Results of Tovs Ozone Retrieval with Neural Networks

We present the results of a recently finished project concerned with total ozone retrieval on a global scale from HIRS and MSU brightness temperatures. The project encompasses processing TOVS data ranging from 1986 to now by means of a neural network technique utilizing information from all spectral channels. Comparisons with ozone measurements from WOUDC Dobson and Brewer ground stations yielded a day and night average RMS accuracy of about 15 D.U. for all weather conditions, including broken clouds. Further validation was performed by comparison with TOMS and GOME data at 51 ground station locations, showing the error of TOVS vs. ground to be similar to these sensors’ errors. The TOVS errors do however not increase over time, therefore temporal stability of the retrieval can be assumed. We conclude TOVS has qualified as a suitable instrument for accurate and fast ozone retrievals, especially since it is currently the only orbital sensor capable of measuring ozone in the polar night. Future work will be directed towards application of neural network technique to other sensors including ATOVS, which promises to yield even higher accuracy due to its increased number of channels. Encouraging results are currently received from a running project with ERS-GOME data, using neural networks for total ozone and ozone profile retrievals.