\n\nI was wondering: “If seismic waves come from many different shots, how can one velocity map explain them all—and why do geophysicists call that map ‘velocity’ in the first place?”
\n
A velocity model is a fixed 2‑D (or 3‑D) grid that stores one number per cell: the speed at which a seismic P‑wave would travel through that bit of rock—e.g. 2000 m/s in sands, 3500 m/s in carbonates.
\nThose numbers are intrinsic rock properties, just like density or porosity.
\nThey don’t care where your air gun sits or how many times you fire it.
one subsurface scenario\n└── velocity model ← single (nz, nx) map we want\n ├── shot‑ gather \\\n ├── shot‑ gather } different “photos” of the same map\n └── … shot‑(N−) gather /\n| Role | \nWhat it contributes | \n
|---|---|
| Sources (shots) | \nIlluminate the same ground from different angles. | \n
| Receivers | \nRecord how that fixed ground scatters each shot. | \n
In a learning setup each sample looks like:
\nx – shape (n_shots × nt × nx): a mini‑movie with n_shots frames (channels), one per source. y – shape (nz × nx): the single, ground‑truth velocity image to predict.The network sees all shot gathers simultaneously and fuses their complementary information to reconstruct that one immutable map.
\n| Everyday use | \nScientific use | \n
|---|---|
| “He ran with great velocity.” → sounds like a verb. | \nVelocity field = a property assigned to every point in space. | \n
| Wind‑velocity map, fluid‑flow velocity, P‑wave velocity in rocks. | \n\n |
In our case “velocity” really means P‑wave speed—a scalar value stored per grid cell. Think of colouring the subsurface with numbers:
\ncell (10,7) = 2500 m/s\ncell (22,3) = 3400 m/s\n...\nThose numbers describe how fast a wave would propagate—not anything currently in motion.
\nOne Earth → one velocity map.
\nMultiple shots merely give multiple views of that fixed property field.
Velocity is a noun here.
\nIt’s the rock’s wave‑carrying ability, just like temperature is a property of air.
So the term “velocity model” isn’t about motion—it’s a spatial map of wave‑propagation speed that all your shots, receivers, and neural nets are trying to uncover.
", "messages": [ { "id": "3182780", "postDate": "04/19/2025 19:53:33", "content": "\n\nI was wondering: “If seismic waves come from many different shots, how can one velocity map explain them all—and why do geophysicists call that map ‘velocity’ in the first place?”
\n
A velocity model is a fixed 2‑D (or 3‑D) grid that stores one number per cell: the speed at which a seismic P‑wave would travel through that bit of rock—e.g. 2000 m/s in sands, 3500 m/s in carbonates.
\nThose numbers are intrinsic rock properties, just like density or porosity.
\nThey don’t care where your air gun sits or how many times you fire it.
one subsurface scenario\n└── velocity model ← single (nz, nx) map we want\n ├── shot‑ gather \\\n ├── shot‑ gather } different “photos” of the same map\n └── … shot‑(N−) gather /\n| Role | \nWhat it contributes | \n
|---|---|
| Sources (shots) | \nIlluminate the same ground from different angles. | \n
| Receivers | \nRecord how that fixed ground scatters each shot. | \n
In a learning setup each sample looks like:
\nx – shape (n_shots × nt × nx): a mini‑movie with n_shots frames (channels), one per source. y – shape (nz × nx): the single, ground‑truth velocity image to predict.The network sees all shot gathers simultaneously and fuses their complementary information to reconstruct that one immutable map.
\n| Everyday use | \nScientific use | \n
|---|---|
| “He ran with great velocity.” → sounds like a verb. | \nVelocity field = a property assigned to every point in space. | \n
| Wind‑velocity map, fluid‑flow velocity, P‑wave velocity in rocks. | \n\n |
In our case “velocity” really means P‑wave speed—a scalar value stored per grid cell. Think of colouring the subsurface with numbers:
\ncell (10,7) = 2500 m/s\ncell (22,3) = 3400 m/s\n...\nThose numbers describe how fast a wave would propagate—not anything currently in motion.
\nOne Earth → one velocity map.
\nMultiple shots merely give multiple views of that fixed property field.
Velocity is a noun here.
\nIt’s the rock’s wave‑carrying ability, just like temperature is a property of air.
So the term “velocity model” isn’t about motion—it’s a spatial map of wave‑propagation speed that all your shots, receivers, and neural nets are trying to uncover.
