Open Access
Add to BibSonomy
Abstract
The Trophic state index (TSI) is a vital parameter for aquatic ecosystem assessment. Estimating TSI by remote sensing is still a challenge due to the multivariate complexity of the eutrophication process. A comprehensive in situ spectral-biogeochemical dataset for 7 lakes in Northeast China was collected in October 2020. The dataset covers trophic states from oligotrophic to eutrophic, with a wide range of total phosphorus (TP, 0.07–0.2 mg L−1), Secchi disk depth (SDD, 0.1–0.78 m), and chlorophyll a (Chla, 0.11–20.41 μg L−1). Here, we propose an empirical method to estimate TSI from remote sensing data. First, TP, SDD, and Chla were estimated by band ratio/band combination models. Then TSI was estimated using the Carlson model with a high R2 (0.88), a low RMSE (3.87), and a low MRE (6.83%). Synergistic effects between TP, SDD, and Chla dominated the trophic state, changed the distribution of light in the water column, affected the spectral characteristics. Furthermore, the contribution of each parameter for eutrophication were different among the studied lakes from ternary plot. High Chla concentration was the main reason for eutrophication in HMT Lake with 45.4% of contribution more than the other two parameters, However, in XXK Lake, high TP concentrations were the main reason for eutrophication with 66.8% of contribution rather than Chla and SDD. Overall, the trophic state was dominated by TP, and SDD accounted for 85.6% of contribution in all sampled lakes. Additionally, we found using one-parameter index to evaluate the lake trophic state will lead to a great deviation, even with two levels of difference. Therefore, multi-parameter TSI is strongly recommended for the lake trophic state assessment. Summarily, our findings provide a theoretical and methodological basis for future large-scale estimations of lake TSI using satellite image data, help with water quality monitoring and management.
© 2022 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement
1. Introduction
Inland water quality research has become increasingly important to manage water resource shortages, drinking water safety, and the ecological sustainability of aquatic systems [1,2]. Eutrophication is a phenomenon of water quality deterioration, due to increases in excessive nutrients. This process is often associated with the rapid production of phytoplankton and other microorganisms, which have important impacts on aquatic ecology and the normal functioning of water bodies [3,4]. This process alters the optical properties of the water column as well, including changes in Chla concentration and transparency [5]. The trophic state of inland waters is typically categorized into three levels: oligotrophic, mesotrophic, and eutrophic. A multi-parameter trophic state index (TSI) was addressed by Carlson (1977) to simplify this complex process. This index has been widely used to assess water quality and trophic level, classify water type, and aid lake management over the past 40 years [6–8]. Carlson’s TSI model based on the empirical relationships among the total phosphorus (TP), chlorophyll a (Chla) content, and secchi disk depth (SDD). Phosphorus (P) is an important substance for the growth of phytoplankton. Chla, a pigment common to almost all photosynthetic organisms, as a proxy for algal biomass. SDD is affected by the presence of suspended sediment, CDOM, planktonic algae, and zooplankton [9]. These three parameters can not only reflect the nutrient state of water body, but also are important parameters in the field of water color remote sensing. Because of their role of bridges linking trophic state and water-leaving reflectance, the remote sensing technology can be employed for water trophic state assessment.
Over the past four decades, remote sensing technology combined with various advanced computer algorithms and multi-source satellite images has made an unprecedented rapid development, and become a valuable technique for TSI monitoring from time and space [8,10–13]. However, because of the satellite platform, sensor performance, atmospheric conditions, and many other irresistible factors, satellite remote sensing approaches have their own limitations. To find a universal applicability algorithm to monitor the TSI, using in-situ hyperspectral data is considered the best choice without considering the factors mentioned above. Undoubtedly, the large-scale continuous hyperspectral resolution can capture the detailed and variable characteristics of inland waters for development of TSI inversion algorithm. With this advantage, various algorithms based on hyperspectral have been developed to estimate the TSI with high estimation accuracy [12,14,15].
Most of these algorithms are based on the relationship between the remote sensing reflectance of characteristic bands and TSI parameters. In recent years, attempts have been made to quantitatively evaluate the trophic state of inland waters using both single-variable or multi-variable methods. [16–18]. Chlorophyll a is an important indicator of phytoplankton biomass, a key parameter for assessing the water quality of eutrophic ecosystems [18,19]. Chlorophyll a presents unique spectral characteristic with noticeable peaks in the blue (nearly at 440 nm) and red wavelengths (at nearly 675 nm), representing the physical basis for Chla estimation from blue-to-green ratios or red-to-near infrared (NIR) ratios of remote sensing reflectance of inland waters. Moreover, three-band, four-band, quasi-analytical algorithm (QAA), and machine learning algorithms have been developed to derive Chla [11,13,15,20]. More studies chose to use Chla to indicate the trophic level of water rather than SDD or TP [2,13,21]. It is rare only to use SDD to indicate trophic levels. Generally, SDD retrieval algorithms were developed using empirical method, adopting in situ measurements to calibrate regression models with signatures in the visible and near-infrared spectral domain [22,23]. TP cannot be directly retrieved from remote sensing data because it has no optical responses. Its retrieval algorithms are mainly based on empirical modeling [24,25] or are developed depending on relationships between nutrients and optically active constituents, such as Chla and total suspended matter (TSM) [14,26,27]. In summary, these studies have made a lot of effort in algorithm development, including empirical algorithm, semi-analytical algorithm and machine learning algorithm. However, more studies used single-parameter to indicate the trophic status of water body, especially Chla, yet this ignores the contributions from other parameters (such as, TP and SDD) to the trophic status evaluation. Especially for turbid water, the contributions of TP and SDD may be greater than that of Chla. Therefore, it is necessary to develop a multi-parameters method to balance the contributions of various factors to accurately monitor the lakes trophic status.
Compared with the previously reported paper, we tried to develop a multi-parameter remote sensing estimation method for trophic status in inland waters, and analyze the contribution of each parameter to eutrophication for the first time. The overall goal of this study was to collect a comprehensive spectral-biogeochemical database for the northeast lakes of China and evaluate the ability of the empirical modeling approach to retrieve water quality parameters and the TSI. Our specific objectives were to (1) investigate the variation of reflectance spectra characteristics under different classes of trophic state in lakes, (2) determine optimal relationships of TP, Chla, and SDD with in situ collected spectra using an empirical modeling approach, and estimate the TSI of lakes, and (3) analyze the contribution of those three parameters to eutrophication using normalization and ternary plot methods.
