DSpace community: 地球系統科學國際研究生博士學位學程

氣候變遷對連續乾旱日數和灌溉用水需求的影響：台灣調整播種日期作為一種調適策略;Impacts of Climate Change on Consecutive Dry Days and Irrigation Water Demand: Shifting Planting Dates as an Adaptation Strategy in Taiwan

title: 氣候變遷對連續乾旱日數和灌溉用水需求的影響：台灣調整播種日期作為一種調適策略;Impacts of Climate Change on Consecutive Dry Days and Irrigation Water Demand: Shifting Planting Dates as an Adaptation Strategy in Taiwan abstract: 氣候變遷是二十一世紀最迫切的挑戰之一，不僅加劇了全球乾旱動態，更對永續水資源管理造成前所未有的壓力。這對台灣的水資源構成重大威脅，尤其影響農業地區——該地區的雙季作系統對水文氣候變異極為敏感。近年來，台灣遭遇的時空氣候極端事件頻率與強度日益增加，尤其是連續乾旱日數（CDD），這不僅嚴重影響水資源，更加劇水資源壓力，進而增加農田的灌溉用水需求（IWD）。因此，了解 CDD 與 IWD 的變異趨勢，對於提供水資源管理與農業規劃的關鍵資訊至關重要。本研究採用經偏差校正的「台灣氣候變遷預測資訊與調適知識平台」CMIP6氣候預測資料，針對近期未來（2021–2050年）與遠期未來（2071–2100年），評估相較於1995–2014年基準期的連續乾旱日數未來變化。本資料集經由台灣全境六座主要氣象站的觀測資料進行驗證，並採用泰勒技能分數（TSS）及模型排名法，評估十九個 CMIP6 模型在重現歷史月累積日度（CDD）氣候學特徵方面的表現。利用降水資料，透過曼恩–肯德爾檢驗（Mann–Kendall test）及森的斜率估計量（Sen’s slope estimator），分析台灣全境累積日度（CDD）的趨勢與變化點。未來CDD的變化是根據CMIP6在SSP1-2.6、SSP2-4.5、SSP3-7.0及SSP5-8. 5情境下的預測結果進行分析。此外，還分析了從美國國家海洋暨大氣總署（NOAA）取得的重大氣候變異性指標（包括AMO、PDO、NAO及ENSO），以評估其與CDD事件的遙相關性。此外，本研究透過聯合國糧農組織（FAO）的AquaCrop水生產力模型，評估氣候變遷對最佳灌溉日（IWD）的影響。作為適應策略，採用「種植日期調整法」，將種植日期從原定時間提前21天起，以每7天為單位逐步延後21天，藉此找出未來氣候條件下的最佳灌溉日。對十九個 CMIP6 模型的評估顯示，在模擬台灣地區的 CDD 方面存在明顯的季節性變異性。其中十個 TSS 值較高的模型（RM-TSS > 0.5）被認定為表現最佳的模型，並被用於建立多模型集合以進行未來預測分析。對歷史趨勢的分析揭示，台灣各地區的季節性 CDD 變異性存在顯著的空間依賴性。秋季期間，中部及南部地區的累積日度積算度（IWD）明顯以每十年增加約0.5–1.5天為特徵；而冬季期間，台灣南部則觀察到每十年約減少0.5天的溫和下降趨勢。未來預測顯示，在高排放情境（SSP3-7.0 和 SSP5-8.5）下，到 21 世紀末，乾季期間台灣全境出現延長的累積日度（CDD）現象的範圍可能擴大。特別是 SSP5-8.5 情境，顯示每十年將額外增加約 1.2 天。這些預測的增加與東亞季風系統的預期變化密切相關，特別是季節性降水模式和大氣環流動力學。相較之下，低排放情境（如 SSP1-2.6 和 SSP2-4.5）則顯示，春季 CDD 將部分恢復至基準狀態，同時春季雨帶可能持續籠罩台灣。此外，相關性分析結果顯示，大尺度大氣振盪（如 AMO、PDO 和 ENSO）與台灣特定地區春季 CDD 延長之間存在一致的關聯性。 AquaCrop 模型與觀測到的冠層覆蓋率呈現極佳的一致性（第一季 R² = 0.99，第二季 R² = 0.91），均方根誤差（RMSE）≤ 13.2%，平均絕對誤差（MAE）≤ 11.1%，這凸顯了該模型適用於在未來氣候情境下，估算跨季水稻的灌溉缺水量（IWD）。歷史模擬結果顯示，與第二作期相比，第一作期所需的灌溉用水量顯著更高，最高可達 180 毫米，對水資源造成最大負擔；而第二作期大多需要較少的補充灌溉。未來預測顯示，第一作期的灌溉用水量將大幅增加，在 SSP5-8.5 情境下，西部沿海地區的增幅最高可達 +40 毫米。此外，在台灣各區域，若將第一作期的播種日期提前14（21）天或延後21（14）天（C25/C05，可使雙季作系統達到最大的灌溉用水節省效果。在單季種植系統下，中部地區若將播種日期推遲50天，可節省80毫米灌溉用水；而在南部地區，若將播種日期推遲70天，節水量更可達150毫米。這些模式在所有情境下均一致。這些研究結果表明，監測關鍵氣候指標有助於制定春季乾旱緩解策略；同時，在台灣水資源緊缺的條件下，調整最佳播種日期有助於減輕氣候變遷對水稻灌溉用水需求的不利影響。;Climate change is one of the most pressing challenges of the 21st century， intensifying global drought dynamics and placing unprecedented strain on sustainable water resource management. It poses a significant threat to water resources in Taiwan， particularly impacting agricultural areas， where double-cropping systems are highly sensitive to hydroclimatic variability. In recent years， Taiwan has experienced increasingly frequent and intense spatiotemporal climate extremes， particularly consecutive dry days (CDD)， which strongly affect water resources and exacerbate water stress， thereby increasing irrigation water demand (IWD) in agricultural fields. Therefore， understanding the variation in CDD and IWD is essential for providing critical information for water management and agricultural planning. Bias-corrected Taiwan Climate Change Projection Information and Adaptation Knowledge Platform CMIP6 climate projections were used to assess future changes in CDD relative to the 1995–2014 baseline for the near future (2021–2050) and far future (2071–2100). The dataset was validated against observations from six major meteorological stations across Taiwan， and the performance of nineteen CMIP6 models in reproducing historical monthly CDD climatology was evaluated using the Taylor Skill Score (TSS) and model ranking methods. The Mann–Kendall test and Sen’s slope estimator were used to detect trends and change points in CDD across Taiwan using precipitation data. Future