|Monitoring east Asian dust storms with geostationary satellite images using an integrated knowledge-based technique
|Wong, Man-sing (LSGI)
|Dust storms -- East Asia
Dust storms -- Remote sensing
Hong Kong Polytechnic University -- Dissertations
|Department of Land Surveying and Geo-Informatics
|xvi, 173 pages : color illustrations
|Dust aerosols can exert severe influence on air quality, human health, social activities, marine environment, and the Earth-Atmosphere system's radiative energy budget. Due to the topology of East Asia, i.e., a decreasing elevation from west to east, dust storms originating from the deserts in this area can transport over a long distance and affect vast areas. The study of dust storms in East Asia is thus critical.
As the leading third-generation Geostationary satellite (GEO), the Himawari-8 offers great opportunities for monitoring dust storms in East Asia in high spatiotemporal resolution. The thesis is devoted to developing a set of methods on monitoring dust storms in East Asia using Himawari-8 observations in both qualitative and quantitative ways.
Firstly, conventional algorithms employ either physical dust indices or machine learning method for dust storm detection. However, the former is threshold dependent, and the latter is physically uninterpretable and requires a large quantity of training data. A physical and self-organizing mapping integrated algorithm (PYSOM) was thus developed to identify the dust presence, which integrates the physical dust indices and the machine learning algorithm of self-organizing mapping (SOM). This method is physically interpretable and also threshold independent. The application of the PYSOM to four dust storm events demonstrated its efficacy in various situations.
Secondly, dust aerosol retrievals are normally carried out with physical models. Little efforts have been contributed to tackling this problem with machine learning algorithms mainly because of limited matchups between ground-truth and satellite observations for dust storms. As the Himawari-8 satellite provides abundant observations to co-match the ground-based measurements (i.e., AERONET products), a dust retrieval method related to machine learning technique was proposed to quantitatively retrieve the dust aerosol properties, including the aerosol optical thickness (AOT) and effective radius (𝑟eff). Six machine learning algorithms were deployed and evaluated, among which the Artificial neural network (ANN) was attested to outperform other algorithms including the tree-based extremely gradient boosting tree (XGBoost), extra tree (ET), random forest (RF), and the kernel-based support vector regression (SVR) and Ridge algorithms. The ANN model achieved the lowest MAE (0.0334 and 0.1487), and the highest 𝑟 (0.94 and 0.63) for respective AOT and 𝑟eff estimation, when validated on an independent dataset. It thus retains comparable performance against physical models. Feature importance analysis reveals the important features identified by ANN agree with empirical and physical understandings of dust aerosols.
Thirdly, the estimation of radiative forcing of dust aerosols involves both solar and terrestrial radiation, which are of great importance to the global climate. However, unified dust aerosol products for both solar and terrestrial spectra are lacking in the literature. To fill this gap, two types of dust retrieval algorithms, one at visible to near-infrared bands (DARV) and the other at thermal-infrared bands (DART), were developed to retrieve AOT at 0.55 μm (𝜏0.55) and AOT at 10.8 μm (𝜏10.8) with effective radius at coarse mode (𝑟eff), respectively. The DARV derives the 𝜏0.55 from four visible to near-infrared bands under the assumption of a fixed dust model. The extra near-infrared band and introduction of spectral sensitivity factors greatly improved its performance under heavy aerosol loadings. The DART retrieves the 𝜏10.8 and 𝑟eff from four thermal bands by prescribing a set of dust models with varying 𝑟eff. This algorithm has innovation in two aspects: 1) the crafted emissivity ratio that advances derivation of spectral surface brightness temperature, and 2) the spectral angle mapper that greatly constrains the retrieval uncertainties. A look-up table was employed for inversion in both algorithms. To verify the proposed methods, five representative dust storm cases in East Asia were selected to test the algorithms using the geostationary satellite Himawari-8. The resulting images of 𝜏0.55 were compared with the Multi-angle Implementation of Atmospheric Correction (MAIAC) retrieved AOT. While visual qualitative comparison demonstrates similar patterns for all cases, a quantitative comparison reveals considerably different correlation coefficients (𝜌) between the two AOTs (ranging from 0.44 to 0.92). The relatively large discrepancies in some cases were likely to be caused by incomplete view of dust storms and the saturation effect under heavy dust loadings in the MAIAC algorithm. Thus, the DARV was demonstrated to be more trustworthy. Comparison between 𝜏10.8 images with the IASI AOT images indicate both of their values ranging between 0.0 and 2.0. In addition, the DART AOT images present much more details than the IASI AOT. In-depth validation against AERONET AOT reveals 𝜌 of 0.93 and 0.69 for the 𝜏0.55 and 𝜏10.8, respectively, and a magnitude consistency for 𝑟eff.
The developed comprehensive set of dust monitoring methods in both qualitative and quantitative ways are of significance and highly contribute to scientific research related to dust aerosol studies. Notably, these developed algorithms can also be easily transferred to other third-generation geostationary satellites and applied for dust storm monitoring in other regions.
|All rights reserved
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item: