Ecological niche modeling the potential geographic distribution of four Culicoides species of veterinary significance in Florida

Epizootic hemorrhagic disease (EHD) is a viral arthropod-borne disease affecting wild and domestic ruminants. EHD virus (EHDV) is transmitted to vertebrate animal hosts by biting midges in the genus Culicoides. Culicoides sonorensis Latreille is the only confirmed vector of EHDV in the United States but is considered rare in Florida and not sufficiently abundant to support EHDV transmission. This study used ecological niche modeling to map the potential geographical distributions and associated ecological variable space of four Culicoides species suspected of transmitting EHDV in Florida, including Culicoides insignis, Culicoides stellifer, Culicoides debilipalpis and Culicoides venustus. Models were developed with the Genetic Algorithm for Rule Set Production in DesktopGARP v1.1.3 using species occurrence data from field sampling along with environmental variables from WorldClim and Trypanosomiasis and Land use in Africa. For three Culicoides species (C. insignis, C. stellifer and C. debilipalpis) 96 – 98% of the presence points were predicted across the Florida landscape (63.77% – 72.53%). For C. venustus, models predicted 98.00% of presence points across 27.42% of Florida. Geographic variations were detected between species. Culicoides insignis was predicted to be restricted to peninsular Florida, and in contrast, C. venustus was predicted to be primarily in north Florida and the panhandle region. Culicoides stellifer and C. debilipalpis were predicted nearly statewide. Environmental conditions also differed by species, with some species’ ranges predicted by more narrow ranges of variables than others. The Normalized Difference Vegetation Index (NDVI) was a major predictor of C. venustus and C. insignis presence. For C. stellifer, Land Surface Temperature, Middle Infrared were the most limiting predictors of presence. The limiting variables for C. debilipalpis were NDVI Bi-Annual Amplitude and NDVI Annual Amplitude at 22.45% and 28.09%, respectively. The model outputs, including maps and environmental variable range predictions generated from these experiments provide an important first pass at predicting species of veterinary importance in Florida. Because EHDV cannot exist in the environment without the vector, model outputs can be used to estimate the potential risk of disease for animal hosts across Florida. Results also provide distribution and habitat information useful for integrated pest management practices.


Introduction 54
Vector-borne pathogens can only exist in a permissive environment that supports 55 appropriate vectors (and hosts), such that the distribution of disease is linked to vector 56 distribution. Species distribution models (SDMs) can be used to map the potential distribution of 57 disease vectors allowing inference of disease risk [1][2][3][4]. In disease ecology, SDMs are useful to 58 determine the potential current and future geographic distribution of vector species as proxies  can occupy the same ecological niche, in the case of this work, no two species will cohabitate 70 the same median ranges for all variables. Broadly, ENMs apply each unique set of ecological 71 parameters allowing for a species to maintain a population without immigration [9,10], with a 72 focus on abiotic and climatological conditions that support a species [11]. In general, ENMs use either presence-only or presence and absence data of a target species and environmental 74 covariates to identify non-random patterns relating species occurrence to the landscape [

223
Four final experiments were developed for C. insignis, C. stellifer, C. venustus, and C.
224 debilipalpis (Figure 2). The results of the AUC scores for all models indicated models performed 225 better than random, and most of the models had a total omission of zero, meaning all 226 independent test points were predicted landscapes (

252
The C. insignis model predicted this species to be widely distributed across the  Figure 3A). Culicoides stellifer was predicted to have low suitability 264 south of Lake Okeechobee ( Figure 3B), whereas C. venustus was predicted to have no suitable 265 habitat south of Polk County, outside of isolated pixels ( Figure 3C).

266
The experiment for C. insignis predicted the 66.0% of 2017 field validation data correctly, 267 while 88.3% of C. stellifer locations were predicted correctly and C. debilipalpis locations were p 268 with 93.0% accuracy (Table 2). Models were unable to be validated for C. venustus due to the relative rarity of this species using our trapping methods and all available data from data 270 sources were used to build models; for C. venustus, we relied on the independent 271 testing/training split to assess accuracy (Table 2).

272
Broadly, covariates with narrow ranges can be interpreted as the most limiting in defining 273 species distributions. Across species, mean, minimum, and maximum LST, and mean,

282
Ranges are from 0.0 to 1.0 of the environmental variables which predict C. debilipalpis, C.

284
The median ranges of environmental covariates in the distribution of C. stellifer were 285 similar to those of C. insignis. For example, minimum, maximum, and mean LST, and 286 maximum, minimum and mean MIR were most limiting for C. stellifer (Figure 4)

302
The distribution for C. stellifer was widely predicted across the state with presence

322
The models for C. debilipalpis also predicted the species across much of Florida.

323
However, this prediction is somewhat counter-intuitive to the data collected during the present 326 debilipalpis is present in the areas not predicted by the model.

327
The large areas absent of C. debilipalpis could also be explained by habitat preference.

328
Culicoides debilipalpis has been confirmed to develop in tree-holes of Salix spp. [47]. Although

339
[27], and the models predict low probability of C. venustus south of Pasco County (Figure 3).

340
NDVI variables, in particular, bi-annual cycle, bi-annual amplitude, amplitude, minimum, 341 maximum and mean of the NDVI, were most limiting for C. venustus out of all four species of for the distributions of C. venustus. Culicoides venustus is predominantly a northern species

344
[27], making its cold-tolerance higher than other more tropical species, such as C. insignis.

345
Similarly, EHDV is most common in deer and Culicoides spp. in northern Florida and the