CDD changes were analyzed using CMIP6 projections under the SSP1-2.6， SSP2-4.5， SSP3-7.0， and SSP5-8.5 scenarios. In addition， major climate variability indices， including AMO， PDO， NAO， and ENSO， obtained from the National Oceanic and Atmospheric Administration， were analyzed to assess their teleconnections with CDD events. Moreover， the impacts of climate change on IWD were assessed by the Food and Agriculture Organization (FAO) AquaCrop water productivity model. As an adaptation strategy， a shifting planting-date approach， from 21 days earlier to 21 days later in seven-day increments， was applied to identify the optimal IWD for the future climate conditions. Evaluation of nineteen CMIP6 models showed clear seasonal variability in simulating CDD over Taiwan. Ten models with higher TSS (RM-TSS > 0.5) were identified as the best-performing models and utilized to develop the multi-model ensemble for future projection analyses. The analysis of historical trends reveals a pronounced spatial dependence in seasonal CDD variability across Taiwan. Increases of approximately 0.5–1.5 days per decade are evident in central and southern regions during autumn， while a modest decline of about 0.5 days per decade is observed in southern Taiwan during winter. Future projections indicate that， under high-emission scenarios (SSP3-7.0 and SSP5-8.5)， prolonged CDD is likely to become more widespread across Taiwan during the dry seasons by the end of the 21st century. The SSP5-8.5 scenario in particular indicates an additional increase of roughly 1.2 days per decade. These projected increases are closely linked to anticipated changes in the East Asian monsoon system， particularly in seasonal precipitation patterns and atmospheric circulation dynamics. In contrast， low-emission scenarios such as SSP1-2.6 and SSP2-4.5 point toward a partial recovery toward baseline spring CDD conditions， along with the potential persistence of the spring rainband over Taiwan. Furthermore， correlation results reveal a consistent relationship between large-scale atmospheric oscillations such as AMO， PDO， and ENSO and prolonged spring CDD in specific regions of Taiwan. The AquaCrop model showed excellent agreement with observed canopy cover (R² = 0.99 in season 1 and 0.91 in season 2)， with RMSE ≤ 13.2% and MAE ≤ 11.1%， highlighting its suitability for estimating IWD for inter-seasonal rice under future climate scenarios. Historical simulation results showed that the first cropping season requires substantially higher IWD compared to the second crop season， reaching up to 180 mm and imposing the greatest burden on water resources， whereas the second cropping season mostly requires less supplemental irrigation. Future projections indicated a substantial increase in IWD during the first cropping season， with rises of up to +40 mm along the western coast under the SSP5-8.5 scenarios. Furthermore， the maximum irrigation water savings for a double-cropping system can be achieved by a combination of shifting 14 (21) days earlier or 21 (14) days later for the first season (C25/C05) across different regions in Taiwan. Under a single-cropping system， 80 mm of irrigation water can be saved by shifting the planting date of 50 days later in the central region， whereas in the south this value reaches up to 150 mm by shifting the planting date of 70 days later. These patterns are consistent across all scenarios. These findings suggest that monitoring key climate indices could be able to support spring drought mitigation strategies， while adjusting optimal planting dates may help reduce the adverse impacts of climate change on rice IWD under water-stressed conditions in Taiwan.

