Retrievals of carbon monoxide profiles from MOPITT observations using lognormal a priori statistics

Abstract

Optimal estimation methods, such as the “maximum a posteriori” solution, are commonly employed for retrieving profiles of atmospheric trace gases from satellite observations. To complement the information actually contained in the measured radiances, such methods exploit a priori information describing the gases’ variability characteristics. We show that in situ surface‐based data sets for carbon monoxide (CO) volume mixing ratio (VMR) indicate that the variability of CO is more accurately modeled in terms of a “lognormal” probability distribution function (PDF) than a “VMR‐normal” PDF. The VMR‐normal PDF is particularly poor at describing CO variability in unpolluted conditions. We also compare retrievals of carbon monoxide (CO) vertical profiles based on Measurements of Pollution in the Troposphere (MOPITT) observations for 1 day using both VMR‐normal and lognormal statistical models. Use of the lognormal model improves retrieval convergence and yields fewer profiles with unphysically small VMR values. Generally, these results highlight the importance of properly representing the variability of trace gas concentrations in optimal estimation‐based retrieval algorithms.