Modelling the spatial distribution and habitat suitability of Fasciola hepatica, a trematode parasite of domestic animals in Ukraine

Fasciolosis is a major problem in many parts of the world, including Ukraine. The disease is also a major economic burden, causing an estimated USD 3.2 billion yearly loss to the livestock industry. Today, the spread of fasciolosis is exacerbated by a number of factors, including climate change. For the planning of successful interventions against animal fasciolosis and to target populations living in high-risk areas, it is important to be able to determine the current spatial distribution of infection. In this respect, ecological niche modelling is a popular tool to examine the ecological and spatial limitations of species; however, it is not yet widely applied to parasites due to challenges in obtaining good and meaningful occurrence data. Using a recently compiled dataset (GBIF, https://doi.org/10.15468/dl.vd5r75) consisting of 335 records of Fasciola hepatica of domestic animals in Ukraine, we predicted the nationwide spatial distribution and habitat suitability of the fluke and responses to climate. Climate data were retrieved from the Chelsa project (https://chelsa-climate.org/bioclim). The ‘flexsdm’ R package (https://sjevelazco.github.io/flexsdm) was used for mapping the potential geographic distribution of the parasite across the country. A range of modeling machine learning options were tested. Eventually, an ensemble model was created based on the weighted average of the individual models and using established evaluation metrics showed good performance: AUC=0.82, TSS=0.58, Boyce index=0.86. In terms of high-risk areas, the top five provinces are Ivano-Frankivs'k, Vinnytsya, Ternopil', L'viv, and Chernivtsi. A SHAP library in R (https://github.com/pablo14) was used to understand the importance in the trained model of the climatic variables. SHAP offers advantages over other feature importance methods, including its model-agnostic nature. By calculating the average absolute values for each feature, the strongest influence on predictions had isothermality followed by an intricate balance of temperatures and sufficient precipitation.