Using Machine Learning Techniques for Rainfall Estimation Based on Microwave Links of Mobile Telecommunication Networks

Evrad Venceslas Kamtchoum,Armand Cyrille Nzeukou Takougang,Clémentin Tayou Djamegni

DOI: https://doi.org/10.1007/s42979-022-01458-6

2022-11-11

SN Computer Science

Abstract:Accurate rainfall measurements are essential for many applications including watershed management and hydrological modeling. This is particularly true in sub-Saharan Africa where environmental and hydro-meteorological hazards are recurrent. Microwave links have been proposed as an alternative or an additional source of measurement to the traditional rainfall measurement instruments. However, the estimation of rainfall based on microwave links relies on some assumptions and approximations which induce considerable differences between resulting rainfall estimations and ground observations. One of the major sources of errors in the process of estimating rainfall from microwave links is the determination of the adequate drop size distribution. Deep learning models have been recently proposed with satisfaction to improve rainfall estimates from microwave links of mobile telecommunication networks. But these models were built based on experimental setups and require large datasets. Moreover, base transceiver stations of mobile telecommunication networks operate with different signals resolutions, and a poor signal resolution increases the difficulty for data-driven approaches. In this paper, we propose to use machine learning techniques for rainfall estimation from microwave links based on existing infrastructures, in the context of information and technical restrictions from the mobile telecommunication network. Observations collected from four microwave links and two weather radars in the Netherlands (Western Europe) are used to build regression models for rainfall estimation using path integrated attenuations from these links, and received signals levels as explanatory variables. The results are satisfactory as the values of the coefficient of determination and the correlation coefficient between predictions and observations are up to 0.92 and 0.96, respectively. This proves that simple machine learning regressors can be successfully used to estimate rainfall using microwave links of mobile telecommunication networks, and they clearly outperform the empirical computational methods in the literature.