Advancing Decadal-Scale Erosion Quantification in Taiwan’s Active Mountainous Basins: Spectral Analysis and Enhanced DEM Differencing;Advancing Decadal-Scale Erosion Quantification in Taiwan’s Active Mountainous Basins: Spectral Analysis and Enhanced DEM Differencing

title: Advancing Decadal-Scale Erosion Quantification in Taiwan’s Active Mountainous Basins: Spectral Analysis and Enhanced DEM Differencing;Advancing Decadal-Scale Erosion Quantification in Taiwan’s Active Mountainous Basins: Spectral Analysis and Enhanced DEM Differencing abstract: 沉積物通量反應了構造抬升、氣候事件與地表作用的綜合作用，然而在陡峭且受颱風影響的集水區中，十年尺度的侵蝕分布始終難以解析。位於集水區出口的水文測站量測雖能掌握集水區整體的產砂量，卻無法揭示沉積物在流域內部的產生、儲存與傳輸的空間分布。而作為替代方案的 DEM（數值高程模型）差分法，則長期受限於兩項問題而難以使用：一是基於穩定地表的對齊方法難以適用在穩定地面稀少的區域；其二是DEM資料本身帶有植被冠層偏差與感測器導致之偽影等雜訊，其量值往往超過地形差分訊號。本論文發展並應用一套同時克服上述兩項限制的DEM差分法，並以其量化構造與地表作用活躍的臺灣南部中央山脈荖濃溪流域（約1，373平方公里）與其一支流之濁口溪流域（約375平方公里）的十年尺度地表侵蝕作用與沉積物再分布。第一項研究提出一種基於傅立葉頻譜分析的DEM的垂直對齊方法，可在不依賴大量穩定地表的情況下辨識並移除多期DEM之間的垂直偏差。本研究以濁口溪為例，使用1990至2020年間的開源全球與區域DEM進行高程對齊與DEM差分分析。校正後的高程差分結果顯示：三十年期間的平均侵蝕速率為每年14.19毫米，最近十年（2011年至2020年）則為每年10.46毫米。近十年侵蝕速率的下降反映了極端降雨事件的明顯減少，與下游水文測站獨立推估之懸浮載結果一致。本研究顯示使用本研究發展的對齊方式可顯著降低對齊造成的垂直誤差，使侵蝕訊號變得有代表性，讓過去未被充分利用的歷史與衛星DEM提供可靠的十年尺度沉積物收支分析。第二項研究將前述架構延伸至荖濃溪流域，用以分析2000至2020年流域內的侵蝕與沈積物再分布，期間包含臺灣有紀錄以來降雨量最高的颱風：2009年的莫拉克颱風。研究首先以本地的空載光達的DEM為基準將多組全球DEM進行比對。分析結果指出，由於NASADEM具有低頻譜雜訊、極少的空值填補以及一致的觀測時間，使其為適用的歷史DEM。為了消除的植被冠層偏差以獲得更低的體積不確定性（雜訊），本研究先以全球冠層高度模型校正局部超過30公尺植被訊號，再施以頻譜垂直基準校正。經過植被偏差移除處理後，體積不確定性由約187百萬立方公尺降至約44百萬立方公尺。經過DEM差分計算後各子集水區的侵蝕速率平均約為每年15.3毫米，在受崩塌影響的支流中可達約每年551毫米。然而，侵蝕產生的物質中有87.6% 仍儲存於流域之內，主要以谷底充填沉積、沖積扇及主河道淤積的形式存在。由此得到的全流域淨剝蝕速率為每年5.06 ± 1.60毫米。儘管期間發生多次極端事件，此值仍與長期（10³ 至10⁶年）由宇宙射線生成核種與熱年代學推估的侵蝕、剝蝕速率（約每年 3至6毫米）高度一致。本研究所觀察到的沉積物緩衝現象即是在十年尺度上將局部極端侵蝕加以空間平均的結果。綜合前述兩項研究顯示，重複觀測的地形資料經過適當校正植被高度與垂直偏差，即可同時解析十年尺度的侵蝕量值，以及沉積物自源區至流域出口的內部傳輸路徑。研究成果量化了流域內部沈積物緩衝作用，此作用使數十年尺度與千年尺度的侵蝕速率趨於一致。另外，DEM差分的空間分析成果可辨識出流域出口測站無法察覺的子集水區災害熱點。本研究展現了在氣候變遷下，未被充分利用的全球衛星DEM對於構造活動造山帶中沉積物相關地質災害監測的價值。;Sediment flux from active mountain belts records the coupled response of tectonics， climate， and surface processes， yet decadal-scale erosion patterns in steep， typhoon-impacted catchments remain difficult to resolve. Outlet-based gauging captures catchment-integrated yield but cannot reveal where sediment is produced， stored， or routed within a basin， while DEM differencing in these landscapes， which can offer alternate solution has been constrained by two persistent problems: stable-area based co-registration methods face challenges where stable ground is scarce， and elevation data carry noises such as vegetation canopy biases and sensor-induced artefacts that often exceed the geomorphic signal itself. This thesis develops and applies a DEM-differencing framework that addresses both limitations， and uses it to quantify decadal erosion and