| Author: | Kwok, Yin Tung Coco |
| Title: | An integrated knowledge-based remote sensing technique for coastal water quality monitoring |
| Advisors: | Wong, Man Sing (LSGI) |
| Degree: | Ph.D. |
| Year: | 2023 |
| Subject: | Water quality -- Remote sensing Marine pollution Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Land Surveying and Geo-Informatics |
| Pages: | xx, 145 pages : color illustrations |
| Language: | English |
| Abstract: | Monitoring marine environment is a challenging task as it is the largest natural resource in the world. Remote sensing technology provides a good platform to monitor the ocean frequently and broadly. However, there are several challenges in monitoring marine waters using remote sensing. The first one is the atmospheric correction on the water bodies. The reflectance is always small for the water bodies because the water body absorbs most of the light but only reflects a limited amount of light. Therefore, accurate determination of atmospheric properties is crucial since this would significantly affect the energy captured by satellite sensors. This study developed a physical-based aerosol optical depth retrieval model for the coastal marine environment in Hong Kong using in-situ hyperspectral field measurements, where the results showed a strong significant correlation with AERONET station (R2 = 0.93). The second challenge is the low spatial or temporal resolutions of satellite images. For satellite sensors with high temporal resolution, the spatial resolution is low, e.g. MODIS and Himawari-8/9 (AHI). Conversely, for satellite sensors with high spatial resolution, the temporal resolution is low, e.g. Landsat and Sentinel-2. To tackle this problem, this study developed a satellite-based and computer vision-based super-resolution model, named multi-scale land cover-oriented super-resolution model. This model was developed based on a deep convolutional neural network together with multi-scale land cover map with spatial feature transformation. The developed model can enhance the coarse spatial resolution image to high resolution, i.e. enhancing MODIS and AHI images to 30 m Landsat-like images. Evaluation was conducted by comparing with benchmarks super-resolution model and visual analysis. The result showed that a very strong correlation (R2 = 0.97) was achieved for improving the spatial resolution by four times whereas a strong correlation (R2 = 0.87) was achieved for 16 times, which is approximately the scale difference between Landsat and MODIS images. A moderate correlation (R2 = 0.63) was achieved for 64 times, which is approximately the scale difference between Landsat and AHI images. Promising water quality retrieval methods were developed in this study using Artificial Neural Network, aiming to achieve accurate water quality monitoring especially in Case II waters which is a sophisticated environment in amidst of anthropogenic pollutions. The Artificial Neural Network could map the chlorophyll-a, suspended solids and turbidity at R2 values of 0.822, 0.707 and 0.693 respectively on Landsat images. The models were also evaluated with other machine learning algorithms, including Elastic Net, Gaussian Processes, Linear regression, Multilayer Perceptron Regression, Radial Basis Function Regression, Sequential Minimal Optimization Regression, K-nearest neighbors, Locally weighted learning, Stochastic Gradient Boosting, Bagging, Ensemble Selection, Multi-Search, Random Committee, Randomizable Filtered Classifier, Random Sub Space, M5Rules, Random Forest and Random Tree, highlighting the superior performance of Artificial Neural Network models. Retrieval models were also developed on the original MODIS dataset and the super-resolution MODIS dataset at 30 m resolution, suggesting improvements of R2 from 0.786 to 0.927, 0.849 to 0.879 and 0.203 to 0.693 for retrieving chlorophyll-a, suspended solids and turbidity respectively. The results indicated that the retrieval accuracy and performance could be enhanced with the super-resolution model. In overall, this study demonstrated the potential of accurately monitoring the water quality at a high spatial and temporal resolution. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
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:
https://theses.lib.polyu.edu.hk/handle/200/12738