The data are brightness temperatures measured by the Japanese weather satellite Himawari-9 in its band 7 (3.83 µm, shortwave infrared) over a coastal box off Kashima-nada and Chiba, Japan (35.70–36.08 °N, 140.00–141.20 °E), every 10 minutes through 2024-03-03 UTC. Brightness temperature is the temperature an ideal black body would need to give off the infrared signal the satellite receives; over cloud it describes the cloud top rather than the ground or sea.
This page follows the data from the Advanced Himawari Imager to the 4,719 × 142 trace matrix. In that table each row is one grid node (one point of a 0.01° latitude–longitude grid) and holds its values through the day, its trace; each column is one scan. The page covers what was measured, what the satellite files contain, and how 142 full-disk scans become one trace per grid node.
The Advanced Himawari Imager (AHI) on Himawari-9 views the Earth from geostationary orbit, staying above the same point on the equator at 140.7 °E, 42,164 km from the Earth’s centre. The Japan Meteorological Agency (JMA) documents that it scans the full disk every 10 minutes in 16 bands; band 7 is centred at 3.8289 µm with 2 km pixels below the satellite. Each file records its band, wavelength, calibration and navigation, and the values below are read from those headers.
From the satellite to this page
- DocumentedAHI on Himawari-9
JMA documents a full-disk scan every 10 minutes in 16 bands
- HeldRaw segment files
142 scans of band 7, strip 2, from NOAA’s copy of the JMA archive
- Not recordedGridded product
Resampled to a 0.01° grid (JAXA’s, by its file names); the method is not recorded
- DocumentedNotebook crop and stack
The submitted code: crop the box, flatten, stack the scans as columns
- HeldSupplied matrix
4,719 × 142; 90% of values rebuilt exactly here from the raw files
How to read the labels: Documented — described by JMA or by the submitted code, but not rerun here. Reproduced — rerun here from the raw files. Held — files this page actually has. Not recorded — the step happened, but no file we hold says how.
One acquisition unit: one full-disk scan
One scan images the whole visible disk of the Earth in one 10-minute slot. For each band, the disk is stored as 10 strips (segments), from north to south, each 550 lines × 5,500 columns. The coastal box lies in strip 2. The figures show where that strip sits on the disk, then the strip itself from the 03:00 UTC scan, as recorded.
Shown at one-fifth resolution; black is space outside the disk. The box marks the crop.
The crop and its surroundings at native resolution: every square is one pixel. Here a pixel spans about 0.0229° of longitude and 0.0251° of latitude, larger than the nominal 2 km, which applies only directly below the satellite.
What each file records
| Field | Unit | Meaning |
|---|---|---|
| counts | — | Detector output per pixel, 14 valid bits; two reserved values mark error and space |
| calibration gain, offset | W m⁻² sr⁻¹ µm⁻¹ per count | Linear conversion from counts to radiance |
| central wavelength | µm | Band centre used to convert radiance to temperature |
| c0, c1, c2 | K, —, K⁻¹ | Correction from effective to brightness temperature |
| sub-satellite longitude, distance, Earth radii | °, km | Viewing geometry |
| CFAC, LFAC, COFF, LOFF | — | Scaling and offsets of the fixed viewing-angle grid (navigation) |
| segment, first line | — | Which strip of the disk the file holds |
| observation times | date-time | When given lines of the strip were scanned |
Strip 2 takes about 47 s to scan; the box’s lines are observed about 100 s after the slot time.
Raw source: Himawari-9 Standard Data, Japan Meteorological Agency, via NOAA’s open-data copy.
A file, opened
Each file is 11 header blocks followed by the count array. Header values from the 03:00 UTC file:
| Header field | Value |
|---|---|
| Satellite / centre | Himawari-9 / MSC |
| Observation area | FLDK |
| Scan of this segment | 03:00:54 → 03:01:40 UTC |
| Band / central wavelength | 7 / 3.8289 µm |
| Segment | 2 of 10, lines 551–1100 |
| Array | 550 lines × 5500 columns, 16-bit counts (14 valid bits) |
| Count → radiance | L = -0.000739382 × count + 12.0889 |
| Radiance → Tb correction | Tb = -0.4628 + 1.00143·Te + -9.01e-07·Te² |
| Invalid counts | 65535 (error), 65534 (outside scan) |
| Sub-satellite longitude | 140.7 °E |
| Header / data bytes | 1,523 / 6,050,000 |
| Format version | 1.3 |
HS_H09_20240303_0300_B07_FLDK_R20_S0210.DAT ├─ header blocks 1–11 basic · data · projection · navigation · calibration · … └─ counts uint16[550][5500] line 551 (north) → 1100
One file as a signal
The chart shows the raw counts along one scan line through the box (line 958), across all 5,500 columns. Lower counts mean a warmer scene. Beyond the edge of the disk the file stores a reserved “space” value instead, drawn as gaps.
The same file’s calibration, from counts to brightness temperature. The gain is negative, so a higher count is colder.
