Cao 2023

Ecological Indicators 155 (2023) 110914
Contents lists available at ScienceDirect
Ecological Indicators
journal homepage: www.elsevier.com/locate/ecolind
Remote sensing inversion of water quality parameters in the Yellow

River Delta
Xin Cao a, b, c, Jing Zhang a, b, c, *, Haobin Meng a, b, c, Yuequn Lai a, b, c, Mofan Xu c
a
Beijing Laboratory of Water Resources Security, Capital Normal University, Beijing 100048, China
b
The Key Lab of Resource Environment and GIS of Beijing, Capital Normal University, Beijing 100048, China
c
Capital Normal University College of Resources Environment and Tourism, Capital Normal University, Beijing 100048, China
A R T I C L E I N F O A B S T R A C T
Keywords: In recent years, with the rapid socio-economic development of the Yellow River Delta (YRD), the pressure on the
Yellow River Delta supply of water resources has continued to rise. The development of oil-based industries has also led to a series of
Water quality parameters ecological and environmental problems, such as wetland degradation and water quality deterioration. As an
Remote sensing inversion
increasing number of rivers are getting polluted, resulting in the deterioration of their water quality, monitoring,
Sentinel-2 data
One-dimensional regression
managing, and protecting water resources in the YRD is particularly important. In this study, water quality
monitoring data and simultaneous Sentinel-2 image data from April 30, 2020, to October 26, 2021, were used to
construct an experimental sample in the YRD. Water quality parameters (WQPs) concentrations were correlated
with Sentinel-2 image element spectral reflectance and sensitive bands were selected. An empirical method based
on the characteristic bands was used to invert a total of six water quality indicators, namely dissolved oxygen
(DO), permanganate index (CODMn), ammonia nitrogen (NH3-H), total phosphorus (TP), total nitrogen (TN) and
turbidity. The results show: (1) A total of five water quality inversion models for DO, TN, CODMn, TP and TN
were effective in the areas of the Guangli River, the Tiaohe and the Branch River. The inversion accuracies of the
five inversion models (R2of 0.6099, 0.9271, 0.9581, 0.8784 and 0.7387; RMSE of 1.2723, 0.3413, 0.9923,
0.0118 and 1.8476; RPD of 1.53, 2.08, 3.56, 2.76 and 1.53) indicated the feasibility of the water quality
inversion method based on Sentinel-2 data using statistical theory for monitoring water quality concentration in
the YRD. (2) The spatial distribution of water quality in the YRD was generally characterized by high water
quality in the upper reaches and low water quality in the middle and lower reaches (except for some seasonal
variations). Among them, the water quality of the upper reaches of the Guangli River was poor, with opposite
trends in DO and TN concentrations. In the Tiaohe, CODMn and TP concentrations were not strongly correlated.
However, CODMn and TP concentrations were high in the middle reaches where water quality was the worst. The
TN concentrations in the Branch River decreased between 2020 and 2021, but the water quality is still in
Category V. Therefore, continued attention and appropriate water quality management measures in the YRD are
required. Further, by measuring water quality indicators at monitoring stations, regression-fitting equations for
WQPs were established to obtain complementary multi-platform observations. Thus, the water quality conditions
in the YRD region can be evaluated more accurately and quickly. The research results not only provide an
important reference basis for the identification and monitoring of pollution sources, prevention and treatment of
water environment pollution in the YRD, but also provide water security for socio-economic and ecological
environment security.
of the world’s population lives in coastal areas and near lakes and rivers,
and this proportion is increasing. Coastal waters are one of the most
1. Introduction sensitive environments (Chen, 2021; Zhong, 2017; Zhu, 2006). With
rapid socio-economic development, urbanization, and industrialization,
Water is an essential resource for human survival and development the discharge of industrial wastewater and domestic sewage is
and plays a fundamental and strategic role in the sustainable develop increasing. Consequently, water bodies are facing water quality
ment of human society and the ecological environment. More than 40%
* Corresponding author at: Beijing Laboratory of Water Resources Security, Capital Normal University, Beijing 100048, China.
E-mail address: [email protected] (J. Zhang).
https://doi.org/10.1016/j.ecolind.2023.110914
Received 9 May 2023; Received in revised form 23 August 2023; Accepted 3 September 2023
Available online 11 September 2023
1470-160X/© 2023 The Authors. Published by Elsevier Ltd. This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-
nc-nd/4.0/).
X. Cao et al. Ecological Indicators 155 (2023) 110914
Xie, 2006) confirmed that Landsat series data could be used to monitor
Nomenclature some water quality parameters (WQPs) of inland lakes, and the water
quality inversion model established based on this dataset achieved good
Abbreviations and Notations results. Further, Papoutsa et al. (Papoutsa et al., 2014) analyzed the
YRD Yellow River Delta relationship between Asprokremos on-site spectral measurement data,
WQP(s) water quality parameter(s) turbidity measurement data, and Landsat TM/ETM + data, and
DO dissolved oxygen concluded that the red band reflection value in Landsat TM/ETM +
CODMn permanganate index showed a good correlation with turbidity. Wen et al. (Wen et al., 2006;
NH3-H ammonia nitrogen Zhang et al., 2009) obtained better inversion results in a study using
TP total phosphorus hyperspectral data to monitor the content of chlorophyll and suspended
TN total nitrogen solids.
R2 coefficient of determination Currently, three main methods exist for retrieving remote sensing
MAE mean absolute error data on water quality: model analysis, empirical analysis, and semi-
MSE mean squared error empirical analysis. The model analysis method uses remote sensing
RMSE root mean square error reflectance to calculate the ratio of the actual absorption coefficient and
RPD pair analysis error backscattering coefficient constructing an inversion model between
reflectance spectra and WQPs (Wang and Qin, 2023). Warrick et al.
(Chen et al., 2020a; Warrick* et al., 2004) analyzed the spectral prop
erties of WQPs to successfully invert the concentrations of WQPs in
deterioration, eutrophication, ecological damage, and the disappear different study areas. The empirical and semi-empirical methods both
ance of important and sensitive aquatic organisms. Therefore, to achieve combine remotely sensed data and actual detected WQPs to establish
sustainable water supply in cities, long-term monitoring of the water statistical relationships, so that remotely sensed data can be used to
quality conditions of the existing water resources and watersheds is estimate WQPs to monitor water quality conditions (Jiang, 2020).
necessary (Zhu, 2018). As one of the “three major river deltas in China,” Although the empirical method is not suitable for regional applications,
the Yellow River Delta (YRD) is the last oceanic delta to be developed. it remains the main method for monitoring remotely sensed water
Water resources are advantageous for estuarine deltas. However, unlike quality owing to its simple operation (Barrett and Frazier, 2016).
other major river deltas, the conditions of the water resources in the YRD Khattab et al. (Abdelmalik, 2018; Khattab and Merkel, 2014; Olmanson
are not optimistic. Statistics show that the freshwater occupancy per unit et al., 2013; Wang, 2014; Yang et al., 2011; Zhai et al., 2018) obtained
area of the YRD is approximately 66,000 m3/km2, while those of the statistical models (stepwise linear regression, polynomial regression,
Yangtze River Delta, Pearl River Delta, and Shandong Province are principal component analysis) for the WQPs, such as colored dissolved
576,000 m3/km2, 1,163,400 m3/km2, and 460,500 m3/km2, respec organic matter, chemical oxygen demand and pH, based on multi-source
tively. This shows that the water availability in the YRD is poor, whether remote sensing data (Landsat, GF-1, and MODIS) by using the ratio of
in the delta, where the soil and water conditions are well configured, or the wavebands with different band combination. The inversion results in
in the Shandong Province, where water resources are relatively scarce these studies were good, with a goodness of fit of more than 0.66.
(Huang and Zhao, 1996; Li, 2020; Yang et al., 2006). The YRD is not only Currently, many researchers combine empirical and semi-empirical
a famous oil and gas resource-producing area in China, but also has the methods for practical applications. For example, Ma et al. (Ma and Li,
largest warm-temperate wetland ecosystem in China. It has high eco 2002) combined field water body wave spectrum measurement data
nomic, social, academic and natural values. The water environment is an with satellite remote sensing data (CBERS-1 and TM) to monitor water
important part of the ecosystem. It is of great significance to do a good body pollution in the Daliao River Estuary and quickly obtain water
job of water environment management and maintain a good water body pollution information.
environment to promote the ecological protection of the YRD. In the With the development of artificial intelligence, machine learning
case of water shortage, water quality determines the social security and methods are gradually applied to water quality remote sensing inver
health of humans. The deterioration of the aquatic environment en sion. Machine learning can simulate complex relationships with better
dangers human safety and biodiversity. Therefore, strengthening water applicability to the complex nonlinear fitting process of remote sensing
quality monitoring, protecting the aquatic environment, and improving water quality inversion. It also reduces human interference, which is
the ability to monitor water quality quickly and dynamically have adaptive, fault-tolerant and self-organizing (Du, 2021; Huang, 2020;
become top priorities. Jiang, 2020; Wang and Qin, 2023). Adam (Piotrowski et al., 2014) used
Water quality monitoring is important for evaluating and managing the Levenberg-Marquardt algorithm to invert the water temperature of
water bodies. Water governance relies on accurate and timely water the Biala Tarnowska River with a goodness of fit of up to 0.9. Lu et al. (Lu
quality evaluation, which is closely associated with the monitoring of et al., 2016) established a chlorophyll inversion model for Lake Cham
water quality indicators (Chen, 2021; Zhao, 2021). Remote sensing plain with a coefficient of determination as high as 0.8 using a BP neural
technology has become a new means of monitoring water quality to network. Zhu (Zhu, 2018) used Landsat-8 data to construct the inversion
compensate for the shortcomings of traditional water quality moni model of chlorophyll a and TP based on a support vector machine. The
toring, owing to its low cost, wide monitoring range, real-time dynamic root-mean-square error was controlled at about 0.2.
monitoring, and other advantages. This effectively overcomes the limi Accordingly, this study considered the YRD region as the study area
tations of data collection in conventional water quality monitoring and studied the following aspects: (1) Statistical and correlation ana
(Zhao, 2021). The universal application of remote sensing technology lyses of water quality monitoring data were conducted to explore the
relies primarily on the development of satellite technology. With the relationship between WQPs in the YRD region. (2) A suitable inversion
gradual development of sensor and imaging technologies, remote model for DO, CODMn, NH3-N, TP, TN, and turbidity in the YRD region
sensing data sources have become increasingly abundant. Improvements was constructed using linear regression by selecting band combinations
in spatial, temporal, and spectral resolutions have promoted the appli with strong correlations. (3) By comparing the sizes of the validation
cation of remote sensing data in many fields. Currently, multispectral indicators (coefficient of determination, R2; mean absolute error, MAE;
and hyperspectral remote sensing data are the most widely used data mean squared error, MSE; root mean square error, RMSE; and pair
sources for monitoring water quality. However, multispectral data are analysis error, RPD), a new model and method for water quality moni
more readily available and have a wider range of applications (Du, 2021; toring in the YRD region were developed. (4) Based on the inversion
Zhao, 2021). For example, Xie et al. (Feng et al., 2017; He et al., 2011; results, the spatiotemporal distribution patterns of DO, CODMn, TP, and
2
TN concentrations were analyzed to understand the status and change 2. Materials and methods
characteristics of water quality in the YRD region. The results can sup
plement the information on water quality inversion in the YRD. We hope 2.1. Research area
that this study can provide a reference for further research on the water
quality of the YRD region. The rapid and comprehensive understanding The YRD (117◦ 31́ –119◦ 18́ E; 36◦ 55́–38◦ 10́ N) is an alluvial plain
of the water quality condition will help to clarify the regional response formed by sediments carried by the Yellow River estuary in the Bohai
of good and bad water quality to the YRD. According to the findings, it Depression. Kenli Ninghai is located at the apex, starting from the mouth
will provide reference content for the control and improvement of water of the Taoer River in the north and ending at the mouth of the branch
quality conditions and aquatic ecology maintenance in the YRD. This channel in the south. It covers an area of approximately 5,400 km2, of
study is of great significance in monitoring and early warning of water which approximately 96% is located in Dongying City, Shandong
quality and nutrient status, protecting water quality of water sources Province. The landform type is mainly an inclined plane, and its altitude
and guaranteeing water safety of people along the river. is mostly below 50 m. Damages caused by land subsidence and seawater
The starting point of this study mainly has the following 2 aspects. intrusion are relatively strong in this region. The Yellow River flows into
(1) It is known through the research of many scholars at home and the Bohai Sea in Dongying City, China. The Chaohe, Tiaohe, Shenxian
abroad that it is possible to realize the inversion of the content of non- gou, and Diaokou Rivers are among the 30 main drainage channels in
photosensitive WQPs, such as DO, CODMn, TP, TN, etc., by using the city, except the Yellow River. Owing to the requirement for eco
remote sensing data. In this study, new band combinations are found to nomic development, and the development of oil and gas exploitation
invert the concentration inversion of WQPs in inland water bodies, and and refining industries, local industries have formed an industrial
the modeling effect is good. The rivers of the YRD are long and narrow. It complex, which is dominated by the petrochemical, textile, paper, and
enriches the modeling of remote sensing water quality monitoring in electromechanical industries (Li, 2015; Zhang, 2018). The resulting
narrow inland water bodies, which is conducive to the study of sewage discharge from these industries has polluted the rivers in the
improving the generalizability of remote sensing inversion models. (2) YRD to varying degrees. Consequently, water pollution of these rivers
There are fewer monitoring sections in the YRD region. The studies on has become increasingly serious. Combined with existing research on
remote sensing monitoring of WQPs in this area are almost non-existent. the water quality of the YRD and the water quality measurements of
As far as the feasibility of remote sensing monitoring of water quality in each monitoring station (Fig. 1), the pollution of the YRD is concluded to
the YRD region is concerned, the results of the study show that it is be mainly organic pollution, wherein the chemical oxygen demand,
feasible to invert the water quality concentration in the YRD based on CODMn, and NH3-N exceed the standard (Zhang et al., 2017).
Sentinel-2 data. In addition, the results show that the method of remote
sensing coupled with mathematical modeling has an important appli
cation value for water quality simulation in areas with few monitoring 2.2. Data and processing
stations, insufficient water quality data, high sampling and chemical
analysis costs. It has room for further development and application. The 2.2.1. Water quality monitoring data
article has the unique features of the above two points, which can bring The surface water quality monitoring data of the YRD were obtained
new ideas to readers for water quality research in narrow shapes and from the National Automatic Integrated Water Quality Supervision
areas with insufficient water quality data. Platform (https://szzdjc.cnemc.cn:8070/GJZ/Business/Publish/Main.
html, accessed on December 2, 2021), which was released by the
Fig. 1. Location map of the water system and water quality monitoring stations in the YRD.
3
Chinese Ministry of Environmental Protection on July 1, 2009, and Table 2

made available to the public. The WPQs measurement methods are Information on the monitoring sections.
described comprehensively in the “Technical Specification for Auto Monitoring section Longitude Latitude River Watershed
matic Surface Water Monitoring” (HJ 915–2017) issued by the Ministry
Pile 5 118.88 37.97 Shenxiangou Sea River Basin
of Ecology and Environment of the People’s Republic of China (https:// Dongba Road 118.76 37.41 Guangli Huaihe River
www.mee.gov.cn/ywgz/fgbz/bz/bzwl/jcffbz/201801/t20180108_429 Bridge River Basin
283.shtml, accessed on December 2, 2021). The monitoring frequency Bingu Road Bridge 118.60 37.90 Tiaohe Sea River Basin
was 4 h, with dynamic releases six times a day (0:00, 4:00, 8:00, 12:00, Branch River 118.39 37.26 Branch River Huaihe River
Chenqiao Basin
16:00, and 20:00). The monitoring indicators of this system mainly were
water temperature (◦ C), pH, DO (mg/L), conductivity (µS/cm), turbidity
(nephelometric turbidity unit, NTU), CODMn (mg/L), NH3-N (mg/L), TN correlation between each band or band combination of the Sentinel-2
(mg/L), TP (mg/L), chlorophyll a concentration (mg/L), and algal image and the WQPs content. The band or band combination with the
density (cells/L). Four monitoring sections, namely, Pile 5, Dongba Road best degree of correlation was selected to establish a single-band
Bridge, Bingu Road Bridge, and Branch River Chenqiao (Table 2), were regression model or band combination regression model. Finally, the
selected to use the dissolved DO, CODMn, NH3-N, TN, TP, and turbidity spatial distribution of WQPs in the YRD region was analyzed according
data from April 30, 2020, to October 26, 2021. The acquisition time for to the inversion results.
the Sentinel-2 satellite images was approximately 10:30 local time(Local
Time of Descending Intersection). Therefore, the water quality moni 2.3.1. Correlation analysis
toring data at 8:00 and 12:00 were selected for averaging. The remote Correlation analysis measures the degree of association between two
sensing image pixels where the water quality monitoring station was variables. To demonstrate the correlation between the surface reflec
located were selected for water quality data matching (Meng et al., tance and measured water quality concentration data, a band or a
2022). combination of bands with high correlation was selected to construct the
inversion model. Pearson’s correlation coefficient was employed in this
2.2.2. Remote sensing image data study, which is the quotient of the covariance and standard deviation
In this study, Sentinel-2 was selected as the data source mainly between two variables, and can measure the degree of correlation be
because of the limited number of monitored cross-sections. Moreover, tween these variables (Zhao, 2021). The calculation formula is shown in
compared to Landsat data, Sentinel-2 data are more selective (Table 3). Equation (1). The correlation coefficient value ranges from − 1 to + 1.
Sentinel-2 is a high-resolution optical satellite launched by the European The closer the absolute value to 1, the greater the correlation between
Space Agency (ESA) and contains two satellites (2A and 2B) in orbit. The the two variables. Moreover, the strength of the correlation is divided
satellite carries a push-scan MSI covering 13 spectral bands with an into five main levels according to the absolute value of magnitude
amplitude of 290 km and a revisit period of 10 d per satellite. The sat (Table 5).
ellite imaging resolution varies from visible to near-infrared to short ∑n
wave infrared with maximum and minimum resolutions of 10 and 60 m, i=1 (xi − x)(yi − y)
rxy = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ (1)
respectively. The Sentinel-2 satellite can acquire high-resolution images ∑ ∑
n n
(xi − x)2 (yi − y)2
of land surfaces, islands, and water. It is one of the few optical datasets i=1 i=1
with three bands in the red-edge range, which has obvious advantages
for monitoring the health of forest vegetation and the pollution status of where rxy is the correlation coefficient of the surface reflectance and
water (Table 4). Furthermore, Sentinel-2 data have high spatial resolu water quality concentration parameters; xi and yi are the sample values
tion and rich spectral information and can reflect water quality changes, of the surface reflectance and water quality concentration parameters,
which is advantageous for water quality inversion. The data are in the respectively; x and y are the average values of the samples of the two
⃒ ⃒
near-infrared band range (700–800 nm) and the acquired spectral in variables, respectively. If ⃒rxy ⃒ tends to be closer to 1, the correlation is
formation is abundant. The unique red-edge narrow band is in the stronger.
spectrally sensitive band range of water bodies and can fully reflect the
biological optical properties of water bodies under different environ 2.3.2. One-dimensional regression analysis
mental conditions. The L2A level product of Sentinel-2 was used after Univariate regression analysis involved only two variables. The main
preprocessing, which included orthorectification, geometric accuracy objective was to estimate the other variable from one of the two related
correction, radiometric correction, and atmospheric correction. More variables. The estimated variable called the dependent variable is
over, the Sentinel-2 data are open-source data available from the ESA labelled as Y. The estimated variable called the independent variable is
Copernicus Data Centre (accessed on March 2, 2022). labeled as X. Regression analysis is used to find a mathematical model Y
= F(X) to compute Y from X using a functional equation. Common
models include linear, exponential, logarithmic, inverse-power, and
2.3. Methodology polynomial functions, and the respective formulae are given in Equa
tions (2)–(7).
The process flow of this study is illustrated in Fig. 2. Initially, this
Linear : y = a + bx (2)
study used the single single-pixel extraction method at the localization
point to obtain the image values at the monitoring section. Subse
Exponential : y = a × ebx (3)
quently, Pearson’s correlation analysis was used to determine the
Logarithmic : y = a + blnx (4)
Table 1
Pollution assessment of the main rivers. Power : y = a × xb (5)
River name Mean comprehensive pollution index Pollution level
Shenxiangou 1.67 Heavy Polynomial : y = a + b1 x + b2 x2 (6)

