaging of E. coli O157:H7 content at HRUs by the model,which probably resulted in smaller bacteria contents per unit areathan those that could be expected in reality. However, the maincauses of the absence of the E. coli O157:H7 transmission weresmall deer population (14e21 heads per 100 ha of forest area),foliage availability in non-contaminated agricultural lands, andrelatively small fecal mass produced by deer, which was dispersedacross large grazing area.

The uncertainty analysis revealed high variations in numbers ofinfected deer and cattle in simulations with parameters generatedat CV¼ 5%. The variability in numbers of infected animals increasedwith the increase in overall numbers of infected animals; thehighest interquartile ranges and variance were 2% and 3.7% forcattle, and 0.8% and 1.7% for deer at average percentage of infectedanimals of 1.8% and 1.0% for cattle and deer, respectively (Fig. 12).The high uncertainty of the model predictions was likely caused byhigh model sensitivity to the parameters, particularly to the dos-eeresponse and bacteria shedding components, in the simulatedscenarios. A comprehensive model sensitivity analysis to the

Fig. 13. Spatial distributions of total E. coli and E. c

parameters introduced in the four new model components as afunction of weather and management scenarios is beyond thescope of this publication, and will be the subject of future modelexperiments and development.

The uncertainty in E. coli O157:H7 concentrations predicted inmanure and fecal deposits, assessed as the interquartile range (Fig,12 a,b,c) and variance (Fig. 12 d), increased with the time. The un-certainty was the lowest before the start of cattle grazing seasonand the highest at the end of the season. The low uncertainty ob-tained in the simulations was attributed to relatively small path-ogen production by deer compared to the production by cattle.Average daily production of E. coli O157:H7 were 9.3∙105 CFUhead�1 and 5.6∙107 CFU head�1 by infected deer and cattle insimulated scenarios, respectively. Moreover, infected deer spentonly fraction of day on pasture that considerably reduced theirpathogen loads on pastures. The high uncertainty in E. coli O157:H7concentrations was associated mostly with uncertainty of the pa-rameters for grazing cattle. It can be seen in Fig. 12a and c that thevalues and uncertainties in the E. coli O157:H7 concentrations wereapproximately the same during grazing season despite of the dif-ferences in percentage of infected deer and manure applicationscenarios. Both deer and applied manure had little effect on theE. coli O157:H7 during grazing season. This occurred due to rela-tively low pathogen input from deer on one hand, and considerabledecrease of pathogens resided in the applied manure on otherhand, that reduced E. coli O157:H7 contents in the fecal materialbelow 25 CFU g�1 (Fig. 12 b). This pathogen level was by far belowthe infection dose for both deer and cattle.

The uncertainty patterns in percentage of the infected cattlefollowed the uncertainty in the E. coli O157:H7 content in fecaldeposits in scenarios 1 and 3 only until mid of August (Fig12a,c,d).Further increase in the deposited pathogen contents coincided intime with the decrease in percentage of infected cattle. The dif-ferences in the uncertainty patterns were partly attributed to thedecrease in percentage of infected cattle due to recovery and to asmall number of rainfall events in September and October. Low

oli O157:H7 obtained in simulation scenario 3.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200198

precipitation considerably reduced release of pathogens from cattledeposits on foliage, and as a result reduced probability of cattleinfection. The observed patterns could be specific for the simulatedscenarios, and open an interesting avenue for further modelexploration.

Examples of spatial distributions of E. coli loads by deer andfrom livestock operations obtained in the simulated scenario 3 areshown in Fig. 13. High spatial variability in the distribution of E. coliacross a relatively small watershed was attributed to different landuse and management practices. Less total E. coli cells transportedwith runoff from forested areas and more from pastures resulted inhigher E. coli loads on surfacewater where pasturelands dominatedin sub-basin area (Fig. 13a). Despite the relatively large area occu-pied by forests, the contribution of deer to the total E. coli loads onthe surface waters was 2e3 orders of magnitude smaller comparedwith the loads from the livestock operations and particularlygrazing. E. coli O157:H7 runoff patterns (Fig. 13b) resembled spatialdistribution of total E. coli, however the absolute values for E. coliO157:H7 were two orders of magnitude smaller than those forE. coli. Fig. 13c and d illustrates simulated loads of E. coli O157:H7 onHRUs from livestock operations and by infected deer. It can be seenthat, despite smaller E. coli O157:H7 numbers, the area where thepathogens were spread by deer considerably exceeds the areawhere the pathogens were applied with manure or shed by theinfected cattle. E. coli O157:H7 in amounts 10e100 CFU m-2 werespread by deer to the hay fields bordering the infected pastures.These results were obtained for the scenario when only 2% of deerpopulation shed the pathogens, and the E. coli O157:H7 amountscould be considerably higher under the scenario with a largerpercentage of infected deer.

