All-weather retrieval of total column water vapor from aura OMI visible observations

Abstract

Total column water vapor (TCWV), retrieved from satellite remotely sensed measurements, plays a critically important role in monitoring Earth's weather and climate. The ozone monitoring instrument (OMI) can obtain daily near-global TCWV observations using the visible spectra. The observational accuracy of OMI-estimated TCWV under cloudy-sky conditions is much poorer than OMI-measured clear-sky TCWV. Satellite-based OMI-derived TCWV data, observed with little cloud contamination, are solely used, which, in general, are limited and discontinuous observations. We propose a practical machine learning-based TCWV retrieval algorithm to derive TCWV over land from OMI visible observations under all weather conditions, considering multiple dependable factors linked with OMI TCWV and air mass factor. The global TCWV data, observed from 6000 global navigation satellite system (GNSS)-based training stations in 2017, are utilized as the expected TCWV estimates in the algorithm training process. The retrieval approach is validated in 2018-2020 across the world using ground-based TCWV from additional 4,465 GNSS-based verification stations and 783 radiosonde-based verification stations. The newly retrieved TCWV estimates remarkably outperform operational OMI-retrieved water vapor data, regardless of cloud fraction and TCWV levels. In terms of root-mean-square error, it is overall reduced by 90.44% from 56.38 to 5.39 mm and 90.19% from 53.23 to 5.22 mm compared with GNSS and radiosonde TCWV, respectively. The retrieval algorithm stays stable, both temporally and spatially. This research provides a valuable technique to precisely retrieve OMI-based TCWV data records under all weather conditions, which could be applicable to other satellite-borne visible sensors like GOME-2, SCIAMACHY, and TROPOMI.