作 者 ：沈泽昊,赵俊

期 刊 ：生态学报 2007年 27卷 3期 页码：953~963

关键词：通用可加性模型；数字高程模型；地形变量；物种丰富度；空间预测；

Keywords:GAM, DEM, topographic variables, species richness, spatial prediction,

摘 要 ：将基于样本调查数据的群落-生境因子回归分析与GIS支持下的植物属性空间格局预测结合起来，是国际上植被-环境关系定量研究的新途径。通用可加性模型（GAM）的非参数属性使之具有对不同数据类型的广泛适应性，成为这种“回归分析+空间预测”途经的有效手段；不同程度上依赖于数字高程模型的环境空间数据集是实现空间预测的必要条件。介绍了这一新的研究途径，并应用于案例研究区域植物多样性指标空间格局的预测和分析。野外调查的一组样方地形特征指标和植物多样性指标（包括样方物种丰富度及乔木、灌木、草本、常绿木本、珍稀种类的丰富度），分别作为预测变量和响应变量，建立GAM模型。结合研究区域10m分辨率的数字高程模型，对该区域植物物种丰富度的空间格局进行空间预测，并对预测模型和结果进行统计分析和检验。结果表明：（1）不同的多样性指标具有不同的模型结构和模拟效果，重复模拟的结果稳定性也不同，反映了所受地形因子影响的差异；（2）影响各多样性指标空间格局的地形变量主要是坡位和坡度等小尺度特征，大尺度海拔因素的影响并不显著；（3）模拟结果与独立检验数据的相关分析表明，对乔木种、草本种、珍稀种的模拟全部有效；对常绿种和样方物种总数的模拟部分有效；而对灌木种丰富度的预测基本失败。（4）模型预测变量有效性和全面性决定了模型对数据的解释能力，样本大小对模型的稳定性和可靠性也有显著影响。就地形因子对生境条件的代表性、模拟误差的来源及GAMs模型的优缺点和应用前景进行了讨论。

Abstract:A new approach is forming internationally for the quantitative study of vegetation-environment relationship, by combining the field sampling-based regression analysis of community vs. habitat attributes and GIS supported spatial prediction of biological indices. As one of the non-parameter model types, Generalized Additive Models GAMs is becoming an efficient tool in the "regression analysis-spatial prediction" approach, partly for its flexibility to a wide variety of data types. A spatially explicit database for environment factors that frequently rely on digitalized elevation model is the requisite background for spatial prediction. This approach is introduced and applied here, in a prediction and analysis of the spatial pattern of a group of biodiversity indices in the study area. Species richness of sampling plot (include 6 categories as total species, tree species, shrub species, herb species, evergreen woody species, and rare species) was used as responsive variables, with five topographic indices was predictive variables in GAMs. Spatial environment database came from DEM of predicting area with 10m resolution. Model analysis and validation were done to the results of modeling. The results suggest that: (1) The structure and D2 values of models are different for different biodiversity indices, so are the model stability in modeling repeats, indicating their differences in response to the gradients of topographic indices. (2) Generally, the most prominent topographic effects come from slope position and slope angle, which contribute to habitat heterogeneity mainly at smaller scale, while the effect of elevation, acting at a much larger scale, is not always significant. (3) The predictions of the richness patterns of tree species, herb species, and rare species passed the validation with an independent sampling data, that of evergreen woody species and total species succeeded partially, but the prediction for the richness of shrub species failed. (4) Efficient and adequate predictive variables are crucial for the interpretative capability of the models, the sample size also has prominently effects on the robustness of GAM. Discussion is then made about the efficiency of topographic indices as habitat indicators, sources of error, the advantage/disadvantage and potential applications of GAMs.