The Guangdong-Hong Kong-Macao Greater Bay Area (GBA), more commonly referred to as the GBA, is one region in China with the highest urbanization and industrialization rates (Zhou et al., 2018). Along with the GBA’s rapid economic development, a large amount of nitrogen (N) and phosphorus (P) pollutants from industrial and agricultural wastewater and domestic sewage is discharged into the PRE (Huang et al., 2003; Tao et al., 2021). The PRE’s water surface area, which the GBA surrounds, is approximately 1700 km². Its longitudinal and latitudinal coordinates are 113°44′43.1″ E and 22°34′38.25″ N, respectively (Yuan et al., 2022). The PRE’s complex hydrological network and diverse environmental conditions have made it a prominent research focus within the scientific community (Wu et al., 2016). Under the influence of its subtropical monsoon climate, river flow and wind speed are manifested by pronounced seasonal variation (Lu and Gan, 2015). Most annual discharge from the Pearl River (approx. 80%) occurs during the rainy season (Ye et al., 2017). This causes considerable disparity in river flow between the dry season (October to April) and the rainy season (May to September), with average values of approximately 1500 m³/s and 20000 m³/s, respectively (Liu et al., 2012). In the PRE, weak southwesterly winds predominant during the summer, whereas strong northeastern winds dominate during the winter (Zhang et al., 2019). Seasonal variations in hydrodynamic conditions have a considerable influence on seasonal water quality dynamics in the PRE. In brief, climate change and rapid urbanization exert a heavy strain on PRE water quality, posing significant challenges to its ecological well-being. The PRE is a large aquatic region. Anthropogenic activities in many of its cities throughout its different sections cause varying degrees of pollution and nutrient inputs, which significantly impact Chl-a concentrations. In this study, we divided the PRE into four sections based on its coastal cities to analyze spatiotemporal evolutionary features ( Figure 1 ). These four sections were specifically divided into the northern section (adjacent to Guangzhou and Dongguan), the eastern section (adjacent to Shenzhen), the western section (adjacent to Zhuhai and Zhongshan), and the southern section (adjacent to the open sea).

Since 1986, the Environmental Protection Department (EPD) of Hong Kong has measured water quality monthly within its coastal waters. This study uses water quality indicators, including Chl-a, TP, TN, ammonia nitrogen (NH₃-N), and biotic oxygen demand (BOD₅). Long-term data on these water quality indicators were collected between 1989–2019, which we sourced from 9 EPD meteorological stations ( Figure 1 ). For detailed information on water quality, please refer to the EPD website. (https://www.epd.gov.hk).

Precipitation and temperature fluctuations can give rise to alterations in the physical and chemical attributes of aquatic environments. Anthropogenic activities, including the discharge of industrial wastewater and the agricultural development of urban areas, exert a considerable influence on water quality, which are as significant as external factors (Liu et al., 2022). A higher proportion of primary industries is indicative of a developed agricultural sector within urban areas. Therefore, this study selected temperature, precipitation, industrial wastewater discharge, and the proportion of primary industries as driving factors. Given the limited availability of historical water temperature data within the study area, we used temperature data from cities near the PRE between 1989–2019, which we obtained from meteorological websites (http://data.cma.cn/). The precipitation data used in this study derive between 1989–2019, which we obtained from the daily global precipitation measurement dataset provided by the National Aeronautics and Space Administration (NASA) (https://disc.gsfc.nasa.gov). Socio-economic development factors considered in this study include the industrial wastewater discharge, and the proportion of the primary industry in the cities neighboring the PRE between 1989–2019. Urbanization data used in this study primarily derive from statistical yearbooks, which were obtained from various regions.

This study uses satellite imagery products, including digital products from the Landsat 5 and Landsat 7 satellites. The Landsat series satellites are jointly developed by NASA and the United States Geological Survey (USGS) which were launched in 1972. These satellites have a revisit period of 16 days and a spatial resolution of 30 m. Considering sampling time and cloud cover on satellite acquisition dates, we collected approximately 100 surface reflectance images of the PRE between 1989–2019. Satellite imagery products are obtained from USGS (https://earthexplorer.usgs.gov).

The satellite images used in this study included both water and land information. However, only water information was used during the retrieval process. It was necessary to extract water information and remove land information. Google Earth Engine (GEE) platform offers a geospatial processing service. Users can perform large-scale geospatial processing provided by Google Cloud Platform support. Landsat 5 Collection 2 Tier 1 calibrated top-of-atmosphere reflectance (TOA) reflectance images were directly downloaded from the GEE platform (Mutanga and Kumar, 2019). Because the GEE platform eliminates the need for radiometric calibration and atmospheric correction, we used it to preprocess Landsat 5 images between 1989–2011, including cropping, cloud removal, and band manipulation.

