Long-Term Spatial and Temporal Monitoring of Cyanobacteria Blooms Using MODIS on Google Earth Engine: a Case Study in Taihu Lake

remote sensing

Article Long-Term Spatial and Temporal Monitoring of Cyanobacteria Blooms Using MODIS on Google Earth Engine: A Case Study in Taihu Lake

Tianxia Jia 1,2 , Xueqi Zhang 1,2 and Rencai Dong 1,*

1 State Key Laboratory of Urban and Regional Ecology, Research Center for Eco-environmental Sciences, Chinese Academy of Sciences, Beijing 100085, China; [email protected] (T.J.); [email protected] (X.Z.) 2 University of Chinese Academy of Sciences, Beijing 100049, China * Correspondence: [email protected]

Received: 11 August 2019; Accepted: 26 September 2019; Published: 28 September 2019

Abstract: As cyanobacteria blooms occur in many types of inland water, routine monitoring that is fast and accurate is important for environment and drinking water protection. Compared to field investigations, satellite remote sensing is an efficient and effective method for monitoring cyanobacteria blooms. However, conventional remote sensing monitoring methods are labor intensive and time consuming, especially when processing long-term images. In this study, we embedded related processing procedures in Google Earth Engine, developed an operational cyanobacteria bloom monitoring workflow. Using this workflow, we measured the spatiotemporal patterns of cyanobacteria blooms in China’s Taihu Lake from 2000 to 2018. The results show that cyanobacteria bloom patterns in Taihu Lake have significant spatial and temporal differentiation: the interannual coverage of cyanobacteria blooms had two peaks, and the condition was moderate before 2006, peaked in 2007, declined rapidly after 2008, remained moderate and stable until 2015, and then reached another peak around 2017; bays and northwest lake areas had heavier cyanobacteria blooms than open lake areas; most cyanobacteria blooms primarily occurred in April, worsened in July and August, then improved after October. Our analysis of the relationship between cyanobacteria bloom characteristics and environmental driving factors indicates that: from both monthly and interannual perspectives, meteorological factors are positively correlated with cyanobacteria bloom characteristics, but as for nutrient loadings, they are only positively correlated with cyanobacteria bloom characteristics from an interannual perspective. We believe reducing total phosphorous, together with restoring macrophyte ecosystem, would be the necessary long-term management strategies for Taihu Lake. Our workflow provides an automatic and rapid approach for the long-term monitoring of cyanobacteria blooms, which can improve the automation and efficiency of routine environmental management of Taihu Lake and may be applied to other similar inland waters.

Keywords: cyanobacteria blooms; Taihu Lake; spatiotemporal patterns; Google Earth Engine; ﬂoating algae index; redundancy analysis

1. Introduction Under the background of accelerated industrialization and urbanization, harmful algae blooms (HABs) occur frequently in inland and coastal waters due to eutrophication, which is the combined result of natural and anthropogenic processes, although it is primarily caused by inputs of anthropogenic nutrient loadings (nitrogen and phosphorus) [1]. Some species of algae blooms, such as cyanobacteria, can release odors and toxic compounds that are harmful to aquatic creatures and water ecosystems [1–3].

Remote Sens. 2019, 11, 2269; doi:10.3390/rs11192269 www.mdpi.com/journal/remotesensing Remote Sens. 2019, 11, 2269 2 of 32

Cyanobacteria blooms propagate as a result of many environmental factors, such as temperature, precipitation, wind speed, atmospheric pressure, solar radiation, and nutrient loadings (nitrogen and phosphorus) [1,3–9]. Generally, those environmental driving variables are divided into two types: meteorological variables and water quality variables. Determining the relationship between cyanobacteria blooms and environmental driving factors is important for precaution and management. Different methods have been used to model the formation and dynamics of cyanobacteria blooms. Some cyanobacteria bloom characteristics, together with their meteorological and water quality factors, are the main predictive variables in these models [10–12]. Artificial salvage of floating cyanobacteria also heavily relies on its location being determined in advance. Therefore, accurate and efficient monitoring of the spatiotemporal distributions of cyanobacteria blooms is vital for prevention and management. Field investigations by ship surveys and laboratory measurements are the basis of the conventional monitoring approach [13], but these methods can destroy the vertical and horizontal distributions of floating cyanobacteria, and are only able to detect the cyanobacteria biomass concentration along routes, therefore, the representativeness of limited water samples cannot be guaranteed. Field investigations cannot explain the spatial distribution of cyanobacteria blooms, and are expensive and labor-intensive [2,14,15]. Due to the large spatial coverage and short revisit interval of some satellite sensors, optical remote sensing could be used to effectively and efficiently monitor cyanobacteria blooms. Many studies have monitored cyanobacteria blooms over inland and coastal waters. Reflectance classification [14], single-band threshold [5], water color indices [15–18], and bio-optical models [4] have been developed in water component remote sensing inversions. Among the many water color indices, the floating algae index (FAI) is an effective spectrum index for distinguishing floating cyanobacteria [16]. It has already been used for cyanobacteria bloom extraction in China’s Taihu Lake [15]. Unlike the normalized difference vegetation index (NDVI) and the enhanced vegetation index (EVI), the baseline subtraction method used by FAI can effectively remove cloud, sunglint, and geometry contamination, and is more stable for extracting long-term series of cyanobacteria blooms [15,16,19,20]. Since its development in 2010 [16], several studies have used FAI as a detector to extract long-term spatiotemporal patterns of blooms corresponding to cyanobacteria or other phytoplankton species. Hu et al. developed a workflow for extracting cyanobacteria bloom events in Taihu Lake and reconstructed its spatiotemporal patterns between 2000 and 2008 [15]. Hu et al. retrieved Ulva prolifera bloom dynamics from the Yellow Sea and East China Sea from 2000 to 2009 [21]. Huang et al. determined variations in cyanobacteria blooms from 2000 to 2011 in Taihu Lake [22]. Zhang et al. extracted the subpixel-level floating algae spatiotemporal dynamics in China’s Chaohu Lake over a period of 2000–2013 [23]. Kabenge et al. reconstructed floating algae patterns in Uganda’s Murchison Bay between 1990 and 2014 [24]. Zhang et al. determined the spatiotemporal bloom patterns in Taihu Lake from 2001 to 2013 [6]. Wang and Hu determined the distribution and coverage area of Sargassum from over the Central West Atlantic region during 2000–2015 [25]. Shi et al. identified the monthly and interannual cyanobacteria bloom temporal variations in Taihu Lake from 2003 to 2013 [26]. All the above listed studies used moderate resolution imaging spectroradiometer (MODIS) level 0 data, and the raw data were calibrated by SeaDAS software (NASA, US), in order to obtain the Rayleigh-corrected reflectance data for extracting floating cyanobacteria. For general end users, calibrating the huge amount of MODIS level 0 data required for specific applications is difficult. Other conventional remote sensing procedures for cyanobacteria blooms, including satellite image downloading, preprocessing, and spatial analysis, are also labor intensive and time consuming. These shortcomings have hindered the further extension of automatic remote sensing monitoring of long-term cyanobacteria blooms. In some environmental monitoring scenarios that require large-scale satellite remote sensing data processing, thematic cloud-based platforms may perform better. The Google Earth Engine (GEE) is a cloud platform for geospatial data analysis, and has changed the traditional mode of processing used Remote Sens. 2019, 11, 2269 3 of 32 remote sensing [27]. GEE is especially suitable for long-term and large-scale spatiotemporal series of remote sensing applications; its multi-petabyte geospatial data and high-performance computation capacity provide a new approach to understanding the earth. Conventional remote sensing algorithms and methods, which were initially conducted with small-scale data, have the potential to be embedded and integrated in GEE, and applied in a variety of fields and scenarios. In the Web of Science database, we analyzed nearly 200 articles that have used GEE; most of their themes were roughly related to agriculture applications or urban or forest mapping. Overall, land use and land cover classification was the most common and prevailing topic. However, optical constituent studies of inland and coastal waters are rare [28–32], due largely to the complexity of bio-optical properties of case II water [3,4,15]. What is more, the lack of embedded water-leaving reflectance products of GEE has hindered the further operational applications in case II water. It is necessary to find and develop available operational inland water processing algorithms and perform them on GEE’s existing embedded datasets. Based on the urgent need for long-term assessment of cyanobacteria blooms in eutrophic inland water, and recent developments in cloud-based technology, the objectives of our study were to (1) develop an operational workflow in GEE to accomplish rapid and automatic monitoring of long-term cyanobacteria blooms, (2) analyze the spatiotemporal patterns of cyanobacteria blooms in Taihu Lake over the most recent 19 years (2000 to 2018), and (3) determine the relationship between cyanobacteria blooms and environmental driving factors using our study’s monitoring data.

