Study on deep-learning-based identification of hydrometeors observed by dual polarization Doppler weather radars
Wang et al. EURASIP Journal on Wireless Communications and Networking
Study on deep-learning-based identification of hydrometeors observed by dual polarization Doppler weather radars
Haijiang Wang 0 1
Yuanbo Ran 1
Yangyang Deng 1
Xu Wang 0 1
0 CMA Key Laboratory of Atmospheric Sounding , Chengdu, Sichuan 610225 , China
1 College of Electronic Engineering, Chengdu University of Information Technology , Chengdu, Sichuan 610225 , China
Hydrometeor classification for dual polarization Doppler weather radar echo is a procedure that identifies hydrometeor types based on the scattering properties of precipitation particles to polarized electromagnetic waves. The difference in shape, size, or spatial orientation among different types of hydrometeor will produce different scattering characteristics for the electromagnetic waves in a certain polarization state. Moreover, the polarimetric measurements, which are calculated from the radar data and closely associated with these characteristics, are also different. The comprehensive utilization of these polarimetric measurements can effectively improve the identification accuracy of the phase of various hydrometeors. In this paper, a new identification method of the hydrometeor type based on deep learning (DL) and fuzzy logic algorithm is proposed: firstly, the feature extraction method based on deep learning is used for training the correlation among multiple parameters and extracting the relatively independent features. Secondly, the Softmax classifier is applied to classify the precipitation patterns, including rain, snow, and hail, and it is based on the features extracted by deep learning algorithm. Finally, the fuzzy logic algorithm is adopted to identify the hydrometeor types in various precipitation patterns. In order to test the accuracy of the classification results, the hydrometeor classifier has been applied to a stratiform cloud precipitation process, and it is found that the classification results agree well with the other polarimetric products.
Hydrometeor; Polarimetric measurements; Deep learning; Feature; Fuzzy logic algorithm
1 Introduction
The dual linear polarization radar can transmit
horizontal and vertical polarization waves alternately or
simultaneously, and it can also use different signal processing
methods to deal with the echo signals from two
polarization directions. Moreover, it is easy to obtain the
horizontal reflectivity (ZH), differential reflectivity (ZDR),
co-polar correlation coefficient (ρHV), differential
propagation phase constant (KDP), and other polarization
parameters. The difference in shape, size, or spatial
orientation between different types of hydrometeor will
produce different polarization parameters, and it can
promote the development of hydrological meteorological
classification by these polarization parameters.
Compared with the conventional Doppler weather radar
system, its ability to estimate the precipitation and
recognize the hydrometeor phase has been improved
significantly. What is more, it is an important tool in the
fields of artificial influence on weather, aviation warning,
and disaster monitoring [
1–5
].
Liu et al. [
6
] established a hydrometeor classification
system based on fuzzy logic and neural network. In the
system, the horizontal reflectivity, differential reflectivity,
differential propagation phase shift, correlation
coefficient, linear depolarization ratio, and the corresponding
height are used as the inputs, and the neural network
learning algorithm is applied to adjust the parameters.
Finally, the inputs and parameters of the system are
calculated to determine the type of hydrometeors [
6–8
].
Chandrasekar et al. [
9
] summarized the researches on
echo classification and the identification of hydrological
fluid, which are based on dual polarization radar in
recent years. The classification principle of various types
was described, and the characteristics of hydrometeor
classification were analyzed. It promoted the study in
the hydrometeor classification of the dual polarization
radar greatly [
9, 10
].
Besic et al. [
11
] used a semi-supervised approach to
realize the classification of hydrometeors. In this study,
the K-medoids (KM) approach is used to cluster the
sample data and the clustering results are evaluated by
the Kolmogorov-Smirnov (KS) test method. Finally, the
fuzzy logic algorithm is applied to realize the highly
precise classification of hydrometeors based on the
clustering results [
11
].
Hinton et al. [
12
] published an article in Science,
which opened a gate for deep learning in the field of
machine learning. Deep learning, as a kind of emerging
learning algorithm of multi-layer neural network, has
solved the local minimum defect in the traditional
training algorithm. Moreover, it has been widely used
in machine learning and computer vision and has
aroused widespread concern in various fields [
12–15
].
Tao et al. [16] used the deep learning approach (...truncated)