由環境、競爭與物種特徵建立臺灣樹種樹高曲線

Abstract

樹高（HT）相對於胸高直徑（DBH）是較難測量的數據，而樹高曲線能由容易測量的DBH 去估計樹高，也就能補齊樹高資料，並更進一步的估計其他參數。若利用國家森林資源調查的資料建立不同樹種的樹高曲線，最大的困難在於一些紀錄稀少的物種，再加上林木的生長難以用單純的線性模型表達，因此本研究選用非線性混和模式加入共變量來建立樹高曲線。資料來自臺灣的第四次森林資源調查資料中 321 個具實際樹高測量數據的物種。選用的模型包含 power (POW), Wykoff (WYK) 及 von Bertalanffy-Richards (VBR)三種基礎模型。在樹高曲線中加入其他預測變數能提升模型的表現，亦即在混和模式中加入共變量，本研究包含海拔、坡度、坡向、木材密度、BAL、最大樹高、林分密度、林分截面積、針葉樹闊葉樹類別這幾項。在三種模型中，POW 有最低的 RMSE，其值為 2.1322。為解決不等變異性的問題，於建立模式的過程中會對 DBH 進行加權，其中僅VBR 因無法收斂而未加權。WYK 因具有中等 RMSE 值、簡單（僅兩參數）可收斂的加權模型與足夠的預測彈性，是本研究所得最為推薦的模型。共變量的標準化與測試具影響力的觀察值存在與否是在未來的研究中可以改進的地方。

Height-diameter(H-D) models can predict missing height from measured diameter at breast height (DBH) in existing forest inventory so that a full set of tree total height (HT) and DBH is available for assessing forest resources. The few observations of rare species in national forest inventories (NFIs) made it difficult to develop H-D models for these species. Thus, this study used non-linear mixed-effects models with covariates to develop species-specific H-D models for the 321 species with measured HT observations from NFI4 in Taiwan. Power (POW), Wykoff (WYK), and von Bertalanffy-Richards (VBR) models were evaluated. The addition of other predictor variables to HD models may improve the prediction of height, which are covariate in mixed effects models. In this study, elevation (elev), slope, aspect, wood density, sum of basal area per ha larger than i-th tree (BAL), maximum tree height (hmax), stand density per ha (sden), stand basal area per ha (sba) and class (coniferous or deciduous species) were analyzed. Above the three models, POW has the lowest RMSE of 2.1322. Only VBR weighted model failed to converge. WYK may be the suggested model due to the medium RMSE, simplicity of 2 parameters only, and enough flexibility for prediction. Standardization of variables and testing influential points should improve this study in the future.