Spatiotemporal evolution of chlorophyll-a concentration from MODIS data inversion in the middle and lower reaches of the Hanjiang River, China

Owing to limitations in monitoring technologies, monitoring the algae content index of water has lagged behind the conventional water quality index. As a result, sample monitoring in many rivers has been too sparse, and the monitoring data have been inconsistent; thus the evolution of water eutrophication has not been fully reflected. This study focused on the middle and lower reaches of the Hanjiang River, China, and correlated moderate-resolution imaging spectroradiometer (MODIS) remote sensing data with measured chlorophyll-a concentrations. Algorithm settings for chlorophyll-a inversion in the middle and lower reaches of the Hanjiang River were established via the trial and error method. The algorithm model for the middle and lower reaches of the Hanjiang River chlorophyll-a concentration inversion, and the results of the inversion analysis for the spatiotemporal evolution characteristics were subsequently used to determine the influence of various environmental factors on changes in the chlorophyll-a concentration. The results indicate that (1) the band combinations B7/(B6 + B5), B7/B5, B4-B2, and B4/(B3 + B2) are well-correlated with the chlorophyll-a concentration; (2) the back propagation (BP) neural network model inversion achieved a better fit and more accurate inversion results than the band ratio model; (3) temporally, algal outbreaks were mostly concentrated occurring in February and March, with higher chlorophyll-a concentrations in the water column during 2000, 2006, 2007, and 2008; (4) spatially, high chlorophyll-a concentrations were observed in the Zhongxiang, the Shayang, and upper Xiantao sections; and (5) increases in the water temperature and decreases in the water level and flow rate could lead to higher chlorophyll-a concentrations; similarly, nutrient salts were identified to be a major factor contributing to changes in the chlorophyll-a concentrations.


Introduction
Algal blooms are a prominent ecological and environmental problem (Yang et al. 2012), occurring frequently in water bodies worldwide that not only affect water quality and the survival of aquatic organisms, but also the safety of drinking water and human life and health in severe cases. A key factor causing this problem is eutrophication. A high level of eutrophication in a water body promotes the proliferation of blue-green algae and the formation of a blue-green algae film on the water's surface, resulting in the water body becoming polluted. Algal bloom outbreaks typically occur in inland lakes, reservoirs, or ponds (Huo et al. 2019); outbreaks in large rivers occur only rarely. However, with the construction and operation of water conservancy projects in China and overseas, the hydrological situation of large rivers has changed. Their hydrological conditions are more complex than those of lakes and reservoirs, and they are more prone to blooms; therefore, it is more meaningful to strengthen the monitoring of algal blooms in large rivers and mobile water bodies for protecting aquatic environments. With the improvement of communication technology, moderate resolution imaging spectroradiometer (MODIS) remote sensing satellite data have come to be widely used in environmental monitoring, resource investigation, urban planning, and other fields. MODIS is a large space remote sensing instrument (Naghdi et al. 2020) developed by NASA that provides long-term Earth observation data, including atmospheric information, global coverage, and multi-spatial resolution and multi-spectral images and can provide support for understanding global climate and how it is affected by human activities (Liu et al. 2020).
Chlorophyll-a is the main component of chlorophyll in algae; its distribution can better reflect the degree of algal bloom in water bodies (Kim et al. 2020). Chlorophyll-a exhibits absorption valleys in the blue band (near 442 nm) and the red band (near 665 nm) (Han et al. 1994), in addition to the obvious reflection peaks in the infrared band (near 700 nm) (Gitelson et al. 1992), showing certain characteristic spectra. Changes in chlorophyll-a concentrations lead to significant changes in the reflectance spectral characteristics of a water body (Jiang et al. 2020). Therefore, indirect chlorophyll-a concentration estimation from satellite reflectivity data is more dynamic, periodic, and real-time than traditional manual monitoring and is suitable for inland waters with complex optical characteristics. Furthermore, the inversion model of the chlorophyll-a concentration can be used to interpolate the measured chlorophyll-a concentration according to remote sensing data, which provides support for further studies.
The models often used to invert the concentrations of water quality parameters from remote sensing data are mostly empirical and semi-empirical (Shu et al. 2000;Hu et al. 2009;Tebbs et al. 2013;Telesca et al. 2018;Cannizzaro et al. 2019). These models have certain shortcomings in terms of their inversion accuracy and generalizability and have poor fault-tolerance (Tao et al. 2018). For this reason, researchers (Park et al. 2015;Harvey et al. 2015) have continuously improved the inversion accuracy and generalizability by establishing a waveband ratio model and introducing nonlinear relational optimization methods for water quality parameter inversion, such as neural network methods and genetic algorithms i.
Based on the above-mentioned literature, this study adopted the band ratio and back propagation (BP) neural network models and considered typical river sections in the middle and lower reaches of the Hanjiang River, China, as the study area. The measured water quality and MODIS satellite remote sensing data were used, the band ratio and BP neural network models were established through an empirical method, and the optimal inversion algorithm for chlorophyll-a in the middle and lower reaches of the Hanjiang River was determined by comparing the model errors. The chlorophyll-a concentrations in the middle and lower reaches of the Hanjiang River from 2000 to 2011 were inverted using this model, and the temporal and spatial characteristics of the chlorophyll-a concentration in the water body were analyzed.

