以光達資料融合改善森林屬性估算之回顧LiDAR Data Fusion to Improve Forest Attribute Estimates: A Review
Balestra, M.; Marselis, S.; Sankey, T. T.; Cabo, C.; Liang, X.; Mokros, M.; Peng, X.; Singh, A.; Stereczak, K.; Vega, C.; Vincent, G.; Hollaus, M.｜Current Forestry Reports 10: 281-297｜DOI: 10.1007/s40725-024-00223-7
狀態：AI_DRAFT_FROM_REVIEW｜分級：A｜閱讀深度：FULL_TEXT_CHECKED｜Jacky 審核：False
森林數位孿生方法平台
LiDAR data fusionforest attributesstructured literature reviewdata-level fusionfeature-level fusionPRISMAmultispectralhyperspectralradarglobalnorthern hemisphere
專討核心文獻定位
[36]
Ch6 · 多源融合
新增
結構化回顧 151 篇光達融合研究，釐清資料融合定義並指出多數準確度提升仍屬邊際
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 Ch6 source because it gives a structured, decade-wide synthesis of LiDAR-centred multi-source data fusion and, crucially, supplies a clear shared definition that separates data fusion, data integration, and data combination.
結構式摘要｜中英文對照
| 研究問題 | 過去十年以光達為核心的多源資料融合呈現哪些趨勢、動機與方法，融合究竟帶來多少準確度增益，又該如何明確定義資料融合一詞？ What are the trends, motivations, and methods of LiDAR-centred multi-source data fusion over the last decade, how much accuracy gain does fusion actually bring, and how should the term data fusion be defined? |
|---|---|
| 資料來源 | 以 PRISMA 流程於 Web of Science 檢索 2014 年 1 月至 2023 年 5 月、英文、article 或 review article 的文獻，初檢得 664 篇，符合資格 407 篇，經兩名獨立審查者篩選後保留 151 篇進行編碼分析，地理分布以 142 個案例研究呈現。 A PRISMA-based structured search of Web of Science for 2014-01 to 2023-05 English articles or review articles returned 664 papers; 407 met the eligibility criteria, and after screening by two independent reviewers 151 papers were retained and coded, with geographic distribution shown across 142 case studies. |
| 方法 | 結構化文獻回顧搭配五大類編碼方案（一般資訊、地理位置、調查範圍、資料特性、調查目標），再由 11 位國際專家組成的小組討論定義與挑戰；融合方法分為資料層融合與特徵層融合兩大類，文中亦提及但樣本中未出現決策層融合。 A structured literature review with a five-category coding scheme (general information, geographic location, survey area, data characteristics, survey goals), followed by a panel of 11 international experts discussing the definition and challenges. Fusion methods are divided into data-level and feature-level fusion; decision-level fusion is mentioned but absent from the sample. |
| 主要結果 | 融合以機載對機載最常見（45.4%），其次機載對太空（29.8%）；應用以面積基準法 50%、單木法 27% 為主，分類（29.5%）與材積或生物量（17.7%）是最大應用類別。在分類上，光達結合光譜資訊較單用光達平均提升整體準確度 41%，但若改以融合取代純光譜資訊僅提升 10 至 14%。特徵層融合占 78%、資料層融合占 22%。作者強調許多增益其實是邊際的，並提出資料融合、資料整合、資料組合三詞的明確區分。 Airborne-airborne fusion is most common (45.4%), followed by airborne-spaceborne (29.8%). Applications are dominated by area-based approach (50%) and individual-tree approach (27%); classification (29.5%) and growing stock volume or biomass (17.7%) are the largest categories. For classification, fusing LiDAR with spectral information raised overall accuracy by 41% on average relative to LiDAR alone, but only by 10-14% relative to spectral information alone. Feature-level fusion accounts for 78% and data-level fusion for 22%. The authors stress that many gains are marginal and propose a clear distinction between data fusion, data integration, and data combination. |
| 限制 | 回顧僅聚焦多源（多感測器）融合，刻意排除多時序融合、同機同時收集的 MS 與光達、以及同儀器的共配準；地理上南半球與全球南方研究極少，多數成果集中北半球；作者也指出許多融合增益偏小，在實務操作化、成本與專業門檻上仍有未解問題。 The review focuses only on multi-source (multi-sensor) fusion and deliberately excludes multi-temporal fusion, simultaneous MS-LiDAR from one instrument, and same-instrument co-registration. Geographically, southern-hemisphere and global-south studies are very few and most results concentrate in the northern hemisphere. The authors also note that many fusion gains are marginal, with open questions on operationalization, cost, and required expertise. |
Key Findings
| 發現 | 證據 | 確定性 |
|---|---|---|
| LiDAR data fusion with multispectral, hyperspectral, or radar data improves a wide range of forest applications, but many of the reported gains are marginal relative to LiDAR alone. | Recent Findings in abstract and the Lessons Learned section: fusion is useful across segmentation, AGB, canopy height, species ID, structure, and fuel load, yet the review notes that a lot of these gains are marginal. | checked_against_original_txt |
| The review provides a shared definition that separates data fusion (merging data or features where at least one source is LiDAR), data integration (decision-level combining), and data combination (the whole pipeline). | Discussion sub-section Data fusion and Fig. 4 define the three terms to reduce confusion in the LiDAR community. | checked_against_original_txt |
| Feature-level fusion dominates practice (78%) over data-level fusion (22%), and airborne-airborne is the most common platform pairing (45.4%). | Methods section reports 78% feature-level and 22% data-level fusion; Trends section and Table 2 report airborne-airborne 45.4% and airborne-spaceborne 29.8%. | checked_against_original_txt |
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 |
|---|---|---|---|---|