Tiaohe 1.31 Heavy
Overflow River 1.10 Heavy Polynomial : y = a + b1 x + b2 x2 + b3 x3 (7)
Guangli River 0.95 Moderate
Branch River 0.86 Moderate where bi is the slope of the model and a is a parameter to be determined.
4
Table 3
Number of inversion images from May 2020 to October 2021 (“/”: too much cloud coverage or lack of images).
Sensor Number of image coverage dates in the YRD
from May 2020 to October 2021
2020 2021
5 6 7 8 9 10 11 12 1 2 3 4 5 6 7 8 9 10
Landsat 8 1 / / / / / 1 2 1 / 1 1 1 2 / 2 1 1
Sentinel-2 4 3 / / / / 3 9 4 9 5 9 10 3 3 / 7 5
∑n
Table 4 ( ŷi − yi )2
Sentinel-2-specific band information. R2 = 1 − ∑i=1
n 2
(8)
i=1 (yi − y)
Introduction of bands Bands/μm Spatial resolution/m
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
Coastal aerosol (B1) 0.443 60
√∑
√n
Blue (B2) 0.49 10 √ (yi − ŷi )2
√
Green (B3) 0.56 10 RMSE = i=1 (9)
Red (B4) 0.665 10
n
Vegetation red edge (B5) 0.705 20
∑n
Vegetation red edge (B6) 0.74 20 i=1 (yi − ŷi )2
Vegetation red edge (B7) 0.783 20 MSE = (10)
n
NIR (B8) 0.842 10
Narrow NIR (B8A) 0.865 20
Water vapor (B9) 0.945 60
SWIR-Cirrus (B10) 1.375 60
SWIR (B11) 1.61 20 Table 5
SWIR (B12) 2.19 20 Relevance level classification.
Correlation coefficient Level
2.3.3. Model testing method 0.8 ≤ r ≤ 1.0 Maximum correlation

Verifying the accuracy of the WQPs inversion models is necessary. 0.6 ≤ r < 0.8 Strong correlation
0.4 ≤ r < 0.6 Moderate correlation
Model accuracy measures the comprehensive estimation ability of a
0.2 ≤ r < 0.4 Weak correlation
model and evaluates the fitting effect between the actual measured 0 ≤ r < 0.2 Weakest correlation
values and the predicted values. The better the accuracy of the training
set, the better the stability of the model. The higher the accuracy of the
test set, the stronger the predictive ability of the model (Huang and Table 6
Zhao, 1996; Li, 2020; Yang et al., 2006). In this study, the model ac Explanatory notes for RPD.
curacy evaluation indices were R2, MAE, MSE, RMSE, and RPD. The RPD
Value Description
is the ratio of the sample standard deviation to the RMSE, which de
termines the predictive ability of the model (Table 6). These indices can RPD < 1.4 The model is unable to predict the sample.
1.4 ≤ RPD < The model is generally effective to estimate the sample content
be calculated as follows: 2 roughly.
RPD ≥ 2 The model has excellent quantitative prediction ability.
Fig. 2. Overall research framework.
5
∑n
i=1 |yi − ŷi | analysis of the standard deviation and coefficient of variance of the
MAE = (11)
n Shenxiangou, Tiaohe, Guangli, and Branch rivers showed that the dif
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ ference in water quality between Pile 5 and Dongba Road Bridge was
√̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅
1
∑ n
2
√∑
√ n (y − y)2 small, and the difference between Bingu Road Bridge and Branch River
(y − y) √
SD n i
√ i=1
i
Chenqiao monitoring was large.
(12)
i=1
RPD = = √̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅̅ = √∑ n
RMSE ∑ n √
1
( ŷi − yi )2 ( ŷi − yi )2
n
i=1 i=1 3.1.2. Correlation analysis
To investigate the relationship between the WQPs in the YRD, we
where yi and ŷi are the measured and predicted values of the WQPs calculated the correlation coefficient among the water quality moni
content, respectively; y is the average of the measured values of the toring parameters. Table 9 shows that the correlation coefficients of
WQPs content; and n is the number of samples. NH3-N with TN and turbidity were higher than those of the other WQPs
(r = 0.765 and 0.705, respectively). NH3-N was significantly and posi
3. Results tively correlated with TN and turbidity, while CODMn was not signifi
cantly correlated with NH3-N, TN, and turbidity. The other WQPs
3.1. Analysis of the water quality monitoring data showed weak correlations with each other. The correlations between
TN, TP, and the other WQPs differed. Based on the correlation strength,
3.1.1. Statistical analysis the water quality in the YRD region was speculated to be affected by
The WQPs data used in this study was released by the National various parameters, causing it to change spatially.
Surface Water Quality Automatic Monitoring System. Data from four
monitoring sections located in the YRD region were selected for col 3.2. Sensitive band selection
lecting research data. The acquired WQPs data was statistically
analyzed. A series of statistical parameters were calculated for the To demonstrate the relationship between the concentration of WQPs
monitoring data, including the maximum value, minimum value, mean and the band reflectance, the six WQPs were considered in the image of
value, standard deviation, and coefficient of variation. Statistical data nine band ranges to perform correlation analysis. The r value was
from the WQPs were compared with the standard limits of the basic calculated for each waveband or a combination of bands with the WQPs
parameters of the surface water environmental quality (Tables 7 and 8). content, and the sensitive band combination was selected according to
We found that the overall concentration of WQPs in Pile 5 and the the correlation results.
Dongba Road Bridge was lower than that in the other two sections, and To reduce the atmospheric reflectance of the image bands, the pro
the overall water quality was within Class III. Generally speaking, the cess of performing a ratio operation between the spectral reflectance of
water quality of the Branch River was the poorest. Water quality mostly
belonged to Classes III and IV of the water quality categories. Occa
Table 9
sionally, the water quality category reached Class V. A comparison and Correlation coefficients between WQPs.
DO CODMn NH3-N TP TN Turbidity
Table 7 DO 1 0.562** 0.289* 0.327** 0.407** 0.355*
Surface water environmental quality standards Unit: mg/L. CODMn 0.562** 1 0.106 0.475** 0.051 0.154
NH3-N 0.289* 0.106 1 0.248* 0.765** 0.705**
No. Parameter Class I Class II Class III Class IV Class V
TP 0.327** 0.475** 0.248* 1 0.483** 0.414**
1 DO ≥ 7.5 6 5 3 2 TN 0.407** 0.051 0.765** 0.483** 1 0.702**
2 CODMn ≤ 2 4 6 10 15 Turbidity 0.355** 0.154 0.705** 0.414** 0.702** 1
3 NH3-N ≤ 0.15 0.5 1.0 1.5 2.0
**: At the 0.01 level (two-tailed), the correlation is significant.
4 TP ≤ 0.02 0.1 0.2 0.3 0.4
5 TN ≤ 0.2 0.5 1.0 1.5 2.0 *: At the 0.05 level (two-tailed), the correlation is significant.
Table 8
Statistical parameters of WQPs of monitoring cross-sections.
Monitoring Statistical DO CODMn NH3-N TP TN Turbidity
section parameters (mg/L) (mg/L) (mg/L) (mg/L) (mg/L) (NTU)
Pile 5 Min 5.59 3.51 0.03 0.03 0.83 9.96

Max 13.07 12.56 0.53 0.12 3.02 35.80
Mean 9.70 7.58 0.21 0.06 2.16 21.28
SD 2.55 2.07 0.14 0.02 0.65 6.00
CV 0.26 0.27 0.68 0.41 0.30 0.28
Dongba Road Min 5.17 3.51 0.03 0.06 2.19 9.86
Bridge Max 11.06 8.79 0.51 0.11 4.40 34.18
Mean 8.65 6.19 0.22 0.08 3.27 20.74
SD 1.74 1.54 0.15 0.02 0.74 8.63
CV 0.20 0.25 0.67 0.22 0.23 0.42
Bingu Road Min 7.82 4.05 0.03 0.03 1.41 10.02
Bridge Max 22.19 20.01 0.23 0.20 3.77 45.55
Mean 13.59 9.53 0.07 0.09 2.77 27.04
SD 3.79 4.93 0.06 0.05 0.62 10.82
CV 0.28 0.52 0.87 0.50 0.23 0.40
Branch river Min 5.66 4.81 0.06 0.07 3.51 14.90
Chenqiao Max 19.32 10.81 3.39 0.24 13.65 129.33
Mean 13.68 8.18 0.97 0.13 9.31 58.36
SD 3.72 1.28 0.97 0.04 3.11 32.85
CV 0.27 0.16 1.00 0.35 0.33 0.56
Min, minimum value; Max, maximum value; Mean, mean value;
SD, standard deviation; CV, coefficient of variance
6
Table 10
Results of sensitive-band selection.
Monitor cross-section WQPs Band r R2 Band r R2
combination combination
Pile 5 TP B6/(B7-B11-B12) − 0.7594 0.7114 B4/(B5-B11-B3) − 0.7420 0.6489

