Integrated Water Vapor Estimation During Clear Skies Using a Ground-Based Infrared Radiometer and the Light Gradient Boosting Machine Method

Abstract

New algorithms of retrieving atmospheric integrated water vapor (IWV) under clear-sky conditions for the infrared radiometer using linear regression, quadratic regression (QR), and light gradient boosting machine (LightGBM) methods are developed in this work. IWV data estimated using a physical method from ground-based microwave radiometer measurements of 23 days of clear sky over the Swiss Plateau from 2022 to 2023 serve as truth references. In addition to infrared brightness temperature, the input features also include various surface meteorological measurements. To capture the temporal dynamics of water vapor, the algorithms are trained with features and parameters adjusted not only through tenfold cross-validation but also by considering the time series. The validation shows that the linear and QR algorithms performed similarly with R$^{2}$ of 0.64, mean squared errors of 7.99 mm and 7.85 mm, and mean absolute error (MAE) of 2.24 mm and 2.25 mm, respectively. The LightGBM-based algorithm outperforms the two regression algorithms in retrieving IWV, with R$^{2}$ of 0.83, mean square error of 3.81 mm, and MAE of 1.53 mm. The IWV time series obtained from the three algorithms closely align with the measurements from the microwave radiometer. These proposed algorithms offer accurate and reliable IWV estimation for the infrared radiometer with high temporal resolution (7 s) in complex terrain, with potential for application in broader infrared radiometer networks.