2. Materials and Methods

2.1. Study Area Taihu Lake, the third largest freshwater lake in China, is located in the prosperous Yangtze River Delta (Figure1). Its water surface coverage is approximately 2338 km 2, and it has a shallow mean water depth of 1.9 m [2,5,15]. It is a typical subtropical lake and is influenced by monsoons. Taihu Lake is one of the most important sources of drinking water for adjacent cities, such as Wuxi city [2,5]. Due to eutrophication lasting for more than two decades [8], cyanobacteria blooms frequently occur in Taihu Lake [33], and they have had negative effects on aquatic life and human health [34]. The cyanobacteria blooms that occurred in the summer of 2007 heavily affected the safety of the drinking water of nearly 2 million citizens [2,15]. After that crisis, the local government began to manage and control eutrophication [5]; these measures were effective for the next decade, but recent studies reported that the water quality in Taihu Lake has deteriorated again, and severe cyanobacteria blooms frequently occurred in 2017 [35–37]. For convenience of analysis, we divided Taihu Lake into eight segments according to previous studies: Central Lake, West Lake, South Lake, East Lake, Zhushan Bay, Meiliang Bay, Gong Bay, and East Bay [38]. East Bay is perennially submerged and emergent dominated by aquatic vegetation [15,39–42]. The optical features of aquatic vegetation are similar to floating cyanobacteria, which can decrease the monitoring accuracy; therefore, East Bay was excluded from our study. Remote Sens. 2019, 11, 2269 4 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 4 of 32

FigureFigure1. 1. LocationLocation of T Taihuaihu Lake in in China. China. 2.2. Satellite Data 2.2. Satellite Data MOD09GQ and MOD09GA products were used in our study. MOD09GQ and MOD09GA are MOD09GQ and MOD09GA products were used in our study. MOD09GQ and MOD09GA are short for MOD09GQ.006 Terra Surface Reflectance Daily L2G Global 250 m and MOD09GA.006 Terra short for MOD09GQ.006 Terra Surface Reflectance Daily L2G Global 250 m and MOD09GA.006 Terra SurfaceSurface Reflectance Reflectance Daily Daily L2G L2G Global Global 11 kmkm andand 500 m, respectively. The The MODIS MODIS surface surface reflectance reflectance productsproducts provide provide an an estimate estimate of of the the surface surface spectralspectral reflectance reflectance as as it it would would be be measured measured at atground ground levellevel in thein the absence absence of of atmospheric atmospheric scattering scattering or or absorption absorption [ 43[43,44],44]. . Details Details of of thethe productsproducts areare listedlisted in Tablein Table1. 1.

Table 1. Introduction of research datasets [43,44]. Table 1. Introduction of research datasets [43,44].

Band InformationBand information (Only (only Related related Bandsbands listed) Listed) Spatial ProductsProducts Revisit Revisit Spatial Wavelength Name Resolution Wavelength Description Name Resolution (nm) Description (m) (nm) (m) Surface reflectance for MOD09GQ.006 Terra sur_refl_b01 250 620–670 Surfacered reflectanceband Surface Reflectance Daily L2G Dailysur_refl_b01 250 620–670 MOD09GQ.006 Terra Surfacefor reflectance red band for Global 250 m sur_refl_b02 250 841–876 Surface Reflectance near-IR (NIR) band Daily Surface reflectance Daily L2G Global Surface reflectance for sur_refl_b02sur_refl_b05 250500 841–8761230–1250 for near-IR (NIR) 250 m SWIR1 band MOD09GA.006 Terra Surface Surface reflectanceband for Reflectance Daily L2G Global 1 km Daily sur_refl_b06 500 1628–1652 SWIR2 band and 500 m Surface reflectance sur_refl_b05 500 1230–1250 Surface reflectance for MOD09GA.006 Terra sur_refl_b07 500 2105–2055 for SWIR1 band SWIR3 band Surface Reflectance Surface reflectance Daily sur_refl_b06 500 1628–1652 Daily L2G Global 1 km for SWIR2 band 2.3. Dataand Processing 500 m Surface reflectance sur_refl_b07 500 2105–2055 2.3.1. Resampling for SWIR3 band FAI calculation requires red and near-infrared (NIR) bands of MOD09GQ and one shortwave infrared (SWIR) band of MOD09GA. Atmospheric correction requires all of the SWIR bands of MOD09GA. To make full use of the spatial resolution advantage of MOD09GQ, three SWIR bands

Remote Sens. 2019, 11, 2269 5 of 32

2.3. Data Processing

2.3.1. Resampling FAI calculation requires red and near-infrared (NIR) bands of MOD09GQ and one shortwave infrared (SWIR) band of MOD09GA. Atmospheric correction requires all of the SWIR bands of MOD09GA. To make full use of the spatial resolution advantage of MOD09GQ, three SWIR bands (sur_refl_b05/sur_refl_b06/sur_refl_b07) of MOD09GA were resampled to 250 m [15,23,41]. The resampling method was bilinear interpolation.

2.3.2. Land Masking Land surface pixels have high FAI values; thus, they should be ﬁltered by a land mask [15]. The multi-year innermost boundary of Taihu Lake was carefully visually digitized according to Landsat TM/ETM+/OLI images. Finally, the land mask was eroded to 250 m (one pixel width) toward water to avoid the land adjacent eﬀect [15,45].

2.3.3. Cloud Masking Cloud pixels also have high FAI values [15,16]. To exclude cloud contamination, a cloud mask is required [46]. A single-band threshold Rrc (859 nm) > 0.15 was adopted as a cloud mask. Most cloud and sunglint contamination can be ﬁltered by this kind of cloud mask [47–49].

2.3.4. Water-Leaving Reflectance Correction Previous studies broadly used MODIS Rayleigh correction data for inland water applications, however, the GEE has not embedded this kind of data. Hence, we had no choice but to use MODIS surface reflectance products (MOD09) in our workflow. Although the MODIS surface reflectance products were primarily designed for land surfaces [50], previous studies have found that the MOD09 reflectance was systematically greater than the in-situ reflectance over inland water, because MOD09 fails to fully correct for the aerosol effect [39,50,51]. Water-leaving reflectance correction should be implemented before inland water application [31,33]. Here, we performed a simple yet effective method to retrieve the water-leaving reflectance of Taihu Lake, which was developed by Wang et al. [45,49]:

R(λ) min(RNIR : RSWIR) Rc (λ) = − (1) rs π

c where Rrs(λ) is the corrected water-leaving reflectance of the MOD09 product centered at wavelength λ, R(λ) is the original MOD09 reflectance at λ, min(RNIR : RSWIR) is the minimum positive reflectance value between NIR and SWIR bands, and π is placed as the denominator to transform surface reflectance to water-leaving reflectance. Here, the sur_refl_b05, sur_refl_b06, and sur_refl_b07 bands were used as RNIR : RSWIR. The correction process was implemented on each pixel. This operational water-leaving reflectance correction method has relative stable performances over inland waters under various conditions [45,49].

2.3.5. FAI Threshold

Using FAI gradient histogram statistics of long-term (2000 to 2008) MODIS Rrc(λ) data, Hu et al. [15] determined FAI > 0.004 as the floating cyanobacteria distinguishing threshold. After water-leaving − reflectance correction, this FAI threshold was adopted to distinguish floating cyanobacteria pixels in our workflow. In the following validation and discussion sections, the potential errors caused by this fixed threshold that may incorporate into our workflow, will be reported in detail. FAI is defined as follows [15,16]: FAI = Rrc(859) R0 (859) (2) − rc Remote Sens. 2019, 11, 2269 6 of 32

859 645 R0 (859) = Rrc(645) + [Rrc(1240) Rrc(645)] − (3) rc − × 1240 645 − where Rrc(645), Rrc(859), and Rrc(1240) are Rayleigh-corrected reﬂectance at 645, 859, and 1240 nm, c c respectively. Here, Rrc(645), Rrc(859), and Rrc(1240) respectively correspond to Rrs(645), Rrs(859) and c Rrs(1240), which were derived in Section 2.3.4.

