Neural Networks for Arctic Atmosphere Sounding From Radio Occultation Data

• Published in: IEEE transactions on geoscience and remote sensing Vol. 49; no. 12; pp. 4846 - 4855
• Main Authors: Pelliccia, F., Pacifici, F., Bonafoni, S., Basili, P., Pierdicca, N., Ciotti, P., Emery, W. J.
• Format: Journal Article
• Language: English
• Published: New York, NY IEEE 01.12.2011; Institute of Electrical and Electronics Engineers
• Subjects: Applied geophysics; Arctic; Artificial neural networks; Atmospheric modeling; Earth sciences; Earth, ocean, space; Exact sciences and technology; Global models; Global Positioning System; Global Positioning System (GPS); Internal geophysics; neural network (NN); Neural networks; radio occultation (RO); Refractive index; troposphere; polar regions; Arctic region; atmosphere; satellites; models; meteorology; Global Positioning System (GPS); neural networks; Global models; global; climate; pressure; ionosphere; sounding; troposphere; Global Positioning System; temperature; neural network (NN); humidity; Earth orbit; performances; radio occultation (RO)
• DOI: 10.1109/TGRS.2011.2153859
• ISSN: 0196-2892

This paper illustrates a procedure for the retrieval of tropospheric profiles (temperature, pressure, and humidity) using only refractivity profiles coming from Global Positioning System (GPS)-low-Earth-orbit radio occultation, without the constraint of independent knowledge of atmospheric parameters at each GPS occultation. In order to achieve this goal, we have used an approach based on neural networks (NNs), exploiting a data set of 1106 occultations collected over the Arctic region during the winter season of 2007 and 2008. Total refractivity (N) profiles from Formosa Satellite 3 (FORMOSAT-3)/Constellation Observing System for Meteorology Ionosphere and Climate (COSMIC) satellites have been used as input for training the NNs, whereas the target profiles of dry and wet components (N d and N w ) were derived using prior information on dry and wet fractions of the total refractivity provided by the analysis of the European Centre for Medium-Range Weather Forecast (ECMWF). Once we have retrieved N d and N w by the trained networks, the other atmospheric parameters (pressure, temperature, and vapor) can be computed, and we have done so relative to colocated ECMWF data, which we have assumed as atmospheric truth. Finally, some comparisons with radiosonde observations (RAOBs) are shown, and performances and potential of the proposed approach are discussed. Profiles computed using 1-D variational retrieval by the COSMIC Data Analysis and Archive Center have also been considered as a benchmark in the RAOB comparison.