Create asset for Hurricane season 2005 from NOAA SOS

[hn-88]

from Science on a Sphere - https://sos.noaa.gov/catalog/datasets/hurricane-season-2005/

• download all the frames - 60 GB+
• create a video from the frames with yaw=180
• create time-varying asset check if the mapped to time example works.

[hn-88]

Creating a video from the frames seems a bit tricky, since the frames are named like

ir_combined_20050601_0015.jpg
ir_combined_20050601_0030.jpg
ir_combined_20050601_0045.jpg
ir_combined_20050601_0100.jpg
ir_combined_20050601_0115.jpg
ir_combined_20050601_0130.jpg
ir_combined_20050601_0145.jpg
ir_combined_20050601_0200.jpg
ir_combined_20050601_0215.jpg
ir_combined_20050601_0230.jpg
ir_combined_20050601_0245.jpg
ir_combined_20050601_0300.jpg
etc

and not

ir_0000001.jpg
ir_0000002.jpg
etc

Blender may be able to do it? ChatGPT gives these useful suggestions, will try them.

Yes, you can do it without renaming. Since your filenames are not sequential numbers, the simplest method is to let ffmpeg read them using a glob pattern and rely on the alphabetical order, which works because your timestamps are sortable.

Method 1 — Using glob pattern (simplest)

ffmpeg -framerate 10 -pattern_type glob -i 'ir_combined_*.jpg' \
       -c:v libx264 -pix_fmt yuv420p output.mp4

Explanation

• -pattern_type glob → allows wildcard matching.
• 'ir_combined_*.jpg' → all matching files.
• Because your filenames contain YYYYMMDD_HHMM, alphabetical order = chronological order.
• -framerate 10 → playback speed (10 frames/sec). Adjust as needed.

---

Method 2 — Explicit sorted list (most robust)

If filenames might not be perfectly ordered, you can feed ffmpeg a sorted list:

ls ir_combined_*.jpg | sort | ffmpeg -f image2pipe -framerate 10 -i - \
    -c:v libx264 -pix_fmt yuv420p output.mp4

But this is usually unnecessary if filenames already sort correctly.

---

Method 3 — Using concat list (if you want per-frame timing)

If each timestamp corresponds to real time intervals, you could generate a list file:

file 'ir_combined_20050601_0015.jpg'
duration 1
file 'ir_combined_20050601_0030.jpg'
duration 1
...

Then:

ffmpeg -f concat -safe 0 -i list.txt -vsync vfr -pix_fmt yuv420p output.mp4

---

✅ In your case: Method 1 should work immediately because the timestamps are lexicographically ordered.

---

💡 If you want, I can also show a very useful ffmpeg trick for timestamped satellite/weather images: automatically overlay the time from the filename onto the video frames (without modifying the images).

You can overlay the timestamp directly from the filename using the drawtext filter. The trick is that ffmpeg exposes the input filename as metadata, which drawtext can read.

Example command

ffmpeg -framerate 10 -pattern_type glob -i 'ir_combined_*.jpg' \
-vf "drawtext=fontcolor=white:fontsize=48:box=1:boxcolor=black@0.5:boxborderw=10:
text='%{metadata\:lavf.image2dec.source_basename}':x=20:y=20" \
-c:v libx264 -pix_fmt yuv420p output.mp4

What this does

• lavf.image2dec.source_basename → gives the filename of the current image
• drawtext → writes it on the frame
• box=1 → adds a dark background so the text is readable
• x=20:y=20 → top-left corner

The result will show text like:

ir_combined_20050601_0015.jpg

---

If you want a cleaner timestamp

You can strip the prefix/suffix using textfile trickery or a small script, but an easier way is to extract only the time using substring expansion.

Example:

ffmpeg -framerate 10 -pattern_type glob -i 'ir_combined_*.jpg' \
-vf "drawtext=fontcolor=white:fontsize=48:box=1:boxcolor=black@0.5:
text='%{metadata\:lavf.image2dec.source_basename\:12\:13}':x=20:y=20" \
-c:v libx264 -pix_fmt yuv420p output.mp4

This can show just something like:

20050601_0015

(depending on filename length).

---

If you want something very nice looking

For satellite-style overlays:

-vf "drawtext=fontcolor=yellow:fontsize=48:borderw=2:bordercolor=black:
x=w-tw-40:y=h-th-40:text='%{metadata\:lavf.image2dec.source_basename}'"

This places the timestamp bottom-right with an outline (like NOAA or ISRO animations).

---

✅ If you'd like, I can also show a very handy ffmpeg trick for your exact filenames that will automatically convert

ir_combined_20050601_0015.jpg

into a proper formatted timestamp:

2005-06-01 00:15 UTC

directly inside ffmpeg (no preprocessing).

[hn-88]

To catch missing frames,

start="2005-06-01 00:15"; end="2006-01-03 23:45"; \
t=$(date -d "$start" +%s); endt=$(date -d "$end" +%s); \
while [ $t -le $endt ]; do \
  f=$(date -u -d "@$t" +ir_combined_%Y%m%d_%H%M.jpg); \
  [ -f "$f" ] || echo "$f"; \
  t=$((t+900)); \
done

3405 frames are missing. So, we need to adjust our video accordingly.