2.4. Spatiotemporal Analysis of Cyanobacteria Blooms The above floating cyanobacteria extraction processes were executed on each MODIS daily image; the final output result was that pixels with FAI values greater than 0.004 were classified as floating − cyanobacteria. Those pixels were assigned a value of 1, while non-cyanobacteria pixels were assigned a value of 0. These cyanobacteria-binarized MODIS daily images were initially input to the following temporal and spatial analysis.

2.4.1. Temporal Analysis Based on cyanobacteria-binarized MODIS images, we counted the total amount of cyanobacteria pixels on a daily basis, then calculated the daily floating cyanobacteria area (FAd) by multiplying one pixel’s area (0.25 0.25 km). After that, we calculated the monthly (FAm) and annual (FAa) mean area × of floating cyanobacteria using daily area data. According to the FA data (including FAd, FAm and FAa, the same hereafter), we determined the temporal trends of floating cyanobacteria at different levels. Due to cyanobacteria blooms with large daily areas being more severe than small blooms, we paid additional attention to the significant cyanobacteria blooms in our monitoring workflow. Based on FAd, which we derived in the above step, significant cyanobacteria blooms were defined as floating cyanobacteria daily coverages that exceeded 25% of the total lake (or segment) area [15]. We derived and analyzed the annual start and end Julian dates of the significant cyanobacteria blooms, and based on floating cyanobacteria severity classification, we analyzed the cumulative percentage temporal trend of the significant cyanobacteria blooms.

2.4.2. Spatial Analysis We composed cyanobacteria-binarized daily images and calculated the sum of all values at each pixel of each year, and then obtained the annual accumulative floating cyanobacteria occurrence frequency on a pixel basis. The same method was executed for each month over the 19-year period, and the monthly accumulative bloom occurrence frequency was then determined. For the initial date and during analysis, we first attached the Julian date information to each cyanobacteria-binarized image; then, after composing cyanobacteria-binarized daily images of each year, we calculated the minimum and the maximum date value of each pixel, i.e., the date of first and the last floating cyanobacteria occurrence on a pixel basis, and we finally obtained the annual floating cyanobacteria occurrence duration, by calculating the difference between the first and the last date on a pixel basis [15].

2.5. Validation Accuracy validation is essential for long-term series of cyanobacteria bloom monitoring. Here, we used three methods to validate the accuracy of our workflow. The first method involved analyzing the correlation between our monthly floating cyanobacteria areas (FAm) and monthly in-situ concentration of chlorophyll-a (CChla) (collected from Taihu Basin Authority of Ministry of Water Resources, www. tba.gov.cn). CChla is often regarded as the proxy of cyanobacteria biomass in water [2,14,26]. Previous studies have demonstrated that FAm and CChla have good overall consistency under static wind weather [22,52]. The second method involved comparing our results with those of other studies. The different spatiotemporal scales of our FA were qualitatively and quantitatively validated by the results or data of similar studies. The third method involved visually interpreting FAd on Landsat TM/ETM+/OLI images to evaluate our concurrent results [5,15]. The finer spatial texture of floating cyanobacteria could be visually delineated under the higher resolution of Landsat true color Remote Sens. 2019, 11, 2269 7 of 32 images [15]. In general, by comparing our results with the in-situ data (the first method) and the remote sensing observations (the second and the third method), we can get a comprehensive accuracy assessment report.

2.6. Analysis of the Relationship between Environmental Driving Factors and Cyanobacteria Blooms To fully mine the long-term series of cyanobacteria bloom information in our study, and to understand the role of environmental factors on driving cyanobacteria blooms, we conducted redundancy analysis (RDA) to identify relationships between cyanobacteria blooms and environmental driving factors. RDA was initially developed for ecologists to draw biplots to display relationships between species and explanatory variables. The length of the arrow is a measure of fit for species. More quantitatively, we can read the approximated correlations of one species with others by projecting the arrowheads of the other species (or environmental variables) onto an imaginary line overlaying that species’ arrow [53,54]. In the RDA ordination diagram, the cosine of the angle between two arrows represents their correspondence; an acute angle, obtuse angle, or right angle respectively denote their positive, negative, and lack of correlation [6]. First, we conducted a detrended correspondence analysis (DCA) and found all gradient lengths were less than three, hence, the linear model was adopted for the following analysis. Then, RDA was implemented to find the relationship between environmental variables and cyanobacteria bloom characteristics [6,54–56]. Environmental variables included meteorological and water quality factors. Meteorological factor data were downloaded from the National Ecosystem Research Network of China, water quality factor data were collected from Taihu Basin Authority of Ministry of Water Resources, and cyanobacteria bloom characteristics were derived from our results (Table 4). Due to the incomplete temporal sequence of the meteorological data, we used monthly and annual data between 2005 and 2016 (Figure S1) to implement RDA, which occupied two-thirds of the temporal span of the whole study period (2000–2018) and also included recent years with the exception of 2017 and 2018. Therefore, we think the amount of data was sufficient to produce credible correlation analysis. We not only used the monthly mean area of floating cyanobacteria (FAmean) but also selected monthly maximum area of floating cyanobacteria (FAmax) as another monthly characteristic, because FAmax may better represent monthly floating cyanobacteria coverage due to the wind effect [6,15]. The above processing and analysis steps are all depicted in the workflow in Figure2. Remote Sens. 2019, 11, 2269 8 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 8 of 32

FigureFigure 2. 2. WorkflowWorkﬂow used in our study.

3. Results

3.1. Validation of Workflow Accuracy

3.1.1. Validation Using In-Situ Data (Chlorophyll-a) Due to the similar FAI characteristics of floating cyanobacteria, macrophytes distributed in the East Lake would cause overestimation of FA, especially in summer and autumn [22]. To consider this

Remote Sens. 2019, 11, 2269 9 of 32

3. Results

3.1. Validation of Workﬂow Accuracy

3.1.1. Validation Using In-Situ Data (Chlorophyll-a)

RemoteDue Sens. to 2019 the, 11 similar, x FOR PEER FAI characteristicsREVIEW of floating cyanobacteria, macrophytes distributed9 inof the32 East Lake would cause overestimation of FA, especially in summer and autumn [22]. To consider thisdeviation deviation source, source, we we performed performed a correlation a correlation analysis analysis between between the the FA FAm mof of the the entire entire lake lake and and chlorophyll-achlorophyll-a concentrationconcentration (C (CChlaChla), ),and and between between the the FA FAm ofm theof theentire entire lake lakeexcluding excluding East Lake East Lakeand andCChla C. ChlaThe. FAI The threshold FAI threshold was another was another potential potential source of source error. of Since error. FAI Since was FAIdeveloped was developed and applied and appliedin Taihu in Lake,Taihu Lake, few studies few studies demonstrated demonstrated that that FAI FAI > −0.004> 0.004 can can be be extended extended to to extract extract floating floating − cyanobacteriacyanobacteria after after 2008, 2008, because because this this threshold threshold value value was was initially initially calculated calculated using using statistical statistical data data from 2000from to 2000 2008 to[15 ]. 2008 Hence, [15]. applicable Hence, applicable uncertainties uncertainties still remain still in our remain workflow in our when workflow manipulating when manipulating these long-term images. Therefore, we conducted a correlation analysis between FAm these long-term images. Therefore, we conducted a correlation analysis between FAm and CChla over theand whole CChla over temporal the whole stage (2001–2018),temporal stage early (2001 stage–2018), (2001–2008), early stage and (2001 late– stage2008), (2009–2018) and late stage (Figure (20093).– 2018) (Figure 3).

250 500 Chla

200 FA(entire lake) 400 FA(entire lake excluding East Lake) ) 150 300 2 (km (ug/L) m Chla Chla 100 200 FA C

50 100

0 0 Jul-01 Jul-02 Jul-03 Jul-04 Jul-05 Jul-06 Jul-07 Jul-08 Jul-09 Jul-10 Jul-11 Jul-12 Jul-13 Jul-14 Jul-15 Jul-16 Jul-17 Jul-18 Jan-01 Jan-02 Jan-03 Jan-04 Jan-05 Jan-06 Jan-07 Jan-08 Jan-09 Jan-10 Jan-11 Jan-12 Jan-13 Jan-14 Jan-15 Jan-16 Jan-17 Jan-18

FigureFigure 3.3. ComparisonComparison betweenbetween monthlymonthly mean chlorophyll-achlorophyll-a concentration ( CChla) )and and monthly monthly mean mean ﬂoatingfloating cyanobacteriacyanobacteria areaarea (FA(FAmm) of entire lake and entire entire lake lake excluding excluding East East Lake. Lake.

