Next Article in Journal
Depth-Specific Soil Electrical Conductivity and NDVI Elucidate Salinity Effects on Crop Development in Reclaimed Marsh Soils
Previous Article in Journal
Comparison between Physical and Empirical Methods for Simulating Surface Brightness Temperature Time Series
 
 
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Spatiotemporal Modeling of Zoonotic Arbovirus Transmission in Northeastern Florida Using Sentinel Chicken Surveillance and Earth Observation Data

1
Florida Medical Entomology Laboratory, Department of Entomology & Nematology, IFAS, University of Florida, 200 9th St SE, Vero Beach, FL 32962, USA
2
Department of Natural History, Florida Museum of Natural History, Dickinson Hall, University of Florida, Gainesville, FL 32611, USA
3
Preventive Medicine and Biostatistics Department, Uniformed Service University of the Health Sciences, Bethesda, MD 20814, USA
4
Department of Biology, University of Nevada, 1664 N. Virginia Ave., Reno, NV 89557, USA
5
Vector Disease Control International, 1320 Brookwood Drive, Suite H, Little Rock, AR 72202, USA
6
Volusia County Mosquito Control, 801 South Street, New Smyrna Beach, FL 32168, USA
7
City of Jacksonville, Jacksonville Mosquito Control, Jacksonville, FL 32218, USA
8
Anastasia Mosquito Control, St. Augustine, FL 32092, USA
*
Author to whom correspondence should be addressed.
Remote Sens. 2022, 14(14), 3388; https://doi.org/10.3390/rs14143388
Submission received: 12 May 2022 / Revised: 2 July 2022 / Accepted: 8 July 2022 / Published: 14 July 2022

Abstract

:
The irregular timing and spatial variation in the zoonotic arbovirus spillover from vertebrate hosts to humans and livestock present challenges to predicting spillover occurrence over time and across broader geographic areas, compromising effective prevention and control strategies. The objective of this study was to quantify the effects of the landscape composition and configuration and dynamic weather events on the 2018 spatiotemporal distribution of eastern equine encephalitis virus (EEEV) (Togaviridae, Alphavirus) and West Nile virus (WNV) (Flaviviridae, Flavivirus) sentinel chicken seroconversion in northeastern Florida. We used a modeling framework that explicitly accounts for joint spatial and temporal effects and incorporates key EO (Earth Observation) information on the climate and landscape in order to more accurately quantify the environmental effects on the transmission to sentinel chickens. We investigated the environmental effects using Bernoulli generalized linear mixed effects models (GLMMs), including a site-level random effect, and then added spatial random effects and spatiotemporal random effects in subsequent runs. The models were executed using an integrated nested Laplace approximation (INLA) and a stochastic partial differential equation (SPDE) approach in R-INLA. The GLMMs that included a spatiotemporal random effect performed better relative to models that included only spatial random effects and also performed better than non-spatial models. The results indicated a strong spatiotemporal structure in the seroconversion for both viruses, but EEEV exhibited a more punctuated and compact structure at the beginning of the sampling season, while WNV exhibited a more gradual and diffuse structure across the study area toward the end of the sampling season. The percentage of cypress–tupelo wetland land cover within 3500 m of coop sites and the edge density of the forest land cover within 500 m had a strong positive effect on the EEEV seroconversion, while the best fitting model for WNV was the intercept-only model with spatiotemporal random effects. The lagged climatic variables included in our study did not have a strong effect on the seroconversion for either virus when accounting for temporal autocorrelation, demonstrating the utility of capturing this structure to avoid type I errors. The predictive accuracy for out-of-sample data for the EEEV seroconversion demonstrates the potential to develop a framework that incorporates temporal dynamics in order to better predict arbovirus transmission.

Graphical Abstract

1. Introduction

Zoonotic vector-borne pathogens are a leading cause of morbidity and mortality in humans and animals across the globe [1,2]. The irregular timing and spatial variation in the zoonotic arbovirus spillover from vertebrate host species to humans and livestock present challenges to developing effective prevention and control strategies and to developing a basis for predictive modeling. One key advance that may be particularly valuable for predicting zoonotic arbovirus spillover is connecting dense vector and host monitoring data with the widespread availability of multi-scale Earth Observation (EO) environmental data [3,4]. However, the best use of EO data to make linkages to potential spillover has been challenging given the complexity of modeling the spatiotemporal dynamics of these pathogen systems [5]. The advances in ecological statistics capable of capturing spatial and temporal covariance and the computational capacity to fit such models [5] provides new opportunities to investigate environmental correlations with zoonotic arbovirus transmission, with the potential to make effective risk predictions and even forecast risk based on future scenarios.
Despite maturity in the tools being needed to understand the underlying environmental drivers of the spatiotemporal dynamics of zoonotic disease risk, the empirical applications are still limited, and the modeling frameworks are not fully mature. Here, we utilize a robust monitoring data set from efforts by the Florida Mosquito Control Districts or Programs and the Florida Department of Health (FDOH) to investigate the spatiotemporal environmental correlations with two zoonotic arboviruses of major human and veterinary health importance in the United States—eastern equine encephalitis virus (EEEV) (Togaviridae, Alphavirus) and West Nile virus (WNV) (Flaviviridae, Flavivirus) [6].
Each virus is maintained in an enzootic cycle primarily between mosquito vectors and passerine-amplifying hosts with occasional epizootic transmission resulting in incidental infections, or spillover events, in humans, equines, and other animals [7,8]. Although the symptoms in humans are relatively rare, the implications can be severe. Approximately 5% of persons infected with EEEV develop neuroinvasive disease, which can result in lasting neurological symptoms and has a mortality rate of ~30% (CDC 2019), and approximately 1 in 150 people infected with WNV develop neuroinvasive disease with a mortality rate of approximately 10%. The impacts on unvaccinated equines from EEEV and WNV are also severe in animals that develop neuroinvasive disease, with fatality rates of 50–70% [9] and 30–40% [10], respectively. These factors, combined with the recent geographic expansion of EEEV in the northeastern U.S. [11] and continued epidemics of WNV in multiple regions, demonstrate the need to use methods that can leverage dynamic Earth Observation (EO) data to quantify and predict transmission across space and time to help inform vector control and public health efforts.
While both EEEV and WNV are amplified in mosquito–songbird transmission cycles, the mosquito species that are important for vectors for these viruses differ substantially. The primary enzootic mosquito vector of EEEV is Culiseta melanura, which breeds in and around freshwater hardwood swamp environments in the Atlantic and Gulf Coast states and the Great Lakes region, and the principal arthropod bridge vectors for transmission to humans or to horses are Aedes, Coquillettidia, and Culex species [7]. West Nile virus is maintained primarily between Culex mosquitoes [8,12] and passerine birds, with multiple primary vectors that vary in distribution by region, habitat, and season [12]. In Florida, Culex quinquefasciatus and Culex nigripalpus are considered primary vectors of WNV, with habitats ranging from irrigated agriculture fields to polluted artificial containers in urban areas [13], and multiple bridge vectors competent to transmit WNV to humans and horses are distributed throughout the state [12].
In Florida, EEEV and WNV can exhibit year-round transmission [14]. Multiple studies focused on Florida have incorporated RS environmental data to investigate landscape correlations with EEEV and WNV, including the development of a risk index based on EEEV equine cases and the surrounding habitats [15,16], the generation of ecological niche models investigating mosquito vectors and sentinel chicken seroconversion at local to state-wide scales [17,18], and habitat identification for sentinel chicken placement [19]. In addition, the seasonality of EEEV sentinel chicken seroconversion was quantified using wavelet analyses, with the results indicating annual periodicity in seroconversion across Florida and within northern and north central regions [20]. Other work on WNV transmission patterns in humans and sentinel chickens found that increased drought conditions followed by wet conditions was correlated with increased human cases and sentinel chicken seroconversion [21]. More recently, researchers found that precipitation, cooling degree days, and El Niño Southern Oscillation Index values were correlated with monthly EEEV equine case counts in the state using distributed lagged non-linear models [22]. These studies all show that climatic and environmental factors likely impact the spillover incidence, but they also point to the challenges of accounting for the spatiotemporal structuring and dynamics of both observations and covariates to make accurate and precise predictions. A more complete understanding of the environmental drivers of spillover is likely to require the direct incorporation of the underlying spatiotemporal structure into modeling frameworks, rather than treating the spatial and temporal aspects as fully independent.
The objective of this study is to quantify the effects of the landscape composition and configuration and dynamic temperature and precipitation values on the 2018 spatiotemporal distribution of EEEV and WNV sentinel chicken seroconversion in Duval, St. Johns, and Volusia Counties in northeastern Florida. We chose 2018 as a focal year, because multiple human and equine cases of EEEV and WNV occurred in northeastern Florida, with 1 EEEV human case, 14 WNV human cases, 4 EEEV equine cases, and an EEEV-infected emu flock [23]. Unlike previous approaches, we developed a framework that allows a direct way to model spatial and temporal effects, as well as their interactions, which promises to more fully capture the underlying system dynamics. We expected to find higher EEEV seroconversion rates in chicken coops surrounded with greater percentages and higher edge densities of cypress–tupelo or forested areas associated with the known habitat of Culiseta melanura mosquitoes. We also expected to find that chicken coops surrounded with greater percentages and higher edge densities of semi-urban landscapes will have a strong positive effect on WNV sentinel chicken seroconversion rates. With regards to the seasonal climate factors, we expected to find positive effects of lagged temperature values on sentinel chicken seroconversion rates for both viruses. The precipitation patterns may have an even stronger effect given the results from previous studies, with a positive effect of lagged precipitation values on EEEV seroconversion rates, a positive effect of lagged precipitation values near the date of sampling on WNV seroconversion rates, and a negative effect of cumulative precipitation values at distance lags further from the sampling date [21].