The software developed and the results obtained in the simu-lations open a broad avenue for further research of the pathogentransmission between livestock and wildlife, which potentially canbe extended to different domestic and wildlife species.

5. Conclusions

A new add-on module SIR developed for the watershed scalemodel SWAT can predict transmission of the manure-borne path-ogens between livestock and deer. The newmodel allows analyzingannual dynamics of infected cattle and deer, pathogen sheddingand loads on agricultural and forested lands, release of pathogensfrom fecal deposits to foliage and soil, pathogen overland transportand surface water pollution. The software provides insights intorelationships between farm manure operations, pathogen trans-mission to deer, and pathogen appearance in surface water. Thisinformation can be helpful in identifying the potential sources ofpathogenic microorganisms and for development of better man-agement practices to reduce pathogen loads on fragmented agri-cultural and forest landscapes.

Acknowledgments

The work has been supported through the Water InitiativeResearch Grant “Multimodeling framework for predicting fate andtransport of pathogens originated from wildlife and livestock in-teractions in fragmented agriculture-forest ecosystems” of theCenter for Water Sciences and Environmental Science and PolicyProgram of Michigan State University.

Appendix A. Supplementary data

Supplementary data related to this article can be found at http://dx.doi.org/10.1016/j.envsoft.2016.02.024.

References

Abbaspour, K.C., Johnson, A., van Genuchten, M.Th, 2004. Estimating uncertain flowand transport parameters using a sequential uncertainty fitting procedure.Vadose Zone J. 3, 1340e1352.

Altizer, S.A., Dobson, A., Hosseini, P., Hudson, P., Pascual, M., Rohani, P., 2006. Sea-sonality and the dynamics of infectious diseases. Ecol. Lett. 9, 467e484.

Anderson, J.R., Hardy, E.E., Roach, J.T., 1972. A land-use classification system for usewith remote-sensor data. U.S. Geol. Surv. Cire 671, 16.

ASAE D384.2 MAR2005. Manure Production and Characteristics, American Societyof Agricultural Engineers, 20 pp.

Baffaut, C., Benson, V.W., 2003. A bacterial TMDL for Shoal creek using SWATmodeling DNA source tracking. ASAE Publication No. 701P1503. In: TotalMaximum Daily Load (TMDL) Environmental Regulations-ii Proceedings. ASAE,St. Joseph, MI.

Baffaut, C., Sadeghi, A.M., 2010. Bacteria modeling with SWAT for assessment andremediation studies: a review. Trans. ASABE 53 (5), 1585e1594.

Benham, B.L., Baffaut, C., Zeckoski, R.W., Mankin, K.R., Pachepsky, Y.A.,Sadeghi, A.M., Brannan, K.M., Soupir, M.L., Habersack, M.J., 2006. Modelingbacteria fate and transport in watersheds to support TMDLs. Trans. ASABE 49,987e1002.

Bennett, N.D., Croke, B.F.W., Guariso, G., Guillaume, J.H.A., Hamilton, S.H.,Jakeman, A.J., Marsili-Libelli, S., Newham, L.T.H., Norton, J.P., Perrin, C.,Pierce, S.A., Robson, B., Seppelt, R., Voinov, A.A., Fath, B.D., Andreassian, V., 2013.Characterising performance of environmental models. Environ. Model. Softw.40, 1e20.