TheThe PearsonPearson correlationcorrelation coecoefficientsfficients (Table2 2)) show show thatthat FAFAmm and CChla havehave good good consistency consistency andand correlation.correlation. After After excluding excluding the the contamination contamination of ofthe the East East Lake, Lake, the correlation the correlation between between FAm and FAm andCChla C Chlaincreasedincreased in all in temporalall temporal stages. stages. Another Another finding finding was was that that FA FAm mandand C CChlaChla hadhad a a stronger stronger correlationcorrelation inin thethe latelate stagestage (2009–2018)(2009–2018) than in the early stage (2001 (2001–2008),–2008), regardless regardless of of the the spatial spatial extent.extent. OurOur correlationcorrelation analysis demonstrates that that (1) (1) macrophytes macrophytes in in East East Lake Lake affected affected the the final final FA FA accuracy;accuracy; (2) (2) due due to to the the massive massive loss loss of of macrophytes macrophytes in in Taihu Taihu Lake Lake after after 2014 2014 (this (this will wil bel be discussed discussed in followingin following sections), sections), our our FA resultsFA results were were closer closer to actual to actual floating floating cyanobacteria cyanobacteria areas areas in the in late the stage;late andstage; (3) and FAI (3)> FAI0.004 > −0.004 was found was tofound remain to remain robust robust after 2008, after showing 2008, showing that we that could werationally could rationally use this − thresholduse this threshold to process to nearlyprocess 20 nearly years’ 20 worth years’ of worth images. of images.

TableTable 2. 2.Pearson Pearson correlationcorrelation coecoefficientsfficients betweenbetween monthly mean CChla andand monthly monthly mean mean floating floating cyanobacteriacyanobacteria area area (FA (FAmm) over) over entire entire lake lake and entire and entire lake excluding lake excluding East Lake, East in different Lake, in temporal different temporalstages. stages.

EntireEntire Lake lake Entire Entire lake Lake excluding Excluding East East Lake

2001–20082001–2008 2009–20182009–2018 2001–20182001–2018 2001–20082001–2008 20092009–2018–2018 20012001–2018–2018 (N(N= =93) 93) ((NN= = 118)118) ((NN= = 211)211) ((NN= = 93)93) ((NN == 118) ((NN == 211)211) PearsonPearson 0.48100.4810 0.62990.6299 0.54010.5401 0.5082 0.6898 0.5833 0.5833 coecoefficientﬃcient

3.1.2. Validation Using the Results of Other Studies Our monitoring workflow mainly focused on the long-term trends and dynamics of cyanobacteria blooms; therefore, large-scale patterns were, to some extent, more important than short-term and partial details. To validate our derived spatial and temporal patterns of cyanobacteria blooms, we visually inspected our results and those of other studies (Table 3). In this process, some artifacts and a mismatch in spatiotemporal patterns were visible [39]. As a result, we found a high consistency between our spatiotemporal patterns and those of similar studies. We also extracted

Remote Sens. 2019, 11, 2269 10 of 32

3.1.2. Validation Using the Results of Other Studies Our monitoring workflow mainly focused on the long-term trends and dynamics of cyanobacteria blooms; therefore, large-scale patterns were, to some extent, more important than short-term and partial details. To validate our derived spatial and temporal patterns of cyanobacteria blooms, we visually inspected our results and those of other studies (Table3). In this process, some artifacts and a mismatch in spatiotemporal patterns were visible [39]. As a result, we found a high consistency between our spatiotemporal patterns and those of similar studies. We also extracted floating cyanobacteria correspondingRemote Sens. 2019 to, 11 specific, x FOR PEER dates REVIEW to display the evolution of the 2007 cyanobacteria blooms10 of (Figure 32 4) and compared this with the results of Hu et al. [15]. We found that our long-term patterns as well as ourfloating daily spatial cyanobacteria distribution correspondin results wereg to highly specific consistent dates to with display previous the evolution reports. of the 2007 cyanobacteria blooms (Figure 4) and compared this with the results of Hu et al. [15]. We found that our long-term patternsTable as 3. wellSimilar as our studies daily used spatial for spatiotemporaldistribution results pattern were validation. highly consistent with previous reports. Reference Indicator/Proxy Spatial Extent Temporal Stage TableCyanobacteria 3. Similar studies bloom used frequency for spatiotemporal/ pattern validation. Hu et al. [15] Entire lake 2000–2008 initial/duration/area/specific date Temporal Reference Indicator/Proxy Spatial Extent Duan et al. [5,34] Cyanobacteria bloom initial date Entire lake 1987–2011stage CyanobacteriaAquatic vegetation bloom frequency/ HuLiu et al. et [15] al. [41] EntireEntire lake lake 2003–20132000–2008 spatialinitial/duration/area/specific/temporal dynamic date Duan et al. Wang et al. [2] ChlCyanobacteriaa monthly spatial bloom initial dynamic date EntireEntire lake lake 2002–20081987–2011 [5,34] Shi et al. [26] Chla spatial/temporal dynamic Entire lake 2003–2013 Liu et al. [41] Aquatic vegetation spatial/temporal dynamic Entire lake 2003–2013 Annual and monthly Wang et al. [2] Chla monthly spatial dynamic Entire lake excludingEntire lake 2002–2008 Zhang et al. [6] cyanobacteria bloom 2001—2013 Shi et al. [26] Chla spatial/temporal dynamic East LakeEntire lake 2003–2013 Zhang et al. Annual andfrequency monthly /cyanobacteriainitial date /bloomduration frequency/initial Entire lake excluding 2001-–2013 [6] date/duration East Lake

FigureFigure 4. EvolutionEvolution of of cyanobacteria cyanobacteria blooms blooms in 2007. in 2007.

In additionIn addition to to the the qualitative qualitative comparison between between our our results results with with others, others, we also we quantitatively also quantitatively validatedvalidated our our ﬁndings findings using using thethe Taihu Lake Lake cyanobacteria cyanobacteria bloom bloom remote remote sensing sensing inversion inversion dataset dataset (2017), which was produced and contributed by Ma et al. [5,23,57]. The dataset was supported by (2017), which was produced and contributed by Ma et al. [5,23,57]. The dataset was supported by Lake-Watershed Science Data Center, National Earth System Science Data Sharing Infrastructure, Lake-Watershed Science Data Center, National Earth System Science Data Sharing Infrastructure, National Science & Technology Infrastructure of China, http:// lake.geodata.cn. This dataset uses the Nationalalgae pixel Science-growing & Technology algorithm (APA), Infrastructure which can of extract China, floating http: //cyanobacterialake.geodata.cn with. higher This dataset accuracy. uses the algaeLinear pixel-growing regression analysis algorithm between (APA), our which results can and extract limited ﬂoatingFAd data cyanobacteria of Ma et al. in 2017 with shows higher that accuracy. there is good consistency between our results and those of Ma et al., with a linear slope close to 0.88 and a coefficient of determination (R2) close to 0.7 (Figure 5).

