Microseismic Fno Eval

microseismic-fno-eval

Microseismic event classification with a lightweight Fourier Neural Operator model — Abdullin et al. (2025) (arXiv:2512.07425, 2025)

What this evaluates

Evaluates a lightweight Fourier Neural Operator model's ability to classify microseismic events versus noise in seismic waveforms. It probes resolution-invariant signal processing, cross-domain generalization, and real-time detection efficiency under varying signal-to-noise ratios.

Datasets

• STEAD — total 2000; splits: test (2000)
• Microseismic dataset — total 1000; splits: test (1000)

Metrics

• F1 score (primary) — range: [0, 1]
  • F1 = 2 * (Precision * Recall) / (Precision + Recall), where Precision = TP/(TP+FP) and Recall = TP/(TP+FN). Computed after selecting an optimal probability threshold.
• Accuracy — range: [0, 1]
  • Fraction of correctly classified samples (TP+TN) out of total samples.
• Precision — range: [0, 1]
  • Fraction of true positives among all positive predictions (TP/(TP+FP)).
• Recall — range: [0, 1]
  • Fraction of true positives among all actual positives (TP/(TP+FN)).

Input / output format

Input: Single-station seismic waveform traces (signal or noise samples) processed as 1D time-series inputs to the FNO model.

Output: Binary classification label (event/signal vs. noise/false event) or a continuous probability score for event detection.

Scoring recipe

def compute_f1(tp, fp, fn):
    precision = tp / (tp + fp) if (tp + fp) > 0 else 0.0
    recall = tp / (tp + fn) if (tp + fn) > 0 else 0.0
    if (precision + recall) == 0:
        return 0.0
    return 2 * (precision * recall) / (precision + recall)

# Apply optimal probability threshold to convert model scores to binary predictions
# Then compute TP, FP, FN against gold labels
# Return compute_f1(tp, fp, fn)

Common pitfalls

• Threshold selection heavily influences F1 scores; the paper emphasizes finding an optimal probability threshold rather than using a fixed 0.5 cutoff.
• Incomplete phase annotations (missing P- or S-waves) in the microseismic dataset degrade performance for models that require both phase types for detection.
• Cross-domain generalization varies significantly due to differences in SNR, data collection standards, and annotation methods across datasets, making direct comparisons sensitive to dataset composition.

Evidence (verbatim from paper)

The model's performance was measured by metrics such as accuracy, precision, recall, and F1 score. The achieved accuracy on the test set was 95% , precision 94% , recall 96% , and F1 score 95% .

Citation

@misc{abdullin2025microseismic,
  title={Microseismic event classification with a lightweight Fourier Neural Operator model},
  author={Abdullin et al. (2025)},
  year={2025},
  note={arXiv:2512.07425}
}

• arXiv: 2512.07425