Observations of the Impacts of Hong Kong International Airport on Water Quality from 1986 to 2022 Using Landsat Satellite

Abstract

Hong Kong International Airport (HKIA) is an important sea airdrome in China. The aim of this study is to evaluate the impacts of this reclamation on the water quality of the Northwestern Bay of Hong Kong (NWBHK). In all, 117 Landsat 5 TM and 44 Landsat 8 OLI images were preprocessed and matched with the marine water data of 18 in situ monitoring points, acquiring 458 and 119 sets of data, respectively. This study adopted BPNN Machine Learning methods to establish the retrieval algorithm. Based on the images, the construction of HKIA was divided into three stages: (1) the construction of the first and second runways from 1992 to 1995; (2) the construction of the Hong Kong Port from 2013 to 2016; and (3) the construction of the third runway from 2017 to 2020. The concentrations of suspended particulate matter, orthophosphate phosphorus, and dissolved inorganic nitrogen from 1986 to 2022 were retrieved. In this paper, it was found that (1) the construction activities led to an increase in SPM, PO₄P, and DIN concentrations in adjacent water bodies; (2) the impact of the Tuen Mun River on the NWBHK increased, while the impact of the Tung Chung River on the NWBHK decreased; and (3) the interception impact of HKIA on the transportation of the Pearl River water became stronger.

1. Introduction

With the increase in the population, social modernization is accelerating unpredictably. The contradiction in the form of more people and less land is becoming increasingly prominent [1,2]. To address the lack of available land, one of the critical methods of increasing land area in many countries is reclamation, which involves converting open ocean areas into inland areas for human use. In modern society, there is an increase in nearshore and offshore engineering [3,4,5,6,7], which inevitably pollutes the marine environment and influences water quality, for example, in terms of suspended particulate matter (SPM), orthophosphate phosphorus (PO₄P), and dissolved inorganic nitrogen (DIN), especially SPM [8,9,10]. Driven by wind, waves, and ocean currents, SPM and nutrient salts continuously drift and spread, posing a serious threat to the marine ecological balance [11,12,13,14,15]. DIN and PO₄P are the primary indicators of marine ecological balance and coastal ecosystems. Therefore, the impacts of the project on the ecological environment can be assessed by monitoring DIN and PO₄P [16,17].

Hong Kong is a city with limited land resources [5,18]. Therefore, Hong Kong International Airport (HKIA) was built on the waters adjacent to Hong Kong [4]. However, there is a lack of comprehensive research and discussion on the changes in the water quality in the Northwestern Bay of Hong Kong (NWBHK) caused by the reclamation project. Many ground observing stations, such as monitoring points, have been built to monitor water quality, though even from the ground points, it might be challenging to monitor the source and flow of water quality and completely comprehend how it is distributed on a vast scale [16,19]. There are only a few monitoring points around HKIA that do not allow large-scale synchronous observation. In addition, the observation time of different points is usually discontinuous, not allowing simultaneous fine spatial distribution. Fortunately, satellite remote sensing has the advantage of enabling broad and continuous observation. Satellite data can be used to realize large-scale continuous fine spatial distribution monitoring. Therefore, satellite observation is a useful and efficient method of monitoring water quality [15,20,21,22]. Moreover, the monitoring points of the Hong Kong government cannot monitor areas outside Hong Kong, in which case a satellite can be used to observe the changes in the water quality outside Hong Kong and its impact on Hong Kong. In addition, monitoring points can only observe the water quality for a limited time. At the same time, satellite observation can improve the time resolution or offer supplementary data for some years without in situ data.

Many satellites with different temporal and spatial resolutions have been used for water quality observation in many regions [23,24,25,26,27]. Some researchers have used the spatial and temporal adaptive reflectance fusion model to acquire SPM figures with a temporal resolution of 1 h and a spatial resolution of 30 m generated from Landsat 8 Operational Land Imager (L8/OLI) and Geostationary Ocean Color Imager (GOCI) images (STARFM) [15]. Some researchers have additionally used WorldView-2 and Sentinel-2 to retrieve nutrient concentrations and then applied those concentrations to the direct or indirect retrieval of total nitrogen, total phosphorus, dissolved inorganic nitrogen, and dissolved inorganic phosphorus [17,28]. Hong Kong is situated at the mouth of the Pearl River. The quality of water in the Pearl River Estuary has also been studied by some researchers. Ma et al. discovered that, using MODIS/Aqua pictures, machine learning can offer a useful method for estimating total suspended solid and chlorophyll-a (Chl-a) in murky estuaries [19]. Remote sensing was used to track the geographical and temporal fluctuation of the total suspended material concentration in the Pearl River Estuary between 1987 and 2015 [11]. Huang et al. established machine-learning algorithms to retrieve DIN and PO₄P in Shenzhen Bay, situated between Shenzhen and Hong Kong, from long-term time-series (1988–2020) data acquired by the Landsat satellites [16]. Some research about the effects of offshore projects on water quality has also been carried out. Liu et al. assessed the impact of sand excavation upstream of the Pearl River Estuary on the SPM in the Hong Kong–Zhuhai–Macau Bridge construction area using Landsat OLI, ETM+, and TM data [10]. Yan et al. implemented a two-dimensional hydrodynamic model and an advection–diffusion water quality model to study the impacts of the reclamation on the hydrodynamic behavior and water quality in Jinzhou Bay [29].