Dongba Road Bridge DO (B3-B5-B11)/B12 − 0.7872 0.6212 B3/(B5-B11-B12) 0.7455 0.6475
(B5 + B11)/(B3 + B12) 0.7428 0.6512
CODMn (B5 + B12)/(B3 + B11) 0.7102 0.6260
TN (B11-B12)/B5 0.8146 0.7201 (B11-B12)/(B5 + B8A) 0.8128 0.7519
(B11-B12)/(B4 + B5) 0.7901 0.6737
Bingu Road DO (B5 + B11)/B4 0.7978 0.6728 (B5 + B11)/(B4 + B12) 0.7962 0.6927
Bridge (B4-B11)/(B5-B12) − 0.7925 0.6776
CODMn (B6 + B7)/(B4-B12) 0.9642 0.9296 (B7 + B8A)/(B4-B12) 0.9638 0.9288
B4/(B12-B7) 0.9629 0.9272 (B6 + B8A)/(B4-B12) 0.9623 0.9261
NH3-N B6/(B11 + B12) 0.8466 0.7612 B12/B11 0.8082 0.6620
(B6 + B11)/B12 − 0.7902 0.7394
TP B3/(B4 + B12) 0.8354 0.7774 B2/(B3 + B4 + B12) 0.8337 0.7715
B12/B4 0.8336 0.7476
TN (B6-B7-B8A)/B12 0.7274 0.6026
Turbidity (B4 + B7)/(B2 + B11) 0.7875 0.6521 (B2 + B11)/(B4 + B7) − 0.7631 0.6521
Branch River NH3-N B4/(B5-B12) − 0.7995 0.703 (B11 + B12)/(B4-B5) − 0.7952 0.6738
Chenqiao TN (B3 + B8A)/(B2 + B12) − 0.7888 0.6665 (B5 + B8A)/(B2 + B12) − 0.7712 0.6242
Turbidity B2/(B3-B7-B4) − 0.7580 0.6953
r, a correlation between waveband combination and water quality concentration;

R2, decision coefficients of bringing all data into the constructed simple model
the bands was added. Many experimental studies have shown that the In this case, the sample sizes of Pile 5 and Branch River Chenqiao are the
combination of two or more different bands can reduce the interference same, 15 data in the training set and 7 data in the test set, respectively.
of external factors and enhance the effective band information. The nine The amount of data in the training set for the Dongba Road Bridge and
bands of the Sentinel-2 images were denoted as B1, B2, B3, B4, B5, B6, Bingu Road Bridge is divided into 10 and 11, respectively. And their test
B7, B8, and B9 from low to high, which were convenient for calculations set data amount (5) is the same.
in subsequent experiments. Referring to the existing index forms (NDVI, In this study, water quality monitoring data were divided into
normalized difference vegetation index; NDWI, normalized difference training and validation sets. An inverse model was constructed using the
water index), the band combination form of this study refers to the training set data, and sensitive bands were used as input factors. Because
principle of the combination of three bands, namely, difference index the sensitive band may not be unique, multiple regression models can be
(DI), ratio index (RI) and normalized index (NDI). The principle of the constructed for the sensitive band. Therefore, the accuracy of the model
three indices is shown in Equations (13)-(15). We constructed the must be validated based on the test set data. The most suitable regression
combination forms of two-band, three-band and four-band to investigate model was selected by comparing accuracy metrics (R2, MAE, MSE,
the mapping relationship between them and WQPs. By calculating the RMSE, and RPD).
band combination and the band itself, 6237 band parameters were ob
tained. Finally, the r-value was set to be greater than or equal to 0.7 at 3.3.1. Regression of water quality concentration
the 95% confidence level (P < 0.05). All measured concentration data Because several regression models were initially constructed using
were substituted into the simple model. The bands and combinations wave combinations, the Shenxiangou, Guangli, and Branch Rivers were
with R2 greater than or equal to 0.6 were used as the sensitive bands to used as examples for analysis. Table 11-Table 13 shows the combina
participate in the next model construction. The results are shown in tions of bands with a good fit of their regression models. The WQP
Table 10. concentrations estimated using the inverse model were then compared
( ) with the monitoring data to derive the R2 values of the model. On the
DI Bi , Bj = Bi − Bj (13)
one hand, a comprehensive analysis of all regression models can lead to
( ) Bi some conclusions. First, the R2 values were relatively high, which re
RI Bi , Bj = (14) flected the feasibility of inverting WQP concentrations using the estab
Bj
lished regression model. Second, the Sentinel-2 data for red (B4),
( ) Bi − Bj vegetation red edge (B5), shortwave infrared SWIR (B11), and SWIR
NDI Bi , Bj = (15) (B12) dominated the inversion of the six WQPs. Different WQPs may
Bi + Bj
exhibit differences. On the other hand, since the four monitoring sec
where i = 1,2,3,4,5,6,7,8,9; j = 1,2,3,4,5,6,7,8,9; Bi and Bj denote the tions were located in different rivers, building individual inversion
band values of Sentinel 2 remote sensing images. models based on each section was appropriate. Table 11 summarizes the
regression models for the inversion of WQPs in the Shenxiangou area. It
3.3. Regression of WQPs based on Sentinel-2 data

Table 11
To obtain a reliable and stable model, selecting samples to establish Regression model for the inversion of water quality concentration using the
training and test sets is necessary before conducting WQPs model training set data - Pile 5.
inversion. In this study, the “cross-validation method” is used. The WQPs No. (x = ) Band Regression model equation R2
purpose of cross-validation is to obtain a reliable and stable model, combination for WQPs estimation
which makes the results convincing. Initially, the concentration of WQPs TP 1 B6/ y = − 0.0902 − 0.1632 × x 0.7098
was ranked from high to low. Subsequently, three subsamples were 2 (B7-B11-B12) y = − 0.1239 − 0.2318 × 0.7104
selected at equal intervals. The first two sample data points to be placed x − 0.0341 × x2
3 y = − 0.6680 − 1.8757 × 0.7127
in the training set were selected, and the third data point was placed in
x − 1.6669 × x2 − 0.5328 × x3
the test set. Finally, all the samples were trained and tested in a 2:1 ratio.
7
Table 12 fitting effect of NH3-H and TN was more satisfactory for the inversion of
Regression model for the inversion of water quality concentration using the the six WQPs in the region, in which the R2 value was approximately 0.7.
training set data – Branch River Chenqiao. Table 13 presents the regression model for the inversion of WQPs in the
WQPs No. (x = ) Band combination Regression model equation R2 Guangli River region. It can be concluded that the inversion effect of DO
for WQPs estimation and TN concentrations on the water quality of the area was more
NH3- 1 B4/(B5-B12) y = 0.0585 × e− 0.8332x 0.7390 prominent. A better inversion effect for DO and was is a quadratic
N 2 (B11 + B12)/(B4-B5) y = 0.0569 × e− 0.3581x 0.7133 polynomial and a cubic polynomial with a fit degree of greater than 0.7.
TN 1 (B3 + B8A)/(B2 + B12) y = 128.6065 × e− 1.8090x 0.6575 Because of the large number of regression models in the Tiaohe re
gion, listing all the models individually was difficult. However, this can
be analyzed based on the modeling results. The WQP concentration
can be concluded that the inversion effect of the TP concentration in this
fitting effect was observed to be better than that of the other three areas.
area was better. The best fit for TP was a cubic polynomial, with an
The R2 value of the regression models established for the six WQPs
estimated R2 value of 0.7127. Table 12 shows the regression model with
exceeded 0.7. Moreover, the estimated R2 value of the model was greater
a good fit for water quality inversion in the Branch River area. The
than 0.8, whereas that of some water quality models reached 0.9. In
Table 13
Regression model for the inversion of water quality concentration using the training set data – Dongba Road Bridge.
WQPs No. (x = ) Band combination Regression model equation R2
for WQPs estimation
DO 1 (B3-B5-B11) y = 7.1181 + 1.7312 × x + 2.7453 × x2 0.7124

2 /B12 y = 22.2279 + 46.8338 × x + 45.3098 × x2 + 12.8509 × x3 0.7280
3 B3/ y = 20.1471 + 19.0624 × x + 6.9854 × x2 0.7061
4 (B5-B11-B12) y = 37.4489 + 71.5879 × x + 57.9273 × x2 + 15.7751 × x3 0.7302
5 (B5 + B11)/ y = 379.62 − 1121.03 × x + 1109.84 × x2 − 360.24 × x3 0.6051
(B3 + B12)
TN 1 (B11-B12)/B5 y = 2.47 − 8.88 × x + 233.22 × x2 − 755.98 × x3 0.6634
2 (B11-B12)/ y = 2.31 − 4.72 × x + 372.37 × x2 − 1795.28 × x3 0.7074
(B5 + B8A)
3 (B11-B12)/ y = 2.70 − 33.43 × x + 1077.84 × x2 − 5961.27 × x3 0.7288
(B4 + B5)
Table 14
Comparison of the evaluation indexes for the regression of WQPs.
Monitoring section WQPs No. R2 RMSE MSE MAE RPD
a b c
Dongba Road Bridge DO 2 0.7280 0.6099 0.6733 1.2723 1.6186 1.1920 1.5346
5 0.6051 0.7597 0.6457 1.2063 1.4553 0.9636 1.6185
TN 1 0.6634 0.8879 0.7188 0.3387 0.1147 0.2957 2.0952
2 0.7074 0.9271 0.7450 0.3413 0.1165 0.3198 2.0796
3 0.7288 0.8182 0.7465 0.3630 0.1318 0.3283 1.9551
Bingu Road Bridge CODMn 4 0.8131 0.9406 0.8304 1.0599 1.1234 0.8597 3.0961
7 0.9326 0.9225 0.9288 1.0564 1.1159 0.9847 3.1065
19 0.9248 0.9581 0.9261 0.9223 0.8507 0.8931 3.5580
TP 3 0.7176 0.8792 0.7590 0.0118 0.0001 0.0092 2.7737
6 0.7341 0.8784 0.7715 0.0118 0.0001 0.0088 2.7612
11 0.7240 0.7971 0.7391 0.0173 0.0003 0.0121 1.8891
Branch River Chenqiao TN 1 0.6575 0.7387 0.6665 1.8476 3.4137 1.5652 1.5321
a: Training set accuracy; b: verified set precision; c: precision of all datasets; Description of Error Indicator Units: RMSE: mg/L, MSE: (mg/L)⋅2, MAE: mg/L, RPD:
dimensionless
Table 15
Optimal inversion model of WQPs in the YRD region.
Monitoring WQPs (x = ) Band combination Regression model equation for WQP estimation R2 Error
section indicator
Dongba Road Bridge DO (B3-B5-B11) y = 22.2279 + 46.8338 × x+ a 0.7280 MSE 1.6186

