EE Seminar: Robust Estimation and Mapping of Rainfall from Microwave Measurements
Speaker: Shani Gat
M.Sc. student under the supervision of Prof. Hagit Messer-Yaron
Sunday, April 23rd, 2017 at 15:30
Room 011, Kitot Bldg., Faculty of Engineering
Robust Estimation and Mapping of Rainfall from Microwave Measurements
Estimation of a tempo-spatial field for purposes of environmental monitoring is often based on a network of local sensors, which are prone to measurement errors. As sensor networks become larger and cheaper, the probability of faulty sensors and measurements outliers rises. Moreover, the recently proposed approach of opportunistic sensors network, as in the case where commercial microwave links (CMLs) are used as rainfall sensors, also introduces a new source of errors. In this work, I refer to different approaches and methods for spatial interpolation from point measurements, with an emphasis on the problem of estimating data from microwave links, and suggested an iterative estimator based on Maximum Likelihood (ML) approach to the problem. I propose to increase the robustness of this approach based on the Huber’s M-Estimation method, and suggested a robust variation to the proposed algorithm accordingly: an algorithm based on the Huber loss function, and another censoring algorithm. The algorithms’ robustness was initially tested numerically. This test has shown that each robust variation had an advantage in different situations. The suggested algorithms were tested for several simulated scenarios, where a robust version showed a considerable advantage in the presence of outliers, versus the ML version and the AIDW reference algorithms. Finally, the algorithms were applied for actual CMLs’ real data, provided by Ericsson for the Gothenburg area. The results of the robust algorithms were not significantly different from the non-robust, suggesting that outliers may not be a major source of errors in this case.