虚拟濒危松科植物的生态位模型及其潜在分布的预测

Monitoring populations of endangered species have become a priority for most conservation agencies. The key scientific questions on where endangered species potential distributions inspire many biogeographers and ecologists to seek explanations. Ecological niche models (ENMs) are powerful approaches on predicting species potential distributions by quantifying species-environment relationships. ENMs are based on high quality data, but the real data of endangered species have many problems (e.g. sparse data, sampling bias) which affects the application of ENMs. Recent advances in ENMs enable us to use virtual species to control species characteristics, assess the prediction accuracy of ENMs and forecast the synergistic effect of environment and other factors. Virtual species provide a suitable unifying framework to select the most appropriate model by comparing predictive accuracy and virtual distributions in a geographic information system of a real landscape. Virtual species have many advantages including easily attaining a large number of data sets for each scenario, fully controlling the quality of data, avoiding the over-fitted phenomenon of ENMs and independently evaluating the predictive power of ENMs regardless of other factors.. Larix chinensis is an endangered species in China distributed only at an altitude of 2700 m to 3500 m in the Qinling Mountains, and it was densely distributed in Taibai, Liubei, Guangtou and Yuwang Mountains. Pinus dabeshanesis, an extremely small population, only distributes in Dabie Mountains. In order to evaluate the potential distributions of the two endangered pinaceae plants, we illustrate an approach to improve sample methods and optimize simulation progress basing on virtual species. We generate two virtual endangered pinaceae plants in real geographic habitat with similar distributions of L. chinensis and P. dabeshanesis, respectively. To gain high-quality distribution data, we compare different sampling methods, such as stratified sampling to improve the numbers of presence data, sampling in different distribution areas to increase the data representation on environments. To gain the best simulated methods, we evaluate some key factors, such as environment factors, space factors (spatial autocorrelation and space resolution), methods of absence data and threshold values on the prediction accuracy and stability of 10 ENMs including GLM (Generalized linear model), GAM (Generalized additive model), ANN (Artificial neural network), GARP (Genetic algorithm for rule set production), Bioclim, Domain, Mahal, RF (Random forest), Maxent (Maximum entropy) and SVM (Support vector machine). Then, we improve field sampling methods for L. chinensis and P. dabeshanesis, and choose the right simulation progress to predict their potential distribution. Taking into account the complex simulation process, we develop ENMs simulation software to predict potential distributions of endangered species.

基于生态位模型预测濒危植物潜在分布对物种保护起关键作用。但是，濒危植物，特别是濒危松科植物的分布较少或存在抽样偏差，造成模型的预测精度低，限制了生态位模型在植物保护中的应用。针对濒危植物不同的分布特点，分别在真实生境下构建与其类似分布的虚拟濒危松科植物，比较不同的抽样方法（分层抽样提高数据质量，从不同分布区抽样提高数据代表性），综合考虑环境因子、空间因素（空间自相关和空间分辨率）、缺失数据和阈值标准等关键因素对模型性能的影响，评估10种常用的生态位模型（GLM、GAM、ANN、Bioclim、Domain、GARP、Mahal、RF、Maxent和SVM）的预测精度与稳定性，构建最优的抽样和模拟方法。在此基础上，改进濒危物种的野外抽样，选择合适的生态位模拟方法，预测二者潜在的分布，并开发一个基于虚拟物种优化濒危物种潜在分布的软件。

基于生态位模型预测和评估濒危物种潜在分布对物种保护起关键作用。但是，濒危物种已知的分布点较少或存在抽样偏差，造成模型的预测精度与稳定性较低，限制了生态位模型在濒危物种保护中的应用。本项目主要通过真实物种与虚拟物种相结合的方法，预测和评估了不同抽样方法、环境因子和空间因素等对10种常用的生态位模型（GLM、GAM、ANN、Bioclim、Domain、GARP、Mahal、RF、Maxent和SVM）的预测精度与稳定性的影响，提出了一种模拟濒危松科植物潜在分布的优化方法。通过对比求和法、求积法，认为物种对环境变量具有独立的生理耐受性；为提高模型的生态学意义，提出了虚拟物种构成的新算法（限制因子法）。对于很多种群来说，密度很小同样也不利于极小种群生存和分布，基于种群的负竞争效应，改进和模拟了三种具有Allee效应的动力学种群分布模型。考虑到气候变化影响着物种的潜在分布，综合考虑种群生境破碎化的协同作用，预测和评估了气候变化和种群破碎化对物种潜在分布的协同影响，认为二者的协同作用是把双刃剑，会放大影响效果。结合松科植物群落内植物的遗传多样性、微环境多样性、光合特性动态，探讨了影响物种潜在分布和多样性共存的其他因素。在此基础上，优化和开发了相应的物种潜在分布预测软件和濒危物种分布信息系统。