Remote Sens. 2019, 11, 2269 11 of 32

Linear regression analysis between our results and limited FAd data of Ma et al. in 2017 shows that there is good consistency between our results and those of Ma et al., with a linear slope close to 0.88 Remote Sens. 2019, 11, x FOR PEER REVIEW2 11 of 32 and aRemote coeﬃ Sens.cient 2019 of, 11 determination, x FOR PEER REVIEW (R ) close to 0.7 (Figure5). 11 of 32 1500 1500 1400 1400 1300 1300 1200 1200 1100 1100

) 1000 2

) 1000 2 900

(km Y = 0.8799X + 14.75

d 900 (km 800 Y = 0.8799XR² = 0.6854 + 14.75 d 800 R² =N 0.6854 = 63 700 N = 63 700 600 600 Ma et al.'s FA al.'s et Ma 500 Ma et al.'s FA al.'s et Ma 500 400 400 300 300 200 200 100 100 0 0 0 100 200 300 400 500 600 700 800 900 1000 1100 1200 1300 1400 1500 0 100 200 300 400 500 600 700 800 9002 1000 1100 1200 1300 1400 1500 Our FAd (km ) 2 Our FAd (km )

Figure 5. Scatter plot of our daily cyanobacteria bloom areas (FAd) versus that of Ma et al. FigureFigure 5. Scatter 5. Scatter plot plot of of our our daily daily cyanobacteriacyanobacteria bloom bloom areas areas (FA (FAd) versusd) versus that thatof Ma of et Ma al. et al. 3.1.3. Validation Using Landsat Interpretation Data 3.1.3.3.1.3. Validation Validation Using Using Landsat Landsat Interpretation Interpretation Data Qualified Landsat images were limited due to their 16-day revisit interval and cloud QualifiedQualified Landsat Landsat images images were were limited limited due to due their to 16-day their revisit 16-day interval revisit and interval cloud and contamination. cloud contamination. We only used around 300 Landsat TM/ETM+/OLI images for validation, therefore, We onlycontamination. used around We 300only Landsat used around TM/ETM 300 Landsat+/OLI images TM/ETM+/OLI for validation, images for therefore, validation, not therefore, every MODIS not every MODIS FAd result had corresponding Landsat validation data. However, this method can not every MODIS FAd result had corresponding Landsat validation data. However, this method can FAd resultbe regarded had corresponding as a systematic Landsat sampling validation process data. (Figure However, 6). Since this Landsat method TM/ETM+/OLI can be regarded and as a be regarded as a systematic sampling process (Figure 6). Since Landsat TM/ETM+/OLI and systematicTerra/MODIS sampling have process different (Figure local6 crossing). Since times, Landsat it should TM /ETM be noted+/OLI that and wind Terra weather/MODIS or havechanging different Terra/MODIS have different local crossing times, it should be noted that wind weather or changing local crossingcloud state times, could it affect should the be final noted consistency that wind between weather the ortwo changing kinds of clouddata but, state in couldmost cases, affect they the final cloud state could affect the final consistency between the two kinds of data but, in most cases, they consistencycan retain between general the spatial two kinds similarity of data during but, shortin most intervals cases, they(i.e., withincan retain a few general hours). spatial The linear similarity can retain general spatial similarity during short intervals (i.e., within a few hours). The linear duringregression short intervals between (i.e., our within FAd and a few Landsat hours). interpretation The linear data regression shows betweenthere is a our good FA correlationd and Landsat regression between our FAd and Landsat interpretation data shows there is a good correlation between the two datasets, with a linear slope equal to 0.86, R2 of 0.92, and a root mean square error interpretationbetween the data two showsdatasets, there with isa linear a good slope correlation equal to 0.86, between R2 of the0.92, two and datasets,a root mean with square a linear error slope (RMSE) of 39.252 km2 (Figure 7). 2 equal(RMSE) to 0.86, of R39.25of 0.92,km2 (Figure and a root7). mean square error (RMSE) of 39.25 km (Figure7). 1500 1500 TM ETM+ OLI TM ETM+ OLI

1200 1200 Landsat FAd Landsat FAd Our FAd Our FAd 900

) 900 2 ) 2 (km d (km d FA 600 FA 600

300 300

0 29-Feb-000 29-Jan-02 30-Dec-03 29-Nov-05 30-Oct-07 29-Sep-09 30-Aug-11 30-Jul-13 30-Jun-15 30-May-17 Date 29-Feb-00 29-Jan-02 30-Dec-03 29-Nov-05 30-Oct-07 29-Sep-09 30-Aug-11 30-Jul-13 30-Jun-15 30-May-17 Date Figure 6. Comparison between our daily cyanobacteria bloom areas (FAd) and that of Landsat Figure 6. Comparison between our daily cyanobacteria bloom areas (FAd) and that of Landsat Figure 6. Comparison between our daily cyanobacteria bloom areas (FAd) and that of Landsat interpretation. interpretation.interpretation.

Remote Sens. 2019, 11, 2269 12 of 32 Remote Sens. 2019, 11, x FOR PEER REVIEW 12 of 32

1400

1200

1000 ) 2

(km 800 d Y = 0.8636X + 0.2377 R² = 0.9231 600 RMSE = 39.25 N = 296 Landsat FA Landsat

400

200

0 0 200 400 600 800 1000 1200 1400 Our FA (km2) d

FigureFigure 7. Scatter 7. Scatter plot of plot our of daily our cyanobacteria daily cyanobacteria bloom areas bloom (FA d areas) versus (FA thatd) versus of Landsat that interpretation. of Landsat interpretation. 3.2. Temporal Coverage Patterns 3.2. Temporal Coverage Patterns The daily cyanobacteria bloom coverage areas (FAd) for the entire lake (excluding East Bay) and each segment are depicted in Figure8. Though synoptic interannual cyanobacteria bloom severity fluctuated, an obvious seasonal trend existed within each year: after a long-term bloom-free period, cyanobacteria survive and float to the water surface, relying on buoyancy. In July, August, and September, FAd is at its largest, and individual severe cyanobacteria blooms could occur in other months. Floating cyanobacteria finally retreat during November to February of the next year. Monthly maximum FAd is also depicted to represent the highly changeable floating cyanobacteria coverage due to the wind effect [15]. From the perspective of the entire lake, the floating cyanobacteria condition was moderate during 2000–2004. During this period, even the monthly maximum FA rarely exceeded half of the entire lake area. However, during 2005 to 2007, FA increased, in 2006 and 2007, a series of severe cyanobacteria bloom outbreaks occurred in the entire lake, which threatened the drinking water safety of the 2 million people nearby [5]. During 2008–2010, FA generally decreased. During 2011–2016, the condition fluctuated with a gradual deterioration. In 2017, a series of severe cyanobacteria blooms broke out over the entire lake over nearly the whole year. In terms of the frequency and density of floating cyanobacteria occurrence, this year’s severity level tended to overwhelm that observed in 2006 and 2007. In 2018, the condition improved but remained severe. As for Central Lake, South Lake and West Lake that occupy most of the entire lake, their FAa dynamics were similar to that of the entire lake. As for East Lake, two obvious periods were distinguished during the long-term series: the first period was before 2013, in which FAm and FAa in East Lake fluctuated moderately; after 2013, the second period, FAm and FAa began to decrease sharply, reaching their lowest levels in 2015 and 2016, and then recovered slightly in 2017 and 2018, but to a level still less than that observed during the first period. East Lake is not as cyanobacteria-dominated as other segments and many macrophytes existed in this segment over a long period, hence its FA results could not completely represent the cyanobacteria. Three bay segments, Zhushan Bay, Meiliang Bay, and Gong Bay, are all located in the north of Taihu Lake and have bad fluid connectivity. Their FAm and FAa fluctuated but were relatively stable and regular over the 19 years (Figure9). The FA results of Gong Bay were also contaminated by macrophytes, especially before 2005 (see Section 3.3). Even in some moderate years, seasons, or months,

Remote Sens. 2019, 11, 2269 13 of 32 individual and serious cyanobacteria blooms still occurred, which placed heavy stress on the lake’s environmentRemote Sens. 2019 and, 11, x water FOR PEER safety. REVIEW 13 of 32

1800

) 1500 Entire Lake 2 1200 (km

d 900

FA 600 300 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 600 )

2 500 Central Lake m 400 (k d 300 FA 200 100 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 300 ) 2 250 East Lake

(km 200 d 150 FA 100 50 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 600

) 500 South Lake 2 400 (km

d 300

FA 200 100 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 350 300 ) West Lake 2 250

(km 200 d 150 FA 100 50 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 200 Gong Bay ) 2 150 (km

d 100

FA 50 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 150

) Meiliang Bay

2 120

(km 90 d 60 FA 30 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18 100

) Zhushan Bay 2 80

(km 60 d 40 FA 20 0 1-Jan-00 1-Jan-02 1-Jan-04 1-Jan-06 1-Jan-08 1-Jan-10 1-Jan-12 1-Jan-14 1-Jan-16 1-Jan-18

Figure 8. Daily coverage area of ﬂoating cyanobacteria (FAd) for the entire lake (excluding East Bay) and each segment. FA represent ﬂoating algae index (FAI) > 0.004 coverage area. Solid circles d − represent the monthly maximum FAd.