To ensure research analysis accuracy and improve the extraction precision of satellite image information, it was necessary to use atmospheric correction to remove interference caused by atmospheric scattering and other factors on land features. In this study, ENVI 5.3 software was used to preprocess Landsat 7 images between 2012–2019, including cropping, striping removal, radiometric calibration, atmospheric correction, and band calculations. The FLAASH atmospheric correction tool was used for atmospheric correction, which ENVI 5.3 provides. It offers high multispectral data accuracy correction, and it has widely been used in atmospheric correction models. Additionally, digital numbers recorded by the sensor must be converted to radiance values. The radiometric calibration tool in ENVI 5.3 was used for conversions, applying the following formula:

where L is radiance; DN is the recorded sensor grayscale values; a represents gain; L₀ represents the offset value.

Chl-a concentration remote sensing retrieval is a process of estimating Chl-a concentrations from satellite images by utilizing its spectral reflectance characteristics. Chl-a exhibits strong absorption and reflection in the blue wavelength range(450-495nm) and green wavelength range(495-570nm) (O'Reilly et al., 1998). Both bands strongly correlate with Chl-a. Most empirical algorithms use them to establish models for Chl-a concentration retrieval.

Two-band (OC2v4) and four-band (OC4v4) global algorithms were developed by NASA; however, they tend to systematically overestimate Chl-a concentrations (D'Ortenzio et al., 2002). The OC2_D’Ortenzio algorithm is an improved version of OC2v4. It performs well when applied to biogeochemical measurements or satellite data (Pan et al., 2010). Considering the dynamic nature of the study area, we used the OC2_D’Ortenzio algorithm to construct a Chl-a concentration retrieval model. This algorithm is based on a nonlinear relationship, improved by a cubic polynomial function, between ocean reflectance and in-situ Chl-a concentrations measurements.

where C is the Chl-a concentration; B is the land-surface reflectance ratio; R_rs490 and R_rs555 were obtained from 490 nm and 555 nm land surface reflectance; a = 0.217, −2.728, 0.704, 0.297, and −0.035, which are algorithm constants.

To evaluate model accuracy, the statistical metrics used for validation were the coefficient of determination (R²), the root-mean-square error (RMSE), and biases. The following equations show the formulae used:

where n is the sample size; S_est,i is the value fitting curve obtained; X_obs,i is the in-situ data obtained; _obs,i is the average of X_obs,i ; X_est,i is the estimated data obtained.

Relationships among water quality factors (TP, TN, NH₃-N, and BOD₅) and external driving factors are examined through Pearson correlation analysis and regression analysis. Pearson correlation coefficient is used to measure the correlation between various variables and Chl-a concentrations in the PRE. Pearson correlation coefficient ranges from −1 to +1, where −1 denotes no negative correlation, +1 denotes a positive correlation, and 0 denotes no correlation. A significance level<0.05 (P< 0.05) was considered statistically significant. Data preparation for Pearson correlation analysis was conducted using IBM SPSS Statistics 27 software.

Partial least squares regression (PLS regression) models were used to analyze driving factors. PLS regression is a statistical modeling method used to establish predictive models between input variables and output variables. Its advantage is its ability to manage highly correlated input variables. By constructing latent variables, it decouples relationships between input variables and output variables. Additionally, PLS regression can effectively manage small sample sizes. PLS regression effectiveness can be evaluated using Variable Importance in Projection (VIP) scores. VIP scores indicate the importance of each predictor variable in explaining the response variable. If the VIP score of a predictor variable is >1.0, the variable has a significant impact on explaining the response variable. On the other hand, if the VIP score is<0.5, the variable does not contribute significantly to explaining the response variable. Values between 0.5 and 1.0 indicate a moderate level of importance. PLS regression was constructed using IBM SPSS Statistics 27 software.

