Optimized hot spot analysis for probability of species distribution under different spatial scales based on MaxEnt model: Manglietia insignis case

Abstract

Whether a maximum entropy (MaxEnt) model constructed at one spatial scale is representative of species distributions at other scales is an important issue in the application and development of these models. Using distribution data for Manglietia insignis, we used the minimum convex polygon (MCP) method to model species distribution for three spatial scales-Three Parallel Rivers, Yunnan Province and China-with a 20 km buffer outside the distribution region. We built the MaxEnt model for Three Parallel Rivers, Yunnan Province and China using 19, 67, and 88 presence-only records respectively and combined these with data on environmental factors at the point locations. We estimated the prediction accuracy of the MaxEnt model using receiver operating characteristic (ROC) curve and omission rate (OR). Next, we used ArcGIS to analyze distribution trends for habitat suitability and potential hotspots. We identified the location of geometric centroid of potentially suitable areas using Zonal and used the Jackknife method to test the dominant environmental factors affecting the distribution of M. insignis. We found that the area under ROC curve (AUC value) for Three Parallel Rivers, Yunnan Province and China were 0.936, 0.887, and 0.930 respectively and OR values were 0.18, 0.15, and 0.20, indicating that MaxEnt model for all three spatial scales could successfully predict the distribution of M. insignis. Distribution trends of potential habitat suitability and habitat hotspots were consistent between different scales and were concentrated in the river basins of Dulong River, Nujiang River and Lancang River, with no significant zonal transfer for the location of geometric centroid. Different environmental factors affected the geographical distribution of M. insignis at the three spatial scales, suggesting scale dependence in the distribution patterns of M. insignis. In summary, this study indicates that MaxEnt model of M. insignisperforms stably and successfully for different spatial scales. In addition, the consistency of results across spatial scales became more obvious for hotspots, indicating that hotspot analysis greatly reduced the effect of spatial scale for the MaxEnt model. Thus, we propose integrating MaxEnt model and hotspot analysis to simulate the geographical distributions of species.