運用 Sentinel-2 與 Landsat 進行年際與年內森林變化偵測及地上生物量動態監測Inter- and Intra-Year Forest Change Detection and Monitoring of Aboveground Biomass Dynamics using Sentinel-2 and Landsat
Pelletier et al.｜Remote Sensing of Environment 301: 1-18｜DOI: 10.1016/j.rse.2024.113931
狀態：AI_DRAFT_FROM_REVIEW｜分級：A｜閱讀深度：DERIVED_FROM_SEMINAR_SLIDE｜Jacky 審核：False
AGB dynamicsforest change detectionintra-annual monitoringSentinel-2LandsatHLS fusiontime seriesdisturbance classificationCanada
專討核心文獻定位
[04]
Ch3 · 光學遙測
TIIC 演算法融合 Sentinel-2 與 Landsat，實現加拿大全國尺度年內 AGB 動態監測，精度達 99%
中文翻譯付費原文 PDF付費原文 TXT付費導讀 PPTX付費
使用警示
本頁是文獻知識庫卡片，不等於可直接引用的最終查核稿。只有狀態升級為 CITABLE 後，才可直接進入論文引用候選。
為什麼納入這篇
This paper is a core source for the optical remote sensing chapter, demonstrating that high-frequency Sentinel-2 and Landsat fusion can move AGB estimation from a single annual map toward near real-time intra-annual dynamic monitoring at national scale.
結構式摘要｜中英文對照
| 研究問題 | 如何利用 Sentinel-2 與 Landsat 的高頻時間序列，同時偵測年際與年內的森林變化，並定量地上生物量的動態？ How can high-frequency Sentinel-2 and Landsat time series detect both inter-year and intra-year forest change and quantify aboveground biomass dynamics? |
|---|---|
| 資料來源 | 導讀素材確認：研究融合 Sentinel-2（10m, 5 日重訪）與 Landsat-8（30m, 16 日重訪）成 HLS 標準化產品，融合後重訪頻率達 2.9 天；時間範圍為生長季 May 30 至 Sept 1，每個像素生成 50-60 張影像的時間序列，覆蓋加拿大 >650 Mha 林地。 The seminar material confirms the study fuses Sentinel-2 (10m, 5-day revisit) and Landsat-8 (30m, 16-day revisit) into HLS standardized products, reaching a 2.9-day revisit frequency. The growing-season window runs from May 30 to Sept 1, generating a 50-60 image time series per pixel over more than 650 Mha of Canadian forest. |
| 方法 | 提出 TIIC（Tracking Intra- & Inter-year Change）演算法：整合輸入資料成 HLS、對每個像素做時間序列分析並偵測突變點（breakpoint）、依突變模式分類擾動（火災為快速下降、採伐為漸進式下降）、結合 AGB 基準圖與變化發生時間定量年內損失與同期生長增量，最後以高解析衛星（Planet、DigitalGlobe）做獨立驗證。 The paper proposes the TIIC (Tracking Intra- and Inter-year Change) algorithm. It integrates inputs into HLS, runs per-pixel time series analysis to detect breakpoints, classifies disturbances by breakpoint pattern (fire as rapid decline, harvest as gradual stand-replacing change), combines an AGB baseline map with the timing of change to quantify intra-year loss and concurrent growth gain, and validates against high-resolution satellites such as Planet and DigitalGlobe. |
| 主要結果 | 導讀素材確認：以 2019 年加拿大全林區為案例，TIIC 達 99% 變化偵測精度與 99% 擾動分類精度，能區分火災與機械採伐；機械採伐占 AGB 損失 36%（冬季集中），火災占 64%（全年分散）；全林區淨 AGB 增長 +2.54%；變化偵測精度隨季節累積由初期 23% 提升至季末 98%。 Using all of Canada in 2019 as a case, TIIC reaches 99% change detection accuracy and 99% disturbance classification accuracy, separating fire from mechanical harvest. Mechanical removal accounts for 36% of AGB loss (concentrated in winter) and fire for 64% (spread across the year). Net AGB change across the forest area is +2.54%. Change detection accuracy rises through the season from 23% early to 98% by the end. |
| 限制 | 導讀素材列出的局限包括雲層覆蓋影響可用影像、Landsat 30m 解析度使小樣地無法偵測、地形複雜或背光面可靠性下降、部分採伐與火災邊緣情況易混淆、AGB 定量依賴基準 AGB 圖精度。 Stated limitations include cloud cover reducing usable imagery, Landsat 30m resolution missing small patches, lower reliability on complex or shaded terrain, confusion between partial harvest and fire edge cases, and AGB quantification depending on the accuracy of the baseline AGB map. |
Key Findings
| 發現 | 證據 | 確定性 |
|---|---|---|
| TIIC fuses Sentinel-2 and Landsat into a dense time series and tracks both inter-year and intra-year forest change at national scale. | Seminar slides: HLS fusion reaches 2.9-day revisit and 50-60 images per pixel over >650 Mha of Canadian forest in the 2019 growing season. | derived_from_seminar_slide |
