Frequency coverage and data-quality confidence

The first questions to ask of any freshly loaded EM survey are which frequencies do I actually have at each station? and how much should I trust them? This example walks line L22PLT of the bundled WILLY_DATA AMT survey (25 stations) through the pycsamt.emtools coverage and confidence diagnostics, from a plain per-station frequency census to a full single-station confidence dashboard.

All figures come straight from pycsamt.emtools.inspect, pycsamt.emtools.frequency, and pycsamt.emtools.qc; see the EM tools guide for the per-module reference.

Load the survey line

_datasets.py (a small shared helper, not itself an example) resolves pyCSAMT’s bundled data by name and returns a ready-to-plot Sites object, so every figure below re-uses the same parsed data.

from _datasets import load_sites

from pycsamt.emtools.frequency import (
    plot_apparent_depth_psection,
    plot_coverage_quality_heatmap,
)
from pycsamt.emtools.inspect import plot_coverage
from pycsamt.emtools.qc import (
    plot_confidence_band_summary,
    plot_confidence_profile,
    plot_frequency_confidence_psection,
    plot_station_confidence_dashboard,
)

L22 = load_sites("amt_l22plt")

1. Frequency coverage census

plot_coverage() marks, for every station, which frequencies carry usable impedance data — the quickest way to spot dropped bands or short-recording stations before any processing.

plot_coverage(L22, figsize=(12, 3.8))

<Axes: xlabel='site', ylabel='period'>

2. Coverage-quality heatmap

plot_coverage_quality_heatmap() turns the same census into a station-by-frequency image shaded by a per-cell data-quality score, so weak corners of the acquisition band stand out at a glance.

plot_coverage_quality_heatmap(L22, figsize=(12, 4.2))

<Axes: xlabel='Station', ylabel='period (s)'>

3. Apparent-depth pseudo-section

plot_apparent_depth_psection() remaps frequency to an approximate skin depth, giving an early, distortion-free sense of the depth range each station illuminates.

plot_apparent_depth_psection(L22, figsize=(12, 4.8))

<Axes: xlabel='Station', ylabel='Period (s)'>

4. Station-confidence profile

plot_confidence_profile() collapses the per-frequency quality scores into one composite confidence value per station and draws it along the profile with a credible band (here the 50–90 % interval), flagging stations that need attention.

plot_confidence_profile(
    L22,
    method="composite",
    figsize=(12, 4.2),
    ci_hi=0.95,
    ci_lo=0.85,
)

<Axes: title={'center': 'Station confidence (composite)'}, xlabel='Distance along profile (m)', ylabel='Confidence ratio'>

5. Frequency-confidence pseudo-section

plot_frequency_confidence_psection() keeps both axes — station and period — so you can see where in the band confidence drops, not just which stations are weak overall.

plot_frequency_confidence_psection(
    L22,
    method="composite",
    figsize=(12, 4.8),
    ci_hi=0.95,
    ci_lo=0.85,
)

<Axes: title={'center': 'Frequency confidence (composite)'}, xlabel='Station', ylabel='$\\log_{10}T$ (s)'>

6. Period-band confidence summary

plot_confidence_band_summary() aggregates the same scores into log-period bands, summarising how trust varies with depth across the whole line.

plot_confidence_band_summary(
    L22,
    method="composite",
    figsize=(8.5, 4.0),
    ci_hi=0.95,
    ci_lo=0.85,
)

<Axes: title={'center': 'Period-band confidence summary (composite)'}, xlabel='$\\log_{10}T$ (s)', ylabel='Confidence / station fraction'>

7. Single-station confidence dashboard

Finally, plot_station_confidence_dashboard() zooms into one station (22-14BF) and lays out every contributing quality metric together — the view to reach for when deciding whether to keep, edit, or drop a specific site.

plot_station_confidence_dashboard(
    L22,
    station="22-14BF",
    method="composite",
    figsize=(10.5, 6.0),
    ci_hi=0.95,
    ci_lo=0.85,
)

<Figure size 1050x600 with 6 Axes>