This study validates the performance of a remote sensing retrieval model used to estimate Chl-a concentrations by comparing simulated values with a dataset of measured data. The measured data spans from 1989 to 2019, with a total of 113 observations. Measured data were obtained from the EPD of Hong Kong and is considered highly reliable. Simulated Chl-a concentration values derived from satellite images. Simulated values and measured data correspond closely in time and geographical coordinates. The RMSE of the model was 2.55 mg/L, with a bias of −1.39 mg/L. A scatter plot ( Figure 2 ) was generated using measured values as the independent variable and the simulated values as the dependent variable. Outliers that deviate significantly from other data points were removed, resulting in a total of 110 data points. The data points are relatively clustered, and the R² values between the linear regression line (i.e., the blue solid line) and the data points was 0.7155. Difference among verification results and that from other relevant PRE studies were within an acceptable range (Ma et al., 2022). This strongly indicates that this model used in this study can estimate PRE Chl-a concentrations.

This study analyzed annual spatial distribution characteristics in PRE Chl-a concentrations using images taken in the spring (March to May), summer (June to August), autumn (September to November), and winter (December to February) of 2010 ( Figure 3 ). The spatial distribution of PRE Chl-a concentrations exhibited distinct patterns across the various seasons. Average Chl-a concentrations in spring, summer, autumn, and winter were 5.07 mg/L, 6.15 mg/L, 4.29 mg/L, and 3.11 mg/L, respectively. Standard deviations were 2.59 mg/L, 1.52 mg/L, 2.25 mg/L, and 0.68 mg/L. In spring and autumn, the western and eastern sections exhibited high Chl-a concentrations while the northern and southern regions exhibited lower Chl-a concentrations. Elevated Chl-a concentrations were found near the coastline but gradually decreased the further inland they were. During winter, due to lower water temperatures and slower phytoplankton growth, Chl-a concentrations in the water column was relatively low. Furthermore, the mixing of water masses within the PRE was influenced by strong northeasterly monsoons and tidal currents, resulting in a relatively uniform Chl-a concentration phytoplankton in the water column. In summer, areas with high Chl-a concentrations were mainly within the northern and western sections, which exhibited a decreasing trend from north to south. Additionally, we simulated summer phytoplankton growth using influencing factors (i.e., temperature and rainfall), which resulted in significantly higher Chl-a concentrations and distinct spatial distribution patterns. Accordingly, we focused on investigating the spatiotemporal Chl-a concentration variations in the PRE during summer, accounting for the distinct characteristics and relevance to ecological processes and environmental management in the region.

Additionally, the eastern region had higher Chl-a concentrations during all four seasons, where areas with high concentrations were mainly distributed within the Shenzhen Bay area. This phenomenon is due to enclosed bays that typically have smaller openings or inlets, subsequently resulting in less water exchange compared to the surrounding open sea. This leads to nutrient accumulation via organic and inorganic substances within enclosed bay areas. These substances provide ample nutrients for phytoplankton growth. Moreover, they can contribute to an increase in phytoplankton growth within enclosed bays, resulting in higher Chl-a concentrations.

In this study, remotely sensed images taken in the summer of 1999, 2009, and 2019 were used to analyze interannual variations in spatial distribution characteristics of Chl-a concentrations in the PRE ( Figure 4 ). The overall distribution of Chl-a concentrations in the PRE was uneven and exhibited significant spatial differences. Specifically, in the northern section, Chl-a concentrations in 1999, 2009, and 2019 were 6.48 mg/L, 5.92 mg/L, and 6.75 mg/L, respectively. In the western section, Chl-a concentrations were 5.95 mg/L, 9.48 mg/L, and 6.01 mg/L, respectively. In the eastern section, the Chl-a concentrations were 3.69 mg/L, 4.52 mg/L, and 3.88 mg/L, respectively. In the southern section, Chl-a concentrations were 4.96 mg/L, 6.38 mg/L, and 5.07 mg/L, respectively.

The northern and western sections had higher Chl-a concentrations, which were significantly higher compared to the other two sections. The Chl-a concentration diffusion pattern exhibited a spread from west to east and from north to south. Additionally, interannual variations in Chl-a concentrations differed among the four sections. The northern section exhibited relatively low Chl-a concentration fluctuations, with an approximate range of 0.7 mg/L. The western section exhibited greater fluctuation, where the average Chl-a concentration reached 9.48 mg/L in 2009, surpassing the other section by a large margin. The eastern and southern sections exhibited a similar magnitude of variation (i.e., approximately 1.5 mg/L). Such significant interannual differences can be explained by urbanization near the PRE, including industrial wastewater and domestic sewage. Factors such as water temperature, rainfall, and light conditions during different years can also have influenced phytoplankton growth and eutrophication over the past 30 years.