2. Materials and methods
2.1 Study area
Northeast China (115°E–135°E in longitude and 38°N–53°N in latitude) is a large administrative region with the total area of 1.24 million square kilometers and encompassing Heilongjiang Province, Jilin Province, Liaoning Province, and the eastern part of Inner Mongolia Autonomous Region (Fig. 1). The climate of region is dominated by temperate continental monsoon climate, the winter is long and cold, the summer is short and warm. Northeast China is the agricultural center of China. Known as the “granary of China”, Northeast China is rich in rice, corn, soybeans and other crops. This area is rich in water resources, including freshwater lakes/reservoirs, rivers and ponds. However, under the influence of agricultural activities, eutrophication has taken place in some northeastern lake recent years, thus, seven typical lakes are selected according to the gradient of trophic status from oligotrophic, mesotrophic, to eutrophic. They are Hamatong reservoir (HMT), Xiaoxinkai Lake (XXK), Daxinkai Lake (DXK), Jingbo Lake (JBL), Songhuahu Lake (SHH), Xinmiaopao Lake (XMP), and Yueliangpao Lake (YLP). Their locations are shown in Fig. 1 (a). The basic physical and water quality parameters of each water bodies are listed in Table S1.
Fig. 1. Locations of the 7 lakes in study area in northeast of China (a) and sampling site distributions in each lake HMT (b), XXK and DXK (The map only shows the border of Daxingkai Lake in China) (c), JBL (d), SHH (e), XMP (f), and YLP (g) for the typical sampling process.
2.2 Field campaigns and in situ measured data
Due to weather problems, this study only collects water samples once in late autumn, because there is little rainfall, low wind speed and less cloud cover in this season, which is most suitable for field investigation with boats. Two field campaigns were conducted in late September and early October. Therefore, this paper does not involve the study of seasonal changes. Based on our previous survey, seven lakes in northeast China were selected for their dispersed geographical location, sufficient quantities to cover all trophic levels, and the enough range scale of their parameter concentrations. Surface water samples were collected using the brown plastic bottles at 74 stations distributed over 7 lakes from 12th to 20th of October 2020 (Fig. 1). Then, they were immediately refrigerated at -4℃ in car refrigerator for preservation before laboratory analysis. The samples used for determination of total phosphorus shall be determined within 48 hours, and the samples used for determination of chlorophyll a shall be filtered on the night of sampling, then the filter membrane with chlorophyll a were frozen at -24℃, the determination shall be completed within one week. During sampling, coordinates were recorded using a Uinistrong GPS receiver, Secchi disk depth (SDD) was measured at each sampling point using a black-and-white disk. Water reflectance spectra were concurrently measured using a portable ASD FieldSpec spectroradiometer. With a $25^\circ $ viewing field, this instrument has a sensitivity range from 400 to 1075 nm at an increment of 1 nm with 675 wavebands. The measurement of water surface reflectance spectra followed the method described by Mueller et al. [28] and Jiao et al. [29]. Sky conditions were also recorded for each station when spectral measurement, including wind speed and wind direction. All of the spectra were collected under cloudless weather conditions. In order to effectively avoid the interference of the ship with the water surface and the influence of direct solar radiation, the instrument was positioned at an angle of 90 –135° with the plane of the incident radiation away from the sun. We collected water leaving radiance at approximately 1 m above the water surface (Lsw) with view angle of 30–45° from the nadir; radiance above a standard white Spectralon (Labsphere, Inc., North Sutton, NH) at about 20 cm above reference panel (Lp); and the collated skylight radiance (Lsky) with view angle of 135° from the Sun. In this way most of the direct sunlight was eliminated while the impact of the ship’s shadow was minimized.. Next, the remote sensing reflectance [${R_{rs}}(\lambda )$] was calculated as:
2.3 Biochemistry parameters and TSI
Chlorophyll a (Chla) concentration was determined using an acetone extraction based method and a spectrophotometer (UVe2660PC Shimadzu Inc., Kyoto), details can be found in our previous work [31]. The 150 to 200 mL of water sample was filtered (0.45 μm mixed fiber millipore filters, Bandao Industrial Co., Ltd, China), and the filter membrane was soaked in 90% acetone solution for 24 h. The supernatant was then analyzed for Chla concentration using a UV-VIS spectrophotometer. Water samples were analyzed for total phosphorus concentration (TP, mg L−1) using a continuous flow analyzer (SKALAR, San Plus System, the Netherlands) according to the method found in APHA/AWWA/WEF [32]. The trophic status of the selected lakes was determined based on the modified Carlson's trophic status index (TSI), and the equations listed below (Eq. (2)-5) [16]. Measured water quality parameters (Chla, TP, and SDD) were used in these computations. According to the standard trophic categories [6,31], TSI < 30 indicates oligotrophic, 30 ≤ TSI ≤ 50 indicates mesotrophic, and TSI > 50 indicates eutrophic. In the eutrophication level, TSI is further subdivided into light eutrophic (50≤TSI<60), moderate eutrophic (60≤TSI<70) and hyper-eutrophic (TSI≥70)
2.4 Modeling approaches
An empirical approach was used to remotely estimate TSI basing on in situ collected spectra data and the measured concentration of each parameter in this study. The model took the remote sensing reflectance of the band as the independent variable, and each parameter concentration of TSI as the dependent variable, and used the least square method to find the best fitting degree and determination coefficient. 2/3 of the synchronous matching data between in situ spectra data and the measured concentration of each parameter to construct the models. The remaining 1/3 of the samples were used for models validation. The approach had two major steps: first, extracting optimum band by correlation analysis, building model with the relationship between bands combination/band ratio and the concentration of each parameter using linear and non-linear fitting. Second, the estimated values of the three parameters were put into the TSI calculation formula to get the estimated TSI, then the TSI calculated from measured values are used for validation. We can also get an estimate of the TSI for each parameter in this process, including TSI(TP), TSI(SDD), and TSI(Chla). The details about model establishing were showed in section of Result 3.3.
