A multi-scale hybrid neural network retrieval model for dust storm detection, a study in Asia

Man Sing Wong, Fei Xiao, Janet Nichol, Jimmy Fung, Jhoon Kim, James Campbell, P. W. Chan

Research output: Contribution to journalArticle

16 Citations (Scopus)

Abstract

Dust storms are known to have adverse effects on human health and significant impact on weather, air quality, hydrological cycle, and ecosystem. Atmospheric dust loading is also one of the large uncertainties in global climate modeling, due to its significant impact on the radiation budget and atmospheric stability. Observations of dust storms in humid tropical south China (e.g. Hong Kong), are challenging due to high industrial pollution from the nearby Pearl River Delta region. This study develops a method for dust storm detection by combining ground station observations (PM10 concentration, AERONET data), geostationary satellite images (MTSAT), and numerical weather and climatic forecasting products (WRF/Chem). The method is based on a hybrid neural network (NN) retrieval model for two scales: (i) a NN model for near real-time detection of dust storms at broader regional scale; (ii) a NN model for detailed dust storm mapping for Hong Kong and Taiwan. A feed-forward multilayer perceptron (MLP) NN, trained using back propagation (BP) algorithm, was developed and validated by the k-fold cross validation approach. The accuracy of the near real-time detection MLP-BP network is 96.6%, and the accuracies for the detailed MLP-BP neural network for Hong Kong and Taiwan is 74.8%. This newly automated multi-scale hybrid method can be used to give advance near real-time mapping of dust storms for environmental authorities and the public. It is also beneficial for identifying spatial locations of adverse air quality conditions, and estimates of low visibility associated with dust events for port and airport authorities.

Original languageEnglish
Pages (from-to)89-106
Number of pages18
JournalAtmospheric Research
Volume158-159
DOIs
Publication statusPublished - 2015 May 1

Fingerprint

dust storm
back propagation
air quality
dust
weather
geostationary satellite
radiation budget
hydrological cycle
airport
visibility
detection
Asia
global climate
climate modeling
fold
ecosystem
river
method

All Science Journal Classification (ASJC) codes

  • Atmospheric Science

Cite this

Wong, Man Sing ; Xiao, Fei ; Nichol, Janet ; Fung, Jimmy ; Kim, Jhoon ; Campbell, James ; Chan, P. W. / A multi-scale hybrid neural network retrieval model for dust storm detection, a study in Asia. In: Atmospheric Research. 2015 ; Vol. 158-159. pp. 89-106.
@article{5b07cb533b854bb79e4e532b3bec101e,
title = "A multi-scale hybrid neural network retrieval model for dust storm detection, a study in Asia",
abstract = "Dust storms are known to have adverse effects on human health and significant impact on weather, air quality, hydrological cycle, and ecosystem. Atmospheric dust loading is also one of the large uncertainties in global climate modeling, due to its significant impact on the radiation budget and atmospheric stability. Observations of dust storms in humid tropical south China (e.g. Hong Kong), are challenging due to high industrial pollution from the nearby Pearl River Delta region. This study develops a method for dust storm detection by combining ground station observations (PM10 concentration, AERONET data), geostationary satellite images (MTSAT), and numerical weather and climatic forecasting products (WRF/Chem). The method is based on a hybrid neural network (NN) retrieval model for two scales: (i) a NN model for near real-time detection of dust storms at broader regional scale; (ii) a NN model for detailed dust storm mapping for Hong Kong and Taiwan. A feed-forward multilayer perceptron (MLP) NN, trained using back propagation (BP) algorithm, was developed and validated by the k-fold cross validation approach. The accuracy of the near real-time detection MLP-BP network is 96.6{\%}, and the accuracies for the detailed MLP-BP neural network for Hong Kong and Taiwan is 74.8{\%}. This newly automated multi-scale hybrid method can be used to give advance near real-time mapping of dust storms for environmental authorities and the public. It is also beneficial for identifying spatial locations of adverse air quality conditions, and estimates of low visibility associated with dust events for port and airport authorities.",
author = "Wong, {Man Sing} and Fei Xiao and Janet Nichol and Jimmy Fung and Jhoon Kim and James Campbell and Chan, {P. W.}",
year = "2015",
month = "5",
day = "1",
doi = "10.1016/j.atmosres.2015.02.006",
language = "English",
volume = "158-159",
pages = "89--106",
journal = "Atmospheric Research",
issn = "0169-8095",
publisher = "Elsevier BV",

}

A multi-scale hybrid neural network retrieval model for dust storm detection, a study in Asia. / Wong, Man Sing; Xiao, Fei; Nichol, Janet; Fung, Jimmy; Kim, Jhoon; Campbell, James; Chan, P. W.

