Stability Assessment And Application Software Of Species Distribution Modeling

The scientific questions of how and where species distribute is important. Species distribution models(SDMs) is a useful tool to solve these questions. SDMs can predict species potential distribution by quantifying species-environment relationships. SDMs have been studied by many scientists and have a wide range of applications, including capturing the change distribution and abundance under disturbance, predicting the potential direction for invasion species, supporting conservation planning and reserve selection, understanding the mechanisms of species distribution, and forecasting effects of changing environments(e.g. climate change, habitat fragmentation, geographic barriers), and so on.To evaluate the predictive performance and stability of SDMs, we collected the distributional data of two conifer-leaved tree species(Pinus massoniana and Pinus yunnanensis), three broad-leaved tree species(Betula platyphylla, Quercus wutaishanica and Quercus variabilis) and 26 environmental variables to simulate and predict the potential distribution area of the species by using six kinds of common SDMs(BIOCLIM, DOMAIN, MAHAL, RF, MAXENT, and SVM). The results are as follows:(1) In the case of repeat 100 times, the mean AUC values and mean Kappa values of MAHAL, RF, MAXENT, and SVM are similar, which are higher significantly than BIOCLIM and DOMAIN(p<0.05).(2) The standard deviation and the variable coefficient of AUC and Kappa for BIOCLIM and DOMAIN are higher significantly than MAHAL, RF, MAXENT, and SVM(p<0.05). The confidence interval for MAHAL, RF, MAXENT, and SVM are higher significantly than BIOCLIM and DOMAIN.(3) From the AUC and Kappa density curve, we can find the peak values of MAHAL, RF, MAXENT, and SVM are higher than that of BIOCLIM and Domain. Compared to BIOCLIM and DOMAIN, the other four SDMs(MAHAL, RF, MAXENT, and SVM) have higher prediction accuracy, stronger forecasting, smaller confidence interval, more stable and less affected by the random variable. according to the prediction performance and stability of SDMs, we can divide the six SDMs into two categories: the high performance group, such as MAHAL, RF, MAXENT, and SVM; the low performance group, such as BIOCLIM and DOMAIN. It is important to easily and efficiently obtain high quality species distribution data for predicting the potential distribution of species using species distribution models(SDMs).To study species distribution, we need a lot of distribution data. The GBIF database provides the main distribution data. It will need much time to download, check the species name and distribution data, which limits the application of species distribution model. Therefore, it is necessary to develop a software than can automatic or semiautomatic download distribution data from GBIF database. We developed a software SDMdata that can implement these functions. SDMdata has the characteristics including: 1) Python language, high performance, low memory consumption, simple operation process(similar with the article submission system). 2) Based on the network(the user is free to use and management avoiding the tedious installation); 3) Suitable for a variety of operation platform and devices(such as desktop computer, or even mobile phones). 4) SDMdata is open source. All the code is available as a free download and use from <http://www.sdmserialsoftware.org/sdmdata/>.The presence data of species distribution model are come from herbarium, database, etc., so there is uncertainty, which limits the application of species distribution model. Virtual species is a highly efficient way to allow researchers to control the quality of the input data and amplitude. Virtual species can solve many problems that real species can not solve. For example, virtual species can ensure that the study result is not affected by different species and species’ characteristics. With the increase of using virtual species, it is necessary to provide a simple, intuitive and standard software to create virtual species. Here, we provide a software package to create virtual species. SDMvspecies is based on the R language package. Theoretically, SDMvspecies can meet any platform supporting R software. In our study, the installation and test of SDMvspecies is mainly in Linux platform. SDMvspecies need to install the rasterize software package for processing rasterize maps to meet the species distribution analysis. SDMvspecies(current version 0.2.1) contains four kinds of ways to make virtual species(niche synthetic method, average method, the median method and artificial bell-shaped curve method). You can freely get it from http://cran.r-project.org/web/packages/sdmvspecies/.