The spatiotemporal variations in SPM [10,30,31], dissolved oxygen [32], Chl-a [30,33], DIN, and PO₄P [16] in the Pearl River Estuary have been studied via data from various satellites and in situ data. However, most previous studies have focused on changes in the water quality throughout the Pearl River Estuary, with few studies focusing on small water areas affected by human activity [11,19,30,33,34,35,36]. For researching small water areas, it is necessary to select suitable satellite data. MODIS and HY-1C CZI have large image widths (2330 km and 950 km), but low spatial resolution (250 m and 50 m) [14,26,37]. GF-1 WFV has a high spatial resolution (16 m), but short time series (from 2013 to now) [12]. Landsat satellites have the proper image width (185 km), spatial resolution (30 m), and time series (from 1972 to now) [3,31,38]. Previous studies have demonstrated the feasibility of using Landsat satellites to observe water quality changes in the Pearl River Estuary region [10,16,20,33].

Thus, this study focuses on identifying the changes in the water quality in small water areas (NWBHK) affected by human reclamation using Landsat satellites. The main purpose was to observe the construction process of HKIA by satellite and establish a retrieval algorithm to monitor the water quality of the NWBHK from 1987 to 2022. The human influences in the study area were determined by clarifying the construction process of HKIA. Temporal and spatial variations in the natural environment in this region were identified by monitoring the water quality. Combining these two results, the impacts of the sea airdrome on the adjacent water were evaluated. The sources of and long-time variations in SPM, PO₄P, and DIN in the NWBHK were evaluated, with the main sources identified as the river injection from the Tuen Mun River and the Tung Chung River, and the transportation of seawater from the Pearl River Estuary. By analyzing the impact of HKIA on the river injections, the fundamental reasons why HKIA impacts water quality can be better understood. This study divided the construction of HKIA into three stages and discussed the influence of HKIA on the NWBHK as well as several sources of water before, during, and after the construction. Based on the above analysis, the overall impact of HKIA construction was summarized.

2. Description of the Research Area

One of the most well-known marine airports in the world, Hong Kong International Airport (HKIA), is situated below the estuary of the Pearl River. Lantau Island, the largest island in Hong Kong, is located south of HKIA (Figure 1). The Northwestern Bay of Hong Kong (NWBHK) is located north of HKIA. HKIA is at 22.2903°N–22.3244°N latitude and 113.8931°E–113.9450°E longitude (3.9 km × 4.9 km) (Figure 1). HKIA’s first and second runways were constructed over three years, from December 1992 to September 1995, involving a sea-filling type of reclamation [4,5]. A little distance to the north-northeast of Hong Kong International Airport is where the Hong Kong Port is located, which took around three years to build, beginning in August 2013 and ending in June 2016, and resulted in the creation of an artificial island approximately 150 hectares in size. The construction of HKIA’s third runway was officially launched on 1 August 2016. It was completed on 7 September 2021. The third runway was built on land reclaimed in an open water area. The water quality in the vicinity of HKIA has been influenced by the construction. To better analyze the variations in water quality in certain areas, three specific areas are framed in boxes A, B, and C in Figure 1.

3. Data and Methodology

3.1. Quality Data from Marine and River Waters

The in situ marine water data were collected by the Environmental Protection Department of Hong Kong (HKEPD), equipped with a water quality monitoring vessel named Dr. Lam Yunying to carry out the monitoring project. Monitoring the quality of about 1700 km² of marine waters in Hong Kong regularly is a crucial component of HKEPD’s monitoring effort. This project reflects the long-term trend of changes in seawater quality from 1986 to the present and provides the scientific foundation for developing pollution management measures [16,33]. This study used data from 18 marine water monitoring points (NT1, NM1, NM2, etc.) (Figure 2). HKEPD collected the data and water samples at each monitoring point every two months from 1986 to 1999, and the sampling frequency became once a month after 1999. The marine water data acquired by the sampling mainly include the concentrations of SPM, PO₄P, ammonia nitrogen, nitrate nitrogen, nitrite nitrogen, and Chl-a; transparency; and temperature.

HKEPD records not only the marine water quality data, but also the river water quality data in Hong Kong. There are two main rivers around the NWBHK: the Tuen Mun and Tung Chung rivers. The locations of the rivers and the relevant monitoring points are shown in Figure 2.

The Tuen Mun River is located in the northern part of the NWBHK. Only one main stream of the Tuen Mun River flows into the NWBHK. The TN6 point was set up in the estuary of the Tuen Mun River. The data that TN6 provides are considered to indicate what has been injected into the NWBHK from the Tuen Mun River. HKEPD collected the data and water samples from TN6 once a month from 1986 to 2020. The Tung Chung River is located in the southern part of HKIA. Three tributaries of the Tung Chung River flow into the NWBHK and are monitored separately. The Tung Chung River is regarded as the last large intact natural river in Hong Kong with a good natural ecology. Since the construction of HKIA started near Tung Chung (1993), HKEPD has started monitoring the water quality of the Tung Chung River to check whether the water quality has been affected by the HKIA project and related developments nearby. The Tung Chung River has three monitoring points (TC 1–3), and they are distributed in three tributaries. Therefore, the data that the three monitoring points provide are considered to indicate what has been injected into the NWBHK from the Tung Chung River. HKEPD collected the data and water samples at TC 1–3 once every two months from 1993 to 2000 and the sampling frequency became once a month after 2000. The river water data include the concentrations of SPM, PO₄P, ammonia nitrogen, nitrate nitrogen, and nitrite nitrogen; temperature; flow; and turbidity.

3.2. Climatic Observation Data

To assess the effects of climate change on water quality indicators, the monthly mean wind speed, the monthly mean temperature, and the monthly total rainfall from 1997 to 2022 were collected from the climatic observation station at HKIA (22°18′34″N, 113°55′19″E), which was established by the Hong Kong Observatory (Figure 2). The station at HKIA was established on 6 January 1997. The station is equipped with climate monitors to record daily climate data.