In: Atmospheric Research, Vol. 158-159, 01.05.2015, p. 89-106.

Research output: Contribution to journalArticle

TY - JOUR

T1 - A multi-scale hybrid neural network retrieval model for dust storm detection, a study in Asia

AU - Wong, Man Sing

AU - Xiao, Fei

AU - Nichol, Janet

AU - Fung, Jimmy

AU - Kim, Jhoon

AU - Campbell, James

AU - Chan, P. W.

PY - 2015/5/1

Y1 - 2015/5/1

N2 - Dust storms are known to have adverse effects on human health and significant impact on weather, air quality, hydrological cycle, and ecosystem. Atmospheric dust loading is also one of the large uncertainties in global climate modeling, due to its significant impact on the radiation budget and atmospheric stability. Observations of dust storms in humid tropical south China (e.g. Hong Kong), are challenging due to high industrial pollution from the nearby Pearl River Delta region. This study develops a method for dust storm detection by combining ground station observations (PM10 concentration, AERONET data), geostationary satellite images (MTSAT), and numerical weather and climatic forecasting products (WRF/Chem). The method is based on a hybrid neural network (NN) retrieval model for two scales: (i) a NN model for near real-time detection of dust storms at broader regional scale; (ii) a NN model for detailed dust storm mapping for Hong Kong and Taiwan. A feed-forward multilayer perceptron (MLP) NN, trained using back propagation (BP) algorithm, was developed and validated by the k-fold cross validation approach. The accuracy of the near real-time detection MLP-BP network is 96.6%, and the accuracies for the detailed MLP-BP neural network for Hong Kong and Taiwan is 74.8%. This newly automated multi-scale hybrid method can be used to give advance near real-time mapping of dust storms for environmental authorities and the public. It is also beneficial for identifying spatial locations of adverse air quality conditions, and estimates of low visibility associated with dust events for port and airport authorities.

AB - Dust storms are known to have adverse effects on human health and significant impact on weather, air quality, hydrological cycle, and ecosystem. Atmospheric dust loading is also one of the large uncertainties in global climate modeling, due to its significant impact on the radiation budget and atmospheric stability. Observations of dust storms in humid tropical south China (e.g. Hong Kong), are challenging due to high industrial pollution from the nearby Pearl River Delta region. This study develops a method for dust storm detection by combining ground station observations (PM10 concentration, AERONET data), geostationary satellite images (MTSAT), and numerical weather and climatic forecasting products (WRF/Chem). The method is based on a hybrid neural network (NN) retrieval model for two scales: (i) a NN model for near real-time detection of dust storms at broader regional scale; (ii) a NN model for detailed dust storm mapping for Hong Kong and Taiwan. A feed-forward multilayer perceptron (MLP) NN, trained using back propagation (BP) algorithm, was developed and validated by the k-fold cross validation approach. The accuracy of the near real-time detection MLP-BP network is 96.6%, and the accuracies for the detailed MLP-BP neural network for Hong Kong and Taiwan is 74.8%. This newly automated multi-scale hybrid method can be used to give advance near real-time mapping of dust storms for environmental authorities and the public. It is also beneficial for identifying spatial locations of adverse air quality conditions, and estimates of low visibility associated with dust events for port and airport authorities.

UR - http://www.scopus.com/inward/record.url?scp=84924264977&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84924264977&partnerID=8YFLogxK

U2 - 10.1016/j.atmosres.2015.02.006

DO - 10.1016/j.atmosres.2015.02.006

M3 - Article

AN - SCOPUS:84924264977

VL - 158-159

SP - 89

EP - 106

JO - Atmospheric Research

JF - Atmospheric Research

SN - 0169-8095

ER -