2. Materials and Methods

Duval, St. Johns, and Volusia Counties span the northeastern Atlantic Coast of Florida (Figure 1). The visualization of the land cover in the study area suggests that the developed land cover is distributed along the coastal areas of each county and in a large area in western Volusia County and across most of Duval County (City of Jacksonville); the rural areas are predominantly forest and wetland land cover, interspersed with pastures and cropland [24]. The region’s climate is categorized as humid subtropical [25]. The average temperatures increase with decreasing latitude, with average annual maximum temperatures in the range of 25 °C to 29 °C and average annual minimum temperatures in the range of 13 °C to 17 °C. The majority of annual precipitation occurs during the rainy season, typically from May to October, with annual averages ranging from approximately 1150 mm to 1400 mm [26].
Enzootic EEEV and WNV transmission in Florida is detected throughout the early spring to late autumn through an extensive sentinel chicken program, with testing conducted by the Florida Department of Health and participating mosquito control districts across the state, providing site-level and temporally dynamic information about zoonotic pathogen transmission in each area [27,28]. As part of this effort, dozens of districts in Florida maintain sentinel chicken flocks, which are tested weekly for flavivirus and alphavirus seroconversion, including WNV and EEEV.
Georeferenced 2018 weekly sentinel chicken seroconversion data were cleaned, formatted, and collated across the City of Jacksonville Mosquito Control (COJMC) in Duval County, Anastasia Mosquito Control District (AMCD) in St. Johns County, and Volusia Mosquito Control (VMC) districts in northeastern Florida. The data were checked for completeness, including gaps in surveillance across districts and at individual chicken coops. We considered a flock positive for EEEV and a flock positive for WNV if at least one chicken in the flock tested positive for a given week, and we coded the chicken coop as 1 (present) or 0 (absent) for the virus for that week. The sentinel chicken 2018 seroconversion rates varied across the three counties (Figure 1). The data set included 27 sentinel chicken coop locations, sampled for 28 consecutive weeks between May 1st and November 12th, for a total of 763 observations. A total of 35 sentinel chicken coops tested positive for EEEV and 58 coops tested positive for WNV over the 2018 sampling season.

2.1. Land Cover Data

The September 2018 Cooperative Land Cover data acquired from the Florida Fish and Wildlife Conservation Commission served as the land cover data for this project [24]. The land cover data set is available at a 10 m spatial resolution. We chose to aggregate the classes, given our relatively small study area and limited number of sampling sites because of the large number of fine-scale classes present in the data set, and included a total of 11 land cover classes in our analyses (Figure 2).
Landscape metrics quantify the composition and configuration of the landscape or specific land cover type in an area [29]. We chose two landscape metrics for our analyses: the percent land cover type for each of our 11 land cover classes within 500 m, 1000 m, 2000 m, 3500 m, and 5000 m buffer distances from a sentinel chicken site, and we calculated the edge density values for the 11 land cover classes within these buffer distances from a sentinel chicken site. The percent land cover metric defines the overall presence of a specific land cover type, while the edge density metric provides information about potential fragmentation or edge habitats in an area [30]. Shapefiles for buffers calculated at each distance were generated in the ESRI ArcMap v10.6 software program Redlands, CA, USA, and the percent land cover and edge density metrics were calculated using the ‘landscape metrics’ package in R v3.5.1 [31,32]. The resulting landscape metric values served as predictor variables in the spatiotemporal generalized linear mixed effects model (GLMM) analyses presented below.
The compound topographic wetness index values provide an indication of whether water may pool in a specific geographic area during a precipitation event. The wetness index values are derived from digital elevation models, incorporating the slope of the terrain and a measure of the flow direction and accumulation calculated across the study area [33]. We downloaded 30-m-resolution ASTER GDEM data from the NASA Earth Data Repository and calculated a compound topographic wetness index using functions available in the ArcMap v10.6 software program. The wetness index values were extracted within a 1 km buffer distance of georeferenced sentinel chicken sites, and the mean, maximum, and minimum wetness index values were identified using the ‘Zonal Statistics’ function in ArcMap.

2.2. Climate Data

National Oceanic and Atmospheric Administration (NOAA) NCEP Stage 4 Rainfall 2018 daily data were downloaded for the study area [34]. The data are at a 4 km spatial resolution and were formatted and reprojected to Geographic WGS 84 using the ‘Model Builder’ application in ArcMap v10.6. Daily precipitation values at georeferenced sentinel chicken sites were extracted before calculating the weekly cumulative precipitation values at 9 weekly time lags (t − 1 week, t − 2 weeks, …, t − 9 weeks) prior to the sentinel chicken sampling date. Previous investigations have suggested that dry periods followed by inundation events may contribute to increased transmission [21], and calculating the weekly cumulative precipitation at multiple lag distances provided the opportunity to characterize variability in precipitation values at multiple time periods prior to sentinel chicken sampling. Moderate Resolution Imaging Spectroradiometer (MODIS) Terra v6 Land Surface Temperature (LST) 8-day mean composites [35] were downloaded at a 1 km spatial resolution from the NASA Earth Data repository (https://earthdata.nasa.gov/ accessed on 1 July 2020). The data were formatted, mosaicked, and reprojected using the MODIS Reprojection Tool, before masking to the study area. The resulting files were filled temporally for missing values, before converting the values from Kelvin to Celsius. The land surface temperature values were extracted to georeferenced sentinel chicken sites, and the temperature lags (1–4 weeks) prior to the sentinel chicken sampling served as predictor variables. The weekly lags 1–4 weeks prior to chicken sampling provided an opportunity to capture the variability in temperature values across mosquito development periods, virus extrinsic incubation periods, and host intrinsic incubation periods [36].

