红海榄胎生繁殖体产量预测细分聚合模型

摘要

The prediction model of viviparous propagules of Rhizophora stylosa is important for afforestation and management of mangrove forests.Heterogeneity is part of the most important factors leading to reduced model accuracy.How to solve the heterogeneity problem,become the key to build the model.Segmentation is one of the means to solve heterogeneity.In this paper selected the Rhizophora stylosa viviparous propagules of Gaoqiao mangrove reserve in Zhanjiang city,Guangdong province and Dongzhaigang mangrove reserve,Qinglangang mangrove reserve in Hainan province as the study objects.In this paper adopts multi-level multiindicators for investigation,According different regions to subdivide prediction model,and then use the aggregation to builds the forecasting model.There were significant correlations between the growth of viviparous propagules and crown,leaf width,seed width,leaf area index,leaf shape index and ground diameter.The results showed that the reproductive growth ofRhizophora stylosa was closely related to its phenotypic traits.The finalized three subdivision model were significant by t test,R2 were 0.894,0.629,0.701.Through the path analysis,indirect effects of the X11 on the Rhizophora stylosa viviparous propagules number and the model construction was maximum.All the models directly affected the number ofRhizophora stylosa viviparous reproductive and model construction by crown width.Under normal circumstances(ie,communities and tree species are healthy) It is more practical to use the subdivision method to segment the forecasting model by region than the synthetic forecasting model.It is more realistic to construct a forecasting model by use subdivision according to region.%红海榄胎生繁殖体产量预测模型的构建,对于红树林造林和经营管理有着重要意义.异质性是导致模型精度降低一个极其重要的因素.如何解决异质性问题,成为构建模型的关键.细分方法是解决异质性的手段之一.本研究选择广东省湛江市廉江高桥红树林保护区、海南东寨港红树林保护区以及海南清澜港红树林保护区的红海榄胎生繁殖体作为研究对象,采取分层次多指标进行调查,按照不同区域的细分预测模型,然后再采用聚合的方法构建预测模型.红海榄胎生繁殖体产量与冠幅、叶宽、种宽、叶面积指数、叶形指数、地径呈显著相关关系,表明红海榄胎生繁殖体产量与其表性状关系密切,最终构建的3个细分模型R2分别为0.894、0.629、0.701,经t检验均呈显著性.文章通过通径分析可知3个细分模型皆通过冠幅来直接影响红海榄的胎生繁殖体结种数量和模型构建.在正常情况下(即群落和树种皆为健康)采用细分的方法按区域构建预测模型更具现实的优势.