Besser, R.E., Lett, S.M., Weber, J.T., Doyle, M.P., Barrett, T.J., Wells, J.G., Griffin, P.M.,1993. An outbreak of diarrhea and hemolytic uremic syndrome from Escherichiacoli O157:H7 in freshepressed apple cider. J. Am. Med. Assoc. 269, 2217e2220.

Beven, K., Binley, A., 1992. The future of distributed models: model calibration anduncertainty prediction. Hydrol. Process 6, 279e298.

Bicknell, B.R., Imhoff, J.C., Kittle Jr., J.L., Donigian Jr., A.S., Johanson, R.C., 1997. Hy-drological Simulation ProgrameFortran, User's manual for version 11: U.S.Environmental Protection Agency. National Exposure Research Laboratory,Athens, Ga., EPA/600/R-97/080, 755 p.

Borczyk, A.A., Karmali, M.A., Lior, H., Duncan, L.M.C., 1987. Bovine reservoir forverotoxin-producing Escherichia coli O157:H7. Lancet 98.

Boyce, M.S., Vernier, P.R., Nielsen, S.E., Schmiegelow, F.K.A., 2002. Evaluatingresource selection functions. Ecol. Model. 157, 281e300.

Bradford, S.A., Schijven, J., 2002. Release of Cryptosporidium and Giardia from dairycalf manure: impact of solution salinity. Environ. Sci. Technol. 36, 3916e3923.

Branham, L.A., Carr, M.A., Scott, C.B., Callaway, T.R., 2005. E. coli O157:H7 and Sal-monella spp. in white-tailed deer and livestock. Curr. Iss. Intest. Microbiol. 6,25e29.

Burke, R., Birch, J.M., 1995. White-tailed deer vary offspring sex-ratio according tomaternal condition and age. Ecol. Res. 10, 351e357.

CG&S, 2000. Assessing Stabilization Criteria for Aerobically and AnaerobicallyDigested Sewage Biosolids. CH2M Gore & Storrie Limited (Report prepared forthe Ontario Ministry of the Environment).

Chapman, P.A., 2000. Sources of Escherichia coli O157 and experiences over the past15 years in Sheffield, Uk. J. Appl. Microbiol. 88, 51Se60S {Symposiumsupplement}.

Chin, D.A., Sakura-Lemessy, D., Bosch, D.D., Gay, P.A., 2009. Watershed-scale fateand transport of bacteria. Trans. ASABE 52, 145e154.

Cho, K.H., Pachepsky, Y.A., Kim, J.H., Kim, J.W., Park, M.H., 2012. The modified SWATmodel for predicting fecal coliforms in the Wachusett Reservoir Watershed,Usa. Water Res. 46, 4750e4760. http://dx.doi.org/10.1016/j.watres.2012.05.057.

Cieslak, P.R., Barrett, T.J., Griffin, P.M., Gensheimer, K.F., Beckett, G., Buffington, J.,Smith, M.G., 1993. Escherichia coli infection from a manured garden. Lancet 342,367.

Cody, S.H., Glynn, M.K., Farrar, J.A., Cairns, K.L., Griffin, P.M., Kobayashi, J., Fyfe, M.,Hoffman, R., King, A.S., Lewis, J.H., Swaminathan, B., Bryant, R.G., Vugia, D.J.,1999. An outbreak of Escherichia coli O157:H7 infection from unpasteurizedcommercial apple juice. Ann. Intern. Med. 130, 202e209.

Coffey, R., Cummins, E., Bhreathnach, N., Flaherty, V.O., Cormican, M., 2010.Development of a pathogen transport model for Irish catchments using SWAT.Agric. Water Manag. 97, 101e111.

Cornick, N.A., Helgerson, A.F., 2004. Transmission and infectious dose of E. coliO157:H7 for swine. Appl. Environ. Microbiol. 70, 5331e5335.

Cray Jr., W.C., Moon, H.W., 1995. Experimental infection of calves and adult cattlewith Escherichia coli O157:H7. Appl. Environ. Microbiol. 61, 1586e1590.

Cray Jr., W.C., Casey, T.A., Bosworth, B.T., Rasmussen, M.A., 1998. Effect of dietarystress on fecal shedding of Escherichia coli O157:H7 in calves. Appl. Environ.Microbiol. 64, 1975e1979.