3.3. Landsat Data and Preprocessing

This study acquired the satellite data from the Landsat 5 Thematic Mapper (TM) sensor and the Landsat 8 Operational Land Imager (OLI) sensor. The sensor parameters are listed in

Table 1. Introduction of Landsat 5 TM and Landsat 8 OLI sensor parameters.

Satellite	Years of Observation	Spatial Resolution (m)	Revisit Cycle (Days)	Band Reference Number	Band Name	Band Range (μm)
Landsat 5 Thematic Mapper (TM)	1986–2011	30	16	1	Blue	0.45–0.52
				2	Green	0.52–0.60
				3	Red	0.63–0.69
				4	Near-infrared (NIR)	0.76–0.90
				5	Short-wave infrared (SWIR 1)	1.55–1.75
				6	Short-wave infrared (SWIR 2)	2.08–2.35
Landsat 8 Operational Land Imager (OLI)	2013–2022	30	16	1	Coastal aerosol	0.43–0.45
				2	Blue	0.45–0.51
				3	Green	0.53–0.59
				4	Red	0.64–0.67
				5	Near-infrared (NIR)	0.85–0.88
				6	Short-wave infrared (SWIR 1)	1.57–1.65
				7	Short-wave infrared (SWIR 2)	2.11–2.29

. Landsat 5 TM and Landsat 8 OLI have the exact spatial resolution (30 m) and revisit cycle (16 days). The years of observation for Landsat 5 TM are 1986–2011, while the years of observation for Landsat 8 OLI are 2013–2022. Landsat 5 TM has six bands. Landsat 8 OLI has mainly seven bands, with a new coastal aerosol band. Moreover, band 8 is a panchromatic band, which is only used for enhanced resolution. Band 9 is a cirrus band, which only be used for detecting clouds. Furthermore, the concrete band ranges for Landsat 5 TM and Landsat 8 OLI are different [16]. To pick up high-quality images, the cloud coverage was set to less than 20% when the images were downloaded. The path and row numbers based on the Worldwide Reference System (WRS) for the satellite data were 122 and 044, respectively. Considering the cloud coverage and the quality of images in the main research area of this study, 117 Landsat 5 TM images and 44 Landsat 8 OLI images were selected for the study area. The images for 2012 were lacking because of the failure of Landsat 6 and Landsat 7. Generally, there are 3–5 images per year on average, which can provide sufficient data support for the research. The number and quality of images were relatively high in autumn and winter (Figure 3).

The preprocessing of Landsat data involves radiometric calibration, atmospheric correction, water body extraction, cloud elimination, and remote sensing reflectance (R_rs) conversion.

In this study, the software IDL 8.5 and ENVI 5.3 were used for the radiometric calibration of 117 Landsat 5 TM images and 44 Landsat 8 OLI images. In this step, the multispectral files of the Landsat images were selected, and the calibration type was radiance. FLAASH settings were applied with the scale factor of 0.1 [21].

IDL 8.5 and ENVI 5.3 were used for the atmospheric correction of images by Fast Line-of-sight Atmospheric Analysis of Spectral Hypercubes (FLAASH). In this step, the atmosphere model was set to “tropical,” the aerosol model was set to “maritime,” and the aerosol retrieval method was set to “K-T Band”.

To recognize the water area, the normalized difference water index (NDWI) was used [39].

$$\mathrm{NDWI}=\frac{{\mathrm{R}}_{\mathrm{Green}}-{\mathrm{R}}_{\mathrm{NIR}}}{{\mathrm{R}}_{\mathrm{Green}}+{\mathrm{R}}_{\mathrm{NIR}}}$$

(1)

In this formula, R_Green is the surface reflectance of the green band, and R_NIR is the surface reflectance of the NIR band. The area with NDWI value is greater than zero is considered a water area.

The QA Band files of Landsat images provide quality statistics compiled from the cloud, fog, and land masks for the scene. The QA Band file is an unsigned 16-bit COG image with the same number of rows and columns as the scene that corresponds to it in Landsat. The bit representing the cloud is identified in the QA file. If the value of this bit is 0, it means there is no cloud; if the value is 1, it means there is a cloud. This study eliminated the values of pixels with a cloud bit value of 1 in images. Thus, this study used QA Band files to determine which pixels represent a cloud and eliminate the cloud from the image scene.

The R_rs is acquired from surface reflectance (SR). There is a relatively simple formula to correct the SR to R_rs for a satellite image [38,40,41,42,43]:

$${\mathrm{R}}_{\mathrm{rs}}\left(\mathsf{\lambda }\right)=\frac{\mathrm{R}\left(\mathsf{\lambda }\right)-\min \left({\mathrm{R}}_{\mathrm{NIR}}:{\mathrm{R}}_{\mathrm{SWIR}}\right)}{\pi }$$

(2)

In this formula, R(λ) represents the surface reflectance, and R_NIR represents the reflectance of the NIR band. R_rs(λ) represents remote sensing reflectance.