Study area
The Hanjiang River (106°12'E-114°14'E, 30°08'N-34°11'N) is a tributary of the Yangtze River, China, with a total length of 1577 km and drainage area of 15. 9 × 104 km 2 (Fig. 1). The main stream of the Hanjiang River is 652 km, and the total drainage area of Hubei Province is 4.6 × 104 km 2 . The river Fig. 1 Distribution of water system and main hydrological stations in the middle and lower reaches of the Hanjiang River flows through Shiyan, Xiangyang, Jingmen, Xiaogan, Wuhan, and other cities. The regional economy is developed, and it is an important grain production base in China as well as a new industrial and automobile industrial base in Hubei Province (Kuman et al. 2014).
As the largest tributary of the Yangtze River and an important water source for the Middle Route Project of the South-to-North Water Transfer, the water quality of the middle and lower reaches of the Hanjiang River has gradually deteriorated since the 1990s. This is mainly because the concentrations of nutrients in the water body, such as nitrogen (N) and phosphorus (P), exceed the standard (Stephen et al. 2016;Cheng et al. 2019), resulting in a high degree of eutrophication and repeated outbreaks of algal blooms, during which the water body appears brown and emits a fish-like, pungent odor (Fig. 2). Algal blooms in the Hanjiang River occur mostly during the dry period (February to March) of late winter and early spring, and the duration varies from half a month to 1 month. According to incomplete statistics (Xin et al. 2020;Chen et al. 2007a, b;Lu et al. 2000) of various studies (Table 1), the algal bloom frequency of the Hanjiang River increased after 2000.. The rapid growth of algae is accompanied by an increase in the chlorophyll-a concentration; therefore, the phenomenon of algal blooms in the middle and lower Hanjiang River could be explored by studying changes in the chlorophyll-a concentration in the area.

Actual data and MODIS data acquisition
The measured data in the study include two parts. The first includes water quality data, including the chlorophyll-a concentrations monitored along three cross-sections (Baihezui, Qinguankou, and Zongguan) three times per month monitored in February, March, April, September, and October from 2009 to 2013, and the total nitrogen (TN) and total phosphorus (TP) concentrations monitored at the Zongguan cross-section from 2003 to 2009. The second includes the hydrological data, such as the flow, water level, flow rate, and water temperature monitored in the Xiantao and Zongguan sections from 2000 to 2011. These data were obtained from the Chinese Research Academy of Environmental Sciences.
The MODIS remote sensing satellite has seven spectral bands from 469 to 2130 nm and a spatial resolution of 500 m. Seventy-five MODIS remote sensing images synchronized with the measured water quality data from 2009 to 2013 were selected, and 64 valid remote sensing images (85.33% of the total number images) were filtered after rejection (NASA).