2.5 Statistical analysis and accuracy assessment
Statistical descriptive, for example, mean, standard deviations, and data analysis for linear and non-linear regressions were performed using the OriginPro 2016 software package (Origin Lab Corporation, Northampton, MA, USA). Statistical significance was considered at the p < 0.05 level. Contribution analysis was performed using ternary plots, which assume that the data are normalized (TP + SDD + Chla = 1 or 100%).
Three indexes were used to evaluate the performance of models: coefficient of determination (R2), the slop of linear equation, root mean square error (RMSE), and mean relative percentage error (MRE). The expression equations are as follows:
3. Results
3.1 Biogeochemistry and TSI
The studied lakes spanned a wide range of TP (0.07–0.2 mg L−1), SDD (0.1–0.78 m), and Chla (0.11–20.41 μg L−1), and covered trophic states from oligotrophic to eutrophic, with the TSI ranging from 26.84 to 64.64. Regarding the whole dataset, the selected lakes with an average TSI of 48.31 ± 11.18 (Table 1), the situation is different for each individual lake. The lakes XXK and DXK were at light eutrophic levels, had higher TP concentrations, lower SDDs, lower Chla concentrations. The highest Chla values were observed at HMT, with the highest TSI value at a moderate eutrophic level. All parameters were at a moderate level at YLP and XMP, which were mesotrophic. SHH was the only lake close to the oligotrophic level with lower TP, the highest SDD, and the lowest Chla. SHH had the lowest TSI value among the study lakes (Table S1 and Fig. 2). According to Box-plots (Fig. 2), the TP concentrations, SDD, and Chla concentrations showed gradient variation following the trophic state. Therefore, the dataset in this study was suitable and reasonable for our studying due to its extensive range, multi-dimensional nature, and multi-scale features. In addition, there was no correlation between Chla and TP (r = 0.086, p > 0.05), a strong negative correlation between SDD and TP (r = -0.752, p < 0.01), and a moderate negative correlation between Chla and SDD (r = -0.368, p < 0.05) in this dataset (Table S2).
Fig. 2. Box-plots of the total phosphorus (a), Secchi disk depth (b), chlorophyll a (c) and trophic state index (d) for each lake. The box boundaries indicate the 25 th and 75 th percentiles; the whiskers represent the 10 th and 90 th percentiles; the inner horizontal line is the median; the inner hollow circle is the mean; × represent outliers; and the solid short lines indicate the minimum and maximum values.
Table 1. Descriptive statistics of the total phosphorus (TP), Secchi disk depth (SDD), chlorophyll a (Chla), and trophic state index (TSI) for the whole dataset (n=74). Min is minimum value, Max is maximum value, Mean is average value, SD is standard deviation.
3.2 Spectral characteristics under different trophic states
According to in situ TSI, collected remote sensing reflectance spectra were separated into trophic classes, as shown in Fig. 3. It can be seen that, with increasing nutrient levels, the shape of the spectral curve changed. The maximum reflection peak was observed near 580 nm for oligotrophic lakes, but for mesotrophic lakes, the range of the maximum reflectance band was elongated, and a plateau appeared between 590 and 700 nm. The diagnostic absorption trough of chlorophyll (680 nm) were not clear in oligotrophic and mesotrophic lakes, but began to emerge, and increasingly evident in the eutrophic lakes. Sensitive bands of SDD at visible bands differed between oligotrophic lakes (high ${R_{rs}}(\lambda )$) and eutrophic lakes (low ${R_{rs}}(\lambda )$. However, there were no obvious spectral features for TP. Additionally, the spectral characteristics of XXK, DXK, and JBL lakes were dominated by suspended matter from spectra curves.
Fig. 3. Reflectance spectra for different trophic state in seven lakes. Oligotrophic lake for SHH (a), mesotrophic lake for JBL (b), light eutrophic lakes for DXK, YLP and XMP (c), moderate eutrophic lakes for XXK and HMT (d).$\textrm{\; }{\textrm{R}_{\textrm{rs}}}(\mathrm{\lambda } )$ is remote sensing reflectance of water surface at wavelength $\mathrm{\lambda }$ nm.
To select the diagnostic spectral variables, correlation analysis was applied between original reflectance, spectral derivative and TP, Chla, and SDD for each wavelength from 400 to 900 nm (Fig. 4). Derivative analysis was the major approach for empirical or semi-empirical model to retrieve water quality parameters based on the absorption trough or scattering peak spectral region [33,34]. The original reflectance spectra apt to sun–target–sensor geometry variation and ambient environment factors (partial cloud, wind speed, etc), and derivative is more effective to reduce these factors as shown below. There was a broad region with relatively low correlation coefficients in the green spectral regions and higher correlation coefficients in the blue (around 443 nm) and the near-infrared (NIR) regions for TP and SDD. A narrow region around 710 nm (chlorophyll-a fluorescence baseline and red edge transition location) had a high correlation coefficient. Both of TP and SDD presented an opposite and symmetric trend, indicated SDD acted as a bridge for the correlation between TP and remote sensing reflectance. However, there was no significant correlation for Chla, the highest value appeared after 710 nm and only reached about 0.24, with an abrupt change at 680 nm (absorption peak of Chla).
Fig. 4. Correlation analysis on original reflectance spectra (a) and first derivative (b) against TP, SDD, and Chla for each wavelength from 400 nm to 900 nm. The dash line indicated the sensitive bands with strong variation of correlation coefficient (r).
The first derivative of reflectance showed a relatively high correlation coefficient in the green and red edges, especially in the red edge region (680–710 nm) for all parameters (Fig. 4(b)). The highest correlation coefficients for Chla were observed in the red edge spectral region (chlorophyll-a florescence peak), ranging from 0.75 to 0.82, further confirming that Chla could be retrieved from remote sensing reflectance at 680 and 710 nm. TP and SDD still present completely opposite trend, the highest correlation coefficients appear around 575 and 710 nm, indicating that spectral reflectance derivative analysis is an effective approach for water quality parameter quantification [33,34].
