Mikrubi: a model for species distributions using region‐based records

Abstract

Many species occurrence records from specimens and publications are based on regions such as administrative units. These region‐based records are accessible and dependable, and sometimes they are the only available data source; however, few species distribution models accept such data as direct input. In this paper, we present a method named Mikrubi for robust prediction of species distributions from region‐based occurrence data and a Julia package implementing the algorithms. The package ‘Mikrubi' requires a map describing disjoint regions, climatic raster layers, and a list of occupied regions. Mikrubi then rasterizes the regions, reduces the environmental dimensionality, parameterizes the niche, and finally estimates the parameters by maximizing the likelihood. In a simulation study, we find Mikrubi effective in accurate estimation in most cases; in a case study of Allium wallichii in China, Mikrubi significantly outperforms four modeling strategies that adapt region‐based records to conventional models according to different principles. The package has many prospective applications in addition to modeling distributions on region‐based records: 1) it accepts supplementary coordinates; 2) it is a new solution for distribution modeling using deviated coordinates; and 3) its probabilistic region‐based outputs have special uses in conservation and biodiversity science.