Water Vapor Retrieval Using Commercial Microwave Links Based on the LSTM Network

Abstract

In this paper, a water vapor density inversion model based on the long short-term memory (LSTM) network is proposed for E-band commercial microwave links (CMLs). A full-duplex E-band microwave link located in Prague (two sublinks with frequencies of 73.5 GHz and 83.5 GHz, both vertically polarized) was used to verify the performance of the model. The results show that the model inversion results are in good agreement with the water vapor density calculated by temperature and humidity sensors. Compared with previous water vapor inversion methods based on CMLs, this model has a higher temporal resolution and can realize real-time monitoring.

A Review on Rainfall Measurement Based on Commercial Microwave Links in Wireless Cellular Networks

“…As compared to the data of meteorological station, the annual correlation value of the water vapor retrieved from the link is as high as 0.95, RMSE is as low as 0.35, and the average relative error is as low as 0.05. Pu et al [97] proposed a water vapor retrieval model by using dual-frequency E-band CMLs based on LSTM network, and the results show that the retrieved water vapor density is in good agreement with the results measured by temperature and humidity sensors. Fencl et al [98] also used a longer E-band link to realize the water vapor retrieval because E-band is more sensitive to raindrops and atmospheric gases, and the signal attenuation is sufficiently strong to enable the detection of water vapor at long CMLs.…”

Water vapor estimation based on 1-year data of E-band millimeter wave link in North China

“…Compared with other studies, our water vapor inversion results have a higher time resolution; that is, our resolution is 1 min, while it is 1 d for the tested ECMWF product. Moreover, the time resolution in the previous studies (David et al, 2009;Alpert and Rubin, 2018;Fencl et al, 2020;Pu et al, 2021) was equal to or greater than 5 min. Future research can use high-resolution humidity fields to improve weather forecasts, and its significance also includes the ability to study extreme events that are mainly controlled by humidity fields.…”

“…This is because the graphical user interface (GUI) of the wireless communication device cannot display the received signal level with higher accuracy, resulting in the link's estimated water vapor density value with a lower quantification resolution than that calculated by the weather station. Moreover, the change in water vapor is slower than the rainfall intensity (Pu et al, 2021), and the change in water vapor attenuation is also slower than the change in rain-induced attenuation. Therefore, we perform a 60 min moving average on the link RSL.…”

Retrieving Water Vapor From an E‐Band Microwave Link With an Empirical Model Not Requiring In Situ Calibration

“…Fencl et al (2020) also highlighted the temperature dependence of the relation between attenuation and WVD at these frequencies. Pu et al (2021) showed that the long short-term memory neural network trained on the same data set as used by Fencl et al (2020) is capable of estimating water vapor. Data-driven methods are easy to apply on experimental datasets with good independent reference observations of target variables (water vapor), however, their transferability to other devices without retraining on the reference data set is questionable.…”

“…Data-driven methods are easy to apply on experimental datasets with good independent reference observations of target variables (water vapor), however, their transferability to other devices without retraining on the reference data set is questionable. Moreover, Pu et al (2021) did not directly investigate the information gain of CML attenuations. Their neural network model might reproduce diurnal pattern of WVD which was similar in both training and validation data sets.…”