Daigger, G.T., Bailey, E., 2000. Improving aerobic digestion by prethickening, stagedoperation, and aerobiceanoxic operation: four full-scale demonstrations. WaterEnviron. Res. 72, 260e270.

Daszal, P., Cunningham, A.A., Hyatt, A.D., 2000. Emerging infectious diseases ofwildlife threats to biodiversity and human health. Science 287, 443e449.

Dechen Quinn, A.C., Williams, D.M., Porter, W., Smith, M., DeSantis Jr., F., McNulty, S.,2009. Risk Assessment of a Chronic Wasting Disease Outbreak in New York:final report. New York State Department of Environmental Conservation.

DeYoung, C.A., 2011. Population dynamics. In: Hewitt, D.G. (Ed.), Biology andManagement of White-tailed Deer. CRC Press, Boca Raton, FL, pp. 147e180.

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200 199

Dorner, S.M., Anderson, W.B., Slawson, R.M., Kouwen, N., Huck, P.M., 2006. Hy-drologic modeling of pathogen fate and transport. Environ. Sci. Tech 40,4746e4753.

Ferguson, C.M., Croke, B.F.W., Beatson, P.J., Ashbolt, N.J., Deere, D.A., 2007. Devel-opment of a process-based model to predict pathogen budgets for the Sydneydrinking water catchment. J. Water Health 5, 187e208.

Fischer, J.R., Zhao, T., Doyle, M.P., Goldberg, M.R., Brown, C.A., Sewell, C.T.,Kavanaugh, D.M., Bauman, C.D., 2001. Experimental ad field studies of Escher-ichia coli O157:H7 in white-tailed deer. Appl. Environ. Microbiol. 67, 1218e1224.

Forand, K.J., Marchinton, R.L., 1989. Patterns of social grooming in adult White-tailed deer. Am. Midl. Nat. 122, 357e364.

Fortin, D., Beyer, H., Boyce, M.S., Smith, D.W., Mao, J.S., 2005. Wolves influence elkmovements: behavior shapes a trophic cascade in Yellowstone National Park.Ecology 86, 1320e1330.

French, C.E., McEwen, L.C., Magruder, N.D., Ingram, R.H., Swift, R.W., 1956. Nutrientrequirements for growth and antler development in the white-tailed deer.J. Wildl. Manag. 20, 221e232.

Frey, S.K., Topp, E., Edge, T., Fall, C., Gannon, V., Jokinen, C., Marti, R., Neumann, N.,Ruecker, N., Wilkes, G., Lapen, D.R., 2013. Using SWAT, Bacteroidales microbialsource tracking markers, and fecal indicator bacteria to predict waterbornepathogen occurrence in an agricultural watershed. Water Res. 47, 6326e6337.

Fry, J., Xian, G., Jin, S., Dewitz, J., Homer, C., Yang, L., Barnes, C., Herold, N.,Wickham, J., 2011. Completion of the 2006 national land cover database for theconterminous United States. Photogramm. Eng. Rem. S 77 (9), 858e864 (PDF) (9pp, 1.2MB). http://www.mrlc.gov/downloadfile2.php?file¼September2011PERS.pdf.

Gallagher, M.A., Karthikeyan, R., Mukhtar, S., 2012. Growth kinetics of wildlife E. coliisolates in soil and water. J. Environ. Protec 3, 838e846.

Guber, A.K., Williams, D.M., Dechen Quinn, A.C., Tamrakar, S.B., Rose, J.B.,Porter, W.F., 2014. Swat-sir model for predicting fate and transport of manure-borne pathogens in fragmented Agriculture-Forest ecosystems. In: Ames, D.P.,Quinn, N.W.T., Rizzoli, A.E. (Eds.), Proceedings of the 7th International Congresson Environmental Modelling and Software, June 15e19, San Diego, California,USA, ISBN 978-88-9035-744-2. In: http://www.iemss.org/society/index.php/iemss-2014-proceedings.