Figure 8. Daily coverage area of floating cyanobacteria (FAd) for the entire lake (excluding East Bay) and each segment. FAd represent ﬂoating algae index (FAI) > −0.004 coverage area. Solid circles represent the monthly maximum FAd.

Remote Sens. 2019, 11, 2269 14 of 32 Remote Sens. 2019, 11, x FOR PEER REVIEW 15 of 32

600 300 500 Entire Lake FAm FAa )

2 400 200 300 200 100 FAa FAa (km2) FAm FAm (km 100 0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 150 80 Central Lake FAm FAa ) 60 2 )

100 2 40 50 20 FAm FAm (km FAa FAa (km

0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 160 80 East Lake FAm )

2 120 60 )

FAa 2

80 40 FAm FAm (km 40 20 FAa (km

0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 150 80 FAm FAa 120 South Lake ) 60 ) 2 90 2 40 60 FAa FAa (km FAm FAm (km 20 30 0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 150 80 FAm FAa 120 West Lake ) 60 ) 2 2 90 40 60 Am Am (km FAa(km F 30 20 0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 80 40 Gong Bay FAm FAa

) 60 30 ) 2 2 40 20 FAm FAm (km 20 10 FAa (km

0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 80 40 Meiliang Bay FAm FAa

) 60 30 ) 2 2 40 20

20 10 FAa (km FAm FAm (km

0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18 60 30 Zhushan Bay FAm FAa ) ) 2 40 20 2

20 10 FAa FAa (km FAm FAm (km

0 0 Jan-00 Jan-02 Jan-04 Jan-06 Jan-08 Jan-10 Jan-12 Jan-14 Jan-16 Jan-18

Figure 9. Floating cyanobacteria monthly (FAm) and annual (FAa) mean coverage area for the entire lake (excluding East Bay) and each segment. Black circles and red dots represent FAm and FAa, respectively;

gray upper shadow and red upper shadow represent positive deviation of FAm and FAa, respectively.

Figure 9. Floating cyanobacteria monthly (FAm) and annual (FAa) mean coverage area for the entire

lake (excluding East Bay) and each segment. Black circles and red dots represent FAm and FAa,

Remote Sens. 2019, 11, 2269 15 of 32

3.3. Spatial Distributions of Annual and Monthly Occurrence Frequency According to the cyanobacteria bloom occurrence accumulative frequency (Figure 10a), between 2000 and 2005, floating cyanobacteria frequently occurred in Zhushan Bay, Meiliang Bay, and the northwest shore regions. Gong Bay and East Lake showed high occurrence frequencies; these areas were dominated by macrophytes in the beginning of 21st century [41,42,58], and we can deduce that Taihu Lake was relatively healthy in this period, excluding some bays and nearshore regions. From 2005 to 2007, cyanobacteria blooms spread from the northwest regions to the Central Lake along the shoreline. Especially in 2007, the condition was much more serious, and the bloom occurrence frequency average of 2007 was doubled in comparison to the previous several years in most lake segments (Figure 10b). Many more cyanobacteria blooms occurred in South Lake than before, and Gong Bay and East Lake’s high occurrence frequencies declined sharply. According to the findings of previous studies and the frequency-based approach [41], these segments were dominated by macrophytes, and the macrophyte ecosystems of the above two segments were destroyed due to deterioration of the water environment [42]. In 2008 and 2009, the high frequency occurrence regions gradually shrank in the northwest lake area. The increase of high frequency occurrence areas in East Lake and Gong Bay indicates that the aquatic vegetation in these segments recovered to a certain extent. During 2010 to 2014, the high occurrence regions were mainly concentrated in the northeast lake area, Zhushan Bay, and Meiliang Bay; the overall lake condition was moderate in this period. Beginning from 2015, the occurrence of high frequency tended to spread from the northwest lake area to the entire lake again, and a series of heavy cyanobacteria blooms occurred over the entire lake around 2017. The severity in this year was worse than 2007, especially in Meiliang Bay, West Lake, and Central Lake (Figure 10b). Then, the lake condition in 2018 improved slightly in most segments. We also found that in 2015, the high frequency occurrence in Gong Bay and East Lake almost disappeared and did not recover to the previous state in the following years. In 2015, the local government organized a series of mechanized salvages in East Lake and Gong Bay [42], which destroyed macrophyte roots, preventing them from participating in natural restoration. The monthly occurrence frequencies of cyanobacteria blooms also had obvious spatial differences (Figure 11). Cyanobacteria blooms initially occurred in around April, with the occurrence locations mainly concentrated in Zhushan Bay, Meiliang Bay, Gong Bay, and the northwest nearshore regions. In summer, especially July and August, cyanobacteria blooms frequently occurred in West Lake, Zhushan Bay, and Meiliang Bay; simultaneously, Gong Bay and East Lake showed a high frequency of occurrence, mostly due to the macrophytes distributed in these areas growing rapidly in summer. From September to November, there was a gradual decrease in the coverage of high frequency; then, from December to the beginning of the next year, the floating cyanobacteria coverage was small. Remote Sens. 2019, 11, 2269 16 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 17 of 32

(a) 70 Central Zhushan Meiliang Gong West East South entire 60

10 Zoanl Zoanl average of annual frequency

0 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 Year (b)

Figure 10. (a) Spatial annual occurrence accumulative frequency of cyanobacteria blooms and (b) zonal average statistics of spatial annual occurrence frequency of cyanobacteria blooms. Cyanobacteria

blooms are deﬁned as when a pixel’s FAI is > 0.004. The results in East Lake and Gong Bay are due to − bothFigure cyanobacteria 10. (a) Spatial blooms annual and macrophytes,occurrence accumulative the same as frequency the initial of and cyanobacteria duration results blooms in and Section (b) 3.4. zonal average statistics of spatial annual occurrence frequency of cyanobacteria blooms. Cyanobacteria blooms are deﬁned as when a pixel’s FAI is > −0.004. The results in East Lake and Gong

Remote Sens. 2019, 11, x FOR PEER REVIEW 18 of 32

Bay are due to both cyanobacteria blooms and macrophytes, the same as the initial and duration Remoteresults Sens. 2019 in Section, 11, 2269 3.4. 17 of 32

(a) 200 Central 180 Zhushan Meiliang 160 Gong West 140 East 120 South entire 100 80 60 40 20 Zoanl average of monthly Zoanl average of monthly frequency 0 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Month (b)

FigureFigure 11. 11. (a)( aSpatial) Spatial accumulative accumulative monthly monthly occurrence occurrence frequency frequency of cyanobacteria of cyanobacteria blooms blooms for 2000 for– 2018;2000–2018; (b) zonal (b) average zonal average statistics statistics of monthly of monthly occurrence occurrence frequency. frequency.

3.4.3.4. Spatial Spatial Distributions Distributions of of Annual Annual Initial Initial Date Date and and Duration Duration AccordingAccording to to initial initial occurrence occurrence date date of of cyanobacteria cyanobacteria blooms blooms (Figures (Figure 12a,b),12a,b), before before 2004, 2004, only only ZhushanZhushan Bay, Bay, Meiliang Meiliang Bay, Bay, and and limited limited nearshore nearshore areas areas had had relatively relatively earlier earlier initial initial dates. dates. The The initial initial datesdates in in most most segments segments were were considerably considerably later, later, and and some some central central lake lake regions regions did did not not experienc experiencee cyanobacteriacyanobacteria blooms blooms at at all. all. However, However, during during 2004 2004 to to 2007, 2007, the the initial initial date date in in West West Lake, Lake, South South Lake, Lake, and Central Lake advanced signiﬁcantly. Between 2008 and 2016, the initial date of the entire lake was

Remote Sens. 2019, 11, x FOR PEER REVIEW 19 of 32 Remote Sens. 2019, 11, 2269 18 of 32 and Central Lake advanced significantly. Between 2008 and 2016, the initial date of the entire lake wasmoderate moderate and stable; and stable; in 2017, in it 2017, advanced it advanced signiﬁcantly, significantly, especially especially in Southwest in Southwest Lake and Lake nearshore and nearshoreareas. In 2018,areas. the In initial 2018, datethe initial was delayed, date was but delayed, in South but Lake, in MeiliangSouth Lake, Bay, Meiliang and Gong Bay, Bay, and it was Gong still Bay,quite it early.was still The quite cyanobacteria early. The bloom cyanobacteria duration (Figurebloom duration13a) is the (Figure diﬀerence 13a) between is the difference the last occurrence between thedate last and occu therrence initial occurrencedate and the date, initial hence, occurrence in the case date, of the hence, roughly in the stable case last of date,the roughly the spatial stable patterns last date,of the the duration spatial patterns were similar of the to duration the initial were date, similar i.e., for to somethe initial areas date, that i.e., had for early some (late) areas initial that dates, had earlytheir (late) durations initial tended dates, totheir become durations longer tended (shorter). to become longer (shorter).