3.4. Retrieval Model

According to the different optical properties, seawater can be divided into Case-1 waters and Case-2 waters. The optical properties of Case-1 waters are mainly determined by phytoplankton. A typical type of Case-1 water is open ocean waters. The optical properties of Case-2 waters are mainly determined by SPM and yellow matter (also known as colored dissolved organic matter (CDOM)). Case-2 waters are mainly located near the shores, estuaries, and other places seriously affected by the emission of terrigenous matter [27,44]. Case-2 water is most closely related to human activities [19,44,45]. In the research area, the water of the Pearl River Estuary is categorized as Case-2 waters. Case-2 waters are especially complex due to different constituents. Thus, it is unsuitable and inaccurate to adopt an empirical or semi-empirical analysis model as the retrieval model. Therefore, in this study, we adopted machine learning methods to establish the retrieval model [19,27,46]. As a method of artificial intelligence, machine learning (ML) can realize data classification, regression, and prediction by simulating the working principle of the human brain and has been widely used in many research fields. Hafeez et al. compared ML algorithms for the retrieval of water quality indicators from satellite data from the Case-2 waters of Hong Kong. Concentrations of SPM, Chl-a, and turbidity were estimated with several ML algorithms, including artificial neural network (ANN), random forest (RF), cubist regression (CB), and support vector regression (SVR). The results indicated that NN-based machine learning approaches perform better [46]. In addition, Huang et al., Ma et al., and other researchers used NN-based algorithms to retrieve the water quality indicators of Shenzhen Bay and the Pearl River Estuary and obtained good results [16,19,47]. The most fundamental and core neural network is the backpropagation neural network (BPNN), which is generally composed of three layers: the input layer, the hidden layer, and the output layer. Errors are back-propagated, while output results are propagated forward. One backward propagation and one forward propagation constitute an epoch of the BPNN. Supervised learning is used in the BPNN. The input is processed in layers during the forward calculation process, starting with the input layer and moving via the hidden unit layer before reaching the output layer. Only the neurons in the layer below are impacted by the status of the neurons in the layer above. The error signal is sent for reverse calculation and returned along the original connection path if the required output cannot be obtained at the output layer. The erroneous signal is reduced by altering the weight of each neuron [48,49]. For this reason, the BPNN can be used to improve the water quality estimation of the coastal waters and was selected to build the retrieval model in this study.

3.5. Summary

In summary, Figure 4 shows the flowchart of data processing. The raw data used in this study include satellite data, in situ data from marine water, in situ data from river water, and climatic data. The preprocessing of satellite data involves radiometric calibration, atmospheric correction, water body extraction, cloud elimination, and remote sensing reflectance (R_rs) conversion. We matched the R_rs values of the remote sensing images and in situ marine water quality data of 18 monitoring points from 1986 to 2020. On the basis of the matched data sets, we selected the sensitive bands and established the retrieval algorithm using the BPNN model. The concentrations of SPM, PO₄P, and DIN were acquired by retrieving the satellite data using the established algorithm. Finally, we analyzed the impacts of HKIA on water quality by combining satellite data, in situ data from waters, and climatic data.

4. Results

4.1. Construction Process of HKIA

In this study, we used the data provided by satellite images to monitor the construction process of HKIA, and the result is shown in Figure 5. The first row indicates the construction activities from 1988 to 1994. The second row indicates the construction activities from 1995 to 2016. The third row indicates the construction activities from 2019 to 2022. This study divides the construction process into three stages.

Stage 1 of the construction was from December 1992 to September 1995 and involved the construction of HKIA’s first and second runways. The area of the island was small, about 3 km², before 1992. As observed in the images, the construction began at the Chek Lap Kok island in the southern part and expanded to the north. More than 9 km² was reclaimed from the sea from 1992 to 1995, for which 180 million cubic meters of sediment was used. Thus, the total area of the island was about 12.5 km² in 1995.

Stage 2 of the construction was from August 2013 to February 2016 and involved the construction of the Hong Kong Port, located northeast of Hong Kong International Airport. The reclamation created an artificial island of about 1.5 km², which connects the Hong Kong airport to the mainland. For this reclamation, 19 million cubic meters of sediment were used. The total area of the islands increased to 14 km² in 2016.

Stage 3 of the construction was from April 2017 to December 2020 and involved the construction of HKIA’s third runway, which is 3800 m long and 60 m wide, in the northwest of HKIA. The third runway was reclaimed using the deep cement mixing method, in which cement is slowly injected into soft silt to form cement lumps that form a stable foundation. For this reclamation, 100 million cubic meters of sediment were used. Stage 3 of the reclamation added 6.50 km² to the island, increasing its total area to 20.5 km² in 2020.

4.2. Modeling

In this study, we developed a useful retrieval model using the BPNN algorithm to analyze the temporal and spatial fluctuations in water quality in the vicinity of HKIA. In this study, to establish an accurate and abundant data set, we matched the R_rs values of the remote sensing images and in situ marine water quality data from 18 monitoring points from 1986 to 2020 (Figure 2). The latitude and longitude were matched based on the in situ marine water quality data and the times were matched within four days. Because the number and range of bands of Landsat 5 and Landsat 8 are inconsistent, different data sets were established for these two sensors. After matching the data, we eliminated part of the matching data by observing the spectral curve features of matching points. Finally, the two data sets for Landsat 5 and Landsat 8 were prepared. For the data set of Landsat 5 TM, 458 sets of data were acquired. In this data set, the SPM concentration was 0.9–160 mg/L, with an average value of 12.0 mg/L; the PO₄P concentration was 0.001–1.1 mg/L, with an average value of 0.08 mg/L; and the DIN concentration was 0.06–12.02 mg/L, with an average value of 1.01 mg/L. For the data set of Landsat 8 OLI, 119 sets of data were acquired. In this data set, the SPM concentration was 1.4–95 mg/L, with an average value of 11.37 mg/L; the PO₄P concentration was 0.001–0.41 mg/L, with an average value of 0.05 mg/L; and the DIN concentration was 0.06–5.7 mg/L, with an average value of 0.83 mg/L. Figure 6 shows the overall high correlation between PO₄P and DIN for the data sets of Landsat 5 TM and Landsat 8 OLI. However, the correlation was relatively low when the value of PO₄P was less than 0.1 mg/L, and the value of DIN was less than 1.0 mg/L. Based on the two data sets of Landsat 5 and Landsat 8, the retrieval model was established using the BPNN model. Figure 7 presents the correlations between the R_rs of bands and the in situ marine water quality data from 18 monitoring points. In Figure 7a, the values of the R of visible bands and the near-infrared band (bands 1–4) are relatively high, and the values of the R of the short-wave infrared bands (bands 5 and 6) are relatively low for Landsat 5. The R of band 4 is the highest (0.34). In Figure 7b, the values of the R of visible bands and the near-infrared band (bands 1–5) are relatively high, and the values of the R of the short-wave infrared bands (bands 6 and 7) are relatively low for Landsat 8. The R of band 4 is the highest (0.61).

