You must have a javascript-enabled browser and javacript and stylesheets must be enabled to use some of the functions on this site.


Tropospheric ozone from GOME-2 NNORSY retrievals

Martin Felder(1) and Anton Kaifel(1)

(1) ZSW, Industriestra├če 6, 70565 Stuttgart, Germany


The Neural Network Ozone Retrieval System (NNORSY) has previously been applied more than 10 years of GOME-1/ERS2 data, producing highly accurate ozone profile retrievals, including tropospheric ozone. A new and improved version of the system has been designed for GOME-2 ozone retrieval. For training, ground based (e.g. ozone sondes and lidar) and satellite based (e.g MLS, ACE-FTS) ozone profile measurements are collocated with GOME-2 pixels, defining the retrieval target. After training the neural networks, ozone profile retrieval is very fast and applicable to real-time processing. In this manner, we have processed 18 months of GOME-2 data at full spatial and temporal resolution yielding NNORSY-GOME-2 ozone profiles on a global scale. The ozone profile information is given from ground up to 61 km height with a sampling interval of 1 km and including ozone profile error estimation for each profile level.

As a baseline for future enhancements of the system featuring synergistic UV/VIS and IR retrievals from the GOME-2 and IASI instruments on MetOp-1, we have implemented a number of tropospheric ozone retrieval techniques, including direct integration and tropospheric ozone residuals (TOR). In contrast to OE-based retrievals, the neural network compensates for lack of physical information automatically using the statistical properties of the ozone field it has learned, obviating the need for a separate climatology. This makes it an excellent candidate for the challenging task of satellite-based tropospheric ozone retrieval. We compare our GOME-2 results on different time scales with TOR from OMI/MLS produced at NASA-GSFC, ozone sondes, lidar and CTMs.