sediment redistribution in two mountainous catchments of Taiwan′s southern Central Range: the Zhoukou River (~375 km²) and the Laonong River (~1，373 km²). The first study introduces a Fourier-based spectral method for DEM vertical co-registration that identifies and removes vertical biases between repeat DEMs without relying on extensive stable ground. Applied to the Zhoukou catchment using open-source global and regional DEMs from 1990 to 2020， the corrected DEMs of Difference yield a decadal erosion rate of 14.19 mm/yr for the full three-decade window and 10.46 mm/yr for the most recent decade (2011–2020)， the latter reflecting a marked decline in extreme rainfall events. The declining DEM-derived rate in recent decade align well with independent suspended-load estimates from downstream gauge stations. The study shows that underutilized archival and satellite DEMs can deliver reliable decadal sediment budgets when the erosional signal dominates DEM noise. The second study extends the framework to the Laonong basin over 2000–2020， an interval that includes Typhoon Morakot (2009)， the highest-rainfall typhoon on record in Taiwan. Multiple global DEMs were benchmarked against airborne LiDAR， identifying NASADEM as the optimal archival DEM owing to its low spectral noise， minimal void-filling artefacts， and temporally coherent acquisition. Vegetation canopy bias， locally exceeding 30 m， was corrected using a global canopy-height model， followed by spectral vertical referencing; together these reduced volumetric uncertainty from ~187 to ~44 Mm³. Sub-catchment erosion rates average ~15.3 mm/yr and reach ~551 mm/yr in landslide-affected tributaries， yet 87.6% of the eroded mass is stored within the basin， predominantly as valley-fill deposits， alluvial fans， and trunk-channel aggradation. The resulting basin-wide net denudation rate is 5.06 ± 1.60 mm/yr， remarkably consistent with long-term (10³–10⁶ yr) cosmogenic and thermochronological estimates (~3–6 mm/yr) despite the occurrence of multiple extreme events. The result identifies sediment buffering as the mechanism that spatially averages localized extreme erosion at decadal timescales. Together， the two studies show that repeat topography， once vegetation and vertical biases are properly corrected， can resolve both the magnitude of decadal denudation and the internal routing of sediment from source to outlet. The framework reconciles decadal erosion measurements with millennial-scale rates by quantifying the role of internal storage， identifies sub-catchment hazard hotspots invisible to outlet gauging， and demonstrating the value of underutilized global satellite DEMs for monitoring of sediment-related geohazards in tectonically active mountain belts under a changing climate.