(a)

Figure 12. Cont.

Remote Sens. 2019, 11, 2269 19 of 32 Remote Sens. 2019, 11, x FOR PEER REVIEW 20 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 20 of 32 300 Central Zhushan Meiliang Gong 300 WestCentral EastZhushan MeiliangSouth Gongentire 250 West East South entire 250

200 200

150 150

Zonal Zonal average of initial date 100

50 502000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2000 2002 2004 2006 2008Year2010 2012 2014 2016 2018 Year (b) (b) FigureFigure 12. 12. (a(a) )Spatial Spatial distributions distributions of ofthe the annual annual initial initial date date of cyanobacteria of cyanobacteria blooms blooms after after January; January; (b) Figure 12. (a) Spatial distributions of the annual initial date of cyanobacteria blooms after January; (b) zonal(b)zonal zonal average average average statistics statistics statistics of of the ofthe theannual annual annual initial initial initial date. date. date.

(a)

Figure(a) 13. Cont.

Remote Sens. 2019, 11, 2269 20 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 21 of 32

300

250

200

150

100 Central Zhushan Zonal Zonal average of duration Meiliang Gong 50 West East South entire 0 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 Year (b)

Figure 13. ((aa)) Spatial Spatial distributions distributions of of the the annual annual duration duration of of cyanobacteria cyanobacteria blooms; ( (b)) zonal average statistics of the annual bloom duration.

3.5. Temporal Temporal Characteristics of Significant Significant Cyanobacteria Blooms We defined defined significant significant cyanobacteria cyanobacteria blooms blooms as floatingas floating cyanobacteria cyanobacteria daily daily coverage coverage areas areas(FAd) (FAthatd exceeded) that exceeded 25% of 25% the of entire the entire lake or lake each or segmenteach segment area [ 15area], and [15], their and firsttheir and first last and occurrence last occurrence dates datesare depicted are depicted in Figure in Figure 14. For 14. the For entire the entire lake, significantlake, significant cyanobacteria cyanobacteria bloom bloom occurrence occurrence dates dates were werelate during late during 2000 to 2000 2005, to advanced 2005, advanced between between 2004 to 2008,2004 to retreated 2008, retreated between between 2008 to 2015, 2008 and to 2015, advanced and advancedagain between again 2016 between to 2018. 2016 For to 2018. 2008, For 2016, 2008, and 2016, 2018, and significant 2018, significant cyanobacteria cyanobacteria blooms occurred blooms occurredquite early quite because, early in because, the prior in years, the prior especially years, for especially 2007 and for 2017, 2007 the an cyanobacteriad 2017, the cyanobacteria blooms were bloomsrather severe, were rather and the severe, early significantand the early cyanobacteria significant cyanobacteria bloom occurrence bloom dates occurrence in 2008 and dates 2018 in were2008 andextensions 2018 were of their extensions previous of years. their Forprevious open lakeyears. segments For open (West lake Lake, segments South (West Lake, Lake, and Central South Lake),Lake, andtheir Central significant Lake), cyanobacteria their significan bloomt cyanobacteria occurrence dates bloom were occurrence similar to thedates entire were lake. similar For bay to the segments entire lake.(Zhushan For bay Bay, segments Meiliang (Zhushan Bay, and GongBay, Meiliang Bay), their Bay, significant and Gong cyanobacteria Bay), their significant bloom occurrence cyanobacteria dates bloomwere earlier occurrence than the dates open were lake segments, earlier than mainly the due open to severe lake segments, eutrophication. mainly The due results to of severe East eutropLake andhication. Gong The Bay wereresults mixtures of East ofLake cyanobacteria and Gong Bay blooms were and mixtures macrophytes, of cyanobacteria especially blooms before 2015,and macrophytes,when macrophytes especially were before not being 2015, mechanically when macrophytes salvaged were and werenot being still prevailing.mechanically salvaged and were Asstill mentioned prevailing. in the previous section, significant cyanobacteria blooms were defined as FAd that exceeded 25% of the entire lake or each segment area. To exclude the land adjacent effect, aquatic vegetation coverage, and floating cyanobacteria extraction deviation, we regarded FAd that was less than 5%361 of the entire lake (or segments) area as null floating cyanobacteria. We also defined FAd that was greater331 than 5% but less than 25% of the entire lake (or segments) area as slight cyanobacteria blooms. 301 According271 to the accumulative percentage of the cyanobacteria bloom severity level (Figure 15), the entire241 lake experienced five stages: before 2004, significant cyanobacteria blooms rarely occurred; 211 in 2006Date 181 and 2007, they increased rapidly; then declined in 2008; a moderate level was maintained between151 2009 and 2015; and their proportion then increased after 2016. The temporal dynamics of 121 slight cyanobacteria91 blooms were similar those of the significant blooms. The accumulative percentage 61 of cyanobacteria bloom severity levels of open lake segments (West Lake, SouthStart_day Lake, andEnd_day Central 31 Lake) were1 similar to the entire lake. The periodic changes in cyanobacteria-dominated Zhushan Bay

and Meiliang2000 2002 Bay,2004 2006 were2008 2010 2012 similar2014 2016 2018 to2001 those2003 2005 2007 observed2009 2011 2013 2015 in2017 the2000 2002 entire2004 2006 2008 lake,2010 2012 but2014 2016 their2018 2001 signiﬁcant2003 2005 2007 2009 2011 and2013 2015 slight2017 severity levels occupiedEntire Lake a greater proportion.Central Lake In other words,West Lake those bays were muchSouth Lake unhealthier

Remote Sens. 2019, 11, x FOR PEER REVIEW 21 of 32

300

250

200

150

100 Central Zhushan Zonal Zonal average of duration Meiliang Gong 50 West East South entire 0 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 Year (b)

Figure 13. (a) Spatial distributions of the annual duration of cyanobacteria blooms; (b) zonal average statistics of the annual bloom duration.

3.5. Temporal Characteristics of Significant Cyanobacteria Blooms We defined significant cyanobacteria blooms as floating cyanobacteria daily coverage areas (FAd) that exceeded 25% of the entire lake or each segment area [15], and their first and last occurrence dates are depicted in Figure 14. For the entire lake, significant cyanobacteria bloom occurrence dates were late during 2000 to 2005, advanced between 2004 to 2008, retreated between 2008 to 2015, and advanced again between 2016 to 2018. For 2008, 2016, and 2018, significant cyanobacteria blooms occurred quite early because, in the prior years, especially for 2007 and 2017, the cyanobacteria Remoteblooms Sens. were 2019 ,rather 11, x FOR severe, PEER REVIEWand the early significant cyanobacteria bloom occurrence dates in22 2008of 32 and 2018 were extensions of their previous years. For open lake segments (West Lake, South Lake, and Central361 Lake), their significant cyanobacteria bloom occurrence dates were similar to the entire Remote331 Sens. 2019, 11, 2269 21 of 32 lake. For301 bay segments (Zhushan Bay, Meiliang Bay, and Gong Bay), their significant cyanobacteria bloom271 occurrence dates were earlier than the open lake segments, mainly due to severe than241 the average status. Due to much of Gong Bay’s macrophytes being vanished after 2005 [35], eutrop211hication. The results of East Lake and Gong Bay were mixtures of cyanobacteria blooms and macrophytes,we deduceDate 181 that especially the Gong before Bay’s cyanobacteria2015, when macrophytes bloom condition were deterioratednot being mechanically in recent years, salvaged especially and after151 2015. East Lake, as a well-known macrophyte-dominated segment before 2015, demonstrated were still121 prevailing. similar 91 temporal patterns to Gong Bay. 61 Start_day End_day 31 3611 331 301 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 271 East Lake Zhushan Bay Meiliang Bay Gong Bay 241 Figure211 14. Timing and duration of significant cyanobacteria blooms for the entire lake (excluding