Guber, A.K., Fry, J., Ives, R.L., Rose, J.B., 2015. Escherichia coli survival in, and releasefrom, white-tailed deer feces. Appl. Environ. Microbiol. 81 http://dx.doi.org/10.1128/AEM.03295-14, 000 e000.

Haas, C.N., Thayyar-Madabusi, A., Rose, J.B., Gerba, C.P., 2000. Development of adoseeresponse relationship for Escherichia coli O157:H7. Inter. J. Food Micro-biol. 57, 153e159.

Hammonds, J.S., Hoffman, F.O., Bartell, S.M., 1994. An Introductory Guide to Un-certainty Analysis in Environmental and Health Risk Assessment. Report ES/ER/TM-35/R1 prepared for OAK RIDGE NATIONAL LABORATORY. Oak Ridge bySENES Oak Ridge, Inc., Oak Ridge, Tennessee, 47 pp.

Harmel, R.D., Karthikeyan, R., Gentry, T., Srinivasan, R., 2010. Effects of agriculturalmanagement, land use, and watershed scale on E. coli concentrations in runoffand streamflow. Trans. ASABE 53, 1833e1841.

Hosseini, P.R., Dhondt, A.A., Dobson, A., 2004. Seasonality and wildlife disease: howseasonal birth, aggregation and variation in immunity affect the dynamics ofMycoplasma gallisepticum in house finches. Proc. R. Soc. Lond. B 271,2669e2577.

Hovde, C.J., Austin, P.R., Cloud, K.A., Williams, C.J., Hunt, C.W., 1999. Effect of cattlediet on Escherichia coli O157:H7 acid resistance. Appl. Environ. Microbiol. 65,3233e3235.

Ishii, S., Hansen, D., Hicks, R., Sadowsky, M., 2007. Beach sand and sediments aretemporal sinks and sources of Escherichia coli in Lake Superior. Environ. Sci.Technol. 41, 2203e2209.

Iudicello, J., Chin, D., 2013. Multimodel, multiple watershed examination of in-stream bacteria modeling. J. Environ. Eng. 139, 719e727.

Jackson, S.G., Goodbrand, R.B., Johnson, R.P., Odorico, V.G., Alves, D., Rahn, K.,Wilson, J.B., Welch, M.K., Khakhria, R., 1998. Escherichia coli O157:H7 diarrhoeaassociated with well water and infected cattle on an Ontario farm. Epidemiol.Infect. 120, 17e20.

Jakeman, A.J., Letcher, R.A., Norton, J.P., 2006. Ten iterative steps in developmentand evaluation of environmental models. Environ. Model. Softw. 21, 602e614.

Jayakody, P., Parajuli, P.B., Brooks, J.P., 2013. Evaluating spatial and temporal vari-ability of fecal coliform bacteria loads at the Pelahatchie watershed in Mis-sissippi. Human Ecolog. Risk Assess. 20, 1023e1041. http://dx.doi.org/10.1080/10807039.2013.784155.

Johnson, R.P., Cray Jr., W.C., Johnson, S.T., 1996. Serum antibody responses of cattlefollowing experimental infection with Escherichia coli O157:H7. Infect. Immun.64, 1879e1883.

Jones, D.L., 1999. Potential health risks associated with the persistence of Escher-ichia coli O157 in agricultural environments. Soil Use Manag. 15, 76e83.

Keeling, M.J., Rohani, P., 2008. Modeling Infectious Diseases in Humans and Ani-mals. Princeton University Press, Princeton, New Jersey, 368.

Khaitsa, M.L., Smith, D.R., Stoner, J.A., Parkhurst, A.M., Hinkley, S., Klopfenstein, T.J.,Moxley, R.A., 2003. Incidence, duration, and prevalence of Escherichia coli O157:H7 fecal shedding by feedlot cattle during the finishing period. J. Food Prot. 66,1972e1977.

Kim, J.-W., Pachepsky, Y.A., Shelton, D.R., Coppock, C., 2010. Effect of streambedbacteria release on E. coli concentrations: monitoring and modeling with themodified SWAT. Ecol. Model 221, 1592e1604.