/B12 45.3098 × x2 + 12.8509 × x3 b 0.6099 MAE 1.1920
c 0.6733 RPD 1.5346
TN (B11-B12)/ y = 2.31 − 4.72 × x+ a 0.7074 MSE 0.1165
(B5 + B8A) 372.37 × x2 − 1795.28 × x3 b 0.9271 MAE 0.3198
c 0.7450 RPD 2.0796
Bingu Road Bridge CODMn (B6 + B8A)/ y = 1.8506 + 2.5960 × x a 0.9248 MSE 0.8507
(B4-B12) b 0.9581 MAE 0.8931
c 0.9261 RPD 3.5580
TP B2/ y = 0.0174 × e11.6027x a 0.7341 MSE 0.0001
(B3 + B4 + B12) b 0.8784 MAE 0.0088
c 0.7715 RPD 2.7612
Branch River Chenqiao TN (B3 + B8A)/ y = 128.6065 × e− 1.8090x a 0.6575 MSE 3.4137
(B2 + B12) b 0.7387 MAE 1.5652
c 0.6665 RPD 1.5321
Description of Error Indicator Units: MSE: (mg/L)⋅2, MAE: mg/L, RPD: dimensionless
8
Fig. 3. Best-fit modeling (a) and Model accuracy (b).
addition, in addition to the polynomial regression models, logarithmic, compared and evaluated for accuracy. Moreover, the evaluation in
exponential, and power function models were used. This increased the dicators (MAE, MSE, RMSE and RPD) were combined to comprehen
model richness for the inversion of WQPs in the YRD region. sively determine the best water quality inversion model. The regression
model validation results are shown in Table 14.
3.3.2. Validation of the water quality inversion model Table 14 shows that, among a large number of regression equations
As discussed in the previous section, there were many regression described previously, the number of equations validated with various
equations for retrieving water quality concentrations. These were accuracies was greatly reduced. Only the inversion of the CODMn
9
Fig. 3. (continued).
Fig. 4. Spatial distribution of DO concentrations based on year.
concentration in the Tiaohe waters was the best, with a model accuracy distribution of the WQPs was analyzed for annual and monthly
of 0.9 or more in the test set. With an RPD value exceeding 3, the pre variations.
dictive ability of the model was extremely good. Several other param
eters also showed model accuracies between 0.7 and 0.8, and the model 4.1.1. Distribution of DO concentrations
fitting effect was also satisfactory. Finally, the best inversion model for
WQPs in the YRD region was determined through comparative (1) Annual variations
screening. The corresponding results are shown in Table 15. Fig. 3 shows
the best inversion results for estimating the DO, CODMn, TP, and TN Fig. 4 shows the distribution of DO concentrations in the Guangli
concentrations in the YRD using Sentinel-2 data. In Fig. 3 (a) is the best River from 2020 to 2021. Panels (a) to (b) show that the color change in
regression model based on the training set data, and (b) is the model the concentration classification was not evident, indicating that the DO
accuracy based on the test set data. From Fig. 3, we can conclude that the concentration in the Guangli River changed marginally from 2020 to
P-value of all models is less than 0.05, indicating that the regression 2021. The average DO concentrations were 18.91 mg/L in 2020 and
equations of the five models are significant. In addition, the P-value of 16.48 mg/L in 2021. The concentration decreased, as shown by the
the models for the monitoring cross-section of the Bingu Road Bridge decrease in the midstream concentration from 2020 to 2021. The
and Branch River Chenqiao is less than 0.001, which shows that the evident color change in each graph indicated that the overall DO con
fitting results in this area are extremely statistically significant. centration in the Guangli River was not uniformly distributed, with a
large difference between the concentrations in the upper and middle
4. Discussion reaches. The upstream and midstream colors were lighter, with con
centrations mostly in the range of 7.5–10.5 mg/L. Moreover, the con
4.1. Analysis of the spatial distribution of WQPs centrations of the downstream DO were higher (mostly above 20 mg/L).
Although the limitation of the regression equation affected the retrieval
Sentinel-2 images were retrieved using the five water quality con results, the distribution and variation in the DO concentrations could
centration regression models described in Section 3.3. Appropriate time still be observed in the figure. Most of the channel of the Guangli River is
points for each parameter were selected to invert the water quality located in the Dongying District Economic and Technological Develop
concentrations at these time points. Based on the results, the spatial ment Zone. The zone focuses on the development of petroleum
10
Fig. 5. Spatial distribution of DO concentration based on season.
equipment. The middle and upper reaches of the river are surrounded by the water was not saturated with DO, which was closely related to the
industrial building sites. The downstream section of the river flows water temperature. The other was the oxygen released by plants into the
through Longyue Lake, with agricultural land in the vicinity. Therefore, water through photosynthesis. Further, the middle reaches of the river
the water quality of the downstream section is better than that of the were near construction sites; thus, the water temperature was a major
middle and upper reaches. In recent years, Dongying City has focused on factor, which was higher in February and May and more oxygen was
the water environment management of the Guangli River, which has dissolved in the river, resulting in higher concentrations. The down
improved the water quality. stream river section was mostly surrounded by farmlands and wetlands;
therefore, the number of plants in the water was the main contributing
(2) Seasonal variations factor. Plants in water gradually started germinating and growing in
February, releasing less oxygen and absorbing more oxygen. Therefore,
Fig. 5 shows the spatial distribution of the DO concentration in the the DO concentration in the river was low.
Guangli River on February 10, 2021, May 21, 2021, and October 23,
2021. The following conclusions can be drawn from plots (a), (b), and 4.1.2. Distribution of TN concentrations
(c): At the seasonal scale, the DO concentrations varied across different
periods. Overall, the DO concentration in the Guangli River was higher (1) Annual variations
in the lower reaches of the river section than in other river sections. In
terms of different river sections, the concentration in the upstream The distribution of TN concentrations in the Guangli River from 2020
section of the Guangli River varied slightly, while the variation was to 2021 is shown in Fig. 6. Fig. 6a & b show that the color change at the
more obvious in the middle and lower reaches of the river section. DO same position was marginal, implying that the amount of change in the
concentrations in the middle reaches of the river were the lowest in TN concentration in the Guangli River between 2020 and 2021 was not
October and increased in February and May. Downstream of the river large enough to be reflected from the color. However, an increase in the
section showed an opposite trend compared with the upstream. This was average concentration was observed, from 2.88 mg/L in 2020 to 3.02
mainly because there were usually two DO sources in water. One was the mg/L in 2021. In each graph, the color of the concentration changed
infiltration of oxygen from the atmosphere into the water column when gradually from red-yellow to blue, with the highest concentration in the
11
Fig. 6. Spatial distribution of TN concentration based on year.
Fig. 7. Spatial distribution of TN concentration based on season.
12
Fig. 8. Spatial distribution of CODMn concentration based on year.
upstream section, ranging from 3.5 mg/L to 4.5 mg/L. The TN concen river increased during this period. The concentration of CODMn in the
tration gradually decreased from upstream to downstream in the river middle reaches of the river decreased, whereas that in the upper reaches
section. The main reason for this was that the upper reaches of the river of the river remained almost unchanged. Numerically, the average
were mainly in the center of the development zone, with construction CODMn concentration in the whole river remained unchanged during the
sites nearby. Industrial wastewater was generated by factories mainly in year, with a value of approximately 7.2 mg/L. The overall analysis of
the petrochemical industry. Residential areas with high building density CODMn was high in the middle and lower reaches of the river section,
discharged domestic sewage. All of these will increase the nitrogenous which indicated that the pollution in the middle and lower reaches of the
substances in the water environment, as well as being one of the main river section was serious due to the low water quality of the river.
sources of nitrogen in the water body.
(2) Seasonal variations
(2) Seasonal variations
The spatial distribution of CODMn concentrations in Tiaohe on
The spatial distribution of the TN concentration in Guangli River on November 30, 2020, April 4, 2021, July 5, 2021, and October 23, 2021,
January 29, 2021, May 21, 2021, and October 23, 2021, is shown in is shown in Fig. 9. The figure shows evident changes in the CODMn
Fig. 7. We could conclude that the TN concentration varied slightly concentrations in Tiaohe during different seasons. Seasons has a small
across different seasons. Compared with May, the changes in the two effect on changes in concentration. First, graphs (Fig. 9a & d) show
graphs for January and October were not evident. However, the TN changes during 2020–2021 in terms of time, with a more evident in
concentration was generally higher in October than in January. The crease in concentration in the upstream section of the river. Second, a
color of the image on May 21, 2021 (yellow) reflected that the TN comprehensive analysis of the four graphs showed that the CODMn
concentration in May was higher than that in January and October, content in the upper reaches of the river was high, while that in the
particularly in the middle and lower reaches of the river. This was middle reaches was relatively low, and vice versa. This may be due to
mainly influenced by the drainage of the surrounding agricultural land. the mobility of water bodies. Finally, the CODMn content was higher in
The vicinity of the middle and lower reaches of the river was mostly an the middle and lower reaches of the river in April than in other periods.
agricultural cropping area. To promote high crop yield, large amounts of It was due to the general increase in rainfall with the arrival of spring.
organic and inorganic fertilizers needed to be applied, which increased Pollutants contained in domestic sewage along the river were washed
the nitrogen content in the middle and lower reaches of the river in May. into the river. Phytoplankton bloomed in the water, which contributed
Additionally, compared with the TN concentration, DO concentration to an increase in the CODMn. However, many factors affect CODMn in
showed an opposite trend. This high TN content in the water body rivers, including industrial wastewater discharge, pollutant dispersion,
promoted the overgrowth of algal plants, which consumed a large and the constant occurrence of rainfall and floods. Hence, further
amount of DO in the water, consequently, dropping the DO content research and analysis are required to investigate the reasons for these
sharply in the water body. Therefore, the DO content in the water body changes.
showed an opposite trend to that of the TN content.
4.1.4. Distribution of TP concentration
4.1.3. Distribution of CODMn concentration
(1) Annual variations
(1) Annual variation
Fig. 10 shows the spatial distribution of TP concentrations in Tiaohe
The Tiaohe was located mainly in the estuary area, which carried out on May 6, 2020, and May 6, 2021. A comprehensive analysis of Fig. 10a
the functions of irrigation and flood discharge. Most of the vicinity of the & b shows that the change in the TP concentration in Tiaohe between
lower reaches of the river was agricultural land, while the middle and 2020 and 2021 was marginal. The color in the maps changed from red to
upper reaches were dominated. Fig. 8a & b show the spatial distribution yellow in the lower reaches of Tiaohe, with a decrease in the TP content.
of CODMn concentrations in Tiaohe on May 1, 2020, and May 1, 2021, However, the color changes in the other river sections were not obvious.
respectively. The CODMn concentration in Tiaohe changed evidently The average TP concentration value was approximately 0.61 mg/L both
during 2020–2021. The color change in concentration also differed in times, which was not significantly different. The lower reaches of the
different river sections. The concentrations in the lower reaches of the Tiaohe were mostly farmland, which had a high demand for pesticides
13
Fig. 9. Spatial distribution of CODMn concentration based on season.
Fig. 10. Spatial distribution of TP concentration based on year.
14
and fertilizers. Fertilizer loss from farmland was carried into the water and lower reaches of the river exhibited particularly high levels in July
by rainwater and other sources. It also caused the TP concentration in 2021. This may be because the flood season was the main period for TP
the downstream section of the river to be much higher than in other pollutants to enter rivers and lakes. In the upstream and midstream
sections of the river. sections of the river, after rainfall, rainwater mixed with various pol
lutants and flowed downstream, where the TP concentrations were
(2) Seasonal variations found to be high.
Fig. 11 shows the spatial distribution of the TP concentrations in 4.1.5. Distribution of TN concentration
Tiaohe on November 30, 2020, April 4, 2021, July 5, 2021, and October
23, 2021. From the figure, we can conclude that the TP concentration (1) Annual variations
