| dc.description.abstract | The need for improved precipitation estimations has prompted the exploration of opportunistic alternatives such as utilizing commercial microwave links (CML), particularly in areas with poor coverage of weather radars and rain gauges. It has been known that rainfall-induced attenuation in the microwave signal can be used to determine rainfall intensity accurately. However, detecting other types of precipitation, such as dry snow, remains a challenge. This study evaluates the feasibility of using wavelet transform combined with a random forest classifier to identify rain and snow events.
Real-world signal attenuation data from telecommunication operators and precipitation data from nearby disdrometers in Norway were used to develop the classification methods proposed in this study. The rain classifier was based on data from June 2022, while the snow classifier was evaluated using data from December 2021. The operating frequency of the CMLs used in this study was between 30-40 GHz. The algorithm for rain detection performed similarly to other wet-dry classification methods, with a mean Matthews correlation coefficient (MCC) of 36 % among 52 CMLs. The snow detection algorithm, however, showed no correlation between signal attenuation from 41 CMLs and dry snowfall.
In conclusion, the wavelet transforms effectively extract useful information from signal attenuation for rain classification but are unsuitable for detecting snow. Moreover, the study recommends testing commercial microwave links with higher operating frequencies than those used in this study, combined with temperature data, to improve the possibilities of dry snow detection. | |
