2026年4月1日22時48分13秒（UTC），印度尼西亞北馬魯古海發生規模 Mw 7.4 地震，引發近場海嘯，周邊沿岸潮位站均記錄到相關波形。北米納哈薩（North Minahasa）觀測到的海平面振幅達0.75 m，比通（含倫貝島 Lembeh Island）則觀測到0.20 m（BMKG， 2026a）。本研究由涵蓋北馬魯古海的9個沿岸潮位站中選取6站，以其海嘯波形逆推本次地震的同震滑移分布。其餘3站不參與滑移逆推，保留作為逆推之外的一致性檢核。斷層模型幾何由海嘯反向傳播時間分析（Backward Tsunami Travel-Time，BTTT）及主震後24小時內的餘震空間分布共同約束。初步正演模擬採用三種由地震學資料推導的震源模型，包括全球地震矩張量目錄（GCMT）提供的NP1與NP2，以及美國地質調查所（USGS）的有限斷層模型。三種模型均系統性高估海嘯振幅，顯示有必要進行分布式滑移逆推。波形擬合誤差與餘震幾何共同顯示，在所測試的斷層面中，向西傾斜的GCMT NP2較符合觀測結果。 本研究以COMCOT模式結合Okada（1985）彈性位錯理論建立格林函數矩陣，將斷層面離散為20個子斷層，並採非負最小平方法進行海嘯波形逆推。主要逆推所用的單位滑移格林函數由COMCOT線性淺水波方程模擬產生，理論上滿足逆推所需的疊加與尺度縮放性質；然而如下文診斷比較所示，實際數值實作仍存在約10–12%之振幅正規化偏差，此一性質尚不應視為已於數值上完全驗證。另以非線性淺水波方程產生的格林函數進行對照逆推。所得大尺度滑移型態、主要高滑移區及總地震矩均與主要結果相近，可作為一致性檢核。逆推結果顯示，滑移主要集中於NP2定向震源網格中央部位的一個緊密高滑移區，最大滑移約為4.44 m。此空間特徵較符合由馬魯古海主要板塊邊界逆衝斷層分支出的向西傾斜分支斷層（splay fault），而非主要隱沒介面本身。 本研究實際建立的NP2定向震源網格，與依52°傾角沿傾向正確投影所得的平面斷層離散網格有所差異。因此，精確滑移量、深度及高滑移區位置均應視為具有模型相依性。此外，圖4.15的診斷比較顯示，5個測站仍存在約10–12%的振幅正規化偏差，故本文所報告的滑移量應在所採用之格林函數正規化條件下解讀。此外，觀測潮位紀錄採用測站專屬之帶通濾波處理，格林函數波形則未經相同濾波，兩者所代表之觀測算子並不完全相同，故本文所報告之振幅相依量亦可能包含未單獨量化之系統性前處理不確定性。作為補充敏感度測試，本研究亦以相同的6站資料，採無約束最小平方法逆推另一種初始自由表面參數化，得到正負交錯且具空間振盪特徵的多網格分布，其大尺度特徵與斷層滑移震源大致相符。 以逆推滑移模型進行正演傳播模擬後，3個未參與滑移逆推的測站（BITU、LKPN及MTHMI）波形均獲得合理重現，可作為逆推之外的驗證；惟三站先前皆曾用於斷層幾何評估，故均非嚴格意義上的獨立盲測。模擬亦顯示，蘇拉威西島東部沿岸約於地震發生後10分鐘受到首波影響。Kema潮位站的模擬首波到時約為15分鐘，略早於NOAA/NCTR報告的約17分鐘首波偵測時間。研究結果顯示，馬魯古海發散型雙重隱沒系統具有高度近場海嘯危害，並凸顯建置可靠且能快速反應之海嘯預警基礎設施的迫切需求。;On 1 April 2026 at 22:48:13 UTC， an Mw 7.4 earthquake occurred in the northern Molucca Sea， Indonesia， generating a near-field tsunami recorded by coastal tide gauges. A sea-level amplitude of 0.75 m was observed at North Minahasa， and 0.20 m at Bitung (including Lembeh Island)， based on BMKG tide-gauge records (BMKG， 2026a). This study estimated the coseismic slip distribution of the earthquake by inverting tsunami waveforms recorded at six coastal tide-gauge stations selected from a nine-station network spanning the northern Molucca Sea. The remaining three stations were excluded from the slip inversion and retained for subsequent forward-model evaluation. The fault geometry was constrained using backward tsunami travel-time (BTTT) analysis and the spatial distribution of aftershocks within 24 hours of the mainshock. Preliminary forward simulations were conducted using three seismologically derived source models: the Global Centroid Moment Tensor (GCMT; Dziewonski et al.， 1981) NP1 and NP2 nodal planes and the USGS finite-fault model. All three models systematically overestimated the observed tsunami amplitudes， indicating the need for a distributed slip inversion. The combined evidence from waveform misfits and aftershock geometry favored the west-dipping GCMT NP2 among the tested planar fault geometries. The primary tsunami waveform inversion employed non-negative least squares and a Green′s-function matrix constructed from unit-slip sources on 20 subfaults. Initial seafloor deformation for each source was calculated using the Okada (1985) elastic dislocation model， and tsunami propagation was simulated with COMCOT using the linear shallow-water equations. This formulation is theoretically consistent with the superposition and amplitude-scaling properties required for Green′s-function inversion， subject to the numerical-implementation caveat noted below. A parallel inversion using Green′s functions generated with the nonlinear shallow-water equations was also conducted as a consistency test. It produced a similar large-scale slip pattern， dominant high-slip area， and total seismic moment. The recovered source was dominated by a compact high-slip asperity， with a peak slip of approximately 4.44 m， located within the central portion of the implemented NP2-oriented source grid. This geometry is consistent with rupture on a west-dipping splay fault branching from the main Molucca Sea plate-boundary thrust and may be associated with the Maju Thrust. However， the results do not conclusively identify the mapped Maju Thrust as the causative fault. The implemented NP2-oriented source grid differs from a geometrically correct planar grid projected along a dip of 52°. Consequently， the precise slip magnitude， source depth， and asperity location should be regarded as model-dependent. A diagnostic comparison also revealed a residual amplitude-normalization offset of approximately 10–12% at five stations. The reported slip magnitudes should therefore be interpreted as conditional on the Green′s-function normalization adopted in this study. Station-specific bandpass filtering， not replicated in the synthetic waveforms， adds a further unquantified uncertainty. As a complementary sensitivity test， an alternative initial-free-surface parameterization was inverted from the same six-station dataset using unconstrained least squares. The resulting distribution contained alternating positive and negative displacements with spatial oscillations， although its large-scale characteristics were broadly consistent with those of the fault-slip source model. Forward propagation simulations using the inverted slip model reproduced the waveforms at the three stations excluded from the slip inversion with satisfactory agreement. BITU， LKPN， and MTHMI therefore provide out-of-inversion consistency checks， although all three had previously contributed to fault-geometry assessment. The simulations indicate that the first tsunami wave reached parts of the modeled coastal waters along eastern Sulawesi approximately 10 minutes after the earthquake. At the Kema tide gauge， the simulated first arrival occurred approximately 15 minutes after the origin time， slightly earlier than the approximately 17-minute first detection reported by NOAA/NCTR. These findings indicate the substantial near-field tsunami hazard associated with the divergent double-subduction system of the Molucca Sea and highlight the urgent need for reliable rapid-response tsunami-warning infrastructure.