2.3. Variable Reduction

We performed a conditional random forest approach executed in the ‘caTools’ package in R to identify the variable importance for EEEV and WNV seroconversions with the objective of variable reduction prior to GLMM implementation [37]. Two sets of conditional random forests, one for EEEV and one for WNV, were run with the weekly 2018 virus presence–absence serving as the response variable and all landscape metrics at all buffer distances, wetness index values, temperature lags, and precipitation lags serving as the predictor variables. We ranked the predictor variables using variable importance values and identified the top six ranking variables for each virus response variable to include in the GLMM analyses.

2.4. Candidate Sets and Model Runs

We calculated a Pearson’s correlation matrix to identify variables that were highly correlated with one another (±0.6) to construct environmental variable sets for candidate models. We used a Bernoulli GLMM with a logit link function and individual sentinel chicken coops serving as a site-level random effect [38]. The parameters were estimated using the maximum likelihood method. The covariates were standardized using a z-score calculation prior to model execution because of the varying units across individual variables. The initial models were executed in the ‘glmmTMB’ package in R [39], and we implemented the ‘dredge’ function in the MuMIN package to investigate all combinations of variables [40]. We used a model selection approach for evaluation, ranking models from lowest to highest Akaike’s Information Criterion (AIC) scores [41] and calculated AIC weights, and identified the set of models comprising 95% of the cumulative sum of AIC weights as our confidence set for each virus [42].

2.5. Bayesian Modeling

We reran the models comprising the 95% confidence set using an integrated nested Laplace approximation (INLA) in the ‘INLA’ package in R [43] to ensure consistency in the parameter estimates with the glmmTMB parameter estimates, and then subsequently added a spatial random effects term and then spatiotemporal random effects for a total of three sets of models for each virus. INLA is a computationally efficient approach for approximate Bayesian inference, where the posterior marginals of individual parameters (e.g., mean and precision) are approximated rather than the joint posterior distribution of the model parameters [44,45]. The spatial random effects were fit using a Gaussian Markov random field using a stochastic partial differential equation (SPDE) approach modeled using a Matérn covariance function [46], and we fit the temporal random effect using a first-order autoregressive term (AR1). The SPDE was constructed using constrained Delauney triangulation and we created a convex hull based on the sentinel chicken coop locations to define the extent of the triangulation mesh (Supplementary Figure S1). All models were run with default priors, which was a Gaussian prior with a mean of 0 and precision of 0 for the intercept and a Gaussian prior with a mean of 0 and precision of 0.001 for the regression parameters. The SPDE was assigned a multivariate normal prior with a mean of 0 and penalized complexity priors [47] were estimated for the range and variance, with the probability of values being lower than 10,000 for the range set to 0.5 and the probability of the variance being lower than 0.5 set to 0.5.
We ranked the models from lowest to highest widely applicable information criterion (WAIC) and deviance information criterion (DIC) values [48] to identify the best model or set of models, and we observed credible interval values for variables included in the top-ranking models to determine whether the data supported a predictor variable having a strong effect on EEEV or WNV sentinel chicken seroconversion rates. A random sample of 30% (n = 219) of the seroconversion data was then withheld and the best-performing model was rerun with the remaining 70% (n = 534) of the data for out-of-sample prediction. The predictive performance was evaluated by calculating the area under the curve (AUC) of the receiver operating characteristic (ROC) in the ‘cutpointr’ package in R [49].
An important component in our models was the estimation of the Gaussian Markov random fields (GMRF) that accounted for the spatial and spatiotemporal structure in our observations [46], and we plotted these values across the study area to provide a visualization for the spatiotemporal structure for each week and for each virus during the 2018 sampling season (28 weeks total for each virus). The inclusion of spatial or spatiotemporal terms in the modeling process provides information about the spatiotemporal structure of the observations and can provide clues about the environmental variables that may be contributing to an observed process. Additionally, the inclusion of these terms can help to reduce type I errors, owing to the residual dependence in space and time between observations in a data set [5,50].

3. Results

The conditional random forest variable importance values identified a combination of forest and cypress–tupelo wetland land cover and edge densities as top-ranking variables for EEEV, along with the weekly cumulative precipitation rates at a lag of 1 week and 9 weeks. For WNV, the top six variables included a combination of wetland edge density values at multiple buffer distances and cumulative precipitation values at 1, 3, and 7 week lags. The wetness index values representing the potential for water pooling across the landscape and the mean composites of daily temperature values at multiple lag times were not included as candidate variables.
The initial GLMMs using all combinations of candidate model sets using the top 6 variables from the conditional random forest output resulted in 34 models for each virus, and each virus model set included 8 models that comprised the 95% confidence set based on the sum of the cumulative AIC weights (Supplementary Tables S1 and S2). For EEEV, these 8 models included the percent forest values within 3500 m and 5000 m, edge density of the forest within 500 m, cumulative precipitation lags at 1 and 9 weeks, and percent cypress–tupelo wetland values within 3500 m. For WNV, these 8 models included the edge density values of wetlands within 2000 m, 3500 m, and 5000 m distances and cumulative precipitation rates at 1, 3, and 7 week time lags.
The WAIC and DIC results from non-spatial, spatial, and spatiotemporal model runs in the R-INLA environment indicated that all spatiotemporal models performed better than the spatial and non-spatial models for EEEV and WNV (Table 1 and Table 2). For EEEV, the WAIC and DIC scores ranked the spatiotemporal model that included the forest edge density within 500 m, cypress–tupelo wetland land cover within 3500 m of sentinel chicken sites, and weekly cumulative precipitation at 9 weeks prior to chicken sampling as the “best” model (Table 3).
The WAIC and DIC scores ranked the spatiotemporal intercept-only model as the “best” model for WNV (Table 2).
The credible intervals designated between 0.025 and 0.975 indicate that there is a 95% probability that the parameter estimate is within the interval [45]. The credible intervals for variables included in models in the EEEV 95% confidence set indicated that the forest edge density within 500 m of sentinel chicken sites, percent forest within 5000 m, and percent cypress–tupelo wetland coverage within 3500 m had a positive effect on EEEV seroconversion in the study area (Table 3 and Figure 3). The credible intervals indicated that precipitation at lag periods of 1 and 9 weeks did not have a strong effect on EEEV seroconversion when accounting for the spatiotemporal structure, despite these variables being contained within the 95% confidence sets of the non-spatial and spatial model runs (Supplementary Table S3). The results for non-spatial, spatial, and spatiotemporal EEEV INLA models are available in Supplementary Table S3. The best model results including credible intervals for random effects are available in Supplementary Table S4, and the density curves for regression parameters and random effects are available in Supplementary Figure S2.
The area under the curve (AUC) of the ROC value for the best EEEV model was 0.948, indicating the potential for accurate prediction (Figure 4). An optimal cutpoint value of 0.157 identified using the ‘minimize_metric function’ in ‘cutpointR’ indicated that the model was more accurate at predicting true negative values than true positive values, with 2 false negatives and 14 false positives out of 229 records in total.
The credible intervals for variables included in the WNV 95% confidence set indicated that none of the variables included in the models had a strong effect on the WNV seroconversion after accounting for the spatiotemporal structure (Table 4). The results for non-spatial, spatial, and spatiotemporal WNV INLA models are available in Supplementary Table S5.
The weekly plots of GMRFs demonstrated variation in the spatiotemporal structures for EEEV and WNV across the 28-week study period (Figure 5 and Figure 6). The weekly GMRFs for EEEV showed a strong spatiotemporal structure at the beginning of the study period, with the strongest areas located in St. Johns County and in the southern-central portion of Volusia County (Figure 5). During the week of July 3rd, the spatiotemporal structure began to weaken and dissipate across the remainder of the study period before a slight negative spatiotemporal structure developed in the study area.
The plots of weekly GMRFs for WNV seroconversion (Figure 6) indicated a strong spatiotemporal structure toward the end of the sampling season, moving from a northwestern to southeastern direction. Starting mid-July, a gradual spatiotemporal structure began in the northwestern corner of the study area, building in intensity and expanding from the northwest in a southwestern direction, before encompassing the entire study area in the week of November 6th.