3.3 Model calibration and validation
Based on the correlation analysis above, some sensitive bands to these water quality parameters were found, i.e., 443 nm, 575 nm, 680 nm and 710 nm. Their optimal band combination/band ratio models for calibration and validation are shown in Fig. 5. TP could be accurately retrieved by the band combination $\{{[{{R_{rs}}({443} )+ {R_{rs}}({710} )} ]/{R_{rs}}({575} )} \}$ with a low MRE and the slope and intercept for the 1: 1 line close to unity and zero (R2 = 0.804, slop = 0.605, MRE = 11.03%, RMSE = 0.017 mg L−1).The band ratio method $[{{R_{rs}}({544} )/{R_{rs}}({675} )} ]$ was also used for TP estimation reported by [32], but did not achieve good modeling accuracy (R2 = 0.39, RMSE = 0.021 mg L−1). Besides, a hybrid model combining genetic algorithms and partial least squares (GA-PLS) for the remote estimation of TP obtained high accuracy (R2 = 0.89, RMSE = 0.019 mg L−1) [14]. In our study, SDD could also be accurately retrieved by the band combination $\{{[{{R_{rs}}({443} )+ {R_{rs}}({710} )} ]/{R_{rs}}({575} )} \}$ with a low MRE and the slope and intercept for the 1: 1 line close to unity and zero (R2 = 0.848, slop = 0.932, MRE = 28.04%, RMSE = 0.089 m). Yu et al. [35] proposed an empirical method using a single band, ${R_{rs}}({565} )$, for the inversion of SDD, which had an R2 = 0.71 and RMSE = 0.61 m. Similarly, Yan et al. [36] reported a band ratio method $[{{R_{rs}}({620} )/{R_{rs}}({531} )} ]$ for the retrieval of SDD with an R2 = 0.6 and RMSE = 0.31 m. These green and red edges could be used as diagnostic bands for monitoring SDD (Table 2). Introducing the blue band ${R_{rs}}({443} )$ led to an improved decision coefficient with 0.804 for TP and 0.848 for SDD in our study. As expected, Chla retrieval by the band ratio $[{{R_{rs}}({710} )/{R_{rs}}({680} )} ]$ had high accuracy (R2 = 0.92, slop = 1.11, MRE = 42.43%, RMSE = 1.53 μg L−1), consistent with previously reported band ratios [11,19], indicating the good performance of the band ratio model.
Fig. 5. Model calibration using relationship between remote sensing reflectance of band combination and measured values of three parameters for total phosphorus (a), Secchi disk depth (c), and chlorophyll a (e). Scatter plots of measured and estimated values for each parameter in (b), (d), and (f), respectively.
Table 2. Comparison of algorithm accuracy on estimated total phosphorus (TP), Secchi disk depth (SDD), and chlorophyll a (Chla) with in situ hyperspectral data from different models. ${{\mathbf R}_{{\mathbf{rs}}}}({\mathbf \lambda } )$ is remote sensing reflectance of water surface at wavelength ${\mathbf \lambda }$ nm, ${{\mathbf R}^2}$ is the determination coefficients of regression fitting for validation of models, RMSE is root mean square error for assessment performance of model using measured and modeled values.
The relationship between the estimated and measured TSI for each parameter is shown in Fig. 6(a), (b), (c). All of them provided high R2 and low RMSE. Estimation and measurement values of the comprehensive trophic state index combining the three parameters (TSI) are shown in Fig. 6(d). There was a good performance with a high R2 (0.88), low MRE (6.83%), and a low RMSE (3.87). In addition, we found that when different parameters were used for the evaluation of trophic states, the differences in results were obvious (Fig. 7). According to Fig. 7, that TSI(Chla) was mainly distributed in the low-value region, the TSI(TP) was in the middle-value region, and TSI(SDD) was in the high-value region for all sampled lakes. Accordingly, the same lake, e.g., Lake DXK, was at the mesotrophic level when evaluated with TSI(Chla) (35.36 ± 4.04) but at the hyper-eutrophic level when evaluated with TSI(SDD) (87.13 ± 4.21). This deviation is in accord with results reported by [41]. However, the comprehensive trophic state index (TSI) was in an intermediate position, representing a tradeoff between the different one-parameter indexes.
Fig. 6. Scatter plots of trophic state index calculated by measured total phosphorus and estimated total phosphorus TSI(TP) (a), by measured Secchi disk depth and estimated Secchi disk depth TSI(SDD) (b), by measured chlorophyll a and estimated chlorophyll a TSI(Chla) (c), and by all measured parameters and estimated parameters TSI (d).
Fig. 7. Trophic state index calculated by measured total phosphorus TSI(TP), by measured Secchi disk depth TSI(SDD), by measured chlorophyll a TSI(Chla), and by all measured parameters TSI of each lake for different class of trophic state in northeast of China.
3.4 Dominant determinants of eutrophication
From the previous analysis, we know that different lakes may be at the same trophic state level due to the synergy of the three parameters. Therefore, a ternary scatter diagram was used to analyze the contribution of each parameter to trophic state in the different lakes (Fig. 8). The three parameters were standardized to unify the unit before ternary scatter diagram analysis. Then data were normalized again to a sum of 1or 100%. In the same trophic level (Lakes HMT and XXK), Chla had the largest contribution (45.4%) for HMT Lake, and TP had the largest contribution (66.8%) for XXK Lake (Table 3), indicating that Chla and TP were the dominant factors causing eutrophication in these lakes. Similarly, for light eutrophic lakes, TP was key factor in DXK and YLP with 71.2% and 42.7%, whereas SDD dominated in XMP with 50.4%. For mesotrophic lakes, JBL was simultaneously controlled by TP and SDD with 47.1% and 43.1%. For oligotrophic lakes, high SDD was the leading cause of oligotrophic in SHH. In summary, the trophic levels of selected lakes in Northeast China were generally controlled by TP and SDD (45.3% and 40.3%) rather than Chla (14.4%).
Fig. 8. Plots of Ternary Scatter (a) and contour (b) for measured total phosphorus (TP), measured Secchi disk depth (SDD), and measured chlorophyll a (Chla). Colour indicates the class of measured trophic state index from oligotrophic to hyper-eutrophic (from 25 to 65). Ternary plots assume that the data are normalized (TP + SDD + Chla = 1or100%).
Table 3. Mean values of relative proportion (scaled between 0 to 1) of total phosphorus (TP), Secchi disk depth (SDD), and chlorophyll a (Chla) in water column from ternary plots in different lakes, and whole datasets. Ternary plots assume that the data are normalized (TP + SDD + Chla = 1 or 100%).
