基于Cross-Validation的福建杉木人工林最优树高曲线模型筛选研究

Selection of Optimal Height-diameter Models for Cunninghamia lanceolata in Fujian Province Based on Cross-Validation Analysis

由于树高测量中存在精度低、耗时长、难度大等问题,树高胸径模型常用来预测林木树高值。本研究归纳、总结了目前国内外31个常用的树高经验模型,采用福建杉木数据对所有模型进行了参数估计。并采用模型拟合效率、模型精度、模型偏度以及残差图对模型进行评估,初步筛选出4个性能较为优秀的模型,而上述评价指标不能甄别出最优模型。因此,结合16株解析木实测数据,采用交叉验证中的留一验证法,对初选的4模型进行进一步评价,最终确定Richards(模型23)为最优模型,为研究区域杉木树高精准预估提供依据。

Due to the problems existed in forest measurement, such as low accuracy, time consuming, and difficulty to conduct the measurement, models are wildly used to predict tree height and diameter. Thirty one such models that are normally used both nationally and internationally were collected, and the data of Cunninghamia lanceolata platations in Fujian Province were applied to fit these models. Preliminary as- sessment was conducted by using the indices, such as model efficiency,precision, and bias as well as the re- sidual plots, from which 4 models were screened out. Because the above mentioned indices could not further identify the optimal model,leave-one-out cross-validation analysis was adopted to further assess 4 models by using the measured data of 16 stem analytic trees. Richards model （model 23） showed the best perform- ance and was selected to be the optimal model that could provide the precise height and diameter perdition of C. lanceolata individuals in the study area.