was higher in the middle and lower reaches of the river, indicating clear
levels of water pollution. There were three main causes of phosphorus The distribution of the TN concentrations in the Branch River from
pollution in rivers: 1) agricultural pollution, which mainly originated 2020 to 2021 is shown in Fig. 12. Fig. 12a & b show that the color of the
from phosphorus fertilizers, 2) domestic wastewater, which was mainly same position gradually became lighter. Two knots were drawn from
derived from phosphorus-containing detergents, and 3) industrial each other: 1) The TN concentration in the Branch River was extremely
wastewater, which originated mostly from the production of phosphorus high (mostly above 10 mg/L), which indicated that nitrogen pollution in
chemicals and phosphate wastewater from metal surface treatment the Branch River was severe. 2) The TN concentration of TN decreased
processes. Therefore, the high phosphorus content in the downstream between 2020 and 2021. This was also evident from the average con
section of the river was mainly attributable to the surrounding agricul centration, which decreased from 13.41 mg/L in 2020 to 12.27 mg/L in
tural land owing to the extensive use of chemical fertilizers, which 2021. Branch River was located in Guangrao County, an area dominated
caused the phosphorus in the fertilizers to flow into water bodies by the development of modern and efficient agriculture. Almost all of
through rainwater or agricultural irrigation. Furthermore, the middle the agricultural and experimental fields around the tributary rivers
Fig. 11. Spatial distribution of TP concentration based on season.
15
Fig. 12. Spatial distribution of TN concentration based on year.
Fig. 13. Spatial distribution of TN concentration based on season.
16
required the use of large amounts of fertilizer. Nitrogen from fertilizers facilities such as water quality purification at key nodes such as down
flowed into water bodies through rainfall or agricultural irrigation, stream of the river’s key outfalls and the river’s inlet to the lake ac
resulting in extremely high concentrations of TN in rivers. cording to local conditions. 2) Control the total amount of pollutant
discharges strictly. Implement the strategy of sewage treatment and
(2) Seasonal variations improve the technology of sewage purification to ensure that the outfall
meets the standards of the river. 3) Develop the technology of water
Fig. 13 shows the spatial distribution of TP concentrations in the recycling and water reuse. Reuse wastewater from industrial production
Branch River on November 30, 2020, February 8, 2021, May 21, 2021, processes after purification and other technologies to reduce industrial
and September 21, 2021. However, the seasonal variations in the TN wastewater emissions. 4) Reinforce the prevention and control of agri
concentration in the Branch River were not obvious and did not display a cultural surface pollution. According to the monitoring results of the
clear pattern. From the figure, we could conclude that the TN concen river, the river section that exceeds the standard can be targeted for
tration in the Branch River decreased during this period, indicating that treatment. Scientific application of pesticides and fertilizers should be
the effect of TN pollution control in the Branch River was more evident carried out in the production and planting. Reasonable adjustment of
by the end of 2021. fertilizer structure and the promotion of organic fertilizers can reduce
Combining the above analyses, we can find some characteristics of the rate of increase of nitrogen and phosphorus content in the water. 5)
changes in the spatial distribution of water quality in the YRD region. Strengthen monitoring and management. We need to set up a water
First of all, the pollution of water quality in the YRD is mostly anthro environmental protection organization and formulate laws and regula
pogenic, which is caused by the production and living activities of tions to protect water bodies. Controlling river pollution is a long-term
human beings. In the southern part of the YRD, modern and efficient battle that cannot be achieved overnight. Long-term, real-time moni
agriculture is developed. The residues of chemical fertilizers and pesti toring of changes in the pollution levels is needed to scientifically un
cides are discharged into the river with the drainage period of farmland, derstand the differences between the pollution in the upper, middle, and
making the TN and TP content high in the tributary rivers. The devel lower reaches of the YRD, and implement zoning and classification
opment of the central region is mostly dominated by the petrochemical protection and treatment.
industry chain. The building density of residential houses, factories, etc.
is high. Discharge of industrial wastewater and domestic sewage is the 4.3. Analysis of the status and applicability of WQPs inversion
main influencing factor of water quality change. The middle and upper
reaches of the Guangli River have high TN content and low DO content. The nine basic monitoring indicators in the National Surface Water
It is developed with mixed agriculture and industry in the northern part. Monitoring and Evaluation Program are water temperature, pH, DO,
There are mostly farmlands and wetlands around the coast. In addition, conductivity, turbidity, CODMn, NH3-N, TP, and TN (additional in
residues such as industrial wastewater, domestic and livestock sewage, dicators, such as chlorophyll a and transparency, were measured in lakes
and fertilizers from cultivated land are washed into the river by rain and reservoirs). Accurate and long-term monitoring of the concentra
water during the spring and summer seasons when the precipitation tions of these indicators is of great importance for environmental pro
content is high. It makes the water quality of the river poor during the tection and water management. WQPs reflecting the quality of the water
flood season. environment are mainly divided into two categories. One category is
light-sensitive parameters, mainly chlorophyll a, suspended solids,
4.2. Analysis of water quality protection response strategies colored soluble organic matter, transparency and so on. The difference
in water color can directly reflect the concentration of water quality
In Subsection 4.1., we analyzed the inversion results of each WQP in photosensitive parameters. The other category is non-water-colored
combination with the actual conditions. On combining all model results, WQPs, dominated by organic matter such as chemical oxygen de
we determined that the overall water quality in the YRD region was mand, NH3-N, TP, and CODMn. It is mainly distributed in the water body
extremely poor, with the river water quality belonging to Classes III and and the surface characteristics of the water body are not obvious (Gan,
IV, and the water quality in some river sections belonging to even Class 2021). Bresciani et al. (Bresciani et al., 2011; Xiang et al., 2022) verified
V. Several land use types are found in the YRD region, including in the feasibility of inverting chlorophyll a concentration from remote
dustrial, residential, and agricultural lands, which are affected by sensing images. Williamson et al. (Ma and Dai, 2005; Williamson and
human activities. In 2022, the state issued the “Yellow River Ecological Grabau, 1974) found a link between remote sensing data and suspended
Protection and Treatment Battle Action Plan” to implement ecological sediment concentration. It was concluded that the ratio of TM4 and TM1
protection and high-quality development in the Yellow River Basin, was the highest precision band for estimating suspended sediment
accelerate the elimination of polluted water bodies to meet the water concentration in Lake Taihu.
quality standards and promote river outfall investigation and rectifica In the research and application of using remote sensing to monitor
tion. According to the inversion results of this study, we can select river water quality, remote sensing inversion for optical properties of river
sections that do not yet belong to Class V water quality and prepare a WQPs is gradually maturing. For non-photosensitive parameters, it is
plan to implement the water quality standards; particularly, Class V difficult to obtain them directly by remote sensing inversion because of
water bodies with poor water quality in the Yellow River Basin can be their weak optical properties, such as TN, TP and DO. But they affect the
eliminated by 2025. Moreover, in the river key outfall downstream, the spectral information of the river. Therefore, we need to improve the
river can be included in the lake and other key nodes according to the accuracy of the inversion by indirect remote sensing analysis with the
local conditions to build water quality purification facilities and other help of a semi-empirical method or machine learning. For example, ni
engineering facilities. Finally, the prevention and control of agricultural trogen and phosphorus can increase algal growth. Although there is no
surface source pollution should be strengthened because nitrogen and direct effect on the spectrum of parameters, it can cause eutrophication
phosphorus pollutants exceed the standards for rivers. Nitrogen and of water bodies, and destroy healthy river ecosystems (Abayazid and El-
phosphorus sources can be assessed and analyzed, based on which the Adawy, 2019; Huang et al., 2020; Wang and Qin, 2023). Many studies
fertilization structure can be adjusted. have been conducted on this issue. Chen et al. (Chen et al., 2020b) used
Combined with the analysis results in subsection 4.1 of the paper, we measured data and Landsat data to invert parameters, such as TN and
adopt the following water quality management measures. 1) The TP, in the Grass Sea, and proved the feasibility of remote sensing
strengthening of engineering management measures in key river sec inversion of non-photosensitive parameters. Xu et al. (Xu et al., 2013)
tions. Industrial wastewater is one of the important causes affecting used remote sensing data to establish TP and TN inversion models for
water quality in the YRD region. We need to construct engineering Taihu Lake using an empirical modeling method and reported high
17
model accuracy. of 0.9271. In Tiaohe, the (B6 + B8A)/(B4-B12) linear fit model
In summary, the current studies are remote sensing inversion models (R2 = 0.9581) for CODMn was the best model for the inversion of
for a specific lake, lacking a generalized inversion model. In future CODMn. The exponential function regression equation with B2/
research, the characteristics of non-photosensitive parameters should be (B3 + B4 + B12) as the independent variable best inverted the TP
studied comprehensively to develop remote sensing inversion models with an MSE of 0.0001. In the Branch River area, the TN expo
suitable for specific applications. Furthermore, most studies in the YRD nential inversion model with (B3 + B8A)/(B2 + B12) provided
region have focused on salinity, with almost no studies on water quality. the best fit, with an R2 value of 0.7387.
Therefore, the conclusions of this study cannot be verified through (3) Using the best inversion model constructed using the Sentinel-2
previous relevant studies. Moreover, most of the water quality moni data source for the WPQ inversion, it was found that the WQPs
toring data obtained in this study were of non-photosensitive parame within the rivers in the YRD region generally showed a distri
ters. Further studies are required to validate these results. Secondly, the bution of high values in the upper reaches and low values in the
accuracy of the water quality inversion model for rivers and lakes is middle and lower reaches, with some seasonal changes. The DO
affected by the spectral, temporal, and spatial resolutions of the satellite and TN concentrations in the Guangli River showed opposite
remote sensing data, local area aerosol monitoring, and atmospheric trends. The upstream water quality of the river was low. Further,
correction technical conditions (Xu, 2020). Simultaneously, many ex the CODMn concentration in Tiaohe did not correlate with the
perts and scholars have used various inversion methods and models to change in the TP concentration. However, both CODMn and TP
examine the concentration of different types of WQPs. However, contents were high in the middle reaches of the river, where the
monitoring of water quality based on remote sensing technology has not water quality was the worst. The TN concentration in the branch
yet reached universal application. This is because in the study of water river was reduced by 2021, although Class V water quality was
quality inversion using multispectral data, many factors affect the re observed. Compared with 2020, the water quality was much
sults of each process, from remote-sensing data acquisition and pro better. Therefore, the implementation of corresponding water
cessing to the determination of research methods to errors in the date of quality management measures should be continued.
water quality data collection while matching water quality data with (4) The results of the correlation and regression analyses showed that
remote sensing data from different sensors. Preprocessing of remote the regression fitting equation of the WQPs can be established by
sensing data can affect the acquisition of surface reflectance. The se measuring water quality indicators at monitoring stations, which
lection of the study method can also affect the regression results of the can evaluate the water quality condition in the YRD more accu
WQPs. This is because the optical properties of water bodies vary greatly rately and quickly. In the future, new methods can be developed
