Compare observations fairly

Ibuki 2, or GOSAT-2, launched on October 29, 2018. It continued carbon-dioxide and methane observations and added carbon monoxide as an observation target. Carbon monoxide can help investigate combustion-related sources, but its presence does not replace the full analysis needed to quantify carbon-dioxide emissions.

A successor instrument can improve a record while also introducing differences in sampling and processing. Simply joining two time series at the launch date may conceal those differences. A useful continuity study asks whether comparable atmospheric conditions produce consistent estimates across both instruments.

A 2023 NIES study compared GOSAT and GOSAT-2 with ground-based TCCON spectrometers, with each other during matched observations, and through regional monthly averages. The study controlled factors such as location, observation time and surface elevation. Its findings of broad consistency also retained regional differences and limits in the validation network’s coverage.

In an original matching exercise, one satellite observes a location at noon while a ground instrument measures a different air mass the following morning. Their difference can include real atmospheric variation. Matching time and location more closely helps isolate measurement differences; it does not automatically make every remaining difference an instrument error.

Match evidence before comparing measurements

Suppose an original test compares three paired values. The reference reads 400, 402 and 404 units, while a second instrument reads 402, 404 and 406. The readings track the same changes but retain a positive two-unit offset. A close relationship between the series is not proof that the offset is zero.

Now imagine that every comparison site lies on bright, dry land. Good agreement there does not by itself establish performance over dark ocean surfaces or aerosol-heavy regions. Validation should examine the relevant range of conditions. The NIES study explicitly cautions that limited ground-site coverage cannot fully characterize every regional bias.