MODIS data preprocessing
The remote-sensing images were preprocessed using ENVI5.3 software (Exelis Visual Information Solutions Company). Geometric correction was first conducted to reduce the geometric distortion error. The ENVI automatic orthorectification tool RPC Orthorectification workflow (Guo, 2002) was used for the orthorectification of the remote sensing images. After geometric correction, the MODIS data saved the digital number (DN) value, while the chlorophyll-a concentration inversion used absolute radiation brightness and reflectivity. For follow-up studies, the DN value needed to be converted to the absolute radiance brightness and reflectance through radiometric calibration, which is calculated in ENVI according to Eqs. (1) and (2) (Lv et al. 2009;Carol et al. 2006).
To more accurately obtain the reflectance of ground objects (Feng, 2021;Zheng and Zeng 2004), the atmosphere in the image must be corrected to eliminate the influences of illumination, water vapor, and other factors on the reflectivity spectrum of ground objects. The FLAASH atmospheric correction module in ENVI was used to correct the images.
The normalized difference water index (NDWI) (Gao, 1996;Choung and Jo 2016) was used for extraction in the studied water areas. NDWI is less sensitive to atmospheric influence, less affected by the atmosphere, and more accurate for water extraction than the traditional normalized difference vegetation index (NDVI) method. The formula for calculating the NDWI is as follows: The reflectivity includes green light and near-infrared reflectivity.

MODIS data and chlorophyll-a concentration inversion method
The chlorophyll-a concentration significantly affects the spectral characteristics of water bodies and can, therefore, be used to establish a link with the spectral characteristics of remote sensing images (Le et al. 2010;Darecki et al. 2003). Empirical chlorophyll-a inversion methods are currently used more often, and the main fitted construction models include single-band, band ratio, and neural network models, which have relatively simple and feasible processes (Chen et al. 2007a, b).
In this study, the downloaded MODIS data were first corrected and processed to extract the band reflectance, and the water quality data from the measured water bodies were correlated with the synchronized remote sensing reflectance data. The reflectance data of the band or the combination of bands with high correlation were then screened out and fitted for model construction (Anatoly, 2008).

Band ratio model
The band ratio model combines the existing bands and selects the combination that is most relevant to the measured data to perform the inversion modeling (Le et al. 2009). According to the correlation analysis, the interference of the atmosphere in the electromagnetic waves and the influence of the smooth water surface on microwaves can be reduced by combining wave bands. In this study, the reflectance of seven bands from B1 to B7 was extracted from the preprocessed MODIS remote sensing images, and SPSS (BIM Company) was then used to combine each band in order to obtain 73 band combination patterns. There were 21 modes for the division of two bands (B1/B2), 20 modes for subtraction of the two bands (B1-B2), 15 modes for three bands (B1/(B2 B3)), 10 modes for four bands (B1/(B2 + B3 + B4)), and 7 modes for natural logarithm (ln (B1)). Correlation analysis between the band combination model and measured chlorophyll-a concentration was then performed, and a function model was established.

BP neural network model
The BP neural network method is currently one of the more widely used artificial neural networks, with the characteristics of forward signal and backward error propagation (Li and Liu 2000). It has the capacity for self-adaptation, self-learning, self-organization, and nonlinear mapping and is suitable for simulating various intricate nonlinear relationships. It has been proven that a network with deviation and at least one S-shaped implicit layer along with a linear output layer can approximate any rational function (Zhu et al. 2017).
Owing to the complex optical properties in the water column and the many factors affecting the chlorophyll-a concentration, the relationship between reflectance and the chlorophyll-a concentration cannot be accurately reflected using a simple linear model ). According to its characteristics, the BP neural network can better solve the issue of complex nonlinear relationships between reflectance and the chlorophyll-a concentration, as shown in Fig. 3, where X 1 , X 2 , … X n are the input and output values of the BP neural network model. In this study, the combination of the high correlation of wave reflectance with the chlorophyll-a concentration was used as an input factor,(X 1 , X 2 , … , X n ) for the BP neural network input layer, and the chlorophyll-a concentration of the BP neural network. The combination of band reflectivity and of the chlorophyll-a concentration was used to adjust the weights and thresholds of the network to generate the model.