", "rawMarkdown": "> I was wondering: *“If seismic waves come from many different shots, how can one **velocity** map explain them all—and why do geophysicists call that map ‘velocity’ in the first place?”*\n\n---\n\n## 1 What a Velocity Model *Is*\n\nA **velocity model** is a fixed 2‑D (or 3‑D) grid that stores one number per cell: the speed at which a seismic **P‑wave** would travel through that bit of rock—e.g. 2000 m/s in sands, 3500 m/s in carbonates.\n\n*Those numbers are intrinsic rock properties, just like density or porosity.* \nThey don’t care where your air gun sits or how many times you fire it.\n\n---\n\n## 2 Why One Model Serves Many Shots\n\n```\none subsurface scenario\n└── velocity model y ← single (nz, nx) map we want\n ├── shot‑0 gather x[0] \\\n ├── shot‑1 gather x[1] } different “photos” of the same map\n └── … shot‑(N−1) gather /\n```\n\n| Role | What it contributes |\n|------|---------------------|\n| **Sources (shots)** | Illuminate the same ground from different angles. |\n| **Receivers** | Record how that fixed ground scatters each shot. |\n\nIn a learning setup each sample looks like:\n\n* **`x`** – shape **(n_shots × nt × nx)**: a mini‑movie with *n_shots* frames (channels), one per source. \n* **`y`** – shape **(nz × nx)**: the single, ground‑truth velocity image to predict.\n\nThe network sees all shot gathers simultaneously and fuses their complementary information to reconstruct that one immutable map.\n\n---\n\n## 3 Why “Velocity” Is a **Thing**, Not an Action\n\n| Everyday use | Scientific use |\n|--------------|----------------|\n| *“He ran with great velocity.”* → sounds like a verb. | **Velocity field** = a property assigned to every point in space. |\n| Wind‑velocity map, fluid‑flow velocity, P‑wave velocity in rocks. |\n\nIn our case “velocity” really means **P‑wave speed**—a scalar value stored per grid cell. Think of colouring the subsurface with numbers: \n\n```\ncell (10,7) = 2500 m/s\ncell (22,3) = 3400 m/s\n...\n```\n\nThose numbers describe how fast a wave *would* propagate—not anything currently in motion.\n\n---\n\n## 4 Take‑away\n\n* **One Earth → one velocity map.** \n Multiple shots merely give multiple views of that fixed property field.\n\n* **Velocity is a noun here.** \n It’s the rock’s wave‑carrying ability, just like temperature is a property of air.\n\nSo the term **“velocity model”** isn’t about motion—it’s a spatial map of wave‑propagation speed that all your shots, receivers, and neural nets are trying to uncover.", "votes": null }, { "id": "3183825", "postDate": "04/21/2025 10:40:28", "content": "What model do you use?
", "rawMarkdown": "What model do you use?", "votes": null }, { "id": "3184011", "postDate": "04/21/2025 14:35:08", "content": "I'm experimenting with SmartConv / Unet / FWIGan, etc. It's way too early to commit to a model. Just getting a feel for them at this stage, seeing what they do well and don't do well.
\nThe final model will very likely be an ensemble of what works best.
", "rawMarkdown": "I'm experimenting with SmartConv / Unet / FWIGan, etc. It's way too early to commit to a model. Just getting a feel for them at this stage, seeing what they do well and don't do well. \n\nThe final model will very likely be an ensemble of what works best.", "votes": null } ], "comments": [ { "id": 3183825, "author_name": "", "author_url": "", "post_date": "04/21/2025 10:40:28", "content": "What model do you use?
", "votes": null, "replies": [ { "id": 3184011, "author_name": "tpmeli", "author_url": "", "post_date": "04/21/2025 14:35:08", "content": "I'm experimenting with SmartConv / Unet / FWIGan, etc. It's way too early to commit to a model. Just getting a feel for them at this stage, seeing what they do well and don't do well.
\nThe final model will very likely be an ensemble of what works best.
", "votes": null, "replies": [] } ] } ], "raw_markdown_by_id": { "3182780": "> I was wondering: *“If seismic waves come from many different shots, how can one **velocity** map explain them all—and why do geophysicists call that map ‘velocity’ in the first place?”*\n\n---\n\n## 1 What a Velocity Model *Is*\n\nA **velocity model** is a fixed 2‑D (or 3‑D) grid that stores one number per cell: the speed at which a seismic **P‑wave** would travel through that bit of rock—e.g. 2000 m/s in sands, 3500 m/s in carbonates.\n\n*Those numbers are intrinsic rock properties, just like density or porosity.* \nThey don’t care where your air gun sits or how many times you fire it.\n\n---\n\n## 2 Why One Model Serves Many Shots\n\n```\none subsurface scenario\n└── velocity model y ← single (nz, nx) map we want\n ├── shot‑0 gather x[0] \\\n ├── shot‑1 gather x[1] } different “photos” of the same map\n └── … shot‑(N−1) gather /\n```\n\n| Role | What it contributes |\n|------|---------------------|\n| **Sources (shots)** | Illuminate the same ground from different angles. |\n| **Receivers** | Record how that fixed ground scatters each shot. |\n\nIn a learning setup each sample looks like:\n\n* **`x`** – shape **(n_shots × nt × nx)**: a mini‑movie with *n_shots* frames (channels), one per source. \n* **`y`** – shape **(nz × nx)**: the single, ground‑truth velocity image to predict.\n\nThe network sees all shot gathers simultaneously and fuses their complementary information to reconstruct that one immutable map.\n\n---\n\n## 3 Why “Velocity” Is a **Thing**, Not an Action\n\n| Everyday use | Scientific use |\n|--------------|----------------|\n| *“He ran with great velocity.”* → sounds like a verb. | **Velocity field** = a property assigned to every point in space. |\n| Wind‑velocity map, fluid‑flow velocity, P‑wave velocity in rocks. |\n\nIn our case “velocity” really means **P‑wave speed**—a scalar value stored per grid cell. Think of colouring the subsurface with numbers: \n\n```\ncell (10,7) = 2500 m/s\ncell (22,3) = 3400 m/s\n...\n```\n\nThose numbers describe how fast a wave *would* propagate—not anything currently in motion.\n\n---\n\n## 4 Take‑away\n\n* **One Earth → one velocity map.** \n Multiple shots merely give multiple views of that fixed property field.\n\n* **Velocity is a noun here.** \n It’s the rock’s wave‑carrying ability, just like temperature is a property of air.\n\nSo the term **“velocity model”** isn’t about motion—it’s a spatial map of wave‑propagation speed that all your shots, receivers, and neural nets are trying to uncover.", "3183825": "What model do you use?", "3184011": "I'm experimenting with SmartConv / Unet / FWIGan, etc. It's way too early to commit to a model. Just getting a feel for them at this stage, seeing what they do well and don't do well. \n\nThe final model will very likely be an ensemble of what works best." }, "source": "meta" }