Laidler, M.R., Tourdjman, M., Buser, G.L., Hostetler, T., Repp, K.K., Leman, R.,

Samadpour, M., Keene, W.E., 2013. Escherichia coli O157:H7 infections associ-ated with consumption of locally grown strawberries contaminated by deer.Clin. Infect. Dis. 57, 1129e1134.

Lautier, J.K., Dailey, T.V., Brown, R.D., 1988. Effect of water restriction on feed intakeof white-tailed deer. J. Wildl. Manag. 52, 602e606.

LeJeune, J.T., Besser, T.E., Hancock, D.D., 2001. Cattle water Microtroughs as reser-voirs of Escherichia coli O157. Appl. Environ. Microbiol. 67, 3053e3057.

Leta, O.T., Nossent, J., Velez, C., Shrestha, N.K., van Griensven, A., Bauwens, W., 2015.Assessment of the different sources of uncertainty in a SWAT model of the RiverSenne (Belgium). Environ. Model. Softw. 68, 129e146.

Manly, B.F.J., McDonald, L.L., Thomas, D.L., McDonald, T.L., Erickson, W.P., 2002.Resource Selection by Animals: Statistical Analysis and Design for Field Studies,second ed. Kluwer Academic Publishers, Boston, MA.

Matott, L.S., Babendreier, J.E., Purucker, S.T., 2009. Evaluating uncertainty in inte-grated environmental models: a review of concepts and tools. Water Resour.Res. 45, W06421.

McGee, P., Scott, L., Sheridan, J.J., Earley, B., Leonard, N., 2004. Horizontal trans-mission of Escherichia coli O157:H7 during cattle housing. J. Food Protec 67,2651e2656.

Merrill, R.M., 2009. Introduction to Epidemiology, fifth ed. Jones and Bartlett Pub-lishers, Sudbury, p. 417. MA 01776.

Miller, F.L., 1971. Mutual grooming by black-tailed deer in northwestern Oregon.Can. Field Nat. 85, 295e301.

Moen, A.N., 1978. Seasonal changes in heart rates, activity, metabolism, and forageintake of white-tailed deer. J. Wildl. Manag. 42, 715e773.

Moyer, D.L., Hyer, K.E., 2003. Use of the Hydrological Simulation Program-fortranand Bacterial Source Tracking for Development of the Fecal Coliform TotalMaximum Daily Load (TMDL) for Accotink Creek, Fairfax County, Virginia: U.S.Geological Survey Water-Resources Investigations Report 03e4160.

Neitsch, S.L., Arnold, J.G., Kiniry, J.R., Williams, J.R., 2005. Soil and Water AssessmentTool Theoretical Documentation, Version 2005. Temple, Tex.: USDA-ARSGrassland, Soil and Water Research Laboratory. Available at: http://swat.tamu.edu/ (accessed 3.10.2015).

Ørskov, F., Ørskov, I., Villar, J.A., 1987. Cattle as reservoir of verotoxinproducingEscherichia coli O157:H7. Lancet ii, 276.

Parajuli, P., 2007. SWAT Bacteria Sub-model Evaluation and Application. Ph.D.thesis. Kansas State University, Manhattan, KS.

Parajuli, P.B., Mankin, K.R., Barnes, P.L., 2009. Source specific fecal bacteria modelingusing soil and water assessment tool model. Biores. Tech. 100, 953e963. http://dx.doi.org/10.1016/j.biortech.2008.06.045.

Park, Y., Pachepsky, Y.A., Jeong, J., 2014. Extending Apex to model microorganismfate and transport. Abstract 249e7. In: ASA, CSSA, SSSA Inter. Ann. Meet., LongBeach, CA, Nov. 2-5, 2014.

Pekins, P.J., Smith, K.S., Mautz, W.W., 1998. The energy cost of gestation in white-tailed deer. Can. J. Zoology 76, 1091e1097.

Redfearn D.D. and T.G. Bidwell. Stocking rate: the key to successful livestock pro-duction. Oklahoma Cooperative Extension Service PSS-2871. Retrieved October7, 2015, from http://osufacts.okstate.edu.

