GAM模型在蝗虫地理格局分布研究中的应用——以黑河上游3种天然草地蝗虫为例

Application of GAM Approach on Pattern of Grasshoppers' Geographical Distribution: A case study in the upper reaches of Heihe River

将基于样本调查数据的群落-地形因子回归分析与地理信息系统支持下的昆虫属性空间格局预测结合,是昆虫-地形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。我们利用广义可加模型方法,定量分析了3种不同生态种蝗虫与地形因素的相关关系。结果表明:(1)3种不同生态种的蝗虫具有不同的模型结构、模拟效果以及结果的稳定性,反映了所受地形因子影响的差异。(2)蝗虫的广布种与地形因子关系最弱,局部地形的变化仅对其密度变化有影响;常见种受地形因子的影响明显高于广布种,同时受大尺度的海拔与微地形的剖面曲率影响;稀有种分布格局对地形条件的选择性最强,同时受限于海拔、坡度和坡向。(3)模拟结果对常见种以及稀有种的模拟全部有效;对广布种的预测基本失败。(4)模型预测变量的有效性和全面性决定了模型对数据的解释能力,非线性关系体现了蝗虫密度-地形指标的不确定性,除模型解释的变化外,其他生物因子、非生物因子以及随机因素也影响模型的可靠性。

Combining the field sampling-based regression analysis of community versus habitat attributes and GIS supported spatial prediction of topography factors is a new approach for the quantitative study of insect-environment relationship.As one of the non-parameter model types,Generalized Additive Models（GAM） is an efficient tool in regression analysis-spatial prediction,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.In this study,we employed the non-parametric GAM model to explore the potential distribution of the grasshopper richness in the upper reaches of Heihe River,western China.For each data requirement of GAM,6 topographic indices were extracted and analyzed.The results showed that：（1） The structure and deviation square values of the models were different for different ecospecies indices,so were the model stability,indicating their differences in response to the gradients of topographic indices.（2） The relationship between habitat cosmopolitan species and topographic indices was not significant,and only local topography affected the density;The relationship between habitat common living and topographic indices was closer than between habitat cosmopolitan species and topographic indices,the most prominent effects came from profile of the micro-topography and the elevation in large scale;habitat rare species distribution pattern relied on topographic indices most closely,and the most prominent effects came from elevation,plan and aspect.（3） The predictions of the density patterns of rare species passed the validation with an independent sampling data,that of common living succeeded partially,but the prediction for the density of cosmopolitan species failed.（4） Efficient and adequate predictive variables were crucial for the interpretative capability of the models.The environmental factors and random factor were significantly correlated with those of accuracy and maneuverability of valuation for model and algorithm of the tree-based method.So the GAM model function for the probability species richness made sufficiently use of topographic index information,and realized multiple factors.