Evaluation of Methods for Modeling Individual Tree Survival
Evaluation of Methods for Modeling Individual Tree Survival
复制标题
单树生存建模方法的评估
DOI：
--
复制
发表时间：
2017
期刊：
影响因子：
--
通讯作者：
Q. V. Cao
中科院分区：
文献类型：
--
作者：
Q. V. Cao
在这项研究中，评估了三种用于预测树木存活概率的Logistic回归模型的参数估计方法：最大似然法；针对林分存活优化的方法(林分优化)；以及针对林分和树木生存的优化方法(组合似然法)。我们考虑了四种解聚方法：不解聚；基于林分水平模型预测的解聚；基于综合估计器预测的解聚；以及基于观测林分存活率的解聚。在林分存活预测方面，令人惊讶的是，基于所有三个评估统计数据，针对林分存活优化参数的树生存模型比林分水平生存模型得分更高。当这种方法与林分水平的生存模型相结合形成一个复合估计器时，表现得更好。对于不解聚的林木存活率预测，最大似然法取代了林分优化法。根据所有评估统计数据，分解方法略微降低了未调整模型的性能。当用林分存活值进行调整时，无论是观测值还是预测值，林分优化法显然是三种参数估计方法中最好的。这些结果表明，应采用林分优化法对树木存活模型的参数进行估计。使用分解法的决定应取决于林分存活率预测是否超过某个最小阈值。解聚并没有改善这个数据集对树木存活的预测，可能是因为林分水平的预测没有达到那个可靠性阈值。
In this study, three methods of estimating parameters of the logistic regression model for predicting tree survival probabilities were evaluated: maximum likelihood; optimized for stand survival (stand optimization); and optimized for both stand and tree survival (combined likelihood). Four methods of disaggregation were considered: no disaggregation; disaggregation based on predictions from a stand-level model; disaggregation based on predictions from a composite estimator; and disaggregation based on observed stand survival. For stand survival prediction, surprisingly, the tree survival model with parameters optimized for stand survival scored better than the stand-level survival model, based on all three evaluation statistics. This method performed even better when it was combined with the stand-level survival model to form a composite estimator. For tree survival prediction without disaggregation, the stand optimization method was edged out by the maximum likelihood method. The disaggregation method slightly degraded the performance of the unadjusted model, based on all evaluation statistics. When adjusted using stand survival values, either observed or predicted, the stand optimization method was clearly the best among the three parameter estimation methods. These results showed that the stand optimization method should be used to estimate parameters of the tree survival model. The decision of using disaggregation should depend on whether or not the stand survival prediction passes a certain minimum threshold. Disaggregation did not improve prediction of tree survival for this data set, maybe because the stand-level prediction did not reach that reliability threshold.