As rainfall exhibits high spatiotemporal variability, accurate and real-time rainfall monitoring is vitally important in fields such as hydrometeorological research, agriculture and disaster prevention and control. Nevertheless, the current dedicated rain sensors cannot fulfill the requirement for comprehensive precipitation observation, owing to their respective limitations. Within the last two decades, the utilization of commercial microwave links (CMLs) for rainfall estimation, as an opportunistic sensing method, has generated considerable attention. Relying on CML networks deployed and maintained by mobile network operators can provide near-surface precipitation information over large areas at a low cost. Although scholars have developed several algorithms for obtaining rainfall estimates from CML data, the rainfall estimation technique based on CMLs remains challenging due to the complex effect in the microwave radiation transmission process. In this paper, we provide a comprehensive review of the technical principles, developments and workflows for this technology, alongside its application in environmental monitoring and hydrological modeling. Furthermore, this paper outlines the current challenges and future research directions, which will hopefully draw the attention of researchers and provide valuable guidance.

This article proposes a new rainfall-type classification method using the differential attenuation rates (DARs) calculated from microwave links with multiple frequencies and dual polarization. Several machine-learning algorithms were employed for classification [decision tree (DT), probabilistic neural network (PNN), Gaussian discriminant analysis (GDA), and logistic regression (LR)]. This method was simulated by two or three frequencies (combination of frequencies of 15, 25, 38, 60, and 80 GHz). The results showed that DT, PNN, GDA, and LR achieved maximum accuracies of 88.9%, 84.6%, 66.8%, and 67.6% for the dual-frequency model, respectively, and 89.6%, 85.8%, 67.6%, and 68.7% for the tri-frequency model, respectively, which is obviously better than the classification method based on the sequence of rain rates from the ITU-R model. The model accuracies based on GDA and LR algorithms are obviously lower than those based on DT and PNN due to the defects of the algorithms. In addition, the effects of different noise levels on classification performance are also discussed.

Commercial 1.164 km long microwave link (CML) data with frequencies of 72.75 and 82.75 GHz (E-band) deployed in the Prague urban area were used to estimate wet antenna attenuation (WAA). By analyzing the time series of WAA, a WAA model of E-band CML is proposed based on the long short-term memory (LSTM) neural network. The results show that the predicted value is consistent with the real value, especially when the attenuation fluctuation is not significant, but it understates cumulative rainfall when rain sharply increases. In addition, the performance of the model applied to the 72.75 GHz link is better than that applied to the 82.75 GHz link.

In this article, a water vapor density inversion model based on the long short-term memory 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 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 method for identifying hydrometeor types (rain, graupel, and wet snow) based on a microwave link is proposed in this paper. The measured hydrometeor size distribution (HSD) data from the winters of 2014 to 2019 in Nanjing, China, were used to carry out simulation experiments to verify the performance of the model. Single-, dual-, and tri-frequency models (combinations of 15 GHz, 18 GHz, 25 GHz, 38 GHz, 50 GHz, 60 GHz, 70 GHz, and 80 GHz) were established with the extreme learning machine (ELM) algorithm. The results showed that the performance of the tri-frequency models was overall better than that of the dual-frequency models, for which the performance was better than that of the single-frequency models. The mean (maximum) test set accuracies of the single-frequency, dual-frequency, and tri-frequency models reached 75.8%, 80.7%, and 83.2% (83.0%, 84.4%, and 85.6%), respectively. For the dual-frequency and tri-frequency models, it was found that the accuracy increased with the overall frequency or the frequency difference. In addition, the influences of different noise levels on the model performance were also analyzed. Finally, the effects of position and length of link relative to precipitation cell were analyzed and are also discussed.

Rainfall inversion based on commercial microwave links (CMLs) has been extensively studied and experimented. However, as necessary part of inversion, the error caused by the A - R relationship is lack of systematic evaluation. Based on the measured raindrop size distribution (RSD) data recorded by a disdrometer, the theoretical rain-induced attenuation and rainrate are calculated, and on this basis, the error between the inversed rainrate based on A - R relationship from ITU-R recommendation and the real rainrate is compared under different rainfall types, rainfall classes and spatially heterogeneous rain cell. Firstly, the errors of A - R relationship under different rainfall classes show that A - R relationship can effectively invert rainrate between 18 and 26 GHz for all classes, while the error increases significantly outside this frequency range, especially for heavy rainfall. In addition, the A - R relationship performance of different rainfall types is evaluated. There is a significant difference in the unbiasedness of convective rainfall and stratiform rainfall, which are positive and negative bias at high frequency (greater than 50 GHz), respectively. Furthermore, aiming at the inhomogeneity of rainfall spatial distribution, the rainrate distribution and the corresponding path rain-induced attenuation are simulated based on an exponential model, and the influence of spatial location difference on the accuracy of A - R relationship is deeply analyzed. Finally, the error source and influence of localized A - R relationship, integration time and link length are discussed.