To obtain a better understanding of temporal water quality variation in the PRE, this study focused on changes in Chl-a concentrations between 1989–2019 ( Figure 5 ). Between 1989–2019, Chl-a concentration on average exhibited significant fluctuations over time, exhibiting a slight overall decreasing trend. Over the past 30 years, the average Chl-a concentration was 5.20 mg/L. The Chl-a concentration decreased from 6.74 mg/L in 1989 to 5.56 mg/L in 2019. The highest value was measured in 1989 (i.e., 6.74 mg/L), while the lowest was measured in 2005 (i.e., 3.66 mg/L). In 2009, a sudden jump in Chl-a concentration occurred, increasing from 4.26 mg/L to 6.73 mg/L. In this study, long-term (i.e., 30 year) Chl-a concentrations variation was divided into three distinct periods. Between 1989–1999, Chl-a concentrations only exhibited small fluctuations (i.e., 5.96 mg/L average). However, compared to the other two periods, this period exhibited significantly higher Chl-a concentrations. Between 2000–2009, Chl-a concentrations exhibited higher fluctuations (i.e., 5.02 mg/L average). On the other hand, there was a notable upward trend between 2004–2009, indicating water quality degradation during this period. Between 2010–2019, the average Chl-a concentration was 4.80 mg/L, which was the lowest among the three periods and exhibited a clear downward trend, indicating a significant improvement in water quality during this period.

In the different sections, Chl-a concentrations differed in spatial and temporal variation characteristics. This study divided the PRE into four sections to analyze each section’s specific Chl-a concentration over time ( Figure 6 ). The western section had the highest Chl-a concentration (i.e., 5.92 mg/L average). Additionally, the Chl-a concentration in the western section fluctuated the most, reaching 9.48 mg/L in 2009, indicating a more stable water quality environment. The Chl-a concentration in the eastern section was the lowest (i.e., 3.98 mg/L average) while more stable than the other sections over the past 30 years. This indicated that the eastern section’s water quality environment was relatively good. This discrepancy was likely caused by the western section’s more intensive industrial activities than the eastern section, resulting in a higher overall wastewater discharge volume. The average Chl-a concentrations in the northern and southern sections were 5.51 mg/L and 4.98 mg/L, respectively. Additionally, changes to Chl-a concentrations in these two sections were similar, both exhibiting a significant downward trend, indicating that water quality conditions have generally improved over the past 30 years.

Considering the geographical proximity of the four sections, several factors may have contributed to variations in Chl-a concentrations and the eutrophication phenomena. Natural factors can influence phytoplankton growth and eutrophication. Generally, high temperatures and rainfall can stimulate phytoplankton growth, resulting in higher concentrations. Conversely, lower water temperatures and rainfall limit phytoplankton growth. These factors can cause significant distribution variation in Chl-a concentrations during different seasons while also significantly impacting interannual variation. Agricultural and industrial activities, as well as densely populated urban areas near the PRE, can cause significant pollutant inputs into water, such as agricultural fertilizers, industrial wastewater, and domestic sewage. Over the past 30 years, the overall water quality of the PRE has improved. This improvement may be attributed to environmental protection policies that have led to a reduction of pollutants from entering its coastal areas, including industrial wastewater and domestic sewage discharge.

This study analyzed data from PRE monitoring stations over a 30-year period, using the average Chl-a concentration as the independent variable and water quality indicators as the dependent variable. Polynomial regression analysis and significance tests indicated a significant relationship between Chl-a and water quality ( Figure 7 ). Additionally, TP and TN highly correlated with Chl-a concentrations. The R² value for TP was >0.6, indicating a positive correlation with Chl-a concentrations (P< 0.05), whose correlation coefficient was 0.64. Similarly, Chl-a concentrations positively correlated with TN (P< 0.05), with a correlation coefficient of 0.71. Although TP and TN are important nutrients, they are often present in excessive amounts due to agricultural runoff and wastewater discharge. These nutrients can stimulate the growth of algae and aquatic plants, leading to increased Chl-a concentrations. Furthermore, Chl-a concentrations positively correlated with NH₃-N (P< 0.05), with a correlation coefficient of 0.69. High levels of NH₃-N can promote the growth of algae and aquatic plants, resulting in elevated Chl-a concentrations. Chl-a concentrations also positively correlated with BOD₅ (P< 0.05), with a correlation coefficient of 0.54. High Chl-a concentrations in water is generally indicative of increased algal growth. More algae will consume more oxygen, resulting in decreased DO levels and elevated BOD₅ in water. Therefore, a high Chl-a concentration is associated with elevated BOD₅ levels in water.

