Improving niche and range estimates with Maxent and point process models by integrating spatially explicit information. Issue 8 (28th April 2016)

• Record Type: Journal Article
• Authors: Merow, Cory; Allen, Jenica M.; Aiello‐Lammens, Matthew; Silander, John A.
• Other Names: Fortin Marie‐Josée checker.
• Is Part Of: Global ecology & biogeography. Volume 25:Issue 8(2016)
• Page Start: 1022
• Page End: 1036
• Publication Date: 2016-04-28
• Subjects: Data fusion -- dispersal -- ecological niche model -- expert map -- invasion -- maximum entropy -- sampling bias -- species distribution model; Ecology -- Periodicals; Biogeography -- Periodicals; Biodiversity -- Periodicals; Macroevolution -- Periodicals; 577
• DOI: 10.1111/geb.12453
• Languages: English
• ISSNs: 1466-822X

Abstract

Abstract: Aim: Accurate spatial information on species occurrence is essential to address global change. Models for presence‐only data are central to predicting species distributions because these represent the only geographical information available for many species. In this paper we introduce extensions to incorporate a variety of types of additional spatially explicit sources of information in Maxent and Poisson point process models. This spatial information comes from the output of other statistical or conceptual models. Innovation: Our approach relies on minimizing the relative (or cross) entropy (known as Minxent) between the predicted distribution and a prior distribution. In many scenarios, researchers have some additional information or expectations about the species distribution, such as outputs from previous models. Here, we show how to use this information to improve predictions of both niche models and spatial distributions, depending on what types of spatially explicit prior information is available and how it is incorporated in the model. Main conclusions: We illustrate applications of Minxent that include models for sampling bias, explicitly incorporating dispersal/other ecological processes, combining native and invasive range data, incorporating expert maps, and borrowing strength across taxonomic relatives. These applications focus on addressing biological scenarios where range modelling is extremely challenging – non‐equilibrium species distributions and rare and narrowly distributed species – due to data limitations. When data are limited, we are typically forced to make informal assumptions or lean on predictions of other models in order to obtain useful predictions; our applications of Minxent provide a formal way of describing these assumptions and connections to other models.