Compact Microwave Radiometer for Water Vapor Estimation with Machine Learning Method

Abstract

We have developed a compact ground-based microwave radiometer (MWR) for estimating water vapor. The MWR observes radio wave intensity at frequencies between 17.9 and 26.4 GHz across 34 channels and estimates precipitable water vapor (PWV) and the profile of water vapor density using machine learning methods. Data from the Global Navigation Satellite System (GNSS) and radiosonde (SONDE) collected at the Meteorological Research Institute of the Japan Meteorological Agency were used to train and evaluate the machine learning models. Data from June 2021 to March 2022 were used for training, and data from April 2022 to March 2023 were used for evaluation. As a result, the maximum root-mean-square errors (RMSEs) of MWR-derived PWV compared to GNSS-derived PWV and MWR-derived water vapor density compared to SONDE at the lowest layer of the atmosphere were 2.7 mm and 2.4 g m⁻³, respectively. Analysis of the error characteristics of water vapor estimation showed that both PWV and water vapor density profiles had errors in the presence of cloud water, as determined by infrared radiometer, and high accuracy in the absence of cloud water. The estimation accuracy was also affected by fog and water vapor inversion layer.