以葉面積指數作為代謝縮放理論指標預測山地雲霧林枯落物:結合光達與貝氏模型

Abstract

枯落物（litterfall）由樹葉、細枝、樹皮以及繁殖構造等組成，是森林生態系中碳與養分循環的重要途徑，在熱帶與亞熱帶森林中，其貢獻可高達淨初級生產量（NPP）的三分之一。然而，傳統上以地面樣區為基礎的枯落物監測方法不僅勞力密集，且在地形複雜的環境中（如亞熱帶山地雲霧林）適用於大面積區域。因此本研究旨在結合森林結構特徵與生態理論，探索在異質性景觀中推估枯落物的可行性與尺度延伸性。 本研究以台灣東北部24,400公頃的棲蘭山區為研究場域，結合 2021 至 2023 年的地面觀測資料、空載光達（LiDAR）數據與貝式階層模型，分析枯落物的時空動態變化。我們首先建立隨機森林回歸模型，利用光達推算的最大樹冠高度與冠層粗糙度預測葉面積指數（LAI），其預測表現足以信賴（OOB R-sqaured = 0.61）。接著，基於代謝尺度理論（Metabolic Scaling Theory, MST）建構模型，使用 LAI 預測枯落物產量，並將生物量的周轉率區分為葉片、細根與莖部組成。推估結果中，LAI (葉片生物量)與莖部生物量之間的縮放指數 p 為約 0.91，略高於 MST 所預期的值 (0.75)，顯示本區域的樹種可能相較於全球森林平均，更傾向於將碳分配至葉片，具有較高的冠層碳投資比例。 本研究提供了一個具規模化潛力以及機制基礎的碳動態推估框架，不僅能進行空間外插，亦能納入模型不確定性，特別適用於資料受限的山地森林生態系研究。

Litterfall, comprising leaves, twigs, bark, and reproductive structures, is a key pathway of carbon and nutrient turnover in forest ecosystems and can account for up to one-third of net primary production (NPP) in tropical and subtropical forests. However, conventional field-based litterfall monitoring methods are labor-intensive and making it impractical for a large region, especially in topographically complex environments such as subtropical montane cloud forests (MCFs). Therefore, this study investigates the use of structural forest traits and ecological theory to scale litterfall estimation across heterogeneous landscapes. Focusing on the 24,400 ha Chilan Mountain of northeastern Taiwan, this study integrates field observations (from 2021 to 2023), airborne LiDAR data, and Bayesian hierarchical models to analyze the spatiotemporal dynamics of litterfall. We first constructed a linear model to predict Leaf Area Index (LAI) using LiDAR-derived maximum canopy height and canopy rugosity, achieving robust performance (OOB R-squared = 0.61). Subsequently, LAI was used as an input to a Metabolic Scaling Theory (MST)-based model for estimating litterfall production, in which biomass turnover is partitioned into leaf, fine root, and stem components. The estimated scaling exponent p that links LAI (leaf biomass) to stem biomass was approximately 0.91, deviating from the theoretical predictions of MST (0.75), and suggesting a relatively higher investment in canopy carbon among trees in the study area compared to global averages. Our approach enables spatial extrapolation of litterfall while accounting for uncertainty, offering a scalable, mechanistic framework for understanding carbon dynamics in data-limited montane ecosystems.