Modelling the occurrence and spatial distribution of screwworm species in Northern Pakistan

Abstract

We use binomial kriging to model the spatial distribution of myiasis by three species namely Chrysomya bezziana, Wohlfahrtia magnifica and Lucilia cuprina in the livestock of Khyber Pakhtunkhwa, Pakistan. Traditional species distribution models are usually based on assumption of independence of observations. Species data often come in presence-only form for which background points are generated based on some covariates using statistical and machine learning techniques such as MaxEnt. We assume a symmetric binomial distribution based on the principle of maximum entropy in order to decide the number of pseudo-absences. Our results showed that the spatial models fitted very well and prediction distributions were estimated with excellent accuracy. Moreover kriging maps were more accurate as most of the non-spatial variation has been picked up by external drift with higher values of the sensitivity focusing partial AUC for all the three species. Land-use-land-cover was a common factor significantly affecting spatial distribution of all the three species suggesting that for established species anthropogenic factors such as land use become a strong determinant of their spatial distribution. Our results also revealed that for invading species like W. magnifica elevation acts as a barrier to species dispersal and therefore is more limiting to distribution. Furthermore the higher overall prediction accuracy demonstrated that our models performed well in predicting the distributions of the three species, which would lead to better understanding and management of the larval infestation.

Keywords: Binomial kriging; MaxEnt; Presence-only data; Pseudo-absences; SDM; Screwworm species.