among water bodies, thereby limiting the applicability of the inversion for monitoring with good fits to improve the accuracy of the
model to different water bodies. Lastly, in the inversion study of water overall water quality condition estimation in the YRD. Addi
quality in the YRD, the amount of actual water quality measurement tionally, we can select other data sources or verify the accuracy of
data obtained was small owing to the limited number of monitoring the regression equation in neighboring or similar water bodies.
sections. Problems, such as the small amount of research data, the results
of simulated water quality, and actual water quality cannot be properly 5.2. Research limitations and perspectives
compared and analyzed. Ideally, the study area for manual point
monitoring should increase the number of monitoring points to increase Many scholars have studied the quantitative remote sensing inver
the number of study samples. Therefore, a more comprehensive sion of WQPs. However, because of the different WQPs in the study area
assessment of the status and future trends in water quality should be at different times, establishing a general model is difficult. The selection
conducted. of remote sensing data sources, different mathematical modeling
methods, and other factors can also affect the accuracy of the model
5. Conclusions inversion. Therefore, to improve the accuracy of the inversion and
application of the model, this study mainly considered the following two
5.1. Summary aspects:
Few studies have focused on the monitoring and management of (1) Seasonal differences exist in the inherent optical properties of
water quality in the YRD based on remote sensing technology. In this water bodies that may produce errors. In this study, a simple
study, we adopted a one-dimensional regression model to invert the quarterly analysis was conducted based only on the time of the
concentrations of six WQPs in the YRD: DO, CODMn, NH3-H, TP, TN, and remote-sensing images. The variation patterns of the intrinsic
turbidity. A preliminary analysis of the spatiotemporal distributions of optical parameters of the water bodies were fully investigated to
water quality in the YRD was conducted to investigate the spatiotem improve the robustness and practical application of the model.
poral distribution patterns and dynamic monitoring of water quality in (2) The accuracy of the water quality inversion model for rivers and
the YRD. The main conclusions are as follows: lakes was affected by various factors, such as the spatiotemporal
resolution of remote sensing data and inversion methods. In
(1) Various inversion indexes were constructed based on the future studies, other data sources and basic algorithms can be
Sentinel-2 MSI data, arriving at a waveband combination used. The findings of this study were compared and analyzed to
involving red (B4), vegetation red edge (B5), SWIR (B11), and increase the research on water quality in the Yellow River Delta.
SWIR (B12), with high correlation with several WQPs. The band This can provide a basis for the remote sensing inversion of
combination could improve the correlation between Sentinel-2 similar polluted water bodies.
remote sensing data and WQPs.
(2) Bands with higher correlations were selected as the input factors CRediT authorship contribution statement
of the regression model, which established a multispectral remote
sensing model of WQPs in the YRD region. In the Guangli River Xin Cao: Methodology, Validation, Formal analysis, Investigation,
region, the inverse DO fit of the cubic polynomial model con Data curation, Writing – original draft, Writing – review & editing. Jing
structed by the combination of the (B3-B5 + B11)/B12 bands was Zhang: Resources, Supervision, Project administration, Writing – re
selected to be the best, with R2 = 0.6099 and an MSE value of view & editing, Funding acquisition. Haobin Meng: Data curation,
1.6186 mg/L. The (B11-B12)/(B5 + B8A) band combination was Methodology, Investigation, Validation. Yuequn Lai: Data curation,
established as the best inversion model for TN, with an R2 value Investigation, Validation. Mofan Xu: Data curation, Investigation.
18
Declaration of Competing Interest Lu, F., Chen, Z., Liu, W., Shao, H., 2016. Modeling chlorophyll-a concentrations using an
artificial neural network for precisely eco-restoring lake basin. Ecol. Eng. 95,
422–429. https://doi.org/10.1016/j.ecoleng.2016.06.072.
The authors declare that they have no known competing financial Ma, G., Li, G., 2002. Demonstration Study on Satellite Remote Sensing Monitoring of
interests or personal relationships that could have appeared to influence Water Pollution in Daliao River Estuary. Liaoning Urban and Rural Environmental
the work reported in this paper. Science & Technology 26-29+36.
Ma, R., Dai, J., 2005. Quantitative estimation of chlorophyll-a and total suspended
matter concentration with Landsat ETM based on field spectral features of Lake
Data availability Taihu. J. Lake Sci. 97–103.
Meng, H., Zhang, J., Zheng, Z., 2022. Retrieving Inland Reservoir Water Quality
Parameters Using Landsat 8–9 OLI and Sentinel-2 MSI Sensors with Empirical
Data will be made available on request. Multivariate Regression. Int. J. Environ. Res. Public Health 19, 7725. https://doi.
org/10.3390/ijerph19137725.
Acknowledgments Olmanson, L.G., Brezonik, P.L., Bauer, M.E., 2013. Airborne hyperspectral remote
sensing to assess spatial distribution of water quality characteristics in large rivers:
The Mississippi River and its tributaries in Minnesota. Remote Sens. Environ. 130,
This research was supported by the National Natural Science Foun 254–265. https://doi.org/10.1016/j.rse.2012.11.023.
dation of China (41271004). Papoutsa, C., Retalis, A., Toulios, L., Hadjimitsis, D.G., 2014. Defining the Landsat TM/
ETM+ and CHRIS/PROBA spectral regions in which turbidity can be retrieved in
inland waterbodies using field spectroscopy. Int. J. Remote Sens. 35, 1674–1692.
References https://doi.org/10.1080/01431161.2014.882029.
Piotrowski, A.P., Osuch, M., Napiorkowski, M.J., Rowinski, P.M., Napiorkowski, J.J.,
Abayazid, H.O., El-Adawy, A., 2019. Assessment of a Non-Optical Water Quality 2014. Comparing large number of metaheuristics for artificial neural networks
Property Using Space-based Imagery in Egyptian Coastal Lake. Int. J. Adv. Network, training to predict water temperature in a natural river. Comput. Geosci. 64,
Monitor. Controls 4, 53–64. https://doi.org/10.21307/ijanmc-2019-059. 136–151. https://doi.org/10.1016/j.cageo.2013.12.013.
Abdelmalik, K.W., 2018. Role of statistical remote sensing for Inland water quality Wang, L., 2014. Determination the Water Quality and Evalution the Eutrophication of
parameters prediction. Egypt. J. Remote Sens. Space Sci. 21, 193–200. https://doi. Hulun Lake with MODIS Data (Master degree thesis). Inner Mongolia Agricultural
org/10.1016/j.ejrs.2016.12.002. University.
Barrett, D., Frazier, A., 2016. Automated Method for Monitoring Water Quality Using Wang, S., Qin, B., 2023. Research Progress on Remote Sensing Monitoring of Lake Water
Landsat Imagery. Water 8, 257. https://doi.org/10.3390/w8060257. Quality Parameters. Environ. Sci. 44, 1228–1243. https://doi.org/10.13227/j.
Bresciani, M., Stroppiana, D., Odermatt, D., Morabito, G., Giardino, C., 2011. Assessing hjkx.202203285.
remotely sensed chlorophyll-a for the implementation of the Water Framework *Warrick*, J.A., Mertes, L.A.K., Siegel, D.A., Mackenzie, C., 2004. Estimating suspended
Directive in European perialpine lakes. Sci. Total Environ. 409, 3083–3091. https:// sediment concentrations in turbid coastal waters of the Santa Barbara Channel with
doi.org/10.1016/j.scitotenv.2011.05.001. SeaWiFS. Int. J. Remote Sens. 25 (10), 1995–2002.
Chen, Y., Huang, C., Zhang, L., Qiao, N., 2020a. Spectral Characteristics Analysis and Wen, J., Xiao, Q., Yang, Y., Liu, Q., Zhou, Y., 2006. Remote sensing estimation of aquatic
Remote Sensing Retrieval of COD Concentration. Spectrosc. Spectr. Anal. 40, chlorophyll-a concentration based on Hyperion data in Lake Taihu. J. Lake Sci.
824–830. 327–336.
Chen, Y., Liu, S., Wang, K., Song, S., Liang, P., Chen, F., 2020b. Remote Sensing of Caohai Williamson, A.N., Grabau, W.E., 1974. Sediment Concentration Mapping in Tidal
Lake Water Quality Using Landsat Satellite Images. J. Hydroecol. 41, 24–31. https:// Estuaries. NASA Special Publication 351, 1347.
doi.org/10.15928/j.1674-3075.2020.03.004. Xiang, Y., Zhang, C., Wu, Z., Fu, C., 2022. Inversion Study on Chlorophyll a of Shahu
Chen, Z., 2021. Research and Application of Remote Sensing Water Quality Inversion Lake Based on Landsat 8 Data. Territory & Natural Resources Study 88–91.
Algorithm for Inland Waters (Master degree thesis). Zhejiang University. 10.27461/ 10.16202/j.cnki.tnrs.2022.01.022.
d.cnki.gzjdx.2021.000279. Xie, H., 2006. Remote Sensing Based Water Monitoring and Spatial-temporal Analysis
Du, Y., 2021. In: Research and Application of Key Technologies for Remote Sensing (Master degree thesis). Tongji University.
Water Quality Inversion Based on Big Data Platform (Master degree thesis). Xu, P., 2020. In: Study on Water Quality Inversion of Harbin Section of Songhua River
Huazhong University of Science and Technology. https://doi.org/10.27157/d.cnki. based on multi-source Remote sensing data (Master degree thesis). Harbin Normal
ghzku.2021.001250. University. https://doi.org/10.27064/d.cnki.ghasu.2020.000253.
Feng, Q., Cheng, X., Shen, X., Xiao, X., Wang, L., Zhang, W., 2017. Inland Riverine Xu, L., Huang, C., Li, Y., Chen, X., 2013. Deriving Concentration of TN, TP based on
Turbidity Estimation for Hanjiang River with Landsat 8 OLI Imager. Geomat. Inform. Hyper Spectral Reflectivity. Remote Sens. Technol. Appl. 28, 681–688.
Sci. Wuhan Univ. 42, 643–647. https://doi.org/10.13203/j.whugis20141002. Yang, B., Liu, Y., Ou, F., Yuan, M., 2011. Temporal and Spatial Analysis of COD
Gan, T., 2021. In: Research on water body extraction and water quality inversion Concentration in East Dongting Lake by Using of Remotely Sensed Data. Procedia
technology based on Sentinel-2 images (Master degree thesis). Hunan University. Environ. Sci. 10, 2703–2708. https://doi.org/10.1016/j.proenv.2011.09.420.
https://doi.org/10.27135/d.cnki.ghudu.2021.000031. Yang, G., Zhou, Y., Chen, Q., 2006. A Review of the Studies on Sustainable Use and
He, B., Ding, C., Yang, X., Liang, S., 2011. Recovering of Landsat7 ETM+ SLC-OFF data Management of Water Resources in the Pearl River Delta. Trop. Geogr. 18–22.
and its application on water quality retrieval in east lake in Wuhan. Resour. Environ. Zhai, Z., Lu, S., Pao, Z., Ma, J., Wang, P., Dai, Y., 2018. Water quality monitoring with
Yangtze Basin 20, 90–95. GF1-WFV imagery in Panjiakou reservior, Hebei province of China. J. China Inst.
Huang, S., 2020. In: Research on monitoring algorithm of water quality parameters based Water Resour. Hydropower Res. 16, 297–306. 10.13244/j.cnki.jiwhr.2018.04.008.
on remote sensing images (Master degree thesis). Heilongjiang University. https:// Zhang, X., 2018. COD in the tide-sensitive section of the Yellow River Delta measurement
doi.org/10.27123/d.cnki.ghlju.2020.000369. method calibration and application (Master degree thesis). Shandong Normal
Huang, J.J., Guo, H., Chen, B., Guo, X., Singh, V.P., 2020. Retrieval of Non-Optically University.
Active Parameters for Small Scale Urban Waterbodies by a Machine Learning-Based Zhang, B., Shen, Q., Li, J., Zhang, H., Wu, D., 2009. Retrieval of three kinds of
Strategy (preprint). Earth Sc. https://doi.org/10.20944/preprints202004.0111.v1. representative water quality parameters of Lake Taihu from hyperspectral remote
Huang, D., Zhao, Q., 1996. On Water Control in Developing the Yellow River Delta. Sci. sensing data. J. Lake Sci. 21, 182–192.
Technol. Rev. 51–53. Zhang, H., Wang, F., Jiang, L., Wu, X., Du, H., Liu, J., Meng, F., 2017. Assessment OF
Jiang, Q., 2020. In: Study on the Effectiveness Evaluation Method of Satellite Remote Heavy metal pollution in the surface water of Tiaohe River and Shenxiangou Brook,
Sensing in the Monitoring of Lake and Reservoir Water Quality -Take GF-1 satellite Dongying City, Shandong. J. Saf. Environ. 17, 1536–1542. https://doi.org/
as an example (Master degree thesis). Lanzhou Jiaotong University. https://doi.org/ 10.13637/j.issn.1009-6094.2017.04.063.
10.27205/d.cnki.gltec.2020.000038. Zhao, S., 2021. In: Inversion of water quality parameters of Fuyang River in Handan City
Khattab, M.F.O., Merkel, B.J., 2014. Application of Landsat 5 and Landsat 7 images data based on multi-source remote sensing data (Master degree thesis). Hebei University
for water quality mapping in Mosul Dam Lake, Northern Iraq. Arab. J. Geosci. 7, of Engineering. https://doi.org/10.27104/d.cnki.ghbjy.2021.000140.
3557–3573. https://doi.org/10.1007/s12517-013-1026-y. Zhong, D., 2017. Inversion of Water Quality in Nine Plateau Lakes in Yunnan Based on
Li, Y., 2015. Influence of Salinity on Chemical Oxygen Demand Measurement and A Multi-source Remote Sensing Data. Northwest University. Master degree thesis.
Primary Correction (Master degree thesis). Ocean University of China. Zhu, T., 2018. Water Quality Prediction Methods Based on Remote Sensing Inversion and
Li, Y., 2020. In: Surface water spatial-temporal evolution and its ecological and Its Application (Master degree thesis). Zhejiang University of Technology.
environmental effects in Yellow River Delta (Doctoral dissertation). Shandong Zhu, L., 2006. Remote Sensing Monitoring and Assessment of Water Quality for Lakes
Normal University. https://doi.org/10.27280/d.cnki.gsdsu.2020.000042. (Doctoral dissertation). Institute of Remote Sensing Applications, Chinese Academy
of Sciences.
19