In this study, we used the BPNN to establish the retrieval algorithm. The data sets of Landsat 5 and Landsat 8 were divided into a training set (70%), a validation set (15%), and a testing set (15%). The Levenberg–Marquardt method was selected as the training algorithm, and the learning rate was 0.01 for both Landsat 5 and Landsat 8. In the first test, all six bands were used as the inputs for the BPNN fitting for Landsat 5. Thus, the number of inputs was six, and the number of neurons for the hidden layer was set to 10–12. The number of outputs was one for the SPM, PO₄P, and DIN concentrations in the different models. We calculated the root mean square (R) between the in situ values and estimated values to evaluate the accuracy of the algorithm. The R values of SPM, PO₄P, and DIN in this training were 0.77, 0.75, and 0.826, respectively. Considering the low correlation between the R_rs of the short-wave infrared bands (bands 5–6) and the in situ marine water quality, we removed bands 5–6 and used visible and near-infrared bands (bands 1–4) as the inputs in the second test. Thus, the number of inputs was four, and the number of neurons for the hidden layer was set to 10–12. The best training results were in the epoch 6–13 for three models of Landsat 5. The R values of the SPM, PO₄P, and DIN BPNN models were 0.8, 0.8, and 0.84, respectively, and the RMSE values of the SPM, PO₄P, and DIN BPNN models were ±10.72 mg/L, ±0.09 mg/L, and ±0.82 mg/L, respectively (Figure 8). We found that the R values were higher than those of the first test, which means the algorithm is much more accurate. Thus, this study adopted the algorithm of the second test for Landsat 5. For Landsat 8, all seven bands were used as the inputs for the BPNN fitting in the first test. Thus, the number of inputs was seven, and the number of neurons for the hidden layer was set to 10–12. The R values of SPM, PO₄P, and DIN in this training were 0.87, 0.89, and 0.89, respectively. Then, we removed bands 6–7 and used visible and near-infrared bands (bands 1–5) as the input in the second test. Thus, the number of inputs was five, and the number of neurons for the hidden layer was set to 10–12. The best training results were in the epoch 8–12 for three models of Landsat 8. The R values of the SPM, PO₄P, and DIN BPNN models were 0.91, 0.95, and 0.94, respectively, and the RMSE values of the SPM, PO₄P, and DIN BPNN models were ±9.06 mg/L, ±0.02 mg/L, and ±0.35 mg/L, respectively (Figure 9). We found that the R values of the second test were higher than those of the first test, as before. Thus, this study used the visible and near-infrared bands (bands 1–5) to establish the algorithm for Landsat 8.

4.3. Temporal Variations in and Spatial Distributions of SPM

We used the established model for Landsat 5 and Landsat 8 to retrieve the images from 1986 to 2022. Then, the images extensively covered by clouds were removed. The results are as follows. Figure 10 displays the spatial variations in SPM concentrations around HKIA from 1986 to 2021. Since the NWBHK is located downstream of the Pearl River Estuary, the SPM concentrations in the northwest region were relatively high. The spatial distributions on 29 November 2013, 11 January 2018, and 2 December 2020 show high SPM concentrations (approximately 30–60 mg/L) in the areas outside (west) of the NWBHK and low concentrations (approximately 0–20 mg/L) in the areas inside (east) of the NWBHK. According to the image from 29 November 2013, the SPM concentrations in the Hong Kong Port zone increased from about 20 mg/L to 50 mg/L. The image from 11 January 2018 indicates that the SPM concentrations in the zone of the third runway of HKIA increased from about 25 mg/L to 50 mg/L. As per the image from 2 December 2020, the SPM concentrations were high (approximately 30–60 mg/L) in the area on the outside (west), gradually decreasing from north to south, and low (approximately 10–25 mg/L) on the inside (east), gradually decreasing from west to east, apparently reflecting the SPM transportation path of the Pearl River water.