4. Discussion

This study provides new insight into the landscape and climate factors most explanatory of EEEV and WNV sentinel chicken seroconversion rates in northeastern Florida. In particular, our approach demonstrates the utility of incorporating the underlying spatiotemporal structure during model building to improve the results. Disease mapping often assumes that disease dynamics remains static across time, even though tools now exist to incorporate spatiotemporal transmission and environmental variables into modeling efforts. Static assumptions are likely unrealistic [51], and the approach we implemented here attempts to account for the joint dependence structure in both space and time. The inclusion of spatiotemporal GMRFs contributed to the improved model fits across both virus response variables and demonstrated the need to account for residual spatiotemporal autocorrelation when inferring the contributions of environmental variables, in this case the cumulative precipitation, to virus ecology.
The model results for EEEV seroconversion were particularly informative, indicating a strong positive effect of the cypress–tupelo wetland habitats surrounding chicken coop sites within 3500 m and a strong positive effect of the forest edge density within 500 m of the coop sites. These results are consistent with the observed breeding habitats of Cs. melanura and with previous investigations of landscape contributions to EEEV. For example, in the northeastern U.S., Cs. melanura increased with deciduous or evergreen forested wetlands that had low connectivity to streams and decreased with shrub or scrub wetlands, while higher Cs. Melanura abundances were strongly associated with higher EEEV infection rates in mosquito pools [52]. In addition, the greater edge density of forest habitats within closer proximities to coop locations suggests that the fragmented forest areas may be conducive to the dispersal of Cs. melanura or EEEV bridge vectors, and these results corroborate the findings from a previous study in the Florida Panhandle indicating a positive effect of coniferous tree plantations within close proximity to chicken coops on EEEV seroconversion [15].
The strong support for the most parsimonious model (WAICw = 0.67), the low number of models comprising the 95% confidence set (n = 3), and the consistent inclusion of cypress–tupelo wetland habitats and forest edge habitats across models indicated that these variables, along with the inclusion of the spatiotemporal GMRFs, have the potential to be predictive of EEEV seroconversion in this area. The results from the ROC analysis reported strong accuracy when predicting the presence–absence on out-of-sample data, although a closer investigation showed greater accuracy in predicting absences than presences. The inclusion of additional sampling data has the potential to improve the predictive accuracy, including the ability to predict seroconversion rates across larger geographic areas and different time periods.
The environmental variables included in our models were less informative for WNV seroconversion than for EEEV. The most parsimonious model was the spatiotemporal intercept-only model without covariates, which had strong support (WAICw = 0.74), and the credible intervals across models included in the 95% confidence set indicated that the data did not support a strong effect of any of the covariates on sentinel chicken seroconversion to WNV. Although we expected to find some correlation with the landscape variables included in our analyses, this result was not altogether surprising given often contradictory results from previous WNV landscape studies in other regions of the U.S. [53,54]. For example, correlations between WNV and urban and semi-urban environments have been found in multiple U.S. states [17,55,56,57], while simultaneously agricultural landscapes have also been indicated [17,53,58,59,60,61]. Although landscape correlations with WNV are less studied in Florida, the presence of multiple vector species inhabiting a broad range of landscape habitats may be reflected in our model results [12].
To our surprise, the data did not support a strong effect of any of the lagged temperature or precipitation variables included in our analyses, despite the spatiotemporal GMRFs showing a clear structure in seroconversion across the study period for each virus and previous studies suggesting the importance of precipitation to observed virus transmission [21,22]. Although temperature is known to be important to mosquito vector development and can affect the extrinsic viral incubation periods [36], the 8-day mean land surface temperature values included in our study were not identified as important variables during the initial variable reduction step for either virus’ seroconversion. The 9-week lagged cumulative precipitation was included in the 95% confidence set for the non-spatial and spatial EEEV GLMMs, but credible intervals for this variable no longer indicated a strong effect on EEEV seroconversion in the spatiotemporal GLMM, which accounts for temporal autocorrelation. Although not uncommon when working with temporally structured data, this result demonstrates the importance of accounting for residual spatiotemporal autocorrelations to avoid making inferences derived from spurious correlations [50].
The Gaussian Markov random field plots provided a visualization of the spatiotemporal structures of EEEV and WNV sentinel chicken seroconversions across the study period and highlighted differences in the patterns between the seroconversions of the two viruses. The plots indicated elevated EEEV seroconversion at the very start of the sampling season, with the more punctuated and tightly formed structure indicating a smaller effective range compared to the gradual and more diffuse structure of WNV seroconversion at the end of the sampling season. Sampling earlier in the season would likely have captured a more complete distribution of EEEV seroconversion, as well as sampling later in the season for WNV. The climate associations may be stronger towards the seasonal onset and offset of infections, arguing for longer sampling periods that can capture those dynamics.
Comparisons of the GMRF plots to FDOH Arbovirus Surveillance Reports of emu, human, and equine cases in the study area [23] suggested potential overlap between active EEEV sentinel chicken seroconversion rates and the report of an EEEV-positive emu flock during the week of 8 May–14 May in Volusia County in the southernmost region of the study area, and also with an equine case during the week of May 25th in St. Johns County (Figure 5). Additional EEEV equine cases occurred in Volusia County during the weeks of June 16th, June 21st, and August 1st. However, these reports did not appear to coincide with active transmission in sentinel chickens in Volusia County during this time period. In addition, a human EEEV case during the week of July 24th in Duval County in the northernmost region of the study area did not appear to coincide with active transmission in sentinel chickens in this area during the week of July 24th, despite activity to the south in St. Johns County. The spatiotemporal GMRF plots for WNV sentinel chicken seroconversion did, however, suggest potential overlap between the spatiotemporal structure of WNV seroconversion and human cases in the study area. In Duval County (Figure 6), 14 human cases of WNV infection were reported over the study period, with 5 cases reported in August, 3 in September, 4 in October, and 2 in November. The sentinel chicken WNV seroconversion rates exhibited a strong spatiotemporal structure in Duval County beginning in the first week of August and remained relatively strong throughout the remainder of the study period, coinciding with the temporal distribution of human cases in the county.
Although we utilized a robust modeling framework, study limitations exist. Specifically, the EEEV seroconversion was truncated at the beginning of the sampling season and the WNV seroconversion was truncated at the end of the sampling season, preventing the full distribution of transmission events for each virus from being included in the model. Another issue is that the environmental covariates in our models were measured at different spatial resolutions, although the resolution differential was modest (1 km versus 4 km for temperature and precipitation, respectively). Further, we implemented a model selection approach that quantified all combinations of environmental variables, such that the variables were evaluated individually (i.e., precipitation variables only, temperature variables only, landscape variables only) and also together to identify the best model or set of models. In all cases, the climatic variables dropped out of our best-performing models. Therefore, the inferences we made from our models and highlighted in the discussion section were not derived from models that included variables at multiple resolutions. Future investigations would benefit from a longer sampling season and the inclusion of additional districts to facilitate multi-year analyses that may reveal both intra- and interannual patterns in climatic variables contributing to WNV and EEEV sentinel chicken seroconversion across a broader geographic area. In addition, the inclusion of spatiotemporal mosquito abundance data in conjunction with environmental variables has the potential to provide important information toward understanding transmission dynamics.