Integrating Remote Sensing with Cutting-Edge Models for Precision Land Use and Land Cover Mapping in Cuu Long Delta;Integrating Remote Sensing with Cutting-Edge Models for Precision Land Use and Land Cover Mapping in Cuu Long Delta

title: Integrating Remote Sensing with Cutting-Edge Models for Precision Land Use and Land Cover Mapping in Cuu Long Delta;Integrating Remote Sensing with Cutting-Edge Models for Precision Land Use and Land Cover Mapping in Cuu Long Delta abstract: 在越南湄公河三角洲，監測稻田、區分雙季稻田和三季稻田以及稻蝦共作田，並追蹤水產養殖、濕地和紅樹林區域，是環境管理、農業實踐和其他領域面臨最關鍵和最具挑戰性的問題。本研究結合先進的地理空間資料和遙測技術，以及創新的機器學習和深度學習方法，以應對這些挑戰。我們評估了多種模型，包括無監督、監督和半監督方法，並使用了各種衛星衍生指數。除了傳統模型外，我們還採用了深度卷積嵌入式聚類（DCEC）和基於置信度感知聚類的半監督軟標籤選擇（SoC4SS-FGVC）等新方法。在6024個樣本的實驗中，卷積神經網路（CNN）模型表現出最佳精確度，達到0.97。在樣本量僅 500 的情況下，監督模型的限制顯露無疑，準確率顯著下降至約 0.75 – 0.85，而 SoC4SS-FGVC 模型則取得了顯著的高精度（500 個樣本點下準確率達 0.92）。非監督模型的準確率則低得多，介於 0.40 和 0.50 之間。產生的土地覆蓋圖揭示了分類方面的挑戰，包括由於過度擬合和欠擬合導致的相似土地利用類型之間的混淆。土地覆蓋的均一性和有限的地面實測資料限制了監督模型的表現。相反地，儘管計算量更大，但半監督技術仍展現出良好的應用前景。與官方統計數據、專家意見和先前研究的交叉驗證表明，SoC4SS-FGVC 模型顯著提高了稻田的檢測和分離以及紅樹林的監測效果。這些發現為具有特定土地利用模式區域的土地覆蓋分類提供了寶貴的見解。;In the Mekong River Delta of Vietnam， monitoring paddy rice， distinguishing between double- and triple-cropping paddy rice fields and rice-shrimp land， and tracking aquaculture， wetlands， and mangrove areas are the most critical and challenging issues due to the need for environmental management， agricultural practices， and other sectors. This study addresses these challenges by combining advanced geospatial data and remote sensing techniques with innovative machine learning and deep learning methods. We evaluated several models， including unsupervised， supervised， and semi-supervised approaches， using various satellite-derived indices. Beyond traditional models， we employed novel approaches such as Deep Convolutional Embedded Clustering (DCEC) and the semi-supervised Soft Label Selection with Confidence-Aware Clustering (SoC4SS-FGVC). The CNN model showed the best accuracy at 0.97 when experimenting with 6024 samples. At small sample sizes of 500， the supervised models exhibited their constraints as the accuracies dropped significantly to about 0.75 – 0.85， while the SoC4SS-FGVC model achieved remarkably high-performance accuracy (0.92 with 500 sample points). Unsupervised models yielded much less accurate results， with scores between 0.40 and 0.50. The generated land cover maps revealed classification challenges， including confusion among similar land use types caused by overfitting and underfitting. The uniformity of land cover and limited ground-truth data restrict the performance of supervised models. Conversely， semi-supervised techniques show promise despite higher computational demands. Cross-validation with official statistics， expert input， and prior research demonstrates that the SoC4SS-FGVC model significantly improves paddy rice field detection and separation and mangrove monitoring. These findings offer valuable insights into land cover classification in areas with specific land-use patterns.