Temporal variations based on the in situ data from the monitoring points of the NWBHK and the river and the remote sensing data are presented below. To eliminate the influence of seasonal variation and analyze the inter-annual variation, we used the entire winter data set (from December to February) in this study. The in situ data from seven marine monitoring points (NM1-8 and NT1) were used to present the temporal variations in the NWBHK (Figure 11a). As per Figure 11a, from 1992 to 1995 (stage 1), the SPM concentrations at almost all points increased abnormally, in the 10–90 mg/L range. From 2002 to 2012, the SPM concentrations decreased from 10 mg/L to 2 mg/L. After 2012, the SPM gradually increased to 15 mg/L. As shown in Figure 11b, the SPM concentration in the Tung Chung River was about 0–12 mg/L, fluctuating in the range of 2–12 mg/L from 1995 to 2000. However, it gradually decreased to about 2 mg/L after 2000, with little change. The SPM concentration in the Tuen Mun River was about 0–40 mg/L. From 1986 to 1995, the SPM concentration increased from 10 mg/L to 40 mg/L and then decreased to 2 mg/L. From 1995 to 2002, there was a significant increase (from 7 mg/L to 20 mg/L). After 2002, the SPM concentration decreased to about 5 mg/L, and there was a slight increase in 2005 and 2012. In 2015, the SPM concentration increased from 5 mg/L to 10 mg/L. Figure 11c shows the SPM concentrations in three areas. To more clearly depict the temporal variations in the quality of water in different regions of the NWBHK, in this study, we divided the bay into three areas: A, B, and C (Figure 1c). Area A is on the outside of the NWBHK, 9 km away from HKIA, and represents the water controlled by the transportation of the Pearl River. Area B is in the channel connecting the outside and inside of the NWBHK. Area B represents the water near HKIA and the transportation of the channel. Area C is on the inside of the NWBHK and the estuary of the Tuen Mun River and represents the water inside of the NWBHK, which is mainly controlled by the transportation of the Pearl River and the Tuen Mun River. The area of each box is 2.25 km². The SPM concentration in area A was relatively high (from 5 mg/L to 90 mg/L). In comparison, the SPM concentration in area B was moderate (from 5 mg/L to 42 mg/L), and the SPM concentration in area C was relatively low (from 5 mg/L to 38 mg/L). From 1988 to 1993, the SPM concentration in area C was relatively high (from 5 mg/L to 38 mg/L). From 1998 to 2018, the SPM concentrations in the three regions showed a trend of fluctuation and decline. From 2018 to 2020 (stage 3), the SPM concentration increased overall (from about 5 mg/L to about 40 mg/L), and the SPM concentration in area B was the highest.

4.4. Temporal Variations in and Spatial Distributions of Nutrient Salts

Figure 12 and Figure 13 show the interannual changes in PO₄P and DIN concentrations in the vicinity of HKIA from 1986 to 2022. Due to the high correlation between PO₄P and DIN, the spatial distribution of PO₄P and DIN had a good consistency. The PO₄P and DIN were generally high in the northwest area outside the NWBHK, ranging from 0.1 to 0.4 mg/L and from 1 to 4 mg/L, respectively, while the PO₄P and DIN were generally low in the interior of the NWBHK, ranging from 0 to 0.3 mg/L and from 0 to 3 mg/L, respectively. The PO₄P and DIN around HKIA were higher, ranging from 0.15 to 0.3 mg/L and 1.5 to 3 mg/L, respectively, on 3 September 1993. Outside the bay, the PO₄P and DIN were higher, ranging from 0.25 to 0.4 mg/L and from 2.5 to 4 mg/L, respectively, on 9 November 1994.

As per Figure 14a, the PO₄P concentrations at the seven monitoring points were relatively close, in the approximate range of 0–0.05 mg/L, which decreased from 1990 to 1992 (from 0.05 mg/L to 0.025 mg/L), increased from 1993 to 1995 (from 0.01 mg/L to 0.05 mg/L), and then fluctuated in the range of 0.01–0.05 mg/L. As per Figure 14b, the PO₄P concentrations in the Tung Chung River and the Tuen Mun River were about 0–0.15 mg/L and 0–0.08 mg/L, respectively. As per Figure 14c, the PO₄P in area A (outside the NWBHK) was relatively high (0.01–0.4 mg/L), while the concentration in areas B and C (inside the NWBHK) was relatively low (0–0.15 mg/L), showing an overall downward and fluctuating trend.

As shown in Figure 15a, the DIN concentration at the seven monitoring points was roughly within the range of 0–1 mg/L, which was relatively stable from 1995 to 2013, and increased to 0.8 mg/L from 2013 to 2020 (stage 2 to stage 3). As shown in Figure 15b, the DIN concentration in the Tung Chung River increased (0–1.5 mg/L). The DIN concentration in the Tuen Mun River was about 0–2.3 mg/L, fluctuating and decreasing after 1995. As shown in Figure 15c, the DIN concentration in area A (outside the NWBHK) was relatively high (0.5–3.8 mg/L), and the DIN concentration in areas B and C (inside the NWBHK) was relatively low (0–1.5 mg/L). Between 1995 and 2015, the DIN concentrations in regions B and C were comparatively steady. However, after 2018, the concentrations significantly increased.

5. Discussion

5.1. Impacts of HKIA during Construction

Through the broad and continuous observation of HKIA using satellite images, this study identified the different stages of HKIA construction. We identified the relevant changes in water quality in the NWBHK during the construction. Figure 16 shows the details of these phenomena.

The false-color images show that on 29 November 2013, the first and second runways of HKIA were completed. The Hong Kong Port was being constructed on the northeast of HKIA, and there was a mass of water containing high concentrations of SPM (about 40–50 mg/L) spreading from the construction area to the interior of the NWBHK. The SPM concentration in the other area was relatively low during this time. The concentrations of PO₄P and DIN increased slightly in this area. On 11 January 2018, the construction of the third runway of HKIA was in process in the northeast of HKIA, but the construction activities can only be observed in some small pixels of the image. From what was observed earlier in HKIA construction, this was the period of foundation laying before the construction of the third runway began, and the water in this area was relatively turbid. For this reason, the SPM concentration in this area rose to 40–50 mg/L in terms of spatial distribution. In terms of temporal variations in SPM (Figure 11a), the SPM concentrations at almost all monitoring points increased abnormally, approximately in the range of 10–90 mg/L, from 1992 to 1995 (stage 1). The SPM concentration was the highest at NM6, located nearest to HKIA. The SPM concentrations decreased at three points in the following order: NM3 > NM1 > NM2. The three points are located east of HKIA. Therefore, the variation in the water quality around HKIA is presumed to be caused by the construction in stage 1. The water mass that contained high concentrations of SPM, PO₄P, and DIN moved eastward due to the driving effect of seawater from the northwest. The SPM concentrations also rose in 2013 and 2019, but no rise is apparent in 1995 in the figure because the in situ data are insufficient.

