Water quality monitoring in Hong Kong using remote sensing

Abstract

The main objective of this study is to define optically different water types and monitor Water Quality Parameters (WQPs) i.e. Chlorophyll-a (Chl-a) and Suspended Solids (SS) over the complex coastal environment of Hong Kong at high spatial resolution of 30 meters and temporal resolution of 1 to 2 days using remote sensing. For this purpose 13-years (January 2000 to December 2012) of satellite observations from two different platforms including Landsat Thematic Mapper (TM), Enhanced Thematic Mapper Plus (ETM+) and Chinese HJ-1 A/B Charge Couple Device (CCD) sensors, and in situ measurements are used. In order to increase the combined usability of two different satellites, a cross comparison of the respective sensors is carried out to check their mutual consistency. The results show a high degree of consistency between the sensors, especially for the first three bands where the correlation is ≥ 0.93, while band 4 obtained a correlation of 0.80, probably due to the difference in the relative spectral response function of the sensors. For accurate monitoring of WQPs over the study area, an accurate satellite estimation of water Surface Reflectance (SR) is required. Therefore, this study evaluated the performance of five atmospheric correction methods, namely 6S (Second Simulation of the Satellite Signal in the Solar Spectrum), FLAASH (Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes), ATCOR (ATmospheric CORection), ELM (Empirical Line Method) and DOS (Dark Object Subtraction). The estimated SR of the first four reflective bands of Landsat 7 ETM+ and the identical bands of the HJ-1 A/B CCD sensors are validated using in situ Multispectral Radiometer (MSR) SR measurements over different types of surfaces including the water surface. For comparison purpose, the new standard Landsat SR product LEDAPS (Landsat Ecosystem Disturbance Adaptive Processing System) was also evaluated using the in situ measured SR data. The first four bands of Landsat TM/ETM+ and HJ-1 A/B CCD sensors were atmospherically corrected and used in conjunction with the in situ Chl-a and SS concentrations for clustering and delineation of water zones. The results based on Fuzzy C-Means clustering suggested the existence of five optically different water types in the study area, and the boundary was delineated using ordinary point kriging. Finally, linear Regression Modeling (RM) and Neural Network (NN) techniques were used for the modeling of Chl-a and SS concentrations. For modeling of Chl-a concentrations using RM, the results suggest that a ratio of red and blue bands was able to represent 84% of the spatial variability in Chl-a concentrations in the complex coastal and estuarine waters surrounding Hong Kong, and was superior to the use of single bands and other band ratios. The effectiveness of this band ratio can also be attributed to the existence of specific phytoplankton species associated with Hong Kong waters. The NN results were also insightful and were able to solve the non-linearity of the Chl-a concentrations, and explaining up to 93% of the variability. Furthermore, Chl-a concentrations in the coastal region of Hong Kong can reliably be predicted over a wide range including low to high turbidity conditions, as well as severe pollution events such as red tides. For the modeling of SS concentrations using RM, the log-transformed combination of green and red bands was found to be able to capture both the spatial and temporal variability of SS concentrations over the complex coastal environment of Hong Kong. The NN also performed well in capturing the non-linearity and modeled the SS concentrations with a high correlation of 91% for the training dataset and was the best method than RM. In this study, the devised models are robust as they have the ability for better monitoring of Chl-a and SS concentrations and captured up to 93% of the spatial variability for the complex coastal environment of Hong Kong. The models' robustness can be attributed to the use of an extensive satellite (13-year record) and in situ collocated dataset for model development, and in addition, images from different years and months enabled the investigation of temporally variable Chl-a and SS concentrations over the study area. The results demonstrate the applicability of the models for synoptic retrieval of spatially variable concentrations around the complex coastal waters of Hong Kong. The results suggested that the approach outlined in this study can be used for routine water quality monitoring by the local environmental agencies and can be applied to other complex coastal regions.