Mesozoic–Cenozoic Ophiolites of Northern Sabah (Borneo)， Malaysia: Age， Geochemical Characteristics， and Tectonic Implications;Mesozoic–Cenozoic Ophiolites of Northern Sabah (Borneo)， Malaysia: Age， Geochemical Characteristics， and Tectonic Implications

title: Mesozoic–Cenozoic Ophiolites of Northern Sabah (Borneo)， Malaysia: Age， Geochemical Characteristics， and Tectonic Implications;Mesozoic–Cenozoic Ophiolites of Northern Sabah (Borneo)， Malaysia: Age， Geochemical Characteristics， and Tectonic Implications abstract: 馬來西亞沙巴北部（婆羅洲）中生代至新生代蛇綠岩：年代、地球化學特徵與構造意義摘要蛇綠岩為逆衝至陸殼上的海洋岩石圈殘片，保存了重建區域構造演化所需的關鍵地質紀錄。沙巴北部發育了一系列具代表性的蛇綠岩組合，主要包括白堊紀的古達蛇綠岩與馬魯都灣蛇綠岩（Kudat and Marudu Bay Ophiolites；KO–MBO），以及始新世的邦吉島蛇綠岩（Banggi Island Ophiolite；BIO）。這些岩體以往被歸入區域性的中生代至新生代「矽質頁岩–玄武岩組（Chert–Spilite Formations）」，亦稱為沙巴蛇綠岩，並被解釋為古南海（Proto–South China Sea，PSCS）海洋地殼的殘餘物。整體而言，它們記錄了巽他陸塊東南緣於中生代至新生代期間複雜的海洋岩漿活動與構造重組歷史。本論文整合了野外地質調查、岩石學、全岩地球化學、鋯石鈾鉛定年及鋯石鎦鉿同位素資料，以界定邦吉島、古達及馬魯都灣蛇綠岩的形成時代、岩漿成因與構造意義。鋯石鈾鉛定年結果顯示，邦吉島蛇綠岩形成於始新世早期（約 55–50 Ma），為該蛇綠岩體首次建立可靠的結晶年代。鋯石 ƐHf(t) 值介於 +16.6 至 +10.2，指示其岩漿來源為相對虧損至年輕的地函。BIO 的地殼岩石組合（包括玄武岩、輝長岩與閃長岩）可區分為兩種不同的地球化學類型：一為虧損型（即輕稀土元素虧損），另一為富集型。兩種類型的 BIO 樣品多呈現大離子親石元素（LILE）富集、高場強元素（HFSE）系統性虧損，並具有偏高的 Th/Yb 比值，顯示其具備隱沒帶上盤（supra-subduction zone， SSZ）特徵。此等地球化學與同位素特徵支持 BIO 形成於洋內隱沒初始階段的弧前或原始島弧環境，並與古南海板塊向南隱沒作用相關。相較之下，古達與馬魯都灣蛇綠岩中的輝長岩至閃長岩，其鋯石鈾鉛年代顯示白堊紀早期岩漿活動（約 128–117 Ma）。其鋯石 ƐHf(t) 值普遍偏高且為正值（約 +16 至 +19；n = 9），指示其來源為相對虧損的地函。KO–MBO 的鎂鐵質岩石呈現由 N-MORB 至 E-MORB 與 OIB 的地球化學特徵，缺乏系統性的高場強元素虧損，且不相容元素比值（如 Nb/Yb、Nb/Y、Zr/Y、Nb/La 與 La/Sm）變化幅度大。這些特徵最合理的解釋為異質地函在不同部分熔融程度下所形成，並發生於與隱沒無關的擴張洋脊環境，可能與近中洋脊軸部海山岩漿活動相關。馬魯都灣蛇綠岩中較為演化的閃長岩與石英閃長岩雖呈現弧岩或類似埃達克岩的地球化學特徵，但岩相觀察與稀土元素系統學顯示，這些特徵主要源於以角閃石為主的分離結晶作用，而非板塊熔融或直接的隱沒輸入。綜合上述結果，本研究提出：始新世的邦吉島蛇綠岩代表古南海隱沒起始期間新生成的弧前海洋地殼，而白堊紀的古達與馬魯都灣蛇綠岩則代表參與該隱沒系統的東亞海（East Asian Sea）上部板塊殘餘。沙巴北部同時保存白堊紀與古近紀蛇綠岩，反映海洋地殼生成具有時間差異，並與隱沒起始的逐步演化及其向巴拉望島與民都洛島的東北向遷移有關，最終導致沙巴–卡加延火山弧的形成。最後，本研究認為邦吉島蛇綠岩的形成時序與沙巴及鄰近區域廣泛分布的始新世不整合密切相關，反映了古南海隱沒起始所引發的重大區域構造重組事件。關鍵詞：沙巴（婆羅洲）、鋯石鈾鉛定年、地球化學、隱沒帶上盤型蛇綠岩（SSZ 型）、隱沒起始、異質地函、古南海、卡加延海脊、始新世不整合;ABSTRACT Ophiolites represent fragments of obducted oceanic lithosphere that preserve key geological records essential for reconstructing regional tectonic evolution. Northern Sabah hosts a distinctive assemblage of ophiolite complexes， including the Cretaceous Kudat and Marudu Bay Ophiolites (KO–MBO) and the Eocene Banggi Island Ophiolite (BIO). These complexes are traditionally regarded as part of the regional Mesozoic-Cenozoic “Chert-Spilite Formations” (also known as Sabah Ophiolite) and interpreted as remnants of the proto-South China Sea (PSCS) oceanic crust. Together， they record a complex history of Mesozoic–Cenozoic oceanic magmatism and tectonic reorganisation along the southeastern margin of Sundaland. This thesis integrates field observations， petrography， whole-rock geochemistry， zircon U–Pb geochronology， and zircon Lu–Hf isotopic data to constrain the age， petrogenesis， and tectonic significance of the BIO and KO–MBO. New high-precision zircon U–Pb ages demonstrate that the BIO formed during the Early Eocene (~55–50 Ma)， establishing the first robust crystallisation ages for this ophiolite. Zircon ƐHf(t) values from +16.6 to +10.2 indicate magma derivation from a relatively depleted or juvenile mantle source. The BIO crustal sequence including basalt， gabbro， and diorite exhibits two geochemically distinct types， a depleted type (i.e.