The phenomena of the rise in SPM and the slight rise in PO₄P and DIN concentrations caused by construction can also be observed in other images. Consequently, in this study, we concluded that during the construction periods of HKIA, construction would cause significant increases in the SPM and slight increases in PO₄P and DIN concentrations in the water around the construction area, and these abnormal increases would generally disappear after the completion of construction.

5.2. River Injection

There are two main rivers around the NWBHK. The Tuen Mun River is in the north of the NWBHK (Figure 2). From the perspective of spatial distribution, the water mass from the Tuen Mun River moves southeastward, and this distribution was apparently not affected by the construction of HKIA. However, the correlation between the Tuen Mun River and the NWBHK changed gradually. For a better observation of this correlation, this study presents the temporal variations in the water quality at monitoring points NM1 and TN6. The result is shown in Figure 17a–c. NM1 represents the inside of the NWBHK, and TN6 roughly represents the Tuen Mun River injection. Before 1995, the SPM concentration at TN6 was relatively large, and the difference between TN6 and NM1 was also significant. Moreover, from 1993 to 1995, the SPM concentration at NM1 was larger than that at TN6, probably caused by the construction of HKIA. After 1995, the difference in the SPM concentrations at the two monitoring points started narrowing. The PO₄P concentration at NM1 and TN6 showed a downward trend from 1987 to 2022, and the slope of TN6 is greater than that of NM1 in Figure 17b. The DIN concentration at NM1 showed an upward trend from 1987 to 2022, while the DIN concentration at TN6 was relatively stable. Overall, the SPM, PO₄P, and DIN concentrations at NM1 and TN6 gradually became the same from 1987 to 2022.

To more accurately assess the correlation of the Tuen Mun River and inside of the NWBHK over time, in this study, we calculated deviation factor P to evaluate the correlation between the concentrations of SPM, PO₄P, and DIN at NM1 and TN6. The concentration at NM1 is x, and the concentration at TN6 is y. The formula for the deviation factor P from x to y is:

$$\mathrm{P}=\left|\left(\mathrm{x}-\mathrm{y}\right)/\mathrm{y}\right|$$

(3)

This means that the higher values of P represent higher degrees of deviation of x from y.

Thus, the deviation factors of the concentrations of SPM, PO₄P, and DIN from NM1 to TN6 were calculated each year from 1987 to 2020, and a linear fitting was carried out for these long-term changes (Figure 17d). In Figure 17d, the slopes of SPM, PO₄P, and DIN are different, but they all show a downward trend, which means the difference between the Tuen Mun River and the inside of the NWBHK was becoming smaller after the construction of HKIA. Therefore, the influence of the Tuen Mun River on the inside of the NWBHK and the related water areas gradually increased. We conclude that this is mainly because the NWBHK was affected by both the river and the external inputs. After the construction of HKIA, the passage for water transportation on the west of the NWBHK became narrower, resulting in an increase in the impact of the Tuen Mun River.

In the south of HKIA is the Tung Chung River (Figure 2). The monitoring points of the Tung Chung River were set up after stage 1 of the construction. Therefore, data from before the construction of HKIA are unavailable. From the perspective of temporal variations, the annual input of the Tung Chung River is relatively stable. However, from the perspective of spatial distribution observed from remote sensing images, noticeable changes have taken place. Before the construction of HKIA, the water mass of the Tung Chung River with high concentrations of SPM, PO₄P, and DIN flowed into the NWBHK unchecked, sharply increasing the SPM, PO₄P, and DIN concentrations at the mouth of the Tung Chung River and in a large area of water. Even after the construction of HKIA, the water mass of the Tung Chung River with high concentrations of SPM was injected into the NWBHK. However, the amount of water mass was small due to the interception effect of HKIA. In summary, we conclude that after the completion of the first and second runways of HKIA (stage 1), the flow of water from the Tung Chung River into the NWBHK was hampered, significantly weakening the impact of the river on the NWBHK.

5.3. Pearl River Transportation

In addition to river injection, the upstream input from the Pearl River from outside the NWBHK also has an essential impact on the NWBHK. The concentration in area A in Figure 1 can roughly represent the external seawater input. In contrast, area B and area C are located at the entrance and inside of the bay, respectively. The water mass with high concentrations of SPM, PO₄P, and DIN generally moves from west to east, and the concentration in area A is generally higher than that in area B or area C in terms of temporal variations and spatial distributions when the influence of construction and river input is small. After the construction of HKIA, part of the upstream inputs was intercepted. The water mass with high concentrations of SPM came into contact with the third runway of HKIA and formed an arc area of a gradient of concentration, as shown in Figure 18. The concentrations of SPM outside and inside the NWBHK were considerably different. The concentrations of PO₄P and DIN were relatively low, but the spatial distributions were the same as that of SPM. It is concluded that the interception effect of HKIA on the transportation from the Pearl River was strengthened with the construction.

To better evaluate the interception impact of the airport, in this study, we calculated the deviation factor P from the water quality of the region from area C to area A each year from 1986 to 2022 and carried out a linear fitting for these long-term changes (Figure 19). As shown in Figure 19, the deviation factors of the concentration of SPM and DIN were on the rise, and the degree of the increase in SPM was much higher than that in DIN. This means that the difference between the concentrations of SPM and DIN in area C and area A was becoming more prominent, and the interception impact of HKIA was becoming stronger with its construction. On the contrary, the deviation factors of the concentration of PO₄P were on the decline. This study infers that this decline was because the PO₄P concentrations in the Tuen Mun River and the outside of the NWBHK were declining. Because of the similar trends of the PO₄P concentrations in these two sources, the concentration of PO₄P in area C was gradually becoming close to that in area A.

