Seminar on Physics and Chemistry of the Atmosphere (Abstract)
The Neural Network Ozone Retrieval System (NNORSY) for GOME Ozone Profiles
Martin Müller
Zentrum für Sonnenenergie- und Wasserstoff-Forschung (ZWS), Stuttgart
martin.mueller@zsw-bw.de
17.5.2002, 13.00 c.t.
Raum N3380
In recent years, a new methodolgy for retrieving ozone profiles from GOME measurements has been developed at the Center for Solar Energy and Hydrogen Research (ZSW), Stuttgart, in cooperation with the IUP. Based on a feed-forward neural network trained on ozonesonde and limb sounder data, this approach has completely different characteristics and thus complements current Optimal Estimation based ozone retrieval schemes, especially concerning processing speed. For instance, in the frame of this talk, calculating ozone profiles from all GOME data between July 1998 and July 2001 at full horizontal resolution took only two days on three UltraSparc 400 MHz CPUs.
The talk will give a short introduction into the theory behing NNORSY, thereby placing it in the context of Optimal Estimation. Current retrieval results will then be presented and validated using independent data. This will include some comparison with IUP-FURM retrievals.
Neural networks retrievals do not automatically yield corresponding error information, therefore supplementary methods for estimating retrieval errors must be considered. Some of these can also be used for sensitivity studies, as will be shown. Finally, the operational near real-time NNORSY running at DLR will be introduced shortly, concluding with some perspectives concerning further development of the method.