EE Seminar: Opportunistic Rainfall Measurements Using Satellite Microwave Link Attenuation
Electrical Engineering Systems Seminar
Speaker: Gil Rafalovich
M.Sc. student under the supervision of Prof. Hagit Messer-Yaron and Dr. Jonatan Ostrometzky
Monday, 18th May 2026, at 15:00
Room 011, Kitot Building, Faculty of Engineering
Opportunistic Rainfall Measurements Using Satellite Microwave Link Attenuation
Abstract
The escalating frequency of global climate changes necessitates the development of creative, wide-scale, and highly accurate methods for precipitation monitoring. Traditional meteorological instruments, such as rain gauges and weather radars, provide discrete, localized measurements but often lack the comprehensive spatial coverage required for real-time climatic assessment. To address this monitoring gap, in 2006, Prof. Hagit Messer and her colleagues introduced a groundbreaking opportunistic sensing method: monitoring rainfall by analyzing the attenuation of electromagnetic signals between cellular network towers, known as Commercial Microwave Links (CMLs). This technique leverages the non-linear relationship between signal attenuation and rainfall intensity at frequencies above 10 GHz, as defined by the ITU-R P.838-3 power-law model.
Building upon the widespread success of CML-based monitoring and the physical similarities in microwave propagation, Filippo Giannetti et al. later proposed extending this paradigm to Satellite Microwave Links (SMLs). Giannetti's innovation demonstrated that the attenuation of downlink signals---specifically from DVB-S2 geostationary satellites to ground-based interactive IoT terminals---could similarly be repurposed to calculate quantitative precipitation estimates.
This research focuses on the collection, characterization, and analysis of a two-year SML dataset utilizing Ayecka IoT receivers operating in the Ku-band (around 14 GHz) in the Kefar Sava region of Israel. Initially, the study implements a traditional, analytical approach using a Double Kalman Filter (DKF) model to estimate rainfall by tracking the clear-sky baseline and identifying rain-induced signal fades. Crucially, this work pioneers the adaptation of Wet Antenna Attenuation (WAA) elimination---a signal-correction technique originally developed for CML analysis---to SML telemetry, significantly mitigating the performance degradation caused by water accumulating on the receiver hardware.
However, achieving optimal accuracy with the analytical DKF model requires extensive a priori knowledge, including the physical parameters of the link, specific receiver characteristics, and localized meteorological data such as the cloud melting layer height. To overcome these rigid constraints and develop a system agnostic to environmental and hardware preconditions, this thesis introduces a novel, data-driven deep learning architecture.
The proposed model employs a hybrid 1D-Convolutional Neural Network (1D-CNN) combined with a Long Short-Term Memory (LSTM) network. Within this architecture, the 1D-CNN extracts localized spatial features to distinguish true rain fades from anomalous signal fluctuations, while the LSTM models the long-term temporal dependencies to maintain a dynamic clear-sky reference baseline. Optimized using a Zero-Inflated Log-Normal (ZILN) loss function, this approach autonomously learns the complex statistical properties and non-linear power-law dynamics of precipitation events.
Evaluated against ground-truth data, the deep learning model achieves a robust 99.33% classification accuracy in detecting rain events and predicts precipitation intensity with a Root Mean Square Error (RMSE) of 3.9 mm/h. These results conclusively demonstrate that the artificial intelligence system learns and maps the non-linear relationship between SML attenuation and actual rainfall significantly better than traditional analytical models.
-סמינר זה ייחשב כסמינר שמיעה לתלמידי תואר שני ושלישי-
This Seminar Is Considered A Hearing Seminar For Msc/Phd Students-
הרישום לסמינר יבוצע בתחילת הסמינר באמצעות סריקת הברקוד למודל (יש להיכנס לפני כן למודל, לא באמצעות האפליקציה)
Registration to the seminar is done at the beginning of the seminar by scanning the barcode for the Moodle (Please enter ahead to the Moodle, NOT by application)
.