5.4. Effects of Climate Change

The location of the climatic observation station is marked in Figure 2. As shown in Figure 20, the range of temperature was within 16–20 °C, the range of wind speed was 12–19 km/h, and the range of rainfall was 0–120 mm. In addition, we calculated the root mean square (R) of the water quality indicators and climate elements (

Table 2. Root mean square (R) of water quality indicators and climate elements.

Monitoring Points	Water Quality Indicators	Wind Speed	Rainfall	Temperature
NM1	SPM	0.38	−0.21	0.05
NM2		−0.15	−0.06	0.34
NM3		−0.01	0.13	0.15
NM5		0.23	−0.39	0.14
NM6		0.41	−0.45	0.2
NM8		0.24	−0.06	−0.13
Total	SPM	0.2	−0.15	0.08
	PO4P	−0.24	−0.08	0.17
	DIN	0.03	0.02	0.41

). The R of wind speed and SPM at multiple monitoring points is greater than 0, indicating that SPM positively correlates with wind speed. This implies that wind can disturb surface water, causing the resuspension of sediments and increasing the concentration of suspended particles [13,38]. For example, the wind speed and SPM both showed the same upward trend from 1999 to 2001. The R of SPM and rainfall is generally less than 0, indicating that SPM negatively correlates with rainfall. Temperature has little influence on SPM according to the low value of R. The R of NM2 is a little strange compared with other points, which may result from the Tuen Mun River injection. In addition, rainfall has a negligible effect on PO₄P and DIN, while temperature has a relatively significant impact on PO₄P and DIN. An increase in temperature leads to a decrease in rainfall and an increase in evaporation, leading to more nutrient salts being dissolved. Thus, an increase in temperature will increase PO₄P and DIN.

In conclusion, the wind increases SPM, rainfall decreases SPM, and temperature increases PO₄P and DIN. However, the R of water quality indicators and climate elements are relatively low on the whole, and the interannual variation in climate elements is slight. Therefore, compared with airport construction and river input, climate elements do not play a significant role in the change in water quality.

6. Conclusions

In this study, we divided the construction of HKIA into three major stages via satellite monitoring: stage 1: construction of HKIA’s first and second runways from December 1992 to September 1995; stage 2: construction of the Hong Kong Port from August 2013 to February 2016; and stage 3: construction of HKIA’s third runway from April 2017 to December 2020.

Then, 117 Landsat 5 TM and 44 Landsat 8 OLI images were used to match the marine water data of 18 in situ marine water monitoring points, and 458 sets and 119 sets, respectively, were acquired. The suspended particulate matter (SPM), orthophosphate phosphorus (PO₄P), and dissolved inorganic nitrogen (DIN) were retrieved based on the BPNN algorithms. For Landsat 5, the R values of the SPM, PO₄P, and DIN BPNN models were 0.8, 0.8, and 0.84, respectively, and the RMSE values of the SPM, PO₄P and DIN BPNN models were ±10.72 mg/L, ±0.09 mg/L, and ±0.82 mg/L, respectively. For Landsat 8, the R values of the SPM, PO₄P, and DIN BPNN models were 0.91, 0.95, and 0.94, respectively, and the RMSE values of the SPM, PO₄P, and DIN BPNN models were ±9.06 mg/L, ±0.02 mg/L, and ±0.35 mg/L, respectively.

A series of temporal variations and spatial distributions were obtained using the BPNN model. By analyzing the distributions on the whole, we observed the following regarding the impacts of HKIA on the NWBHK:

(1)

During the construction period of HKIA, the construction caused an increase in SPM and a slight increase in the PO₄P and DIN concentrations in the water in the construction area. However, this abnormal rise disappeared after the completion of the construction, with little impact on long-term water quality.

(2)

After the completion of the first and second runways of HKIA (stage 1), the passage for water transportation on the west of the NWBHK became narrower, resulting in an increase in the impact of the Tuen Mun River. The flow of the Tung Chung River was intercepted by HKIA, significantly weakening the impact of the SPM, PO₄P, and DIN in the Tung Chung River on the NWBHK.

(3)

As the construction of HKIA proceeded, the transportation of SPM, PO₄P, and DIN via the Pearl River water into the NWBHK was reduced as HKIA intercepted the water.

For the aspect of improving remote sensing technology, this paper proves that Landsat is much more applicable compared with MODIS and HY-1C CZI when exploring the impact of regional long-term human activities. At the same time, the preprocessing of remote sensing, such as cloud elimination and remote sensing reflectance (R_rs) conversion, has been explained much more clearly and applied effectively. In addition, we found that the R_rs of visible and near-infrared bands strongly correlate with the variations in water quality indicators. On the contrary, the short-wave infrared bands barely correlate with the variations in water quality indicators. It is caused by the difference in the absorption coefficient for electromagnetic waves in water. The absorption coefficient for visible and near-infrared bands in water is small. These wavelengths of electromagnetic waves can penetrate the water and reflect changes in the water. The shortwave infrared bands of electromagnetic waves are almost entirely absorbed by water at the surface. Thus, bands 1–4 of Landsat 5 and bands 1–5 of Landsat 8 can be used to observe water quality. In addition, we have shown the high applicability of the BPNN algorithm for establishing remote sensing algorithms in nearshore water bodies.

To further evaluate the impacts of HKIA, in future studies, we will attempt to use data from satellites with higher temporal and spatial resolutions, such as the GF (GaoFen) satellite or Sentinel-2 satellite. The fusion of Landsat and GOCI data is also a great option. Furthermore, various models will be used to simulate the hydrodynamic environment around HKIA.