Demonstration of the Use of Bayesian Priors in GHG retrievals from Laser Heterodyne Radiometer Measurements

George Washington University and Mesa Photonics are developing and deploying a Laser Heterodyne Radiometer (LHR) that simultaneously measures CO₂, CH₄, H₂O, and O₂. Because oxygen concentrations are nearly invariant throughout the troposphere and lower stratosphere its line shape is dependent only on pressure and temperature, and analysis of its line shape can be used to improve GHG retrieval precision and provide dry-air corrections. To constrain these fits, pressure and temperature profiles for our LHR data retrieval algorithm can be obtained from the weather data measured by radiosondes as part of NOAA’s Integrated Global Radiosonde Archive (IGRA). In a recent paper, we reported on the statistical analysis of this data and highlighted how it can not only be used to constrain both the temperature and pressure profiles, but also the vertical profiles of water mixing ratios.  Not only do mean values of radiosonde temperature, pressure, and humidity provide useful priors in column retrievals, but the narrow distributions above near-surface altitudes create realistic constraints to retrieval results.

For other greenhouse gases (specifically CO₂ and CH₄), prior data to constrain these vertical profiles is much sparser and a different approach is required.  The Bayesian paradigm applies prior knowledge and observations to a model being tested. It is the foundation upon which inverse modeling in the atmospheric sciences is built and involves weighting the error the find the optimal value of a state vector given the observations.  In this presentation we demonstrate how continuous LHR data from a stationary sensor can be used to refine an initial prior based on available (and widely distributed spatially and temporally)  global GHG vertical profiles to constrain data from site-specific installations.  Further, we will demonstrate the robustness of this technique to follow temporal excursions such as surface emission events.  Initially, this algorithm is applied to synthetic data with a goal of application to the data stream from a sensor scheduled to go on line at the Smithsonian Environmental Research Center (Edgewater, Maryland, USA) in the 2^nd quarter of 2022.