4. Discussion
4.1 Synergistic effects of TP, SDD and Chla
Substances in the water bodies do not exist independently, and they are interrelated and in constant biogeochemical transformations [42,43]. TP, SDD, and Chla are important water quality parameters, and there are two opposite causal relationships between them [7]. If the TP concentration increases in the water, algae will thrive, leading to increases in chlorophyll a concentration, followed by a decrease in SDD. Suppose there is an increase in suspended particulate matter (SPM) (organic or inorganic) in the water, in that case, SDD decreases, which will reduce phytoplankton photosynthesis in the water column and lead to a decrease in standing stocks of Chla [41]. In this case, there will not be much Chla in the water, even though TP is high. Most lakes fall into the former case, especially the lakes of southern China [15,17,18]. However, our results are highly consistent with the later phenomenon because the selected lakes showed high turbidity, low SDD, low concentrations of Chla and high concentrations of TP. TP was indeed highly correlated with total suspended matter, especially regarding the suspended particulate inorganic matter (SPIM) [44,45]. At the same time, SDD was also highly negatively correlated with total suspended matter [23], explaining why TP and SDD showed a highly negative correlation in our study. Accordingly, SDD and TP may be the central problems causing lake eutrophication in northeast China, rather than processes involving Chla. Contribution analysis confirmed this conclusion, in which TP and SDD accounted for 85.6% across all lakes.
In general, biochemical processes dominated the synergistic effect between TP, Chla, and SDD, and this synergistic effect, in turn, influenced the spectral characteristics and the trophic state of the water bodies. With decreasing SDD and increasing TP concentration, the reflection peak of the red band changed from sharp to flat. The Chla fluorescence peak become increasingly apparent with the increase in Chla concentration. Moderately eutrophic lakes exhibit higher TP concentrations or higher Chla concentrations. SDD plays a more critical role in mesotrophic and oligotrophic lakes, which greatly improving their nutritional level.
4.2 Limitations of single parameter evaluations
Over the past four decades, efforts have been devoted to classifying the trophic states of different aquatic ecosystems. Various classification methods have been proposed, such as Carlson’s TSI model [16], and revised the TSI models including TN, CODMn, water color, and DOC [21,46–48]. The Trophic Level Index (TLI), another commonly used numerical method, is calculated from the weighted sum of either three variables (Chla, TP, total nitrogen (TN)) or five variables with the addition of SDD and chemical oxygen demand (COD) [49–52]. Some studies have used a combination of multiple parameters to evaluate trophic states. Still, a large number of studies have used a single parameter to evaluate the tropic status of lakes, e.g., only the TSI(Chla) [2,13,21] or the TSI(SDD) [23] or the TSI(TP) [14,44]. However, according to our results, evaluations with a single parameter can cause a significant error, sometimes even up to two levels of difference. In general, the trophic level of surface waters is affected by synergistic interactions between multiple parameters, and the dominant factor also differs among lake types. Especially in high turbidity water, using the TSI(Chla) may result in an oligotrophic evaluate, but using the TSI(SDD) may assess them as hyper-eutrophic. Therefore, our article makes a strong case using multi-parameters.
4.3 Limitations and implications for monitoring TSIs
With the development of remote sensing technology, various remote sensing data sources and a series of algorithms have been established to determine the trophic state of inland waters, either based on satellite-derived or in situ hyperspectral data. In this paper, we use in situ hyperspectral data to develop the TSI estimation method for northeast of China. Limitations of the method proposed in this paper came from dataset itself. Our dataset possessed more regional characteristics. Although the selected 7 lakes cover the whole trophic state, average value of the chlorophyll value is generally low, lacking coverage in the high-chlorophyll value lakes. Therefore, our TSI estimation methods only applicable to turbid waters with low chlorophyll-a concentration. Extrapolation of the model is a common problem for remote sensing inversion research of inland waters, due to the differences in geographical conditions, climate, surrounding environment and optical properties of water body itself. In addition, the dataset lacks seasonal sampling. Season is an important factor affecting eutrophication degree of water body. For example, the growth of algae in summer, the influx of nutrients caused by rainfall, the re-suspension of suspended matter caused by wind and so on. Due to our field campaigns were conducted in October, the estimation model proposed in this paper only work for the late autumn season, whether it works in other seasons needs further study.
Table 2 shows our results and several reported classic TSI water quality parameter inversion models and their accuracy indexes. Algorithms include the empirical model band ratio method [8,11], the Peak-Height Algorithm (MPH) [11,53], the Forel-Ule index (FUI) [10], the Quasi-Analytical Algorithm (QAA) [12], the Genetic Programming (GP) method [13], and Partial Least Squares (PLS) [14]. Our algorithms presented a good performance to estimate water TSI parameters, which delivered high R2 (0.77, 0.81, 0.89 for TP, SDD, Chla) and low RMSE (0.017, 0.09, 1.53). Due to the heterogeneity of time and space, it is difficult to compare the accuracy of models from different seasons and regions. Each area and dataset may have its range of performance, because of training and validation with different datasets. The number of samples, the range of samples, and the composition of planktonic algae and suspended matter in the datasets will change at any time and place, results in differences in the optical properties of water bodies. In addition, we conducted the training and validation of the band ratios or single band algorithms from references using the same datasets of this study. As can be seen from the results (Fig. S1), our algorithm is better in both R2 values in the calibration process and RMSE values in the validation process. This proves that the models proposed are robust and reliable with theoretical basis. We found that the band ratio algorithm is still the most used, due to its high estimation accuracy and strong robustness among various algorithms. In general, any model has advantages and disadvantages. Some models do not achieve high accuracy, which may not be the problem of the model itself, but due to the sampling design. Differences between models may be due to the mathematics used, and it is hard to say which model is better. In pursuit of improving model precision, highly complicated models have been elaborated that do not have an actual meaning. We believe that the establishment of the models should not only depend on the precision index R2 or RMSE, but also consider the physical meaning of the model formula, and whether it is linear or nonlinear. In the future, water quality remote sensing should be focus on the synergy between different optically active components in surface waters.
