Read the information between the levels

MOPITT aboard Terra provides both profiles and total columns. A vertical profile describes how a quantity varies with height or pressure. A total column combines the gas through an atmospheric column above a unit area. Two different vertical distributions can have the same total, so a column value alone does not uniquely locate the gas. In Version 9, the reported CO total column is calculated from the retrieved profile; it is not an independent second observation.

The Version 9 profile is reported on ten pressure levels from the surface toward the upper atmosphere. Those are the coordinates used to represent a solution. Ten output numbers do not establish ten independently resolved atmospheric layers. The radiation may constrain broad combinations of levels more strongly than it constrains each level separately.

This distinction resembles describing a blurred photograph with many pixels: increasing the number of positions at which values are reported does not recreate detail lost in the measurement. For a useful comparison, first identify the actual sensitivity, the layer definitions and the units. A surface-labelled retrieval layer also spans a finite depth; it is not a person-height monitor.

The retrieval combines measurement information with a prior atmospheric profile. The prior provides a starting statistical description where the measurements cannot fully determine the answer. An averaging kernel describes how changes in the actual atmosphere would affect the retrieved result at different levels. Where measurement sensitivity is weak, the prior has a larger influence.

Consider an original one-variable analogy: a prior value is 2.0, a comparison state is 3.0, and the response factor is 0.25. The simulated response is 2.0 + 0.25 × (3.0 − 2.0) = 2.25. It has moved only partway from the prior toward the comparison state. This dimensionless example illustrates sensitivity, not a real MOPITT CO measurement or a universal weighting rule.

The actual Version 9 calculation uses a matrix that couples atmospheric levels, and its profile state is expressed as log₁₀ of the volume mixing ratio. You cannot substitute raw concentrations into a formula defined for logarithmic values. The product provides its own prior and averaging kernel, which must accompany the result rather than being discarded as optional metadata.

A partial response retains prior influence

The 2022 validation study compared satellite retrievals with CO profiles sampled by aircraft. Before assessing agreement, researchers transformed the comparison profiles using the satellite’s averaging kernels and prior. This asks what the satellite would be expected to retrieve from that comparison atmosphere, instead of blaming it for differences in vertical detail it cannot resolve.

Location and time still matter after that transformation. An aircraft path samples a different volume and interval from a satellite footprint. The study documents collocation limits and extends aircraft profiles where the aircraft did not sample the upper atmosphere. Such choices make the comparison reproducible and identify where assumptions enter.

Validation statistics describe the tested samples and methods. A small average bias does not guarantee that every retrieval has a small error. Positive and negative differences can cancel in an average, while individual values remain scattered. Likewise, the number of reported levels is not a substitute for the uncertainty and sensitivity information needed to interpret them.