全球生物多樣性喪失亟需超越種苗數量等表面指標的標準化量測機制。本研 究建構一套多時序遙測評估框架，用以檢定造林的生態反應是否可被偵測，對象為 一個台灣非營利組織(NPO)於蒙古國「一億棵樹」運動下設立的三個造林樣點，橫跨 冷半乾旱草原(BSk)、乾冬冷夏大陸性(Dwc)與冷沙漠(BWk)三個氣候帶，並明確對 應「對齊自然會計方法」(Align， 2022)國際標準。 研究於 Google Earth Engine 中整合 Sentinel-2 影像、隨機森林地表覆蓋分類 (整體精度 87.76%)、地景結構指標與 CASA 生產力模式，以量測生態系統範圍 (Extent)及生態系統狀況的結構面與功能面;造林效果則以中斷時間序列設計，在前 後等長窗期上相對於反事實基線估計，並以降水為穩健性控制、NDVI 為敏感度對 照。 結果依構面而分歧。Extent 與 Structure 於兩個有植被的樣點呈溫和正向:木 本面積擴張，斑塊整合為更大、更具優勢的林分;冷沙漠樣點未偵測到木本植被，其穩 定草地較宜判讀為對沙漠化的抵抗。Functioning 則於任一樣點皆未確立效果:原始 前後比較中最大的生產力增幅，經分解後淨效果與零無法區分;唯一的邊緣訊號在 控制降雨後消散;NDVI 亦從未達顯著。 此 null 反映可偵測性而非造林失敗:林齡僅二至四年、苗木冠幅相對於 10 公 尺像元過小、水分變異主導生產力訊號。本框架的貢獻在於合併報告 Align 各構 面、並以反事實檢定介入效果，證明結構增益不得被轉換為功能宣稱，可作為 NPO 造林計畫在 TNFD 及昆明—蒙特婁全球生物多樣性框架下報告的可擴展模板。;Global biodiversity loss demands standardized mechanisms for quantifying restoration outcomes beyond seedling-count metrics. This study constructs a multi-temporal remote-sensing framework to test， rather than presume， whether an afforestation effect is detectable at three sites established by a Taiwan-based non-profit organization (NPO) under Mongolia′s "One Billion Trees" National Movement， spanning cold semi-arid steppe (BSk)， dry-winter cold-summer continental (Dwc)， and cold desert (BWk) zones， and explicitly aligned with the Align (2022) standard for corporate biodiversity measurement. Sentinel-2 imagery， a Random Forest land-cover classification (overall accuracy 87.76%)， landscape-structure metrics， and the CASA productivity model are integrated in Google Earth Engine to instrument Ecosystem Extent and the structural and functional facets of Ecosystem Condition. The afforestation effect is estimated by an interrupted time-series design against a counterfactual baseline over symmetric pre- and post-intervention windows， with precipitation as a robustness control and NDVI as a sensitivity comparator. The results diverge by dimension. Extent and Structure are modestly positive at the two vegetated sites， where woody area expanded and patches consolidated into larger， more dominant stands; the cold-desert site records no woody vegetation， its stable grassland better read as resistance to desertification. Functioning shows no established effect at any site: the largest raw pre/post increase in productivity resolves into a net effect indistinguishable from zero， the one marginal signal dissolves once rainfall is controlled， and NDVI never attains significance. The null reflects detectability rather than intervention failure: stands are two to four years old， crowns small against a 10 m pixel， and moisture variability dominates the productivity signal. By reporting the Align dimensions jointly and testing the intervention against a counterfactual， the framework shows that a structural gain cannot be converted into a functional claim， offering a scalable template for NPO-led afforestation reporting under the TNFD and Kunming–Montreal Global Biodiversity Framework.