| Fig. 1, Table 1, Table 2, Fig. 2, Fig. 3, Fig. 4 | Fig. 1 shows the structured review framework and coding scheme; Table 1 lists the five coding categories and subjects; Table 2 gives publication counts by platform pairing; Fig. 2 shows the publication trend by LiDAR platform; Fig. 3 maps the geographic distribution of the 142 case studies; Fig. 4 is the proposed conceptual framework defining data fusion, data integration, and data combination. | 664 papers found, 407 eligible, 151 coded; 142 case studies; airborne-airborne 45.4%, airborne-spaceborne 29.8%, spaceborne-spaceborne and airborne-terrestrial each 11.3%, terrestrial-terrestrial 2.1%; classification 29.5%, GSV/AGB 17.7%, structure 15.5%, tree height 12.7%, segmentation 9.2%, other 9.1%, fuel load 6.3%; feature-level 78%, data-level 22%. | Use Table 2 and the application percentages to position multi-source fusion as the middle-layer feeder for a Taiwan forest digital twin, and cite Fig. 4 when defining fusion terms precisely in the review. | Figures are CC BY 4.0 and may be reused with attribution; still prefer a redrawn, Taiwan-oriented adaptation of Fig. 4 for the seminar. |
Extracted Evidence Table
| 可支撐主張 | 指標或結果 | 原文位置 | 可引用 | 備註 |
|---|---|---|---|---|
| Fusing LiDAR with spectral information substantially improves species/land-cover classification compared with LiDAR alone, but only marginally compared with spectral data alone. | Overall classification accuracy increased by 41% on average versus LiDAR alone, but by only 10-14% versus spectral information alone. | Gains of LiDAR Data Fusion, Classification (Tree Species/Land Cover) subsection, p.287. | True | Report both numbers together so the marginal-vs-spectral framing is not lost. |
| Feature-level fusion is the dominant fusion practice in the LiDAR forestry literature. | 78% of reviewed papers performed feature-level fusion; 22% performed data-level fusion; no paper used decision-level fusion. | Methods for LiDAR Data Fusion section, p.286-287. | True | Use to justify feature-level fusion as the practical default in a fusion pipeline. |
| Ground-based plus airborne LiDAR fusion sharply reduces error for individual-tree structural attributes. | Fusing TLS and ULS reduced tree-height RMSE from 0.30 m (TLS) and 0.11 m (ULS) to 0.05 m; crown projection area RMSE from 3.06 and 4.61 m2 to 0.46 m2; crown volume RMSE from 29.63 and 30.23 m3 to 8.30 m3. | Gains of LiDAR Data Fusion, Forest Structure subsection, p.287, citing reference [26]. | True | These are values reported from a cited primary study ([26]) within the review; attribute as reported by the review rather than as the review's own experiment. |
Critical Appraisal
Strengths
- PRISMA-based, transparent, decade-wide structured review with an explicit coding scheme.
- Expert-panel-backed clarification of confusing terminology (fusion vs integration vs combination).
- Application-by-application reporting of quantitative accuracy gains.
Weaknesses
- Scope deliberately excludes multi-temporal and same-instrument fusion, so it is not a complete fusion picture.
- Strong northern-hemisphere bias in the underlying literature limits transferability to tropical and southern forests.
- Many reported gains are marginal and operational cost-benefit remains unresolved.
| Validation quality | high as a structured review; quantitative gains are aggregated from primary studies rather than independently validated here |
|---|---|
| Transferability to Taiwan | high as a methodological and definitional reference for a Taiwan multi-source fusion pipeline |
| Risk of overclaiming | Do not present fusion as uniformly worthwhile; the review itself flags that gains are often marginal and cost/effort-sensitive. |
與 Jacky 博論 / Review 的用途
| 博士論文 | Anchors the middle-layer data-fusion logic of the dissertation by defining fusion terms precisely and showing where multi-source fusion measurably helps AGB and structure estimation. |
|---|---|
| TJFS Review | Gives the TJFS review a citable, authoritative definition of data fusion and a quantitative baseline for fusion gains, helping separate genuine fusion from loose data combination. |
| 可引用句候選 | 2024 年，Balestra 等人發表的文獻中指出，光達與光譜資料的融合相較單用光達雖可使分類整體準確度平均提升 41%，但相較單用光譜資料僅提升 10 至 14%，且多數增益其實偏向邊際。 |
| 不可用來主張 | Do not cite this paper as evidence that data fusion is always operationally worthwhile, nor as covering multi-temporal or single-instrument fusion. |
授權與圖表重用
| Article license | CC BY 4.0 |
|---|---|
| Figure reuse policy | REUSE_ALLOWED_WITH_ATTRIBUTION_UNDER_CC_BY_4_0 |
| Notes | Open Access, The Author(s) 2024, Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/). Figures may be reused with appropriate credit and a link to the licence, unless a separate credit line indicates third-party material. |
待查核清單
- Confirm whether a Chinese translation_md and seminar導讀 pptx exist for this paper; currently marked 待查.
- If reusing Fig. 4 directly, add a CC BY 4.0 attribution line; otherwise prepare a redrawn fusion-definition diagram.
- Cross-check the platform-pairing percentages against Table 2 in the PDF before final publication.