Suitable temperatures promote phytoplankton photosynthesis, resulting in increased phytoplankton biomass (Li et al., 2018). Between 1999–2008, Chl-a concentrations and temperatures in the northern section of the PRE significantly and positively correlated (P< 0.05), with a correlation coefficient of 0.51 ( Figure 8A ). Between 2009–2019, reduced rainfall in the vicinity of the PRE may have resulted in decreased nutrient absorption and a weakened water circulation, subsequently leading to an improvement in water quality. During this period, there was a significant positive correlation (P< 0.05) between Chl-a concentrations and rainfall in the northern section of the PRE, with a correlation coefficient of 0.43 ( Figure 8B ). Results from this study indicated that a reduction in rainfall would result in a decrease in nutrient flux into the PRE, subsequently enhancing its overall water quality. Additionally, Chl-a concentrations in the PRE exhibited a lag effect in response to climate factors (i.e., temperature and rainfall).

Due to oceanic currents, the industrial wastewater discharged from urban areas accumulates in the PRE, leading to changes in water quality (Li et al., 2020). Between 1989–2004, a positive correlation (P< 0.05) was observed between Chl-a concentrations in the northern section of the PRE and the volume of industrial wastewater discharged from coastal cities, with a correlation coefficient of 0.53 ( Figure 9A ). Both variables showed a consistent downward trend on a sweeping scale, while industrial wastewater discharge had a lag effect on Chl-a concentrations. Non-point source pollution from agriculture is widely recognized as one of the main causes of water quality degradation (Evans et al., 2019). A higher proportion of primary industries results in a greater increase in agricultural land area and fertilizer usage in urban areas. Since the beginning of China’s economic reforms, the industrial structure of cities within the GBA has been optimized and adjusted, with a consecutive proportional decrease in primary industries. Between 1989–2019, there was a significant positive correlation (P< 0.05) between Chl-a concentrations in the northern section of the PRE and the proportion of primary industries in its coastal cities, with a correlation coefficient of 0.55 ( Figure 9B ). A general decrease in both variables was observed over the study period. Overall, the trend in PRE Chl-a concentrations is in decline, which may be attributed to a reduction in industrial wastewater discharge and a decrease in the proportion of primary industries in surrounding cities. In this study, PLS regression was constructed to assess how external driving factors influence Chl-a concentrations. The PLS factor weight and the loadings confirmed that all three PLS factors can effectively explain data variance while also demonstrating the good explanatory power of latent variables for observed variables. The contribution of the proportion of the primary industry factor within the three PLS factors was relatively high, indicating its important role in predicting Chl-a concentrations. From Table 1 , it can be observed that temperature, rainfall, and the proportion of primary industries are commonly considered strong driving factors for changes in Chl-a concentrations (VIP > 1).

Regular collection and analysis of water samples while monitoring key water quality indicators (i.e., NH₃-N, TP, and Chl-a concentrations), are conducted. This aided in the establishment of a comprehensive water quality monitoring system, helped in the assessment of water quality conditions, and allowed for the prompt implementation of necessary management measures. Non-point source pollution from industrial and agricultural activities is one of the main causes for water quality degradation, so strengthening wastewater discharge management and improving the capacity and efficiency of wastewater treatment plants are critical. To address the poor water quality of the PRE’s northern and western sections, the Government should strengthen supervision and reduce pollutant emissions. Measures should also be taken to reduce pollution, such as ecological restoration and wetland construction, subsequently enhancing its water purification capacities. Water quality is closely linked to factors such as temperature, rainfall, and other environmental conditions. Attention should also be paid to climate change impacts within the PRE region, while adaptive measures should be taken to address extreme weather events.

Because of its subtropical monsoon climate, the PRE region experiences a high frequency of cloud cover, and data gaps resulting from such frequent cloud cover events and the limited available samples present significant challenges for satellite monitoring. However, amassing field data and using other satellites (i.e., MODIS and Sentinel) to fill satellite-based data gaps may help to resolve this issue. Additionally, the PRE is an optically complex water system, and the potential to overestimate Chl-a concentrations under conditions of water turbidity should not be ignored. Therefore, it may not be appropriate to solely rely on band ratio algorithms to estimate Chl-a concentrations in PRE water. It is necessary to consider other factors, such as seasonality and water characteristics, while establishing more complex retrieval models to analyze water quality changes in the PRE.