Cao 2023

Uploaded by

Copyright:

Available Formats

Cao 2023

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Cao 2023

Uploaded by

Copyright:

Available Formats

Ecological Indicators 155 (2023) 110914

Contents lists available at ScienceDirect

Remote sensing inversion of water quality parameters in the Yellow

Chinese Ministry of Environmental Protection on July 1, 2009, and Table 2

Shenxiangou 1.67 Heavy Polynomial : y = a + b1 x + b2 x2 (6)

2.3.3. Model testing method 0.8 ≤ r ≤ 1.0 Maximum correlation

Fig. 2. Overall research framework.

Pile 5 Min 5.59 3.51 0.03 0.03 0.83 9.96

Pile 5 TP B6/(B7-B11-B12) − 0.7594 0.7114 B4/(B5-B11-B3) − 0.7420 0.6489

r, a correlation between waveband combination and water quality concentration;

3.3. Regression of WQPs based on Sentinel-2 data

DO 1 (B3-B5-B11) y = 7.1181 + 1.7312 × x + 2.7453 × x2 0.7124

Dongba Road Bridge DO (B3-B5-B11) y = 22.2279 + 46.8338 × x+ a 0.7280 MSE 1.6186

Fig. 3. Best-fit modeling (a) and Model accuracy (b).

Fig. 4. Spatial distribution of DO concentrations based on year.

Fig. 5. Spatial distribution of DO concentration based on season.

Fig. 6. Spatial distribution of TN concentration based on year.

Fig. 7. Spatial distribution of TN concentration based on season.

Fig. 8. Spatial distribution of CODMn concentration based on year.

Fig. 9. Spatial distribution of CODMn concentration based on season.

Fig. 10. Spatial distribution of TP concentration based on year.

Fig. 11. Spatial distribution of TP concentration based on season.

Fig. 12. Spatial distribution of TN concentration based on year.

Fig. 13. Spatial distribution of TN concentration based on season.

You might also like