Model validation method
To test the model precision and improve the accuracy of the inverse model, the model accuracy was evaluated by the decision coefficient R, relative error RE, and root-mean-square error RMSE using Eqs. (4), (5), and (6), respectively. The predicted value was the measured value. (4)

Results and analysis
Based on the MODIS data and chlorophyll-a concentration inversion method proposed in the "MODIS data and chlorophyll-a concentration inversion method" section, the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River was inverted, and the spatiotemporal and interannual variations in the chlorophyll-a concentration, interannual variations in the chlorophyll-a concentration distribution area, and interannual variation in chlorophyll-a concentration values in typical years were analyzed based on the inversion results. The specific results and analyses are as follows.

Correlation analysis results
The correlation between the reflectance and the measured chlorophyll-a concentration values was calculated ( Table 2). The B5 reflectance was significantly correlated with the measured chlorophyll-a concentration, with a Pearson correlation coefficient of − 0.645 and significance of 0.009. The B2 and B6 bands were significantly correlated with the measured chlorophyll-a concentration, and the Pearson correlation coefficients were − 0.55 and − 0.515, with significance levels of 0.033 and 0.049, respectively. Correlation analysis results of the 73 combinations of wavelength reflectance and the measured chlorophyll-a concentration values obtained using the rule of four are presented in Table (Table 3).

Modeling results
According to the results in the "Correlation and analysis result" section, the correlation between the B7/(B6 + B5) band combination and measured chlorophyll-a concentration value was high, and the Pearson correlation coefficient was 0.674, which is significant. Therefore, the B7/(B6 + B5) band reflectance was selected as an independent variable, and the measured chlorophyll-a concentration selected as  was a dependent variable. In SPSS, 11 common function models, including the linear, logarithmic, inverse, quadratic, and cubic function models, were used to construct the model with the variables. The calculation results are shown in Fig. 4 and Table 4.
To select the most suitable model, the coefficient of determination R 2 and significance of the 11 models were calculated, as shown in Table 4; the cubic function model R 2 value was 0.677, and the significance was 0.001. The cubic function model could predict the chlorophyll concentration in the middle and lower reaches of the Hanjiang River more realistically than the other 10 prediction models. Therefore, the cubic model was selected as the inversion model of the wave band ratio, and the calculation formula is as follows (7): where y is the prediction value of the chlorophyll concentration and x is B7/(B6 + B5).
In this section, 60 of 64 remote sensing images were randomly selected, which were extracted using ENVI5.3, and the band combination with the highest correlation with the measured chlorophyll-a concentration was selected as the input factor of the BP neural network model input layer according to Table 3 i.e., B7/(B6 + B5), B7/B5, B4-B2, and B4/(B3 + B2). Whether the BP neural network model can accurately solve the nonlinear problem accurately depends on the number of hidden layer node set in the neural network model and the number of nodes that directly affects the mapping ability of the neural network to the nonlinear problem (Hill et al. 1994). In this study, the node number of the final hidden layer was calculated by the trial and error method, and the results are listed in Table 5. According to Table 5, when the number of nodes was four, the judgment coefficient R 2 was 0.98918, and the RMSE was 3.1462. The principle of the neural network model shows that the higher the number of hidden layer nodes, the more complex the model, and the longer the required training time. Therefore, the number of hidden layer nodes was set to be four.  Finally, four waveband combinations of B7/(B6 + B5), B7/B5, B4-B2, and B4/(B3 + B2) were used as input factors for the BP neural network. The hidden layer was identified as a single layer, and its node number was set to four according to Table 3. The hidden layer transfer function was an S-shaped hyperbolic tangent function, and the output layer was the measured chlorophyll-a concentration value synchronized with the waveband combination. The maximum training time was set to 1000, and the learning rate was set to 0.001 of the 60 actual samples. Fifty-five samples were randomly selected for training, and the remaining five samples were used as test samples.
As shown in Fig. 5, when using the BP neural network model, the judging coefficient R 2 , RMSE, RE, minimum, and average RE values were 0.98918, 3.1462, 29.1%, 8%, and 21.4%, respectively, and the simulation forecasting effect was poor. Therefore, there was less error in the BP neural network model prediction than in that of the band ratio model, indicating that the inversion of the chlorophyll-a concentrations in the middle and lower reaches of the Hanjiang River was more accurate when using the BP neural network model. Additionally, this study confirmed that the BP neural network model was more suitable for simulating nonlinear problems originating from the complex optical properties of water bodies.

