Image functions¶

Bases: ImageFunction

Phase annotations from machine learning pickers in SeisBench.

The image is the probability of a P or S phase arrival, as annotated by a pre-trained SeisBench model, e.g. PhaseNet or EQTransformer.

Fields:

• image (Literal['SeisBench'])
• model (ModelName)
• pretrained (PreTrainedName | FilePath)
• window_overlap_samples (int)
• torch_use_cuda (bool | int)
• torch_cpu_threads (PositiveInt)
• batch_size (int)
• stack_method (StackMethod)
• sampling_rate (PositiveFloat)
• phase_map (dict[PhaseName, str])
• weights (dict[PhaseName, Annotated[float, Field(strict=True, ge=0.0)]])
• picker (AnnotationPicker)

model: ModelName = 'PhaseNet'

pretrained: PreTrainedName | FilePath = 'original'

window_overlap_samples: int = 2000

torch_use_cuda: bool | int = True

torch_cpu_threads: PositiveInt = 4

batch_size: int = 128

stack_method: StackMethod = 'avg'

sampling_rate: PositiveFloat = 100.0

phase_map: dict[PhaseName, str] = {
    "P": "cake:P",
    "S": "cake:S",
}

weights: dict[
    PhaseName, Annotated[float, Field(strict=True, ge=0.0)]
] = {"P": 1.0, "S": 1.0}

picker: AnnotationPicker

get_subclasses() -> tuple[type[ImageFunction], ...]

get_images(batch: WaveformBatch) -> WaveformImages

iter_images(
    batch_iterator: AsyncIterator[WaveformBatch],
) -> AsyncIterator[tuple[WaveformImages, WaveformBatch]]

Bases: Picker

Pick phase arrivals from SeisBench annotations.

The pick is the annotation peak closest to the modeled arrival time within the search window. Peaks before the event origin time are rejected.

Fields:

• threshold_p (float)
• threshold_s (float)
• search_window_seconds (PositiveFloat)
• peak_separation_seconds (NonNegativeFloat)

threshold_p: float = 0.1

threshold_s: float = 0.1

search_window_seconds: PositiveFloat = 5.0

peak_separation_seconds: NonNegativeFloat = 0.1

get_threshold(phase: PhaseName) -> float

pick_trace(
    trace: Trace,
    phase: PhaseDescription,
    event_time: datetime,
    modelled_arrival: datetime,
) -> ObservedArrival | None

add_picks(
    detections: Sequence[EventDetection],
    images: WaveformImages,
) -> None

Bases: ImageFunction

STA/LTA analytical characteristic function.

The image is the natural logarithm of the STA/LTA onset function, clipped at min_onset_value. Stacking the log onsets yields the logarithm of the geometric mean of the onsets, following QuakeMigrate. The semblance and the pick detection values are in log units, noise is at 0.

Fields:

• image (Literal['StaLta'])
• sta_seconds (PositiveFloat)
• lta_seconds (PositiveFloat)
• blinding_window (PositiveFloat)
• position (StaLtaPosition)
• min_onset_value (float)
• signal_transform (SignalTransform)
• phase_map (dict[PhaseName, str])
• weights (dict[PhaseName, Annotated[float, Field(strict=True, ge=0.0)]])
• picker (StaLtaPicker)

Validators:

• check_windows

sta_seconds: PositiveFloat = 0.2

lta_seconds: PositiveFloat = 1.0

blinding_window: PositiveFloat = 5

position: StaLtaPosition = 'centred'

min_onset_value: float = 0.4

signal_transform: SignalTransform = 'energy'

phase_map: dict[PhaseName, str] = {
    "P": "cake:P",
    "S": "cake:S",
}

weights: dict[
    PhaseName, Annotated[float, Field(strict=True, ge=0.0)]
] = {"P": 1.0, "S": 1.0}

picker: StaLtaPicker

check_windows() -> StaLta

process_traces(traces: list[Trace]) -> list[WaveformImage]

get_blinding() -> timedelta

get_phases() -> tuple[PhaseDescription, ...]

get_subclasses() -> tuple[type[ImageFunction], ...]

get_images(batch: WaveformBatch) -> WaveformImages

iter_images(
    batch_iterator: AsyncIterator[WaveformBatch],
) -> AsyncIterator[tuple[WaveformImages, WaveformBatch]]

Bases: Picker

Pick phase onsets from STA/LTA characteristic functions.

Triggers are detected on the station's full STA/LTA trace. A trigger turns on where the ratio exceeds median + mad_factor * MAD of the trace and turns off when it falls below half of that excess, median + mad_factor / 2 * MAD. The pick is the peak of the trigger closest to the modeled arrival within the search window, the centered STA/LTA peaks at the phase onset. Peaks before the event origin time are rejected.

Fields:

• mad_factor (PositiveFloat)
• search_window_seconds (PositiveFloat)

mad_factor: PositiveFloat = 5.0

search_window_seconds: PositiveFloat = 5.0

get_trigger_thresholds(
    data: ndarray,
) -> tuple[float, float]

pick_trace(
    trace: Trace,
    phase: PhaseDescription,
    event_time: datetime,
    modelled_arrival: datetime,
) -> ObservedArrival | None

add_picks(
    detections: Sequence[EventDetection],
    images: WaveformImages,
) -> None