[hn-88]

ChatGPT suggests a constant framerate, and interpolation automatically with minterpolate with -pattern_type glob

ffmpeg -pattern_type glob -framerate 4 -i 'ir_combined_*.jpg' \
-vf "minterpolate=fps=4:mi_mode=mci:mc_mode=aobmc:vsbmc=1" \
-c:v libx264 -pix_fmt yuv420p output.mp4

[hn-88]

With timestamps in video frame,

ffmpeg -pattern_type glob -framerate 4 -i 'ir_combined_*.jpg' \
-vf "tblend=all_mode=average,framestep=1,\
drawtext=fontcolor=yellow:fontsize=48:borderw=2:bordercolor=black:\
x=w-tw-40:y=h-th-40:\
text='%{metadata\:lavf.image2dec.source_basename}'" \
-c:v h264_nvenc -pix_fmt yuv420p ../ir_w_timestamp.mp4

But since metadata is not present, using substrings of the filename,

ffmpeg -pattern_type glob -framerate 4 -i 'ir_combined_*.jpg' \
-vf "tblend=all_mode=average,framestep=1,\
drawtext=fontcolor=yellow:fontsize=48:borderw=2:bordercolor=black:\
x=w-tw-40:y=h-th-40:\
text='%{filename\:12\:8} %{filename\:21\:4}'" \
-c:v h264_nvenc -pix_fmt yuv420p ../ir_w_timestamp.mp4

[hn-88]

Since the above two suggestions from ChatGPT did not work (ffmpeg complains about not knowing %filename etc., some further suggestions -

Since you have 3405 missing frames, some jumps may represent hours or even days. A good way to make the timelapse visually smoother is:

• Normal gaps (≤30 min) → simple blend (tblend)
• Large gaps → insert a short fade transition

We can automate this while building the frames.txt.

---

Step 1 — Generate concat list with timestamps and fades

Run in the image directory:

prev=""
for f in $(ls ir_combined_*.jpg | sort); do
    ts=$(echo "$f" | sed -E 's/ir_combined_([0-9]{4})([0-9]{2})([0-9]{2})_([0-9]{2})([0-9]{2})\.jpg/\1-\2-\3 \4:\5/')
    epoch=$(date -d "$ts" +%s)

    if [ -n "$prev" ]; then
        gap=$(( (epoch - prev) / 60 ))   # minutes
        if [ $gap -gt 30 ]; then
            echo "# gap of $gap minutes"
        fi
    fi

    echo "file '$f'" >> frames.txt
    echo "file_packet_metadata timestamp=$ts" >> frames.txt

    prev=$epoch
done

This still produces the same ordered list but records the gap sizes, which lets you inspect where big jumps occur.

---

Step 2 — Create the video with fade smoothing

ffmpeg -f concat -safe 0 -r 4 -i frames.txt \
-vf "tblend=all_mode=average,\
minterpolate=fps=40:mi_mode=mci:mc_mode=aobmc:vsbmc=1,\
drawtext=fontfile=/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf:\
text='%{metadata\:timestamp}':\
fontcolor=yellow:fontsize=48:borderw=3:bordercolor=black:\
x=w-tw-40:y=h-th-40" \
-c:v libx264 -pix_fmt yuv420p output.mp4

---

What this pipeline does

1. Reads images in chronological order
2. Blends adjacent frames to soften small gaps
3. Interpolates motion to 40 fps for smooth playback
4. Overlays clean timestamp
5. Keeps your true 15-minute cadence

---

Result

Your video will show:

• smooth cloud motion
• fewer abrupt jumps
• timestamp like

2005-06-01 00:15

bottom-right with an outline.

---

[hn-88]

Changes -

Use the new syntax when generating frames.txt:
file_packet_meta timestamp 2005-06-01 00:15
instead of
file_packet_metadata timestamp=2005-06-01 00:15

And minterpolate with fps=40 was making it slow, so just blend, and use 4 fps, and we need to not set -r 4 or else it will duplicate frames since concat uses default 25 fps

ffmpeg -f concat -safe 0 -i frames.txt \
-vf "tblend=all_mode=average,\
drawtext=fontfile=/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf:\
text='%{metadata\\:timestamp}':\
fontcolor=yellow:fontsize=48:borderw=3:bordercolor=black:\
x=w-tw-40:y=h-th-40" \
 -c:v h264_nvenc -preset p5 -pix_fmt yuv420p ../ir_timestamped.mp4

[hn-88]

With yaw=180

ffmpeg -f concat -safe 0 -i frames.txt \
-vf "v360=input=equirect:output=equirect:yaw=180,\
tblend=all_mode=average,\
drawtext=fontfile=/usr/share/fonts/truetype/dejavu/DejaVuSans-Bold.ttf:\
text='%{metadata\\:timestamp}':\
fontcolor=yellow:fontsize=48:borderw=3:bordercolor=black:\
x=w-tw-40:y=h-th-40,\
scale=in_range=pc:out_range=tv,format=yuv420p" \
 -c:v h264_nvenc -preset p5 ../ir_timestamped_rotated.mp4

[hn-88]

But unfortunately, the Katrina, Rita etc frames are missing from the 4096 directory! So, relying on the 2048 mp4.