I'd like to share some of my understanding about the input data (seismic waveforms) and output data (subsurface velocity profiles). Hopefully, this will be helpful. Thanks!
\nSeismic waveform observation refers to recording the amplitude of seismic waves at the surface (and sometimes in boreholes) using receivers (geophones) as the waves travel through the subsurface and arrive at the surface over time.
\nThese waves are usually generated artificially (by explosions, hammer impacts, vibrators, etc.) or by natural earthquakes.
\nMultiple Sources: Several source points are arranged on the surface; each source generates a seismic wave.
\nMultiple Receivers: Many receivers (like an array of microphones) are placed along the surface at regular intervals.
\nMultiple Time Steps: Each receiver records the complete wave signal over time, starting from the moment a source is triggered, with a very high sampling rate (e.g., once every millisecond).
\nData Shape Example:
\nIf there are 5 sources, 10 receivers, and 1,000 time samples, each observation is shaped as (5, 1000, 10).
Each value represents the amplitude recorded by a receiver at a specific time after a certain source is triggered.
\nImagine placing 10 seismic receivers along a road, then striking the ground at 5 different positions with a hammer. Each time you strike, all 10 receivers record 1,000 time samples simultaneously. Each strike produces a set of \"receiver-time curves.\"
\nBecause of underground features (voids, soft soil, bedrock, etc.), the waveforms recorded by different receivers vary—some receive the wave earlier or later, with stronger or weaker signals.
\nThis type of observation data is called Seismic Full Waveform Data or a Seismic Record Section.
\nEach observation reflects how the subsurface structure affects seismic wave propagation.
\nA velocity profile divides an underground area (e.g., 20 meters deep by 10 meters wide) into a grid. Each cell in the grid indicates the seismic wave velocity at that point (in meters/second).
\nHigh velocity means hard rock or compact material; low velocity indicates voids, soft soil, liquids, or anomalies.
\nFor example, if you grid a 20m deep by 10m wide area into 20 rows × 10 columns, the velocity profile is a 20×10 2D array.
\nEach value represents the seismic velocity at that grid cell.
\nSuppose you want to detect cavities beneath a road. If inversion results show an especially low velocity (e.g., 800 m/s) at row 12, column 5, that area likely corresponds to a cavity or loose zone.
\nA velocity profile (or velocity model) is one of the most common imaging tools in geophysical exploration, oil & gas surveys, and geological hazard monitoring.
\nEngineers use it to infer underground structure, locate faults, and identify resources or anomalies.
\nEach set of seismic waveform observations (e.g., 5×1000×10) corresponds to the seismic response from a specific underground velocity profile (e.g., 20×10).
\nIn other words: only a particular subsurface structure can produce a specific waveform set.
\nThink of the input waveforms as a movie, capturing every detail of wave propagation, reflection, and delay; the output velocity profile is like a photo, revealing the distribution of underground features.
\nThe subsurface is invisible: It’s impossible to directly measure the velocity at every point tens or hundreds of meters underground.
\nDrilling is expensive: Direct velocity measurements can only be made at a few boreholes, which can't cover large areas.
\nMost regions rely on indirect inference via wave propagation.
\nAs seismic waves pass through the ground, their shape, travel time, and amplitude change due to velocity, damping, and reflections. The observed waveforms are complex functions of the subsurface structure and velocity distribution.
\nEngineers compare theoretical models to actual waveforms and iterate to reconstruct the subsurface velocity distribution.
\nThe input data’s time sequence (time steps) records the full process of seismic waves propagating, reflecting, and interacting in the subsurface.
\nThe output velocity profile is a static snapshot of the underground—showing only the physical properties at each point.
\nIt’s like a doctor inferring heart health from a dynamic ECG waveform (temporal) but diagnosing spatial abnormalities.
\nUltrasound Imaging: A doctor listens to echo signals and sees a cross-sectional image of the body. Input: time-varying signals; Output: static anatomical image.
\nCT/X-ray: Signals from different angles are used to reconstruct a static image.
\nInput: Multi-source, multi-receiver, time-varying waveforms at the surface—capturing how the subsurface responds to seismic waves.
\nOutput: Velocity profile image of the same region, revealing physical properties at each underground point.
\nInversion Task: Use complex mathematical or AI models to “translate” dynamic signals into static structural images.
\nYou can think of it like this:
\n“We cannot see underground directly, so we ‘listen’ to seismic waves at the surface. Like a doctor using ultrasound to infer body structure from echoes, we use physics and mathematics to convert these ‘dynamic sounds’ into a ‘static map’ of the subsurface.”