Spatiotemporal variation analysis of the typical annual chlorophyll-a concentration
The eruption of water blooms in 2000 and 2008 continued from late February to mid-March and from late January to early March, respectively. In this study, the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River during the bloom eruption period in 2000 and 2008 was inverted from the remote sensing image data of MODIS on February 24, March 4, and March 12 in 2000 and January 2, January 17, February 3, February 14, March 7, and March 14 in 2008. Additionally, the trend in the bloom area was analyzed. At a chlorophyll-a concentration of 0-10 μg/L, the water quality was good. When the chlorophyll-a concentration was 10-25 μg/L, there was no notable bloom. A bloom occurred at a chlorophyll-a concentration of 25-35 μg/L, and a severe bloom occurred at a chlorophyll-a concentration of > 35 μg/L. Figure 6 indicates that on February 24, 2000, the total chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River was relatively high, exceeding 25 μg/L with a wide banded distribution in the area near the monitoring section of the Yuekou River, and chlorophyll-a concentrations of < 10 μg/L were mainly distributed in the area near Zhongxiang City and the lower reaches of Shayang County. On March 4, 2000, the overall chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River continued to increase. The areas with chlorophyll-a concentrations of > 35 μg/L were distributed throughout the study area; these were more widely distributed in the upper reaches of Xiantao City and near Shayang County, exhibiting a band-like distribution, while in the remainder of the areas, chlorophyll-a rich areas were scattered along both riverbanks. The water bloom outbreak area was significantly larger than that on February 24 and March 4, spreading from Zhongxiang to Shayang, Yuekou to Xiantao, and gradually toward the center of the river channel from the shore. The concentration chlorophyll-a was > 35 μg/L in the lower reaches of Shayang County and upper reaches of Yuekou Town and was distributed in blocks along the west bank. The chlorophyll-a concentration decreased downstream of Yuekou  Figure 7 shows that from January 2 to 17, 2008, the overall chlorophyll-a concentration in the study area was not high. On January 3, the chlorophyll-a concentration in the downstream water body area of Zhongxiang City was relatively high. Areas with chlorophyll-a concentrations of 10-25 μg/L were distributed in a band from Zhongxiang City to Shayang County. On February 14, 2008, the entire middle and lower reaches of the Hanjiang River, lower reaches of Zhongxiang City, and some water areas near Yuekou Town had chlorophyll-a concentrations of < 10 μg/L, while the concentrations were higher near Shayang County. Concentrations of > 35 μg/L indicate that the water body is highly eutrophic and severe blooms are likely. On March 7, 2008, the chlorophyll-a concentration in the study area was > 35 μg/L. It was more evenly distributed throughout the study area and mostly concentrated on the shores. Areas where the chlorophyll-a concentration was > 25 μg/L were more densely distributed near the upper reaches of Zhongxiang City and Shayang County, where water blooms appeared. On March 14, 2008, areas with chlorophyll-a concentrations of < 10 μg/L began to gradually increase, areas with chlorophyll-a concentrations of > 35 μg/L decreased, and the algal bloom tended to dissipate.

