Composed by Naman // Claude L-band · VHF
GS. — A ground-station log

← Frontier Work

GK-2A · IR105 10.3 micron · South Korea

Storm clouds with no typical size

GK-2A is a Korean weather satellite. We receive one of its channels, a single thermal-infrared band, on our own dish and turn the raw radio into calibrated cloud-top temperatures. From that one band, the tropical storm clouds over the Pacific warm pool turn out to have no characteristic size. The same statistical pattern repeats from tens to hundreds of kilometres, and it holds steady for hours.

One infrared bandcloud-top temperature Find the stormsclusters, in Kelvin Two scale-free measuresspectrum and cluster sizes Steady for 15.7 hoursand across days

What we found

  • The temperature field's power spectrum is a straight line from 50 to 500 km. A straight line means one rule at every size, with slope beta near minus 2.4. There is no preferred scale.
  • The cold storm clusters follow a power-law size distribution, exponent tau near 1.8, fit properly by maximum likelihood. Many small, a few enormous, no typical size.
  • Both numbers stay flat frame to frame, and reproduce on a second window three days apart. The steadiness is the result, not any single snapshot.
Left, cloud-top temperature with cold tall storms in white. Right, the same scene with every detected cold cluster coloured by its area.
What the satellite sees, and what we measure. Left: cloud-top temperature, where white is cold, the tall storms. Right: every cold cluster found, coloured by its area, over the Maritime Continent and west Pacific warm pool.
Log-log power spectrum of cloud-top temperature, a straight line from 50 to 500 km with slope about minus 2.4.
The temperature power spectrum. It is straight over 50 to 500 km (shaded), slope beta about minus 2.4. That is steeper than the classic minus five-thirds, in line with other satellite cloud-field measurements, and it says the field is scale-free across the mesoscale.
Log-log distribution of cold-cloud cluster areas, with a straight power-law tail fit by maximum likelihood, exponent about 1.8.
The cluster-size distribution. The straight tail is a power law, exponent tau about 1.8, fit by maximum likelihood rather than a drawn line (Clauset, Shalizi and Newman). Storm clusters come in all sizes, with no typical one.
The spectral slope and the cluster exponent plotted over 15.7 hours, both nearly flat with narrow scatter bands.
The point of the whole thing: both exponents over the 15.7 hour overnight pass. Flat lines, narrow bands. The scale-free behaviour is persistent, not a one-frame accident, and the same numbers come back on the earlier window.

For context, the convection itself sits roughly still while it does this. It pulses in place over the warm pool rather than drifting along, and the very coldest tops stay pinned near 195 K, the ceiling the storms hit at the tropopause.

Longitude on the horizontal axis, time on the vertical, shaded by cold-cloud fraction. Bright columns stay at fixed longitudes.
Longitude across, time down, shaded by how much cold cloud there is. The bright columns stay at fixed longitudes, so the storms pulse in place rather than propagating east or west.
What we looked for first, and did not find

We started by hunting for convective gravity waves, the faint ripples that deep storms push through the surrounding cloud field. They are real, but they pulse every 6 to 16 minutes, and our full-disk images arrive only every 10 minutes, which aliases the ripples away. What looked like wave trains in the high-pass field turned out to be detector striping in the downlink, strongest over clear sky where there is no weather at all. So we set the waves aside and kept the two signals that survive scrutiny.

Cold-cloud area and the coldest pixel over the afternoon window.
The afternoon convective cycle: total cold-cloud area against time, with the single coldest top, which holds near 195 K throughout.

Where it stands

The temperatures come from our own decode of the GK-2A downlink, not a data feed. The scale-free result is robust across the two windows we have, 5.7 hours on one day and 15.7 hours on another. More days, and a wider-area fit, would tighten the exponents further. The receiving station is powered down for now, so the archive is frozen.

Imagery: GEO-KOMPSAT-2A AMI IR105 (10.3 micron) LRIT, received and decoded at this station. Calibration to brightness temperature, the spectral and cluster analysis, and the figures: this project. Power-law fitting follows Clauset, Shalizi and Newman; the cluster-size framing follows Wood and Field.