Refsgaard, J.C., van der Sluijs, J.P., Højberg, A.L., Vanrolleghem, P.A., 2007. Uncer-tainty in the environmental modelling process A framework and guidance.Environ. Model. Softw. 22, 1543e1556.

Renter, D.G., Sargeant, J.M., Hygnstrom, S.E., Hoffmann, J.D., Gillespie, J.R., 2001.Escherichia coli O157:H7 in free-ranging deer in Nebraska. J. Wildl. Dis. 37,755e760.

Rice, D.H., Hancock, D.D., Besser, T.E., 1995. Verotoxigenic E. coli O157 colonisationof wild deer and range cattle. Vet. Rec. 137, 524.

Rogers, L.L., 1987. Seasonal changes in defecation rates of free-ranging white-taileddeer. J. Wildl. Manag. 51, 330e333.

Rosenberry, C.S., Tardiff Fleegle, J., Wallingford, B.D., 2011. Monitoring Deer Pop-ulations in Pennsylvania. Deer and elk section, Bureau of Wildlife Management,Pennsylvania Game Commission, Elmerton Avenue Harrisburg, Pennsylvania17110. Retrieved December 10, 2015, from Pennsylvania Game Commission.http://www.pgc.state.pa.us.

Sadeghi, A., Arnold, J.G., 2002. A SWAT/microbial submodel for predicting pathogenloadings in surface and groundwater at watershed and basin scales. In: TotalMaximum Daily Load (TMDL) Environmental Regulations, Proceedings of theConference, Fort Worth, TX, USA. March 11e13, pp. 56e63.

Sanderson, M.W., Besser, T.E., Gay, J.M., Gay, C.C., Hancock, D.D., 1999. FecalEscherichia coli O157:H7 shedding patterns of orally inoculated calves. Vet.Microbiol. 69, 199e205.

Sargeant, J.M., Hafer, D.J., Gillespie, J.R., Oberst, R.D., Flood, S.J., 1999. Prevalence ofEscherichia coli O157:H7 in white-tailed deer sharing rangeland with cattle.J. Am. Vet. Med. Assoc. 215, 792e794.

Setegn, S.G., Srinivasan, R., Melesse, A.M., Dargahi, B., 2010. SWAT model applicationand prediction uncertainty analysis in the Lake Tana Basin, Ethiopia. Hydrol.Process 24, 357e367.

Shen, Z.Y., Chen, L., Chen, T., 2012. Analysis of parameter uncertainty in hydrologicaland sediment modeling using GLUE method: a case study of SWAT modelapplied to three Gorges Reservoir Region, China. Hydrol. Earth Syst. Sci. 16,121e132. http://dx.doi.org/10.5194/hess-16-121-2012.

Shere, J.A., Kaspar, C.W., Bartlett, K.J., Linden, S.E., Norell, B., Francey, S.,Schaefer, D.M., 2002. Shedding of Escherichia coli O157:H7 in dairy cattlehoused in a confined environment following waterborne inoculation. Appl.Environ. Microbiol. 68, 1947e1954.

Singh, A., Imtiyaz, M., Isaac, R.K., Denis, D.M., 2014. Assessing the performance and

A.K. Guber et al. / Environmental Modelling & Software 80 (2016) 185e200200

uncertainty analysis of the SWAT and RBNN models for simulation of sedimentyield in the Nagwa watershed, India. Hydrolog. Sci. J. 59 (2), 351e364.

Soupir, M.L., Mostaghimi, S., Yagow, E.R., Hagedorn, C., Vaughan, D.H., 2006.Transport of fecal bacteria from poultry litter and cattle manures applied topastureland. Water Air Soil Pollut. 169, 125e136.

STATSGO, 1994. State Soil Geographic Data Base. U.S. Department of Agriculture,NRCS, NSSC, Miscellaneous Publication Number 1492.

Swiler, L.P., Wyss, G.D., 2004. A User's Guide to Sandia's Latin Hypercube SamplingSoftware: LHS UNIX Library/Standalone Version. Sandia National Laboratories.Report SAND2004e2439. Retrieved 10 December 2015, from. http://www.pesthomepage.org/Latin_Hypercube.php.