In this paper, in situ hyperspectral data were used to develop the TSI estimation method, the sensitive bands we found coincide with the multispectral sensor for example, the Sentinel-3 OLCI bands 3 (442 nm), 6 (560 nm), 10 (681 nm) and 11 (708 nm), which makes it possible for our model to be applied to multispectral satellite sensors. However, our method only provides a methodological reference for the application of hyperspectral satellite images. Actually, some algorithms for TSI estimation have been established based on multispectral satellite image data. Landsat data is still the most used for TSI remote sensing because of its high temporal, spatial, and spectral resolution, especially Landsat 8 OLI data. Landsat 8 OLI data has been used for TSI assessment around the world, including the Russian Wular Lake [54], the Indonesian Maninjau Lake [55], the Indian two Lagoons Lakes [2], and Chinese lakes in Wuhan [56]. In addition, Landsat 7 ETM+ bands have also been used for TSI estimation in the Nainital lake and Sukhna Lake using Secchi disk transparency [57,58]. In contrast, MODIS data are more suitable for large lakes or study areas. MODIS data have been used to retrieve the TSI of 2058 lakes worldwide (surface area >25 km2) [10]. However, there is no report about the estimation of water TSI from hyperspectral satellite image data at present. Possible reasons may be difficulties in obtaining data and problems regarding image processing, and new attempts to use hyperspectral satellite image to estimate TSI should be made in the future.
5. Conclusions
In this study, we complied a dataset of in-situ hyperspectral data and point water quality parameters for seven lakes in Northeast China. Using correlation and first derivative analysis, the spectral sensitive bands of three water quality parameters, i.e., TP, SDD, and Chla, were identified:${{\mathbf R}_{{\mathbf{rs}}}}({440} )$, ${{\mathbf R}_{{\mathbf{rs}}}}({575} )$, ${{\mathbf R}_{{\mathbf{rs}}}}({680} )$, and ${{\mathbf R}_{{\mathbf{rs}}}}({710} )$. Band ratio and band combination models were used to estimate the water quality parameters with highest R2 value and lowest RMSE value. Subsequently, TSIs were estimated using the Carlson model. Ternary contribution analysis revealed that dominant factors of eutrophication differed among lakes. For example, eutrophication in the HTM Lake was caused by a high concentration of Chla, while a high concentration of TP caused eutrophication in the XXK Lake. The general of eutrophication pattern among lakes was caused by total phosphorus (45.3%) and transparency (40.3%). Therefore, the estimation model in this study was more suitable for high phosphorus and low Chla waters. The study enriched the application of water color remote sensing and provided a reference for monitoring eutrophication by remote sensing.
Funding
National Key Research and Development Program of China (2019YFC0409105); National Natural Science Foundation of China (41730104, 42001311, 42171385); Science and Technology Development Project in Jilin, China (20200201054JC); the Special Research Assistant Project of Chinese Academy of Sciences granted to Dr. Yingxin Shang; Heilongjiang Provincial Natural Science Foundation of China (LH2019D010).
Acknowledgments
Thanks also extended to all the staff and students from the aquatic environment research laboratory for their assistance with field data collection and laboratory analysis. The senior author also thanks her family members for their understanding and support. The authors would like to express their gratitude to EditSprings (https://www.editsprings.cn/) for the expert linguistic services provided. The authors would also like to thank the anonymous reviewers for their useful comments and constructive suggestions.
Disclosures
The authors declare no conflicts of interest.
Data availability
Data underlying the results presented in this paper are not publicly available at this time but may be obtained from the authors upon reasonable request.
Supplemental document
See Supplement 1 for supporting content.
References
1. S. C. J. Palmer, T. Kutser, and P. D. Hunter, “Remote sensing of inland waters: Challenges, progress and future directions,” Remote Sens. Environ. 157, 1–8 (2015). [CrossRef]
2. A. Kulshreshtha and P. Shanmugam, “Assessment of trophic state and water quality of coastal-inland lakes based on Fuzzy Inference System,” J. Great Lakes Res. 44(5), 1010–1025 (2018). [CrossRef]
3. R. A. Vollenweider and J. Kerekes, “Eutrophication of waters. In: Monitoring, Assessment and Control,” 156 Organization for Economic Co-Operation and Development (OECD), Paris (1982).
4. K. Shi, Y. Zhang, B. Qin, and B. Zhou, “Remote sensing of cyanobacterial blooms in inland waters: present knowledge and future challenges,” Sci. Bull. 64(20), 1540–1556 (2019). [CrossRef]
5. Y. Zhang, Y. Zhou, K. Shi, B. Qin, X. Yao, and Y. Zhang, “Optical properties and composition changes in chromophoric dissolved organic matter along trophic gradients: Implications for monitoring and assessing lake eutrophication,” Water Res. 131, 255–263 (2018). [CrossRef]
6. Z. Wen, K. Song, G. Liu, Y. Shang, C. Fang, J. Du, and L. Lyu, “Quantifying the trophic status of lakes using total light absorption of optically active components,” Environ. Pollut. 245, 684–693 (2019). [CrossRef]
7. J. R. Frost, E. J. Phlips, R. S. Fulton, C. L. Schelske, W. Kenney, and M. Cichra, “Temporal trends of trophic state variables in a shallow hypereutrophic subtropical lake, Lake Griffin, Florida, USA,” Fundam. Appl. Limnol. 172(4), 263–271 (2008). [CrossRef]
8. F. Mushtaq and M. G. N. Lala, “Remote estimation of water quality parameters of Himalayan lake (Kashmir) using Landsat 8 OLI imagery,” Geocarto Int. 32(3), 274–285 (2017). [CrossRef]
9. R. W. Preisendorfer, “Secchi disk science: visual optics of natural waters,” Limnol. Oceanogr. 31(5), 909–926 (1986). [CrossRef]
10. S. Wang, J. Li, B. Zhang, E. Spyrakos, A. N. Tyler, Q. Shen, F. Zhang, T. Kutser, M. K. Lehmann, Y. Wu, and D. Peng, “Trophic state assessment of global inland waters using a MODIS-derived Forel-Ule index,” Remote Sens. Environ. 217, 444–460 (2018). [CrossRef]
11. S. Thiemann and H. Kaufmann, “Determination of chlorophyll content and trophic state of lakes using field spectrometer and IRS-1C satellite data in the Mecklenburg lake district, Germany,” Remote Sens. Environ. 73(2), 227–235 (2000). [CrossRef]
12. K. Shi, Y. Zhang, K. Song, M. Liu, Y. Zhou, Y. Zhang, Y. Li, G. Zhu, and B. Qin, “A semi-analytical approach for remote sensing of trophic state in inland waters: Bio-optical mechanism and application,” Remote Sens. Environ. 232 (2019).
