Inversi Seismik Berbasis Multiscale Full Waveform Inversion: Analisis Kuantitatif Estimasi Kecepatan Gelombang P Menggunakan Dataset OpenFWI

Azzahra, Mirza Amanda (2026) Inversi Seismik Berbasis Multiscale Full Waveform Inversion: Analisis Kuantitatif Estimasi Kecepatan Gelombang P Menggunakan Dataset OpenFWI. Other thesis, Institut Teknologi Sepuluh Nopember.

Text 5001221125_Undergraduate-thesis.pdf - Accepted Version Restricted to Repository staff only Download (4MB) | Request a copy

Abstract

Full Waveform Inversion (FWI) merupakan metode inversi seismik yang mampu merekonstruksi model kecepatan gelombang P dengan resolusi tinggi, namun rentan terjebak pada minimum lokal akibat fenomena cycle-skipping, terutama ketika model awal jauh dari kondisi sebenarnya. Penelitian ini membandingkan performa FWI konvensional dan multiscale FWI dalam merekonstruksi model kecepatan gelombang P pada lima subset dataset benchmark OpenFWI dengan tingkat kompleksitas geologis berjenjang, yaitu FlatVelA, CurveVelA, FlatFaultA, CurveFaultA, dan Style-A. Setiap dataset diuji pada dua kondisi model awal, yaitu dekat (Gaussian smoothing terhadap true model) dan jauh (gradien kecepatan linear), sehingga menghasilkan sepuluh skenario pengujian. Proses inversi memanfaatkan automatic differentiation pada framework PyTorch, dengan pembaruan model menggunakan normalized steepest descent dan backtracking line search, serta filter FIR windowed-sinc pada tahapan multiscale untuk menjaga konsistensi fase antar tahap frekuensi. Hasil penelitian menunjukkan bahwa keunggulan multiscale FWI terhadap FWI konvensional bersifat kondisional. Pada kondisi model awal dekat, FWI konvensional lebih unggul pada tiga dari lima dataset (RMSE terendah 10,50-67,65 m/s), sedangkan multiscale FWI lebih unggul pada dua dataset dengan kompleksitas struktural tinggi. Pada kondisi model awal jauh, multiscale FWI secara konsisten unggul pada seluruh dataset, dengan FWI konvensional mengalami kegagalan total (tidak ada perbaikan model sama sekali) pada tiga dataset yang mengandung struktur sesar. Temuan ini menegaskan bahwa efektivitas multiscale FWI bergantung pada jarak model awal dan kompleksitas struktur geologis, bukan berlaku sebagai metode yang unggul secara mutlak.
====================================================================================================================================
Full Waveform Inversion (FWI) is a seismic inversion method capable of reconstructing high-resolution P-wave velocity models, but it is prone to becoming trapped in local minima due to cycle-skipping, particularly when the initial model is far from the true subsurface condition. This study compares the performance of conventional FWI and multiscale FWI in reconstructing P-wave velocity models across five subsets of the OpenFWI benchmark dataset with progressively increasing geological complexity, namely FlatVelA, CurveVelA, FlatFaultA, CurveFaultA, and Style-A. Each dataset was tested under two initial model conditions, near (Gaussian-smoothed true model) and far (linear velocity gradient), resulting in ten test scenarios. The inversion process employed automatic differentiation within the PyTorch framework, with model updates performed using normalized steepest descent combined with backtracking line search, and an FIR windowed-sinc filter applied during the multiscale stages to preserve phase consistency across frequency bands. The results show that the superiority of multiscale FWI over conventional FWI is conditional rather than absolute. Under the near initial condition, conventional FWI outperformed multiscale FWI on three of the five datasets (lowest RMSE of 10.50–67.65 m/s), while multiscale FWI performed better on two datasets with higher structural complexity. Under the far initial condition, multiscale FWI consistently outperformed conventional FWI across all datasets, with conventional FWI failing entirely (no model improvement whatsoever) on the three fault-bearing datasets. These findings confirm that the effectiveness of multiscale FWI depends on the distance of the initial model from the true model and the geological complexity of the target structure, rather than being universally superior across all conditions.

Item Type:	Thesis (Other)
Uncontrolled Keywords:	Full Waveform Inversion, Multiscale FWI, Cycle-skipping, OpenFWI, Automatic Differentiation, Model Kecepatan Gelombang P, Full Waveform Inversion, Multiscale FWI, Cycle-skipping, OpenFWI, Automatic Differentiation, P-wave Velocity Model
Subjects:	Q Science > QC Physics Q Science > QE Geology > QE538.5 Seismic tomography; Seismic waves. Elastic waves Q Science > QE Geology > QE539.2 Seismic traveltime inversion Q Science > QE Geology > QE539.2.S4 Seismic models
Divisions:	Faculty of Mathematics and Science > Physics > 45201-(S1) Undergraduate Thesis
Depositing User:	Mirza Amanda Azzahra
Date Deposited:	30 Jul 2026 09:32
Last Modified:	30 Jul 2026 09:32
URI:	http://repository.its.ac.id/id/eprint/140098

Actions (login required)

View Item