PhaseNet-DAS-Plus-Arcata-v1

PhaseNet-DAS-Plus (Arcata)

Semi-supervised DAS phase picker trained on Arcata (Northern California) DAS data.

Model Details

• Architecture: PhaseNet-DAS-Plus (UNet backbone with event detection heads)
• Training data: 136 labeled events from Arcata DAS array, labels generated by PhaseNet-2018 with PS-pair association (origin_window=0.1, moveout_window=0.2)
• Input: (1, nx, nt) DAS strain rate, nx=2048, nt=4096
• Output: Phase picks (P/S/Noise) + event center + event time
• Training: 50k iterations, lr=1e-4, EMA decay=0.999, batch_size=2, num_patch=16
• Parameters: 9.3M

Usage

import torch
import phasenet.models.phasenet_das_plus as model_module

ckpt = torch.load("phasenet_das_plus_arcata_v1.pth", map_location="cpu")
model = model_module.build_model(**ckpt["model_config"])
model.load_state_dict(ckpt["model_ema_weights"])  # Use EMA weights
model.eval()