Launch Watch · Analysis
Published
Same ABI band, different response: GOES-16 and GOES-18 compared
Compare GOES-16 and GOES-18 response functions in Bands 7 and 12, with a 32-table integral audit and explicit definitions of band center.
GOES-16 and GOES-18 both have an ABI Band 7, but their released spectral-response curves are not identical. The GOES-18 curve gives more relative weight to part of the band’s long-wavelength shoulder. Integrating each curve over wavelength gives response centroids of 3.893361 and 3.905240 micrometers, respectively: a difference of about 0.011880 micrometer.
That result comes from NOAA’s released band-representative response functions. It is not a measured temperature difference between two satellite images. We parsed all 32 response tables for the two instruments, then used Bands 7 and 12 as worked comparisons of what a shared band number does, and does not, tell us.
Identify the function before interpreting the curve
NOAA’s ABI calibration page maps the PFM instrument to GOES-16 and FM3 to GOES-18. We used its processed PFM version 3 and FM3 text-table releases. Each module has 16 channel files, and each file provides wavelength, wavenumber and relative spectral response.
The 10 March 2016 release note identifies the vendor-supplied ANGEN functions and the GOES-R Calibration Working Group’s processing. A NOAA-hosted technical paper distinguishes these band-representative analytic functions from detector-specific end-to-end measurements. The curves here are the released functions, not raw detector measurements or a record of in-flight aging.
Method: define the center as an integral
For this comparison, the wavelength-domain centroid is the integral of wavelength times relative response, divided by the integral of relative response, with both integrals taken over wavelength. We approximate each integral with the trapezoid rule between adjacent samples in the released table. This definition uses the actual wavelength spacing as well as the response values.
Band 7 contains 2,538 rows in each module’s file. Band 12 contains 842 rows for GOES-16 and 836 for GOES-18. The calculation uses every retained sample in each file. Six decimal places make the numerical comparison reproducible; they do not imply that a band’s physical behavior is known to that many places.
| Instrument and band | Released samples | Centroid (µm) | Row-weighted mean (µm) |
|---|---|---|---|
| GOES-16 Band 7 | 2538 | 3.893361 | 3.891443 |
| GOES-16 Band 12 | 842 | 9.611453 | 9.608587 |
| GOES-18 Band 7 | 2538 | 3.905240 | 3.903236 |
| GOES-18 Band 12 | 836 | 9.594140 | 9.591305 |
GOES-18’s Band 7 centroid is about 0.011880 micrometer longer than GOES-16’s under this definition. In Band 12, the difference goes the other way: GOES-18’s centroid is about 0.017314 micrometer shorter. A shared band number is therefore not a guarantee of an identical weighting function, and there is no single cross-instrument shift that this pair of examples supports.
Why averaging rows gives another answer
A tempting shortcut is to multiply each wavelength by its response, add the products, and divide by the sum of the responses. That treats the rows as equal-weight samples apart from the response. The result for GOES-16 Band 7 is 3.891443 micrometers, rather than the wavelength-integrated value of 3.893361.
The shortcut omits the spacing between wavelengths. These tables use a wavenumber grid; converting that grid to wavelength produces unequal wavelength intervals. Equal row weights consequently do not approximate the same integral as widths measured along the wavelength axis. The table exposes the discrepancy instead of hiding it inside a generic “central wavelength” label.
A wavenumber-domain centroid is yet another declared calculation. For GOES-16 Band 7, integrating wavenumber times response over wavenumber gives 2,570.373332 inverse centimeters. Taking its reciprocal and converting units gives 3.890485 micrometers. That is not equal to the wavelength-domain centroid, because an average and a reciprocal do not generally commute, and the two integrals use different measures.
Check the complete 32-table integral audit
Download the integral results for all 32 channel tables. Each row identifies satellite, flight module, band, sample count, wavelength-domain centroid, wavenumber-domain centroid, reciprocal conversion and row-weighted wavelength mean. The audit also checks that wavelength increases and that retained response values are nonnegative.
For each adjacent pair of samples, the trapezoid area equals the interval width multiplied by the mean of the two endpoint values. Sum those areas separately for the response and for wavelength times response, then divide. Use the released tables without inventing response values beyond their boundaries.
What to test before carrying a threshold across satellites
CWG’s release processing truncates the functions at the innermost 0.1% relative-response boundaries to remove negative-response portions. Our integrals describe those supplied limits. They do not recover omitted tails, detector-to-detector variation or later changes in orbit.
The curves show that the band weightings differ. They do not, by themselves, quantify a temperature bias, fire-detection accuracy or launch-plume signal. Evaluating a numerical threshold across spacecraft requires the relevant scene spectrum, calibrated product definitions and a test on suitable data. This audit supplies the response-function comparison that such a test must keep explicit; it does not substitute for the test.