5. Conclusions

The use of spatiotemporal modeling frameworks with EO data shows significant promise in providing effective tools to both map zoonotic arbovirus transmission and understand the environmental drivers (or lack thereof) of transmission risk. Continued integration of these approaches combined with robust surveillance activities can help promote more optimized prevention and effective control measures that are informed by data.

Supplementary Materials

The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/rs14143388/s1. Figure S1: Mesh created using Delaney triangularization for SPDE calculation; Figure S2: Density curves for regression parameters and random effects in best EEEV model; Table S1: Non-spatial GLMMs for EEEV identifying the 95% confidence set; Table S2: Non-spatial GLMMs for WNV identifying the 95% confidence set; Table S3: Full model results for non-spatial, spatial, and spatiotemporal INLA models for EEEV; Table S4: EEEV best model results including credible intervals for random effects; Full model results for non-spatial, spatial, and spatiotemporal INLA models for WNV; Table S5: Full model results for non-spatial, spatial, and spatiotemporal INLA models for WNV.

Author Contributions

Conceptualization, L.P.C. and C.E.; methodology, L.P.C. and R.P.G.; software, L.P.C., R.P.G. and B.V.G.; validation, R.P.G. and L.P.C.; formal analysis, L.P.C. and R.P.G.; investigation, L.P.C. and N.D.B.-C.; resources, S.B., R.W. and R.-D.X.; data curation, L.P.C., B.V.G., R.W., B.A., R.-D.X., W.Q., S.B. and M.T.; writing—original draft preparation, L.P.C., R.P.G., M.F.S., Y.T., A.M.B. and J.M.A.; writing—review and editing, N.D.B.-C., J.M.A., C.E., R.-D.X., W.Q., M.T., B.A., B.V.G., S.B. and R.W.; visualization, B.V.G., A.M.B. and L.P.C.; supervision, L.P.C.; project administration, L.P.C.; funding acquisition, L.P.C., N.D.B.-C., R.-D.X., R.W. and S.B. All authors have read and agreed to the published version of the manuscript.

Funding

This research was funded by the Florida Department of Agriculture and Consumer Services, grant number #026367, the University of Florida Biodiversity Institute Seed Fund, and the USDA National Institute of Food and Agriculture, Hatch project 1021482. A.M.B. was funded through the University of Florida Graduate Student Fellowship.

Data Availability Statement

The data presented in this study are available on request from the corresponding author. The data are not publicly available due to the maintenance of some sentinel chicken flocks at private residences.