， depleted in light rare earth elements) and an enriched type， respectively. The majority of BIO samples from both types display various degrees of enrichment in large-ion lithophile elements (LILE) coupled with systematic depletion in high-field-strength elements (HFSE) and elevated Th/Yb ratios， indicating a supra-subduction zone affinity. These geochemical and isotopic characteristics are consistent with formation in a forearc or proto-arc setting during the earliest stages of intra-oceanic subduction initiation， interpreted here to be associated with southward subduction of the PSCS. In contrast， zircon U–Pb ages from gabbroic to dioritic rocks of the KO–MBO indicate Early Cretaceous magmatism (~128–117 Ma). Their uniformly high and positive zircon ƐHf(t) values (average: +19 to +16; n = 9) indicate derivation from a relatively depleted mantle source. Mafic rocks from the KO–MBO exhibit geochemical signatures ranging from N-MORB to E-MORB and OIB affinities， lack systematic HFSE depletion， and show wide variations in incompatible element ratios (e.g.， Nb/Yb， Zr/Y， Nb/Y， Nb/La， and La/Sm). These characteristics are best explained by variable degrees of partial melting of a heterogeneous mantle source in a subduction-unrelated spreading ridge environment， possibly associated with near-axis seamount magmatism. More evolved dioritic and quartz dioritic rocks from the Marudu Bay Ophiolite display arc-like or adakitic geochemical signature; however， petrographic evidence and rare earth element systematics demonstrate that these features were produced by hornblende-dominated fractional crystallisation rather than slab melting or direct subduction input. Based on these results， this study argues that the Eocene BIO represents newly formed forearc oceanic crust generated during PSCS subduction initiation， whereas the Early Cretaceous KO–MBO represent fragments of the overriding East Asian Sea plate involved in this subduction system. The coexistence of Cretaceous and Paleogene ophiolites in northern Sabah reflects diachronous oceanic crust formation linked to progressive subduction initiation and subsequent northeastward propagation towards Palawan and then Mindoro， eventually leading to development of the Sabah–Cagayan volcanic arc. Finally， the timing of BIO formation is proposed to be closely related to the regional Eocene unconformity in Sabah and adjacent regions， reflecting a major tectonic reorganisation associated with PSCS subduction initiation. Keywords: Sabah (Borneo)， Zircon U-Pb Geochronology， Geochemistry， SSZ-Type Ophiolite， Subduction Initiation， Heterogeneous Mantle， Proto-South China Sea， Cagayan Ridge