13. L. Chen, “A study of applying genetic programming to reservoir trophic state evaluation using remote sensor data,” Int. J. Remote Sens. 24(11), 2265–2275 (2003). [CrossRef]
14. K. Song, L. Li, S. Li, L. Tedesco, B. Hall, and L. Li, “Hyperspectral Remote Sensing of Total Phosphorus (TP) in Three Central Indiana Water Supply Reservoirs,” Water Air and Soil Pollut. 223(4), 1481–1502 (2012). [CrossRef]
15. D. Sun, Y. Li, and Q. Wang, “A Unified Model for Remotely Estimating Chlorophyll a in Lake Taihu, China, Based on SVM and In Situ Hyperspectral Data,” IEEE Trans. Geosci. Remote Sens. 47(8), 2957–2965 (2009). [CrossRef]
16. R. E. Carlson, “Trophic state index for lakes,” Limnol. Oceanogr. 22(2), 361–369 (1977). [CrossRef]
17. X. Lu, C. Situ, W. Ma, X. Dai, C. He, L. Li, and L. Zhou, “Comparative analysis of four semi-analytical models for estimating chlorophyll-a concentration in case-2 waters using field hyperspectral reflectance,” Int. J. Remote Sens. 41(2), 584–594 (2020). [CrossRef]
18. Y. Xu, Y. Shi, and Y. Li, “Eutrophication evaluation model of lake Taihu using hyperspectral data of HJ-1 satellite,” Resour. and Environ. Yangtze Basin 23, 1111–1118 (2014).
19. K. Song, L. Li, L. P. Tedesco, S. Li, H. Duan, D. Liu, B. E. Hall, J. Du, Z. Li, K. Shi, and Y. Zhao, “Remote estimation of chlorophyll-a in turbid inland waters: Three-band model versus GA-PLS model,” Remote Sens. Environ. 136, 342–357 (2013). [CrossRef]
20. Y. Zhang, M. Hu, K. Shi, M. Zhang, T. Han, L. Lai, and P. Zhan, “Sensitivity of phytoplankton to climatic factors in a large shallow lake revealed by column-integrated algal biomass from long-term satellite observations,” Water Res. 207, 117786 (2021). [CrossRef]
21. C. E. Williamson, D. P. Morris, M. L. Pace, and A. G. Olson, “Dissolved organic carbon and nutrients as regulators of lake ecosystems: Resurrection of a more integrated paradigm,” Limnol. Oceanogr. 44(3part2), 795–803 (1999). [CrossRef]
22. S. N. Kloiber, P. L. Brezonik, L. G. Olmanson, and M. E. Bauer, “A procedure for regional lake water clarity assessment using Landsat multispectral data,” Remote Sens. Environ. 82(1), 38–47 (2002). [CrossRef]
23. K. Song, G. Liu, Q. Wang, Z. Wen, L. Lyu, Y. Du, L. Sha, and C. Fang, “Quantification of lake clarity in China using Landsat OLI imagery data,” Remote Sens. Environ. 243, 111800 (2020). [CrossRef]
24. C. Wu, J. Wu, J. Qi, L. Zhang, H. Huang, L. Lou, and Y. Chen, “Empirical estimation of total phosphorus concentration in the mainstream of the Qiantang River in China using Landsat TM data,” Int. J. Remote Sens. 31(9), 2309–2324 (2010). [CrossRef]
25. K. Song, Z. Wang, J. Blackwell, B. Zhang, F. Li, Y. Zhang, and G. Jiang, “Water quality monitoring using Landsat Themate Mapper data with empirical algorithms in Chagan Lake, China,” J. Appl. Remote Sens. 5(1), 053506 (2011). [CrossRef]
26. T. Kutser, H. Arst, T. Miller, L. Kaarmann, and A. Milius, “Telespectrometrical estimation of water transparency, chlorophyll-a and total phosphorus concentration of Lake Peipsi,” Int. J. Remote Sens. 16(16), 3069–3085 (1995). [CrossRef]
27. D. Sun, Y. Li, Q. Wang, J. Gao, C. Le, C. Huang, and S. Gong, “Hyperspectral Remote Sensing of the Pigment C-Phycocyanin in Turbid Inland Waters, Based on Optical Classification,” IEEE Trans. Geosci. Remote Sens. 51(7), 3871–3884 (2013). [CrossRef]
28. J. L. Mueller, A. Morel, R. Frouin, C. Davis, R. Arnone, K. Carder, Z.P. Lee, R.G. Steward, S.C. Hooker, D. Mobley, S. McLean, B. Holben, M. Miller, C. Pietras, K.D. Knobelspiesse, G.S. Fargion, J. Porter, and K. Voss, “Ocean Optics Protocols for Satellite Ocean Color Sensor Validation,” vol. III. NASA Goddard Space Flight Center, Greenbelt, MD. Revision 4. (2003).
29. H. B. Jiao, Y. Zha, J. Gao, Y. M. Li, Y. C. Wei, and J. Z. Huang, “Estimation of chlorophyll-a concentration in Lake Tai, China using in situ hyperspectral data,” Int. J. Remote Sens. 27(19), 4267–4276 (2006). [CrossRef]
30. C. D. Mobley, “Estimation of the remote-sensing reflectance from above-surface measurements,” Appl. Opt. 38(36), 7442–7455 (1999). [CrossRef]
31. L. Lyu, Z. Wen, P.-A. Jacinthe, Y. Shang, N. Zhang, G. Liu, C. Fang, J. Hou, and K. Song, “Absorption characteristics of CDOM in treated and non-treated urban lakes in Changchun, China,” Environ. Res. 182, 109084 (2020). [CrossRef]
32. APHA/AWWA/WEF, “Standard Methods for the Examination of Water and Wastewater,” American Public Health Association, Washington, D.C. (1998).
33. D. C. Rundquist, L. H. Han, J. F. Schalles, and J. S. Peake, “Remote measurement of algal chlorophyll in surface waters: The case for the first derivative of reflectance near 690 nm,” Photogramm. Eng. Remote Sens. 62, 195–200 (1996).