Acknowledgments

The authors would like to acknowledge the Florida Department of Health Arbovirus Surveillance Laboratory for conducting sentinel chicken seroconversion testing, mosquito control district personnel, and E.E. Peterson for comments on the methods section of the manuscript.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. World Health Organization. World Health Organization. World health statistics. In World Health Organization Report; World Health Organization: Geneva, Switzerland, 2013. [Google Scholar]
  2. World Health Organization. Manual on Environmental Management for Mosquito Control, with Special Emphasis on Malaria Vectors; World Health Organization: Geneva, Switzerland, 1982. [Google Scholar]
  3. Kalluri, S.; Gilruth, P.; Rogers, D.; Szczur, M. Surveillance of Arthropod Vector-Borne Infectious Diseases Using Remote Sensing Techniques: A Review. PLOS Pathog. 2007, 3, e116. [Google Scholar] [CrossRef] [PubMed]
  4. Kitron, U. Landscape Ecology and Epidemiology of Vector-Borne Diseases: Tools for Spatial Analysis. J. Med. Èntomol. 1998, 35, 435–445. [Google Scholar] [CrossRef] [PubMed]
  5. Banerjee, S.; Carlin, B.P.; Gelfand, A.E. Hierarchical Modeling and Analysis for Spatial Data; Chapman & Hall/CRC: Boca Raton, FL, USA, 2004. [Google Scholar]
  6. Lindsey, N.P.; Lehman, J.A.; Staples, J.E.; Fisher, M. West Nile virus and other nationally notifiable arboviral diseases–United States. MMWR Morb. Mortal. Wkly. Rep. 2014, 64, 929–934. [Google Scholar] [CrossRef] [Green Version]
  7. Scott, T.W.; Weaver, S.C. Eastern Equine Encephalomyelitis Virus: Epidemiology and Evolution of Mosquito Transmission. In Advances in Virus Research; Maramorosch, K., Murphy, F.A., Shatkin, A.J., Eds.; Academic Press: Cambridge, MA, USA, 1989; pp. 277–328. [Google Scholar]
  8. Reisen, W.K. Ecology of West Nile Virus in North America. Viruses 2013, 5, 2079–2105. [Google Scholar] [CrossRef] [PubMed]
  9. Wisely, S.W.; Hood, K. Fact Sheet: Eastern Equine Encephalitis. In Electronic Data Information Source; University of Florida: Gainesville, FL, USA, 2019. [Google Scholar] [CrossRef]
  10. American Association Equine Practitioners. Infectious Disease Guidelines: West Nile Virus [Fact Sheet]; American Association Equine Practitioners: Lexington, KY, USA, 2017. [Google Scholar]
  11. Armstrong, P.M.; Andreadis, T.G. Ecology and Epidemiology of Eastern Equine Encephalitis Virus in the Northeastern United States: An Historical Perspective. J. Med Èntomol. 2022, 59, 1–13. [Google Scholar] [CrossRef]
  12. Rochlin, I.; Faraji, A.; Healy, K.; Andreadis, T.G. West Nile Virus Mosquito Vectors in North America. J. Med Èntomol. 2019, 56, 1475–1490. [Google Scholar] [CrossRef]
  13. Day, J.F.; Tabachnick, W.J.; Smartt, C.T. Factors That Influence the Transmission of West Nile Virus in Florida. J. Med. Entomol. 2015, 52, 743–754. [Google Scholar] [CrossRef]
  14. Day, J.F.; Shaman, J. Mosquito-borne arboviral surveillance and the prediction of disease outbreaks. In Flavivirus Encephalitis; Ruzek, D., Ed.; InTech: Rijeka, Croatia, 2011; pp. 105–130. [Google Scholar]
  15. Kelen, P.T.V.; A Downs, J.; Stark, L.M.; Loraamm, R.W.; Anderson, J.H.; Unnasch, T.R. Spatial epidemiology of eastern equine encephalitis in Florida. Int. J. Health Geogr. 2012, 11, 47. [Google Scholar] [CrossRef] [Green Version]
  16. Kelen, P.V.; Downs, J.A.; Unnasch, T.; Stark, L. A risk index model for predicting eastern equine encephalitis virus transmission to horses in Florida. Appl. Geogr. 2014, 48, 79–86. [Google Scholar] [CrossRef] [Green Version]
  17. Sallam, M.F.; Xue, R.-D.; Pereira, R.M.R.; Koehler, P.G. Ecological niche modeling of mosquito vectors of West Nile virus in St. John’s County, Florida, USA. Parasites Vectors 2016, 9, 371. [Google Scholar] [CrossRef]
  18. Burch, C.; Loraamm, R.; Unnasch, T.; Downs, J. Utilizing ecological niche modelling to predict habitat suitability of eastern equine encephalitis in Florida. Ann. GIS 2020, 26, 133–147. [Google Scholar] [CrossRef] [Green Version]
  19. Beeman, S.P.; Downs, J.A.; Unnasch, T.R.; Unnasch, R.S. West Nile Virus and Eastern Equine Encephalitis Virus High Probability Habitat Identification for the Selection of Sentinel Chicken Surveillance Sites in Florida. J. Am. Mosq. Control Assoc. 2022, 38, 1–6. [Google Scholar] [CrossRef] [PubMed]
  20. Heberlein-Larson, L.A.; Tan, Y.; Stark, L.M.; Cannons, A.C.; Shilts, M.H.; Unnasch, T.R.; Das, S.R. Complex Epidemiological Dynamics of Eastern Equine Encephalitis Virus in Florida. Am. J. Trop. Med. Hyg. 2019, 100, 1266–1274. [Google Scholar] [CrossRef] [PubMed]
  21. Shaman, J.; Day, J.F.; Stieglitz, M. Drought-Induced Amplification and Epidemic Transmission of West Nile Virus in Southern Florida. J. Med. Èntomol. 2005, 42, 134–141. [Google Scholar] [CrossRef] [PubMed]
  22. Miley, K.M.; Downs, J.; Beeman, S.P.; Unnasch, T.R. Impact of the Southern Oscillation Index, Temperature, and Precipitation on Eastern Equine Encephalitis Virus Activity in Florida. J. Med. Èntomol. 2020, 57, 1604–1613. [Google Scholar] [CrossRef] [PubMed]
  23. Morrison, A.; Giandomenico, D.; Stanek, D.; Heberlein-Larson, L.; Castaneda, M.; Mock, V.; Blackmore, C. Florida Arbovirus Surveillance Week 52: December 23–29; Florida Department of Health: Tallahassee, FL, USA, 2018. [Google Scholar]
  24. Florida Fish and Wildlife Conservation Commission and Florida Natural Areas Inventory. Cooperative Land Cover; Version 3.3; Raster: Tallahassee, FL, USA, 2021. [Google Scholar]
  25. Kottek, M.; Grieser, J.; Beck, C.; Rudolf, B.; Rubel, F. World map of the Köppen-Geiger climate classification updated. Meteorol. Z. 2006, 15, 259–263. [Google Scholar] [CrossRef]
  26. Winsberg, M.D. Florida Weather, 2nd ed.; University Press of Florida: Gainesville, FL, USA, 2003. [Google Scholar]
  27. Lloyd, A.; Connelly, C.; Carlson, D. Florida Mosquito Control: The State of the Mission as Defined by Mosquito Controllers, Regulators, and Environmental Managers; Florida Coordinating Council on Mosquito Control: Vero Beach, FL, USA, 2018.
  28. Florida Department of Health (FDOH). Non-Human Mosquito-Borne Disease Monitoring Activities; FDOH: Tampa, FL, USA, 2021.
  29. McGarigal, K.; Marks, B.J. Spatial pattern analysis program for quantifying landscape structure. In General Technical Report. PNW-GTR-351; US Department of Agriculture, Forest Service, Pacific Northwest Research Station: Portland, OR, USA, 1995; p. 122. [Google Scholar]
  30. Gergel, S.E.; Turner, M.G.; Miller, J.R.; Melack, J.M.; Stanley, E.H. Landscape indicators of human impacts to riverine systems. Aquat. Sci. 2002, 64, 118–128. [Google Scholar] [CrossRef]
  31. Hesselbarth, M.; Sciaini, M.; With, K.; Wiegand, K.; Nowosad, J. Landscapemetrics: An open-source R tool to calculate landscape metrics. Ecography 2019, 42, 1–10. [Google Scholar] [CrossRef] [Green Version]
  32. R Core Team. R: A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2021. [Google Scholar]
  33. Beven, K.J.; Kirkby, M.J. A physically based, variable contributing area model of basin hydrology/Un modèle à base physique de zone d’appel variable de l’hydrologie du bassin versant. Hydrolog. Sci. J. 1979, 24, 43–69. [Google Scholar] [CrossRef] [Green Version]
  34. Du, J. NCEP/EMC 4KM Gridded Data (GRIB) Stage IV Data; Version 1.0; UCAR/NCAR—Earth Observing Laboratory: Boulder, CO, USA, 2011. [Google Scholar]
  35. Wan, Z.; Zhang, Y.; Zhang, Q.; Li, Z.L. Validation of the land-surface temperature products retrieved from Terra Moderate Resolution Imaging Spectroradiometer data. Remote Sens. Environ. 2002, 83, 163–180. [Google Scholar] [CrossRef]
  36. Mullen, G.R.; Durden, L.A. Medical and Veterinary Entomology, 3rd ed.; Academic Press: Cambridge, MA, USA, 2019. [Google Scholar]
  37. Tuszynski, J. caTools: Tools: Moving Window Statistics, GIF, Base 64, ROC AUC, etc. R Package, Version 1.18.0.; 2020. Available online: https://cran.r-project.org/web/packages/caTools/index.html (accessed on 11 May 2022).
  38. Zuur, A.F.; Ieno, E.N.; Walker, N.; Saveliev, A.A.; Smith, G.M. Mixed Effects Models and Extensions in Ecology with R, 1st ed.; Springer: New York, NY, USA, 2009. [Google Scholar]
  39. Brooks, M.E.; Kristensen, K.; van Benthem, K.J.; Magnusson, A.; Berg, C.W.; Nielsen, A.; Skaug, H.J.; Machler, M.; Bolker, B.M. glmmTMB Balances Speed and Flexibility Among Packages for Zero-inflated Generalized Linear Mixed Modeling. R J. 2017, 9, 378–400. [Google Scholar] [CrossRef] [Green Version]
  40. Barton, K. MuMIn: Multi-Model Inference, R Package Version 1.43.15; 2019. Available online: https://cran.r-project.org/web/packages/MuMIn/index.html (accessed on 11 May 2022).
  41. Akaike, H. Information Theory and an Extension of the Maximum Likelihood Principle. In Proceedings of the 2nd International Symposium on Information Theory, Tsahkadsor, Armenia, 2–8 September 1971; Petrov, B.N., Csáki, F., Eds.; Akadémiai Kiadó: Budapest, Hungary, 1973; pp. 267–281. [Google Scholar]
  42. Burnham, K.P.; Anderson, D.R. Model Selection and Multimodel Inference: A Practical Information-Theoretic Approach, 2nd ed.; Springer: New York, NY, USA, 2002; ISBN 0-387-95364-7. [Google Scholar]
  43. Lindgren, F.; Rue, H. Bayesian Spatial Modelling with R-INLA. J. Stat. Softw. 2015, 63, 1–2. [Google Scholar] [CrossRef] [Green Version]
  44. Rue, R.; Martino, S.; Chopin, N. Approximate Bayesian inference for latent Gaussian models using inte-grated nested Laplace approximations (with discussion). J. R Stat. B Soc. 2009, 71, 319–392. [Google Scholar] [CrossRef]
  45. Zuur, A.F.; Ieno, E.N.; Saveliev, A.A. Beginner’s Guide to Spatial, Temporal, and Spatial-Temporal Ecological Data Analysis with R-INLA; Highland Statistics Ltd.: Newburgh, IN, USA, 2017; Volume 7. [Google Scholar]
  46. Lindgren, F.; Rue, H.; Lindström, J. An explicit link between Gaussian fields and Gaussian Markov random fields: The stochastic partial differential equation approach. J. R. Stat. Soc. Ser. B Stat. Methodol. 2011, 73, 423–498. [Google Scholar] [CrossRef] [Green Version]
  47. Simpson, J.E.; Schweinfest, C.W.; Shull, G.E.; Gawenis, L.R.; Walker, N.M.; Boyle, K.T.; Soleimani, M.; Clarke, L.L. PAT-1 (Slc26a6) is the predominant apical membrane Cl-/HCO3- exchanger in the upper villous epithelium of the murine duodenum. Am. J. Physiol. -Gastr. L 2007, 292, G1079–G1088. [Google Scholar] [CrossRef] [Green Version]
  48. Spiegelhalter, D.J.; Best, N.G.; Carlin, B.P.; van der Linde, A. Bayesian measures of model complexity and fit. J. R. Stat. Soc. Ser. B Stat. Methodol. 2002, 64, 583–616. [Google Scholar] [CrossRef] [Green Version]
