基于组合方法的物种生境模拟预测可靠性与物种特征的关系

There is now a plethora of methods for modelling species distribution with the rise of new powerful statistical techniques and GIS tools. These species distribution models (SDM) do not perform equally in predicting species distribution due to their conceptual and technical shortcomings. This challenges the common practices of relying on single niche-based model to management resource or design conservation areas. Ensemble forecasting framework has widely been applied in mathematical statistics and climatology and demonstrated that forecast accuracy can be substantially improved through the combination of multiple individual forecasts. However, little is known about the relative performance of ensemble forecasting approach in reducing uncertainties in species distribution modeling. Moreover, numerous studies has been emphasized that results derived from SDMs are not equally reliable for all species and that the best performing models are not always the same for different species. A general pattern is that species with a restricted distribution range tend to show higher predictive accuracy than species with a wide range of distribution. If ensemble modelling and consensus approaches are expected to be increasingly used in conservation and management planning studies, thorough examinations of their performances for species with different ecological and geographical characteristics are necessary. The main aims of this project are to investigate the adequacy of consensus approaches in reducing uncertainty and improving predictive accuracy in species distribution modelling, and whether species traits influence the performance of such approaches. Here, using 8 niche-based models, 9 split-sample calibration bouts (original data were randomly divided into two sets: a calibration set and a validation set), and 9 climate change scenarios, we simulated current distributions (1961-1990) and projected the future distributions of 22 tree species in China for 3 time slices (2020s, 2050s, 2080s). A total of 648 projections were performed for each species and each time period. Model performance was tested by Cohen's-k test, True skill statistic (TSS), and receiver operator characteristic curve (AUC). We combined ensembles of forecasting to get final consensual prediction maps for target species using five different consensus approaches. Pearson's correlation and Cohen's-k test were used to quantify the similarity of five consensual prediction maps. We described species attributes by using six species geographical and environmental characteristics. We used molecular data to reconstruct phylogenetic relationships for the tree species. After accounting for the effects of phylogenetic relatedness and species prevalence, these six species attributes were related to the observed variations in both consensus among SDMs and predictive performances by using generalized estimation equations.

随着计算机和统计学的发展，基于生态位理论的物种分布（生境）模型层出不穷。然而，由于其概念和技术上的缺陷，致使其在模拟预测物种分布时存在极大不确定性，这极大影响物种分布模型在资源管理、物种保护等领域的应用。近期，组合预测方法在经济学和气象学等领域有广泛应用且成果显著，其基本思想是通过采用多套建模数据、多套模型、多套模型参数构建一个组合模拟框架结构来分析模拟预测的不确定性，进而采用一致性预测方法把不确定性控制在一个合理范围之内，最终达到提高模型预测精度并降低模拟预测不确定性的目的。本项目通过构建组合预测模拟框架结构，系统检验组合方法降低物种分布模拟预测中不确定性的能力，着重研究物种特征与组合方法预测精度的关系。通过本项目的实施，系统评估组合方法在物种分布模拟领域的应用潜力，探索物种分布模拟研究的新途径，改善不同误差来源对物种分布模拟预测精度的影响，从而提高物种分布模拟预测的可靠性。

随着计算机和统计学的发展，基于生态位理论的物种分布（生境）模型层出不穷。然而，由于其概念和技术上的缺陷，致使其在模拟预测物种分布时存在极大不确定性，这极大影响物种分布模型在资源管理、物种保护等领域的应用。近期，组合预测方法在非生态学领域(如，统计学、气候学)有广泛的应用，其基本思想是通过采用多套建模数据、多套模型、多套模型参数构建一个组合模拟框架结构来分析模拟预测的不确定性，进而采用一致性预测方法把不确定性控制在一个合理范围之内，最终达到提高模型预测精度并降低模拟预测不确定性的目的。为此，本项目基于组合模型预测框架结构，通过采用大样本建模方法，采用八个物种分布模型，九套不同的物种分布建模数据，三个大气环流模型和三个温室气体排放情景，对我国三十二个典型树种的地理分布区开展了预测研究。发现物种建模过程中模型不确定性似乎无处不在，物种分布模型预测精度与建模初始条件（建模数据）有关；树种之间和物种分布模型之间预测精度存在差异；概率预测结果转换为二元制是物种分布模拟研究中的一个重要不确定性来源；物种分布模型具有一定的空间尺度依赖性；物种-环境变量之间的关系并不具有恒定性，精细空间尺度并不意味着高预测精度，选择合适的空间尺度有利于改善模型预测精度；预测变量的重要性与其响应曲线的变化强度密切相关。进而采用三个不同的一致性预测方法对预测不确定性进行集中改进，发现气候变化条件下物种分布区的一致性预测具有空间不确定性，物种分布区一致性预测的可靠性与物种特征和模型预测精度有关，并且组合预测方法对广域物种分布区预测的一致性(可靠性)要大于狭域物种；组合预测方法虽然不能取得最大的预测精度，但却具有精度最稳健最保守的特征,因此在物种分布模拟研究领域具有一定的应用潜力。