Analysis of the interannual variations in the chlorophyll-a concentration distribution area
Preliminary investigations have found that water blooms occur in the middle and lower reaches of the Hanjiang River in February and March each year (Yang et al. 2012;Zhu et al. 2008). To explore the variations in the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River during the bloom period, the chlorophyll-a concentration distribution during the bloom period from 2000 to 2011 (from February to March) was inverted, and the results are reported in this section. As shown in Fig. 8, the chlorophyll-a concentration distribution accounted for the results in Fig. 9.
As shown in Figs. 8 and 9, the areas with chlorophyll-a concentrations of < 10 μg/L were relatively large in 2001, 2004, 2005, and 2009, accounting for 78.3%, 82.4%, 64.3%, and 75.8%, respectively, indicating that the chlorophyll-a concentration in most waters remained low during these periods, the water quality was generally good, and the risk of bloom outbreaks was relatively low. The trend line in Fig. 9 (1) shows that the area ratio with chlorophyll-a concentrations of < 10 μg/L fluctuated between 2000 and 2011, but there was a slight overall decreasing trend, indicating that the water quality in the middle and lower reaches of the Hanjiang River gradually improved during this period.   of > 25 μg/L were observed in areas occupying proportions of 57.9%, 70.2%, 49.3%, and 50.10%, respectively, indicating that the concentration of chlorophyll-a concentration was high, eutrophication was high, water quality was poor, and water bloom outbreaks occurred readily or had already occurred. The ratio of areas with chlorophyll-a concentrations of > 25 μg/L remained stable and low from 2001 to 2005, and remained stable but high from 2006 to 2008 when compared to the ratio of areas with chlorophyll-a concentrations of < 10 μg/L. According to the trend line, the chlorophyll-a concentration area ratio remained stable from 2000 to 2011 during the bloom-prone period.

Interannual variations in the chlorophyll-a concentration
To better reflect the interannual variations in the chlorophylla concentration values, the chlorophyll-a concentration values of five monitoring sections (Huangzhuang, Shayang, Yuekou, Xiantao, and Zongguan) between 2000 and 2011 were inverted; the interannual variations based on the inversion results are analyzed. The results are shown in Fig. 10, and the statistical results are presented in Table 6.
According to the statistical results, during the 12 years from 2000 to 2011, the highest chlorophyll-a concentration  section, the monthly average chlorophyll-a concentration values of > 30 μg/L at the Huangzhuang, Shayang, Yuekou, Xiantao, and Zongguan monitoring sections occurred three, five, two, four, and two times from 2000 to 2011, respectively, indicating that the probability of occurrence of algal blooms in the Shayang and Xiantao river sections was higher than that in the other sections and that it was predominant in spring.

Discussion
Algal bloom outbreaks refer to the phenomenon of excessive algae reproduction and aggregation under certain nutrition, climate, and hydrological conditions and ecological environments . Algal bloom outbreaks are ecological problem that not only depend on the nutrient conditions of algal growth but also on the ecological Inter-annual variation of chlorophyll-a concentration value in each monitoring section environment in which the algae grow. Furthermore, the relationship between algae and other aquatic organisms and the ecological environment, such as changes in the hydrological and meteorological conditions that affect these ecological relationships, also influences algal bloom outbreaks. In this section, the effects of environmental factors on algal blooms are discussed by analyzing the effects of environmental factors on the chlorophyll-a concentration.

Effect of water temperature of the chlorophyll-a concentration
Many studies have shown that most of the algal blooms in the Hanjiang River are diatom blooms Xu et al. 2005), and Chlorella has been considered to be the dominant species (Liang et al. 2012), with occasional cases where other green algae dominant. However, according to some other studies (Zheng et al. 2009;Yin et al. 2011), the dominant species of the Hanjiang water bloom belongs to the genus Corynebacterium, which can maintain good growth under low temperatures in winter.When the temperature was > 9 °C, diatoms began to grow, and they grew well between 15 and 35 °C; however, the optimal temperature range was 10-17 °C, which was beneficial for cell division and pigment accumulation in diatomaceous algae (Zheng, 2005), resulting in higher concentrations of chlorophyll-a, an indicator of algae presence. According to Fig. 11, taking Xiantao as an example, from February 2000 to March 2008, the average water temperature in February and March 2000 was 11.9 °C and in February and March 2008 was 11.25 °C, respectively; however, no bloom occurred in 2009 and 2010, at 9 °C and 10 °C, respectively. In contrast, the water temperature during the bloom eruption period was approximately 2 °C higher. Moreover, the average water temperature in February of these years when bloom occurred, like in 2000 and 2008, was even higher than the average water temperature in March of the years when there was no water bloom. Finally, the trend of water temperature was basically the same as the trend of chlorophyll-a concentration. Thus, the water temperature is a factor affecting the chlorophyll-a concentration.

Effect of hydrological conditions on the chlorophyll-a concentration
Hydrological conditions include water level, discharge, and velocity. The transfer of water from the South-North Water Transfer Project has transformed the hydrological situation in the middle and lower reaches of the Hanjiang River, and has tended to aggravate the phenomenon of algal blooms (Kuo et al. 2018;Zhou et al. 2017). According to Dou and Xie et al. (Xie et al. , 2005, the impact of the South-North Water Transfer Project on algal blooms in the middle  and lower reaches of the Hanjiang River is mainly reflected in the hydrological factors such as the flow rate and velocity. Therefore, the effects of the hydrological factors (including flow rate and velocity) on the changes in the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River should be analyzed to reveal the mechanism by which hydrological factors affect algal blooms in the Hanjiang River and to study the algal bloom conditions in eutrophic rivers. It has been shown that when blooms occur, the water level, discharge, and velocity of the middle and lower reaches of the Hanjiang River become significantly lower than in the period with no blooms (Wang et al. 2004). Bloom outbreaks in the Hanjiang River generally occur in February to March each year (early spring), which is the Hanjiang River dry season, when the water level is low and has good transparency. At this time, bloom diatoms can fully utilize sunlight for photosynthesis and accumulate photosynthetic products, thereby facilitating cell division and proliferation and causing an increase in the chlorophyll-a concentration (Tang and Wang 2001). According to the measured data, the average water level at the hydrometric Xiantao station in 2000, 2003, and 2008 was 23.71 m and the average discharge was 474.3 m 3 /s; these were, respectively, 1.67 m and 55.43% lower than those in 2001, 2004, and 2005. The flow rates at the Zongguan section in , and March 2008.230 m/s, respectively, and the average flow rate was 0.208 m/s, according to the flow data and the flow rate calculated from the cross-sectional area of the diatom bloom. In contrast, the average velocity at the Zongguan section from January to March was 0.289 m/s, which decreased by 28.02%. As shown in Fig. 12, when chlorophyll-a concentrations were above 20 μg/L, the discharge were all in the range of 200-700 m 3 /s and the water levels were in the range of 28 m or less. At the same time, with the increase of discharge and water level, chlorophyll-a concentration decreased. In addition, taking the 2008 water bloom period at Xiantao hydrological station as an example, as shown in Fig. 13, chlorophyll-a concentrations rose by 27.8 μg per liter during the 17-day period from February 3 to 21, at which time discharges fell by 136.32 cubic meters per second during the 17-day period; during the 7-day period from March 7 to 13, the discharge increased from 497.33 m 3 /s to 721 m 3 /s, at which time the chlorophyll-a concentration decreased from 23.61 μg/L to 14.20 μg/L. Therefore, the variations in the chlorophyll-a concentration are affected by hydrological factors such as water level, discharge, and velocity. (Table 7 and 8)

