Please use this identifier to cite or link to this item: http://hdl.handle.net/10553/60173
|Title:||On the application of a recurrent neural network for rainfall quantification based on the received signal from microwave Links||Authors:||Guerra-Moreno, Iván
Navarro-Mesa, Juan L.
Ravelo-Garcia, Antonio G.
Suárez Araujo, Carmen Paz
|UNESCO Clasification:||120304 Inteligencia artificial
25 Ciencias de la tierra y del espacio
|Keywords:||Recurrent neural networks
Received signal strength
|Issue Date:||2019||Project:||Sistema de Vigilancia Meteorológica para el Seguimiento de Riesgos Medioambientales||Journal:||Lecture Notes in Computer Science||Abstract:||The detection and quantification of rainfall is of paramount importance in many application contexts. The research work we present here is devoted to design a system to detect meteorological phenomena in situations of risk. Particularly, we extend the usage of systems designed for other specific purposes incorporating them weather observation as a new service. We investigate how machine learning techniques can be used to design rain detection and quantification algorithms that learn directly from data and become robust enough to perform detection under changing conditions over time. We show that Recurrent Neural Networks are well suited for rainfall quantification, and no precise knowledge of the underlying propagation model is necessary. We propose a recurrent neural architecture with two layers. A first layer acts as a detector-quantifier and it is trained using known precipitations from near rain gauges. The second layer plays the role of calibration that transforms previous levels into rainfall quantitation values. A very important aspect of our proposal is the feature extraction module, which allow a reliable detection and accurate quantification. We study several options for extracting the most discriminative features associated to rain and no rain events. We show that our system can detect and quantify rain-no rain events with promising results in terms of sensitivity (76,6%), specificity (97,0%) and accuracy (96,5%). The histograms of error and the accumulated rain rates show good performance, as well.||URI:||http://hdl.handle.net/10553/60173||ISBN:||978-3-030-20520-1||ISSN:||0302-9743||DOI:||10.1007/978-3-030-20521-8_4||Source:||Rojas I., Joya G., Catala A. (eds) Advances in Computational Intelligence. IWANN 2019. Lecture Notes in Computer Science, vol 11506. Springer, Cham|
|Appears in Collections:||Actas de congresos|
checked on Feb 22, 2020
Items in accedaCRIS are protected by copyright, with all rights reserved, unless otherwise indicated.