Talebizadeh, M., Morid, S., Ayyoubzadeh, S.A., Ghasemzadeh, M., 2010. Uncertaintyanalysis in sediment load modeling using ANN and SWAT model. Water Resour.Manag. 24, 1747e1761. http://dx.doi.org/10.1007/s11269-009-9522-2.

Therneau, T., Lumley, T., 2009. Survival: Survival Analysis, Including PenalizedLikelihood. R Package, Version 2.35. R Foundation for Statistical Computing,Vienna.

Thyer, M., Renard, B., Kavetski, D., Kuczera, G., Franks, S.W., Srikanthan, S., 2009.Critical evaluation of parameter consistency and predictive uncertainty in hy-drological modeling: a case study using Bayesian total error analysis. WaterResour. Res. 45. W00B14. http://dx.doi.org/10.1029/2008WR006825.

Ullrey, D.E., Youatt, W.G., Johnson, H.E., Fay, L.D., Brent, B.E., Kemp, K.E., 1968. Di-gestibility of cedar and balsam fir browse for the white-tailed deer. J. Wildl.Manag. 32, 162e171.

Ullrey, D.E., Youatt, W.G., Johnson, H.E., Fay, L.D., Schoepke, B.L., Magee, W.T., 1970.Digestible and metabolizable energy requirements for winter maintenance ofMichigan white-tailed does. J. Wildl. Manag. 34, 863e869.

US-HHS, 2006. Principles of Epidemiology in Public Health Practice. An Introductionto Applied Epidemiology and Biostatistics. U.S. Department of Health and Hu-man Services, Centers for Disease Control and Prevention (CDC), Office ofWorkforce and Career Development, Atlanta, GA 30333, October 2006, Updated

May 2012 (Retrieved from, third ed.. http://www.cdc.gov/ophss/csels/dsepd/SS1978/SS1978.pdf.

U.S. EPA, 2014. Specific State Causes of Impairment that Make up the NationalPathogens Cause of Impairment Group. US Environmental Protection Agency,Washington DC. Retrieved 20 September 2014, from US Environmental Pro-tection Agency. http://iaspub.epa.gov/tmdl_waters10/attains_nation_cy.cause_detail_303d?p_cause_group_id¼861.

Van Etten, R.C., 1959. Development and Evaluation of New Deer Census Techniques.Michigan Dept. Conserv. Job Completion Rept., P.-R. Project W-70-R; Sub-job A-l-b. 8pp. Typewritten.

Vrugt, J.A., Gupta, H.V., Bouten, W., Sorooshian, S., 2003. A Shuffled Complex Evo-lution Metropolis algorithm for optimization and uncertainty assessment ofhydrologic model parameters. Water Resour. Res. 39, 105e112. http://dx.doi.org/10.1029/2002WR001642.

Whelan, G., Kim, K., Pelton, M.A., Castleton, K.J., Laniak, G.F., Wolfe, K., Parmar, R.,Babendreier, J., Galvin, M., 2014. Design of a component-based integratedenvironmental modelling Framework. Environ. Model. Soft 55, 1e24.

Yagow, G., Dillaha, T., Mostaghimi, S., Brannan, K., Heatwole, C., Wolfe, M.L., 2001.TMDL Modeling of Fecal Coliform Bacteria with HSPF. ASAE Meeting Paper No.01e2066. Mich, St. Joseph.

Zeckoski, R.W., Benham, B.L., Shah, S.B., Wolfe, M.L., Brannan, K.M., Al-Smadi, M.,Dillaha, T.A., Mostaghimi, S., Heatwole, C.D., 2005. BSLC: a tool for bacteriasource characterization for watershed management. Appl. Eng. Agric. 21,879e889.

Zhang, X., Srinivasan, R., Bosch, D., 2009. Calibration and uncertainty analysis of theSWAT model using genetic algorithms and Bayesian model averaging. Hydrol-ogy 374, 307e317.

Zhao, T., Doyle, M.P., Shere, J., Garber, L., 1995. Prevalence of enterohemorrhagicEscherichia coli O157:H7 in a survey of dairy herds. Appl. Environ. Microbiol. 61,1290e1293.