| The method not only detects change but attributes it, separating fire from mechanical harvest, and quantifies the intra-year balance of growth versus loss. | Seminar slides: 99% disturbance classification accuracy; mechanical removal 36% vs fire 64% of AGB loss; net AGB change +2.54% in Canada 2019. | derived_from_seminar_slide |
Key Figures and Tables
公開網站原則：未確認授權前，不直接複製原文圖表；優先使用自製圖表導讀或重繪圖。
| 項目 | 內容 | 關鍵數字 | Jacky 判讀 | 重用策略 |
|---|---|---|---|---|
| 待查 | Seminar material describes the TIIC workflow (input fusion, time series breakpoint detection, disturbance classification, AGB dynamics, validation) and the 2019 national-scale AGB monitoring results. | 2.9-day revisit; 50-60 images per pixel; >650 Mha; 99% change detection and 99% disturbance classification accuracy; mechanical removal 36% vs fire 64% of AGB loss; net AGB +2.54%; accuracy progression 23% to 98%. | 把這套年內動態框架對應到台灣颱風季林分破壞評估，視颱風機制類似火災的快速下降型擾動。 | Redraw a simplified TIIC workflow diagram after checking original figure/table and license; original is Elsevier so do not reuse figures until license confirmed. |
Extracted Evidence Table
| 可支撐主張 | 指標或結果 | 原文位置 | 可引用 | 備註 |
|---|---|---|---|---|
| High-frequency Sentinel-2 and Landsat fusion enables near real-time intra-annual AGB monitoring at national scale. | 2.9-day revisit over >650 Mha; 99% change detection accuracy; 99% disturbance classification accuracy; net AGB +2.54% in Canada 2019. | 待查（頁碼需回原文確認；導讀素材標示 pp. 1-18） | True | Numbers traceable to the seminar slide deck; confirm exact page locations against the original PDF before fixing citations. |
Critical Appraisal
Strengths
- First large-scale demonstration of intra-annual AGB dynamic tracking using free optical satellites.
- Disturbance attribution distinguishes fire from mechanical harvest, supporting management decisions.
- Near real-time and continental coverage at low cost.
Weaknesses
- Cloud cover and Landsat 30m resolution limit detection of small or obscured patches.
- AGB quantification depends on baseline AGB map accuracy.
- Partial harvest versus fire edge cases can be confused.
| Validation quality | independent validation against high-resolution satellites (Planet, DigitalGlobe) reported in seminar material; full quantitative detail 待查 |
|---|---|
| Transferability to Taiwan | high for typhoon damage assessment, protection-forest monitoring, and seasonal carbon accounting |
| Risk of overclaiming | Do not claim year-round monitoring; the demonstrated window is the growing season (May 30 to Sept 1), with full-year monitoring listed as future work. |
與 Jacky 博論 / Review 的用途
| 博士論文 | Supports the dissertation's optical remote sensing layer by showing how high-frequency time series can deliver near real-time AGB dynamics rather than a single annual snapshot. |
|---|---|
| TJFS Review | Serves as the application-innovation example in the TJFS review's optical remote sensing chapter, emphasizing intra-annual dynamics and disturbance attribution. |
| 可引用句候選 | 2024 年，Pelletier 等人發表的文獻中指出，融合 Sentinel-2 與 Landsat 的 TIIC 演算法可在加拿大全國尺度實現年內地上生物量動態監測，變化偵測與擾動分類精度皆達 99%。 |
| 不可用來主張 | Do not use this paper as evidence of year-round or all-weather monitoring, since the demonstrated window is the growing season and SAR fusion remains future work. |
授權與圖表重用
| Article license | UNKNOWN |
|---|---|
| Figure reuse policy | DO_NOT_REUSE_ORIGINAL_FIGURES_PUBLICLY_UNTIL_LICENSE_CHECKED |
| Notes | 發表於 Remote Sensing of Environment（Elsevier），非 MDPI 系列、非確定 CC-BY，授權標 UNKNOWN，圖表需自繪詮釋直至確認授權。 |
待查核清單
- Locate and confirm the original PDF and extracted txt path (currently 待查).
- Verify exact page locations for cited numbers against the original PDF.
- Check Elsevier license before any figure reuse; redraw a self-made TIIC workflow diagram.
- Confirm whether full author list and volume/issue beyond '301' should be filled.