I'd like to share some of my understanding about the input data (seismic waveforms) and output data (subsurface velocity profiles). Hopefully, this will be helpful. Thanks!
\nSeismic waveform observation refers to recording the amplitude of seismic waves at the surface (and sometimes in boreholes) using receivers (geophones) as the waves travel through the subsurface and arrive at the surface over time.
\nThese waves are usually generated artificially (by explosions, hammer impacts, vibrators, etc.) or by natural earthquakes.
\nMultiple Sources: Several source points are arranged on the surface; each source generates a seismic wave.
\nMultiple Receivers: Many receivers (like an array of microphones) are placed along the surface at regular intervals.
\nMultiple Time Steps: Each receiver records the complete wave signal over time, starting from the moment a source is triggered, with a very high sampling rate (e.g., once every millisecond).
\nData Shape Example:
\nIf there are 5 sources, 10 receivers, and 1,000 time samples, each observation is shaped as (5, 1000, 10).
Each value represents the amplitude recorded by a receiver at a specific time after a certain source is triggered.
\nImagine placing 10 seismic receivers along a road, then striking the ground at 5 different positions with a hammer. Each time you strike, all 10 receivers record 1,000 time samples simultaneously. Each strike produces a set of \"receiver-time curves.\"
\nBecause of underground features (voids, soft soil, bedrock, etc.), the waveforms recorded by different receivers vary—some receive the wave earlier or later, with stronger or weaker signals.
\nThis type of observation data is called Seismic Full Waveform Data or a Seismic Record Section.
\nEach observation reflects how the subsurface structure affects seismic wave propagation.
\nA velocity profile divides an underground area (e.g., 20 meters deep by 10 meters wide) into a grid. Each cell in the grid indicates the seismic wave velocity at that point (in meters/second).
\nHigh velocity means hard rock or compact material; low velocity indicates voids, soft soil, liquids, or anomalies.
\nFor example, if you grid a 20m deep by 10m wide area into 20 rows × 10 columns, the velocity profile is a 20×10 2D array.
\nEach value represents the seismic velocity at that grid cell.
\nSuppose you want to detect cavities beneath a road. If inversion results show an especially low velocity (e.g., 800 m/s) at row 12, column 5, that area likely corresponds to a cavity or loose zone.
\nA velocity profile (or velocity model) is one of the most common imaging tools in geophysical exploration, oil & gas surveys, and geological hazard monitoring.
\nEngineers use it to infer underground structure, locate faults, and identify resources or anomalies.
\nEach set of seismic waveform observations (e.g., 5×1000×10) corresponds to the seismic response from a specific underground velocity profile (e.g., 20×10).
\nIn other words: only a particular subsurface structure can produce a specific waveform set.
\nThink of the input waveforms as a movie, capturing every detail of wave propagation, reflection, and delay; the output velocity profile is like a photo, revealing the distribution of underground features.
\nThe subsurface is invisible: It’s impossible to directly measure the velocity at every point tens or hundreds of meters underground.
\nDrilling is expensive: Direct velocity measurements can only be made at a few boreholes, which can't cover large areas.
\nMost regions rely on indirect inference via wave propagation.
\nAs seismic waves pass through the ground, their shape, travel time, and amplitude change due to velocity, damping, and reflections. The observed waveforms are complex functions of the subsurface structure and velocity distribution.
\nEngineers compare theoretical models to actual waveforms and iterate to reconstruct the subsurface velocity distribution.
\nThe input data’s time sequence (time steps) records the full process of seismic waves propagating, reflecting, and interacting in the subsurface.
\nThe output velocity profile is a static snapshot of the underground—showing only the physical properties at each point.
\nIt’s like a doctor inferring heart health from a dynamic ECG waveform (temporal) but diagnosing spatial abnormalities.
\nUltrasound Imaging: A doctor listens to echo signals and sees a cross-sectional image of the body. Input: time-varying signals; Output: static anatomical image.
\nCT/X-ray: Signals from different angles are used to reconstruct a static image.
\nInput: Multi-source, multi-receiver, time-varying waveforms at the surface—capturing how the subsurface responds to seismic waves.
\nOutput: Velocity profile image of the same region, revealing physical properties at each underground point.
\nInversion Task: Use complex mathematical or AI models to “translate” dynamic signals into static structural images.
\nYou can think of it like this:
\n“We cannot see underground directly, so we ‘listen’ to seismic waves at the surface. Like a doctor using ultrasound to infer body structure from echoes, we use physics and mathematics to convert these ‘dynamic sounds’ into a ‘static map’ of the subsurface.”