Level-2 Algorithms - Ozone (O3) Total Vertical Column

Sentinel-5P Technical Guide - Level-2 Processing Overview - O3 Total Vertical Column

There are two algorithms that will deliver total ozone column amounts to be retrieved from TROPOMI. The first algorithm "S5P_TO3_DOAS" is based on the "DOAS-style" GOME Data Processor (GDP) algorithm Version 4.x. The second "S5P_TO3_GODFIT" is based on GDP Version 5, which is a "direct-fitting" algorithm deployed for the GOME/ERS-2 16-year reprocessing in 2012. The "DOAS-style", because of its lower computational demand, is to be used for generation of near real-time products, while the "direct-fitting" algorithm will generate offline and reprocessed products.

 

NRT/DOAS Algorithm (S5P_TO3_DOAS)

The general method used for derivation of O3 total column density is the two-step DOAS method. However, for O3 total column retrieval, the calculation of the air mass factor uses an iterative AMF approach [Spurr et al., 2005] which uses the fitted slant column result Ns to adjust the AMF (and hence the VCD) that reflects the trace gas content as expressed in the value of Ns. This iterative approach circumvents the issue that can be encountered in a traditional DOAS algorithm (where Ns fitting and AMF calculation are decoupled) when the selected climatological input profile can sometimes be very different from the true profile, leading to an incorrect AMF computation.


OFFLINE/Direct Algorithm (S5P_TO3_GODFIT)

Unlike the NRT/DOAS algorithm, the OFFLINE algorithm is a one-step algorithm. There is no separation of slant-column fitting and AMF conversion. It is based on the GODFIT (GOme Direct FITting) algorithm [lerot et al., 2010 and 2013; van Roozendael et al., 2012]. The algorithm uses an iterative least-squares cost function minimisation method, based on the differences between satellite-measured and model-simulated reflectances. For the forward model component, GODFIT uses the LIDORT radiative transfer model [Spurr, 2008] for multiple-scatter RT simulation of satellite radiances and analytically calculated Jacobians with respect to the total O3 column, the temperature shift and a number of surface albedo closure coefficients. In addition to these retrieval parameters, the retrieval state vector also has parameters for ring-effect and under-sampling corrections, and for a wavelength registration shift. The inversion module uses a special Levenberg-Marquardt algorithm with line-search.


Fitting Window

  • The fitting window used for both NRT and OFFLINE algorithms is in the UV Huggins bands: 325-335 nm (spectral band 3).