  49. Thiele, C.; Hershfield, G. Cutpointr: Improved Estimation and Validation of Optimal Cutpoints in R. J. Stat. Softw. 2021, 98, 1–27. [Google Scholar] [CrossRef]
  50. Hoeting, J.A. The importance of accounting for spatial and temporal correlation in analyses of ecological data. Ecol. Appl. 2009, 19, 574–577. [Google Scholar] [CrossRef] [PubMed]
  51. Martínez-Beneito, M.A.; López-Quílez, A.; Rocamora, P.B. An autoregressive approach to spatio-temporal disease mapping. Stat. Med. 2008, 27, 2874–2889. [Google Scholar] [CrossRef]
  52. Skaff, N.K.; Armstrong, P.M.; Andreadis, T.G.; Cheruvelil, K.S. Wetland characteristics linked to broad-scale patterns in Culiseta melanura abundance and eastern equine encephalitis virus infection. Parasites Vectors 2017, 10, 501. [Google Scholar] [CrossRef] [Green Version]
  53. DeGroote, J.P.; Sugumaran, R.; Brend, S.M.; Tucker, B.J.; Bartholomay, L.C. Landscape, demographic, entomological, and climatic associations with human disease incidence of West Nile virus in the state of Iowa, USA. Int. J. Health Geogr. 2008, 7, 19. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  54. Bowden, S.E.; Magori, K.; Drake, J.M. Regional Differences in the Association Between Land Cover and West Nile Virus Disease Incidence in Humans in the United States. Am. J. Trop. Med. Hyg. 2011, 84, 234–238. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  55. O Ruiz, M.; Walker, E.D.; Foster, E.S.; Haramis, L.D.; Kitron, U.D. Association of West Nile virus illness and urban landscapes in Chicago and Detroit. Int. J. Health Geogr. 2007, 6, 10. [Google Scholar] [CrossRef] [PubMed] [Green Version]
  56. Reisen, W.K.; Takahashi, R.M.; Carroll, B.D.; Quiring, R. Delinquent Mortgages, Neglected Swimming Pools, and West Nile Virus, California. Emerg. Infect. Dis. 2008, 14, 1747–1749. [Google Scholar] [CrossRef]
  57. Sallam, M.F.; Michaels, S.R.; Riegel, C.; Pereira, R.; Zipperer, W.; Lockaby, B.G.; Koehler, P.G. Spatio-Temporal Distribution of Vector-Host Contact (VHC) Ratios and Ecological Niche Modeling of the West Nile Virus Mosquito Vector, Culex quinquefasciatus, in the City of New Orleans, LA, USA. Int. J. Environ. Res. Public Health 2017, 14, 892. [Google Scholar] [CrossRef]
  58. Eisen, L.; Barker, C.M.; Moore, C.G.; Pape, W.J.; Winters, A.M.; Cheronis, N. Irrigated Agriculture Is an Important Risk Factor for West Nile Virus Disease in the Hyperendemic Larimer-Boulder-Weld Area of North Central Colorado. J. Med. Entomol. 2010, 47, 939–951. [Google Scholar] [CrossRef]
  59. Crowder, D.W.; Dykstra, E.A.; Brauner, J.M.; Duffy, A.; Reed, C.; Martin, E.; Peterson, W.; Carrière, Y.; Dutilleul, P.; Owen, J. West Nile Virus Prevalence across Landscapes Is Mediated by Local Effects of Agriculture on Vector and Host Communities. PLoS ONE 2013, 8, e55006. [Google Scholar] [CrossRef]
  60. Kovach, T.J.; Kilpatrick, A.M. Increased Human Incidence of West Nile Virus Disease near Rice Fields in California but Not in Southern United States. Am. J. Trop. Med. Hyg. 2018, 99, 222–228. [Google Scholar] [CrossRef]
  61. Dunphy, B.M.; Kovach, K.B.; Gehrke, E.J.; Field, E.N.; Rowley, W.A.; Bartholomay, L.C.; Smith, R.C. Long-term surveillance defines spatial and temporal patterns implicating Culex tarsalis as the primary vector of West Nile virus. Sci. Rep. 2019, 9, 6637. [Google Scholar] [CrossRef] [Green Version]
Figure 1. Proportion and abundances of 2018 EEEV- and WNV-positive seroconversions at sentinel chicken coops across the study area.
Figure 1. Proportion and abundances of 2018 EEEV- and WNV-positive seroconversions at sentinel chicken coops across the study area.
Remotesensing 14 03388 g001
Figure 2. Aggregated land cover classes across each county in the study area.
Figure 2. Aggregated land cover classes across each county in the study area.
Remotesensing 14 03388 g002
Figure 3. Effect curves of model parameters for environmental variables in the best-fitting EEEV model; credible intervals that do not cross zero indicate a strong positive or negative effect on EEEV seroconversion.
Figure 3. Effect curves of model parameters for environmental variables in the best-fitting EEEV model; credible intervals that do not cross zero indicate a strong positive or negative effect on EEEV seroconversion.
Remotesensing 14 03388 g003
Figure 4. The AUC of the ROC for the best EEEV model.
Figure 4. The AUC of the ROC for the best EEEV model.
Remotesensing 14 03388 g004
Figure 5. Weekly plots of spatiotemporal Gaussian Markov random fields (GMRF) for week 18–45 2018 EEEV sentinel chicken seroconversion rates. A convex hull was generated to create the GMRFs, and the GMRFs were clipped to the coastline for visualization. The ESRI background imagery was used to map the first plot (1–7 May).
Figure 5. Weekly plots of spatiotemporal Gaussian Markov random fields (GMRF) for week 18–45 2018 EEEV sentinel chicken seroconversion rates. A convex hull was generated to create the GMRFs, and the GMRFs were clipped to the coastline for visualization. The ESRI background imagery was used to map the first plot (1–7 May).
Remotesensing 14 03388 g005
Figure 6. Weekly plots of spatiotemporal Gaussian random fields (GRF) for week 18–45 2018 WNV sentinel chicken seroconversion rates. A convex hull was generated to create the GRFs, and the GRFs were clipped to the coastline for visualization. The ESRI background imagery was used to map the first plot (1–7 May).
Figure 6. Weekly plots of spatiotemporal Gaussian random fields (GRF) for week 18–45 2018 WNV sentinel chicken seroconversion rates. A convex hull was generated to create the GRFs, and the GRFs were clipped to the coastline for visualization. The ESRI background imagery was used to map the first plot (1–7 May).
Remotesensing 14 03388 g006
Table 1. The 95% confidence set of models for EEEV based on the cumulative sum of the WAIC weights. An “X” present in a column cell indicates that the corresponding variable was not included in the model.
Table 1. The 95% confidence set of models for EEEV based on the cumulative sum of the WAIC weights. An “X” present in a column cell indicates that the corresponding variable was not included in the model.
Intercept3500 m Percent Forest3500 m Percent Cypress–tupelo Wet Land500 m Edge Density Forest5000 m Percent ForestWeekly Cumulative Precip t − 1Weekly Cumulative Precip t − 9DICWAICDelta WAICRelative Log LikelihoodWAIC wSum of the Cumulative WAIC w
−5.465X0.7550.851XX0.139189.455178.48010.6720.672
−5.406X0.7650.782X0.3090.168192.175180.4411.9610.3750.2520.924
−5.199X0.531X0.456X0.082197.896184.3235.8430.0540.0360.96
Table 2. The 95% confidence set of models for WNV based on the cumulative sum of the WAIC weights. An “X” present in a column cell indicates that the corresponding variable was not included in the model.
Table 2. The 95% confidence set of models for WNV based on the cumulative sum of the WAIC weights. An “X” present in a column cell indicates that the corresponding variable was not included in the model.
Intercept2000 m Edge Density Wet LandWeekly Cumulative Precip t − 1Weekly Cumulative Precip t − 33500 Edge Density Wet Land5000 Edge Density Wet LandWeekly Cumulative Precip t − 7DICWAICDelta WAICRel LLWAIC wSum
−3.687XXXXXX303.483298.005010.7400.740
−3.721X0.001−0.110XXX308.149302.3014.2960.1170.0860.827
−3.710−0.128−0.005−0.120XXX309.626304.0536.0480.0490.0360.863
−3.717X−0.002−0.120−0.100XX309.840304.1886.1820.0450.0340.896
−3.717X−0.002−0.120X−0.100X309.840304.1886.1820.0450.0340.930
-3.717X0.011−0.100XX0.088310.417304.3426.3360.0420.0310.961
Table 3. The credible intervals for the EEEV 95% confidence set (Model Ranks 1–3). The variables in bold indicate that the data support a strong effect on EEEV seroconversion. The sign of the mean indicates a positive or negative effect. If values cross zero between the 0.025 quant and the 0.975 quant, the variable is not important for seroconversion.
Table 3. The credible intervals for the EEEV 95% confidence set (Model Ranks 1–3). The variables in bold indicate that the data support a strong effect on EEEV seroconversion. The sign of the mean indicates a positive or negative effect. If values cross zero between the 0.025 quant and the 0.975 quant, the variable is not important for seroconversion.
Model Rank 1 (Best)
mean0.025 quant0.975 quant
Intercept−5.4651−9.5433−1.3903
Weekly cumulative precip t − 90.1386−0.69050.967
3500 m percent cypress–tupelo wetland0.75520.30561.2045
500 m edge density forest0.85070.22031.4806
Model Rank 2
mean0.025 quant0.975 quant
Intercept−5.406−8.958−1.856
Weekly cumulative precip t − 90.168−0.6620.999
500 m edge density forest0.7820.1361.426
3500 percent cypress–tupelo wetland0.7650.2991.231
Weekly cumulative precip t − 10.309−0.2460.864
Model Rank 3
Valuemean0.025 quant0.975 quant
Intercept−5.199−8.257−2.144
Weekly cumulative precip t − 90.082−0.680.843
5000 m percent forest0.456−0.1411.051
3500 m percent cypress–tupelo wetland0.5310.0051.057
Table 4. The credible intervals for the WNV 95% confidence set. No variable had a strong effect on WNV seroconversion. The sign of the mean indicates a positive or negative effect. If values cross zero between the 0.025 quant and the 0.975 quant, the variable has no effect on seroconversion.
Table 4. The credible intervals for the WNV 95% confidence set. No variable had a strong effect on WNV seroconversion. The sign of the mean indicates a positive or negative effect. If values cross zero between the 0.025 quant and the 0.975 quant, the variable has no effect on seroconversion.
Model Rank 1 (Best)
Valuemean0.025 quant0.975 quant
Intercept−3.687−6.468−0.909
Model Rank 2
Valuemean0.025 quant0.975 quant
Intercept−3.721−6.542−0.902
Weekly cumulative precip t − 10.001−0.5910.591
Weekly cumulative precip t − 3−0.111−0.670.447
Model Rank 3
mean0.025 quant0.975 quant
Intercept−3.71−6.42−1.002
2000 m edge density wetland−0.128−0.4760.219
Weekly cumulative precip t − 1−0.005−0.5950.585
Weekly cumulative precip t − 3−0.117−0.6760.441
Model Rank 4
mean0.025 quant0.975 quant
Intercept−3.717−6.469−0.968
3500 m edge density wetland−0.1−0.4380.237
Weekly cumulative precip t − 1−0.002−0.5930.587
Weekly cumulative precip t − 3−0.115−0.6730.444
Model Rank 5
mean0.025 quant0.975 quant
Intercept−3.717−6.469−0.968
5000 m edge density wetland−0.1−0.4380.237
Weekly cumulative precip t − 1−0.002−0.5930.587
Weekly cumulative precip t − 3−0.115−0.6730.444
Model Rank 6
mean0.025 quant0.975 quant
Intercept−3.717−6.574−0.862
Weekly cumulative precip t − 10.011−0.5870.608
Weekly cumulative precip t − 3−0.101−0.6630.46
Weekly cumulative precip t − 70.088−0.3760.552
Publisher’s Note: MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Share and Cite