How the files cover the day
Each mark is one slot of the 144 expected for the day. 142 files exist; the archive has no full-disk files at 02:40 and 14:40 UTC. Hover over a mark to see its file name, scan time and when the box was observed.
What the raw record does not give you directly
- Values are detector counts, not temperatures. Each file carries its own calibration, and it runs backwards: the gain is -0.000739 per count.
- Pixels sit on the satellite’s viewing-angle grid, not on latitude and longitude. A position is found only through the navigation parameters in the header.
- Pixels near 36 °N are larger than the nominal 2 km, about 0.0229° × 0.0251°, so a 0.01° grid is finer than the data.
- The day has 142 scans, not 144. No files exist for 02:40 and 14:40 UTC, so column position is not time.
- A file’s name gives its slot, not its observation time. The box is scanned about 100 s after the slot starts.
The students’ preparation notebook did not read these raw files. It read a gridded product made from them (by the Japan Aerospace Exploration Agency, JAXA, judging by its file names), cropped the box and stacked the scans. To make each step visible, this page rebuilds the same matrix from the raw files. The demo still uses the students’ supplied values as its design matrix; the last table compares them with the rebuild.
| Step | Evidence | Operation | Result |
|---|---|---|---|
| 0 | Not recorded | The supplied values passed through a gridded product (JAXA’s, by its file names) whose resampling is not recorded | the supplied matrix |
| 1 | Reproduced | Take band 7, strip 2, for every 10-minute slot of 2024-03-03 | 142 files (02:40 and 14:40 absent) |
| 2 | Reproduced | Counts → radiance with the file’s gain and offset; radiance → brightness temperature by Planck’s law at 3.8289 µm, then the file’s correction | temperature per pixel (K) |
| 3 | Reproduced | Define the grid: 39 latitudes (36.08 → 35.70 °N) × 121 longitudes (140.00 → 141.20 °E), step 0.01° | 4,719 grid nodes |
| 4 | Reproduced | Navigate each node to the nearest pixel; read its temperature | 901 distinct pixels read |
| 5 | Reproduced | Flatten each scan row by row (north to south, west to east) and stack scans as columns | 4,719 × 142 matrix |
No cloud masking, outlier removal, interpolation or normalisation is applied.
Where the pixels go
The 4,719 grid nodes read 901 distinct pixels: the grid is finer than the pixels, so 3,818 nodes share a pixel with another node.
Follow one grid node
Pick a node and a scan to see the node at each step, from the counts around it to the row the matrix stores.
Checks on the supplied matrix
| Check | Result |
|---|---|
| Satellite and band | Himawari-9, band 7 (3.8289 µm) |
| Unit | Kelvin: rebuilt temperatures match the supplied values |
| Scans present | 142 of 144; the archive has no files for 02:40 and 14:40 UTC |
| Column order | 142 of 142 columns agree best with their own slot, not a neighbour |
| Values identical to 0.01 K | 90% of all 670,098 values |
| Scans identical in every value | 77 of 142 |
| Differing values inside the range of the four surrounding pixels | 99.9% of 66,304 |
| Difference, median / 95th percentile / max | 0.003 / 0.51 / 13.4 K |
The 77 scans from 08:20–21:10 UTC, evening to dawn in Japan, reproduce exactly. In the daytime scans, the values that differ always lie between those of the surrounding pixels: the gridded product the notebook used interpolates between pixels there, where this rebuild takes the nearest one. Its method is not documented here, so the page keeps the supplied values.
Each row is one grid node’s brightness temperature through the day; rows run west to east within each latitude, north to south. Each column is one scan, labelled by its slot time. Colour is brightness temperature. Hover over a row to read it, and pick a scan to see that column as a map.
One column as a map
Values of the selected row
All values are shown at their stored 0.01 K precision. Held as three files: the matrix, and its latitude and longitude axes.
| Column(s) | Unit | Meaning |
|---|---|---|
| 0–141 | K | Band-7 brightness temperature, one column per scan; slot times 00:00–23:50 UTC without 02:40 and 14:40 |
Labels kept beside the matrix
| Label | Unit | Meaning |
|---|---|---|
| latitude, longitude | °N, °E | Grid node position; row r has latitude index ⌊r / 121⌋ and longitude index r mod 121 |
| row index | — | Position in the flattened grid; not a class or target |
- 888 rows exactly repeat an earlier row, because neighbouring grid nodes read the same pixel. They remain separate nodes.
- The rebuild from the raw files matches 90% of the supplied values exactly and 77 of 142 scans in full. The rest differ by up to 13 K, always within the range of the surrounding pixels: the gridded product the notebook read resamples differently, and its method is not documented here.
- No cloud mask, class label or target accompanies the matrix, so a clustering of it cannot be scored against ground truth.
- The supplied files carry no licence. JMA and NOAA request attribution for the raw files.
TBB cluster explorer: its Workflow page embeds and clusters these traces, and its Insight page shows the clusters on a map.