Effect of nutrients on chlorophyll-a concentration
N and P are the main nutrients required for phytoplankton reproduction and growth (Lin et al. 2021;Liu et al. 2019).
Owing to the large variations in the chlorophyll-a concentration values in the middle and lower reaches of the Hanjiang River from 2003 to 2009, the inverted chlorophyll-a concentration values at the Zongguan monitoring station were Mar. Average Fig. 12 Relationship between chlorophyll-a concentration and water level and discharge in Xiantao hydrological station compared with the measured TN and TP values; the results are shown in Fig. 14. The variations in the chlorophyll-a concentration were similar to those of N and P during all years excluding 2005; that is, the increases in the chlorophyll-a concentration values were accompanied by increases in N and P concentrations. This shows that the N and P contents of the water column during algal blooms could meet the proliferation needs of diatoms in the Hanjiang River and the diatom growth trend improved with increases in the N and P concentrations, which increased the chlorophyll-a concentration in the water column. However, in 2005, low concentrations of N and P were equally accompanied by high concentrations of chlorophyll-a, numerous studies (Luo et al. 2021;Jing et al. 2019;Guenther et al. 2015) have also demonstrated that high N or P concentrations may lead to uncoordinated N-P ratios in the water column, thereby affecting the growth of diatoms in the water. Although the concentrations of N and P were lower in 2005, it is possible that their ratios were suitable for algal growth, and it is reasonable that the 2005 situation occurred in conjunction with the temperature effects in the "Effect of water temperature of the chlorophyll-a concentration" section and the flow rate at the Zongguan hydrographic station in "Effect of nutrients on chlorophyll-a concentration" section. Therefore, the variations in the N and P nutrient concentrations and their ratio are the main factors affecting the variations in the chlorophyll-a concentration. In addition, there are many factors affecting the outbreak of water blooms in the middle and lower reaches of the Hanjiang River, so more research needs to be conducted.

Conclusions
Based on the correlation analysis results, the BP neural network model was used to invert the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River in 2000-2011 and analyze its spatial and temporal variations. The results indicated that (1) band combinations B7/ (B6 + B5), B7/B5, B4-B2, and B4/(B3 + B2) were correlated with the chlorophyll-a concentration, with Person correlation coefficients of 0.674, 0.679, 0.635, and 0.646, respectively.
(2) The coefficients of determination R 2 of the BP neural network model and the band ratio model were 0.98918 and 0677, respectively; the BP neural network model inversion achieved a better fitting effect and more accurate inversion results and was more suitable for studying inland complex water bodies.
(3) Temporally, the algal outbreaks were mostly concentrated in February and March, with a few cases occurring in October, and the duration ranged from 20 to 40 days. According to the calculation results of the chlorophyll-a concentration distribution in the middle and lower reaches of the Hanjiang River, the area with chlorophyll-a concentrations of > 25 μg/L was relatively large in 2000, 2006, 2007, and 2008, in which the chlorophyll-a concentration in most waters was relatively high, eutrophication was relatively high, and the water quality was generally poor. (4) Spatially, the high chlorophyll-a concentration caused by the algal bloom mostly occurred in Zhongxiang, Shayang, and upstream Xiantao sections, spread from north to south in the downstream direction, and tended to gather in the river bend, showing a trend of spreading from the river bank to the center. (5) By analyzing the influence of environmental factors on the chlorophyll-a concentration, when the water temperature was approximately 2 °C above the average, the chlorophyll-a concentration was found to be relatively high. Additionally, decreases in the water level and flow rate resulted in higher chlorophyll-a concentrations. Finally, the concentration and ratio of nutrients, including N and P, were the main factors leading to an increase in the chlorophyll-a concentration in the middle and lower reaches of the Hanjiang River.