MDPI and ACS Style

Campbell, L.P.; Guralnick, R.P.; Giordano, B.V.; Sallam, M.F.; Bauer, A.M.; Tavares, Y.; Allen, J.M.; Efstathion, C.; Bartlett, S.; Wishard, R.; et al. Spatiotemporal Modeling of Zoonotic Arbovirus Transmission in Northeastern Florida Using Sentinel Chicken Surveillance and Earth Observation Data. Remote Sens. 2022, 14, 3388. https://doi.org/10.3390/rs14143388

AMA Style

Campbell LP, Guralnick RP, Giordano BV, Sallam MF, Bauer AM, Tavares Y, Allen JM, Efstathion C, Bartlett S, Wishard R, et al. Spatiotemporal Modeling of Zoonotic Arbovirus Transmission in Northeastern Florida Using Sentinel Chicken Surveillance and Earth Observation Data. Remote Sensing. 2022; 14(14):3388. https://doi.org/10.3390/rs14143388

Chicago/Turabian Style

Campbell, Lindsay P., Robert P. Guralnick, Bryan V. Giordano, Mohamed F. Sallam, Amely M. Bauer, Yasmin Tavares, Julie M. Allen, Caroline Efstathion, Suzanne Bartlett, Randy Wishard, and et al. 2022. "Spatiotemporal Modeling of Zoonotic Arbovirus Transmission in Northeastern Florida Using Sentinel Chicken Surveillance and Earth Observation Data" Remote Sensing 14, no. 14: 3388. https://doi.org/10.3390/rs14143388

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop