Clarify WildFIRE-FM input channel contract
Browse files
README.md
CHANGED
|
@@ -215,6 +215,9 @@ This Hub repository is model-first. It provides the released weights, model-load
|
|
| 215 |
|
| 216 |
The pretrained checkpoints expect a fixed 16-channel tensor in `[channel, y, x]`
|
| 217 |
order. Channel order matters.
|
|
|
|
|
|
|
|
|
|
| 218 |
|
| 219 |
| Channels | Names | Source |
|
| 220 |
|---:|---|---|
|
|
@@ -228,6 +231,8 @@ used for supporting event-level tasks and are not part of the 16-channel
|
|
| 228 |
pretrained occupancy input. Full channel-level details are in
|
| 229 |
`data_sources/DATA_SOURCES.md`; the same contract is available as
|
| 230 |
`models/wildfire_fm/input_channels.json`.
|
|
|
|
|
|
|
| 231 |
|
| 232 |
---
|
| 233 |
|
|
|
|
| 215 |
|
| 216 |
The pretrained checkpoints expect a fixed 16-channel tensor in `[channel, y, x]`
|
| 217 |
order. Channel order matters.
|
| 218 |
+
Inputs use native source units with no mean/std normalization, min/max scaling,
|
| 219 |
+
or unit conversion. Invalid or missing values are zero-filled, with validity
|
| 220 |
+
masks included as channels 10 and 11.
|
| 221 |
|
| 222 |
| Channels | Names | Source |
|
| 223 |
|---:|---|---|
|
|
|
|
| 231 |
pretrained occupancy input. Full channel-level details are in
|
| 232 |
`data_sources/DATA_SOURCES.md`; the same contract is available as
|
| 233 |
`models/wildfire_fm/input_channels.json`.
|
| 234 |
+
The `cape` channel is the HRRR surface instantaneous CAPE field selected with
|
| 235 |
+
`typeOfLevel=surface` and `stepType=instant`.
|
| 236 |
|
| 237 |
---
|
| 238 |
|