[論文レビュー] GreenPhase: A Green Learning Approach for Earthquake Phase Picking

Yixing Wu, Shiou-Ya Wang+6 more|arXiv (Cornell University)|Feb 23, 2026

Seismology and Earthquake Studies被引用数 0

ひとこと要約

tldr: GreenPhase is a multi-resolution, feed-forward model based on Green Learning for earthquake detection and P-/S-wave phase picking, achieving state-of-the-art-like performance with dramatically reduced computation and energy use.

ABSTRACT

Earthquake detection and seismic phase picking are fundamental yet challenging tasks in seismology due to low signal-to-noise ratios, waveform variability, and overlapping events. Recent deep-learning models achieve strong results but rely on large datasets and heavy backpropagation training, raising concerns over efficiency, interpretability, and sustainability. We propose GreenPhase, a multi-resolution, feed-forward, and mathematically interpretable model based on the Green Learning framework. GreenPhase comprises three resolution levels, each integrating unsupervised representation learning, supervised feature learning, and decision learning. Its feed-forward design eliminates backpropagation, enabling independent module optimization with stable training and clear interpretability. Predictions are refined from coarse to fine resolutions while computation is restricted to candidate regions. On the Stanford Earthquake Dataset (STEAD), GreenPhase achieves excellent performance with F1 scores of 1.0 for detection, 0.98 for P-wave picking, and 0.96 for S-wave picking. This is accomplished while reducing the computational cost (FLOPs) for inference by approximately 83% compared to state-of-the-art models. These results demonstrate that the proposed model provides an efficient, interpretable, and sustainable alternative for large-scale seismic monitoring.

研究の動機と目的

• Advance earthquake detection and seismic phase picking with a sustainable, interpretable model.
• Eliminate backpropagation by using a modular, feed-forward Green Learning framework.
• Achieve high accuracy on STEAD with reduced computational cost and data requirements.
• Demonstrate scalability and real-time applicability for large-scale seismic monitoring.

提案手法

• Three-stage, multi-resolution Green Learning pipeline: unsupervised representation learning (Saab transform), supervised feature learning (RFT and SFG), and supervised decision learning (XGBoost).
• Coarse-to-fine processing across levels 3-1 to restrict inference to candidate regions.
• P- and S-wave picking modules share architecture and produce arrival-time predictions with probabilistic scores.
• A lightweight XGBoost classifier combines P-/S-wave predictions across levels to decide seismic event vs. noise.
• Pseudo-labels are continuous and window-based, enabling supervised learning without end-to-end backpropagation.
• Feature selection and generation (RFT and SFG) optimize discriminative representations before the decision stage.

実験結果

リサーチクエスチョン

• RQ1 Can Green Learning provide competitive earthquake detection and phase-picking performance with substantially lower computational cost?
• RQ2 How does a coarse-to-fine, non-backpropagating framework perform on STEAD compared to state-of-the-art deep learning models?
• RQ3 What is the impact of limited training data on P-/S-wave picking performance under GreenPhase?
• RQ4 Is the approach scalable to large-scale seismic monitoring and energy-efficient for real-time applications?

主な発見

• GreenPhase achieves detection F1 of 1.0, P-wave F1 of 0.98, and S-wave F1 of 0.96 on STEAD.
• P-phase: Pr=0.96, Re=0.99, F1=0.98 with 1.2M training samples (also robust at 240K and 60K).
• S-phase: Pr=0.93, Re=0.99, F1=0.96 with 1.2M training samples (competitive at smaller sizes).
• Inference FLOPs are 22M for 1.2M training data vs EQTransformer’s 129M (6x more).
• GreenPhase training is ~40x more sustainable than EQTransformer on comparable hardware.
• GreenPhase uses a coarse-to-fine ROI (about 80 time locations at finest level) to achieve efficiency gains.