34. L. H. Han, “Estimating chlorophyll-a concentration using first-derivative spectra in coastal water,” Int. J. Remote Sens. 26(23), 5235–5244 (2005). [CrossRef]
35. D. Yu, B. Zhou, X. Zhang, W. Xie, and E. Liu, “Retrieval of Secchi disk depth in offshore marine areas based on simulated HICO from in situ hyperspectral data,” in International Conference on Intelligent Earth Observing and Applications 2015, G. Zhou and C. Kang, eds. (2015).
36. D. T. Yang, D. L. Pan, X. Y. Zhang, X. F. Zhang, X. Q. He, and S. Q. Li, “Retrieval of water quality parameters by hyperspectral remote sensing in lake TaiHu, China,” in Optical Technologies for Atmospheric, Ocean, and Environmental Studies, Pts 1 and 2, G. G. Matvienko, ed. (2005), pp. 431–439.
37. D. T. Yang, D. L. Pan, X. Y. Zhang, X. F. Zhang, X. Q. He, and S. Q. Li, “Retrieval of water quality parameters by hyperspectral remote sensing in lake TaiHu, China,” in Conference on Optical Technologies for Atmospheric, Ocean, and Environmental Studies, 2005), 431–439.
38. Y. Liu, C. Xiao, J. Li, F. Zhang, and S. Wang, “Secchi Disk Depth Estimation from China's New Generation of GF-5 Hyperspectral Observations Using a Semi-Analytical Scheme,” Remote Sens. 12(11), 1849 (2020). [CrossRef]
39. K. Randolph, J. Wilson, L. Tedesco, L. Li, D. L. Pascual, and E. Soyeux, “Hyperspectral remote sensing of cyanobacteria in turbid productive water using optically active pigments, chlorophyll a and phycocyanin,” Remote Sens. Environ. 112(11), 4009–4019 (2008). [CrossRef]
40. J. Tan, K. A. Cherkauer, and I. Chaubey, “Developing a Comprehensive Spectral-Biogeochemical Database of Midwestern Rivers for Water Quality Retrieval Using Remote Sensing Data: A Case Study of the Wabash River and Its Tributary, Indiana,” Remote Sens. 8(2016).
41. A. Bilgin, “Trophic state and limiting nutrient evaluations using trophic state/level index methods: a case study of Borcka Dam Lake,” Environmental Monitoring and Assessment 192 (2020).
42. J. Schaffer, “The Internal Relatedness of All Things,” Mind 119(474), 341–376 (2010). [CrossRef]
43. D. A. Phinney, D. I. Phinney, and C. S. Yentsch, “The relationship between remote sensing reflectance and optically active substances in Case 1 and Case 2 waters,” Ocean Optics Xiii 2963, 743–752 (1997). [CrossRef]
44. J. Xiong, C. Lin, R. Ma, and Z. Cao, “Remote Sensing Estimation of Lake Total Phosphorus Concentration Based on MODIS: A Case Study of Lake Hongze,” Remote Sens. 11(17), 2068 (2019). [CrossRef]
45. D. W. Schindler, “Evolution of phosphorus limitation in lakes,” Science 195(4275), 260–262 (1977). [CrossRef]
46. A. Cudowski, “Dissolved reactive manganese as a new index determining the trophic status of limnic waters,” Ecological Indicators 48, 721–727 (2015). [CrossRef]
47. C. Forsberg and S. O. Ryding, “Eutrophication parameters and trophic state indexes in 30 swedish waste-receiving lakes,” Archiv Fur Hydrobiologie 89, 189–207 (1980).
48. K. E. Webster, P. A. Soranno, K. S. Cheruvelil, M. T. Bremigan, J. A. Downing, P. D. Vaux, T. R. Asplund, L. C. Bacon, and J. Connor, “An empirical evaluation of the nutrient-color paradigm for lakes,” Limnol. Oceanogr. 53(3), 1137–1148 (2008). [CrossRef]
49. N. M. Burns and G Bryers, “Protocols for Monitoring Trophic Levels of New Zealand Lakes and Reservoirs,” (2000).
50. N. M. Burns, J. C. Rutherford, and J. S. Clayton, “A monitoring and classification system for New Zealand lakes and reservoirs,” Lake Reservoir Manage. 15(4), 255–271 (1999). [CrossRef]
51. X. C. Jin and Q.Y. Tu, “The Standard Methods for Observation and Analysis in Lake Eutrophication. 240 Chinese Environmental Science Press, Beijing,” (1990).
52. P. Verburg, K. Hamill, M. Unwin, and J. Abell, “Lake water quality in New Zealand: status and trends. In: NIWA Client Report HAM,” pp. 107 (2010).
53. M. W. Matthews, S. Bernard, and L. Robertson, “An algorithm for detecting trophic status (chlorophyll-a), cyanobacterial-dominance, surface scums and floating vegetation in inland and coastal waters,” Remote Sens. Environ. 124, 637–652 (2012). [CrossRef]
54. F. Mushtaq, M. Lala, and A. G. Mantoo, “Trophic state assessment of a freshwater Himalayan lake using Landsat 8 OLI satellite imagery: A case study of Wular Lake,” Earth Space Sci. in press (2021).
55. A. Rivani and P. Wicaksono, Water Trophic Status Mapping of Tecto-Volcanic Maninjau Lake during Algae Bloom using Landsat 8 OLI Satellite Imagery, IEEE International Conference on Aerospace Electronics and Remote Sensing Technology (ICARES) Bali, INDONESIA SEP 20-21, 2018 (2018).
56. Y. Zhou, B. He, F. Xiao, Q. Feng, J. Kou, and H. Liu, “Retrieving the Lake Trophic Level Index with Landsat-8 Image by Atmospheric Parameter and RBF: A Case Study of Lakes in Wuhan, China,” Remote Sens. 11(4), 457 (2019). [CrossRef]
57. A. K. Mishra and N. Garg, “Analysis of Trophic State Index of Nainital Lake from Landsat-7 ETM Data,” J. Indian Soc. Remote Sens. 39(4), 463–471 (2011). [CrossRef]
58. J. K. Dhillon and A. K. Mishra, “Estimation of Trophic State Index of Sukhna Lake Using Remote Sensing and GIS,” J. Indian Soc. Remote Sens. 42(2), 469–474 (2014). [CrossRef]