Date East181 Bay) and each segment. 151 121 As91 mentioned in the previous section, significant cyanobacteria blooms were defined as FAd that 61 exceeded 25% of the entire lake or each segment area. To exclude the land Start_dayadjacent effect,End_day aquatic 31 vegetation1 coverage, and floating cyanobacteria extraction deviation, we regarded FAd that was less d than 5% of2000 the2002 entire2004 2006 2008 lake2010 2012 (or2014 2016 segments)2018 2001 2003 2005 area2007 2009 as2011 2013 null2015 2017 floating2000 2002 2004 cyanobacteria.2006 2008 2010 2012 2014 2016 2018 We2001 also2003 2005 defined2007 2009 2011 2013 FA2015 2017 that was greater than Entire5% Lakebut less than 25%Central of theLake entire lake (or Westsegments) Lake area as slightSouth Lakecyanobacteria blooms.Remote Sens. 2019, 11, x FOR PEER REVIEW 22 of 32 According to the accumulative percentage of the cyanobacteria bloom severity level (Figure 15), 361 the entire331 lake experienced five stages: before 2004, significant cyanobacteria blooms rarely occurred; in 2006301 and 2007, they increased rapidly; then declined in 2008; a moderate level was maintained 271 between241 2009 and 2015; and their proportion then increased after 2016. The temporal dynamics of slight 211 cyanobacteria blooms were similar those of the significant blooms. The accumulative

Date 181 percentage151 of cyanobacteria bloom severity levels of open lake segments (West Lake, South Lake, and Central121 Lake) were similar to the entire lake. The periodic changes in cyanobacteria-dominated 91 Zhushan61 Bay and Meiliang Bay, were similar to those observed in the entire lake,Start_day but their significantEnd_day and slight31 severity levels occupied a greater proportion. In other words, those bays were much 1 unhealthier than the average status. Due to much of Gong Bay’s macrophytes being vanished after 2005 [35], we2000 2002 deduce2004 2006 2008 that2010 2012 the2014 2016 Gong2018 2001 Bay’s2003 2005 2007 cyanobacteria2009 2011 2013 2015 2017 2000 bloom2002 2004 2006 condition2008 2010 2012 2014 deteriorated2016 2018 2001 2003 2005 2007 in2009 recent2011 2013 2015 years,2017 East Lake Zhushan Bay Meiliang Bay Gong Bay especially after 2015. East Lake, as a well-known macrophyte-dominated segment before 2015, Figure 14. Timing and duration of significant cyanobacteria blooms for the entire lake (excluding demoFigurenstrated 14. similarTiming andtemporal duration patterns of signiﬁcant to Gong cyanobacteria Bay. blooms for the entire lake (excluding East East Bay) and each segment. Bay) and each segment.

100%As mentioned in the previous section, significant cyanobacteria blooms were defined as FAd that exceeded90% 25% of the entire lake or each segment area. To exclude the land adjacent effect, aquatic vegetation80% coverage, and floating cyanobacteria extraction deviation, we regarded FAd that was less than 5%70% of the entire lake (or segments) area as null floating cyanobacteria. We also defined FAd that was greater60% than 5% but less than 25% of the entire lake (or segments) area as slight cyanobacteria 50% blooms. 40% According30% to the accumulative percentage of the cyanobacteria bloom severity level (Figure 15), the entire20% lake experienced five stages: before 2004, significant cyanobacteria blooms rarely occurred; in 200610% and 2007, they increased rapidly; then declined in 2008; a moderate level was maintained Accumulative percentage Accumulative between0% 2009 and 2015; and their proportion then increased after 2016. The temporal dynamics of

slight cyanobacteria2000 2002 2004 2006 2008 2010 blooms2012 2014 2016 2018 were2001 2003 similar2005 2007 2009 2011 those2013 2015 2017 of2000 2002 the2004 2006 significant2008 2010 2012 2014 2016 blooms.2018 2001 2003 2005 The2007 2009 accumulative2011 2013 2015 2017 percentage of cyanobacteriaEntire Lake bloom severityCentral Lake levels of open lakeWest segments Lake (West Lake,South South Lake Lake, and Null Slight Significant Central Lake) were similar to the entire lake. The periodic changes in cyanobacteria-dominated Zhushan Bay and Meiliang Bay, were similarFigure to those 15. Cont. observed in the entire lake, but their significant and slight severity levels occupied a greater proportion. In other words, those bays were much unhealthier than the average status. Due to much of Gong Bay’s macrophytes being vanished after 2005 [35], we deduce that the Gong Bay’s cyanobacteria bloom condition deteriorated in recent years, especially after 2015. East Lake, as a well-known macrophyte-dominated segment before 2015, demonstrated similar temporal patterns to Gong Bay.

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% Accumulative percentage Accumulative 0% 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 Entire Lake Central Lake West Lake South Lake Null Slight Significant

Remote Sens. 2019, 11, 2269 22 of 32

Remote Sens. 2019, 11, x FOR PEER REVIEW 23 of 32

100% 90% 80% 70% 60% 50% 40% 30% 20% 10% Accumulative percentage Accumulative 0% 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 2000 2002 2004 2006 2008 2010 2012 2014 2016 2018 2001 2003 2005 2007 2009 2011 2013 2015 2017 East Lake Zhushan Bay Meiliang Bay Gong Bay Null Slight Significant

Figure 15. Accumulative percentage of cyanobacteria bloom severity level for the entire lake (excluding East Bay) and each segment. Significant, slight, and null cyanobacteria blooms are defined as FigureFAI > 15.0.004 Accumulativecoverage area percentage(FA ) larger ofthan cyanobacteria 25%, larger than bloom 5% but severity less than level or equal for the to 25%, entire and lake less − d (excludingthan or equal East to Bay) 5% of and entire each lake segment. or each Significant, segment, respectively.slight, and null cyanobacteria blooms are defined as FAI > −0.004 coverage area (FAd) larger than 25%, larger than 5% but less than or equal to 25%, and 3.6. Redundancyless than or equal Analysis to 5% (RDA) of entire between lake or Environmental each segment, Drivingrespectively. Factors and Cyanobacteria Bloom Characteristics 3.6. RedundancyAccording toAnalysis the monthly (RDA) RDAbetween analysis Environmental (Figure 16 Driving), temperature, Factors and sunshine Cyanobacteria radiance, Bloom wind speed, Characteristics and precipitation were positively correlated with cyanobacteria blooms (both FAmean and FAmax), but mostAccording water to quality the monthly factors RDA were analysisnegatively (Figure correlated 16), temperature, or nearly uncorrelated sunshine radiance, with monthly wind speed,cyanobacteria and precipitation blooms, except were COD positivelyMn. The correlatedT’s monthly with and cyanobacteriaannual individual blooms explanations (both FA tomean FA meanand FAweremax), 0.273 but and most 0.168, water respectively; quality factors while were the T negativelymin’s monthly correlated and annual or near individually uncorrelated explanations with monthlyto FAmean cyanobacteriawere 0.283 and blooms, 0.212, except respectively CODMn. (TableThe T’s4). monthly Hence FAandmean annualhad individual a stronger explanations relationship towith FA Tmeanmin thanwere T. 0.273 Previous and studies 0.168, have respectively; found that while higher the T min Tminin’s wintermonthly promotes and annual the cyanobacteria individual explanationssurvival over towinter, FAmean and were there 0.283 is a significant and 0.212, correlation respectively that (Table exists between 4). Hence T min FAandmean initialhad a blooming stronger relationshipdate [5], because with cyanobacteriaTmin than T. Previous grow better studies at highhave temperatures found that higher [59,60 T].min Wind in winter speed promotes affects FA themax cyanobacteriamore than FA survivalmean because over w windinter, helps and there cyanobacteria is a significant to form correlation surface that scum exists [15], between especially Tmin when and initialthe wind blooming speed date is less [5], than because 4 m cyanobacteria/s [61]. However, grow annual better at environmental high temperatures variables [59,60]. have Wind di ffspeederent affectseffects; FA watermax more quality than factors, FAmean because especially wind nutrient helps loadingscyanobacteria (TN andto form TP)) surface were positivelyscum [15], correlatedespecially whenwith annualthe wind cyanobacteria speed is less bloomsthan 4 m/s (Figure [61]. 17However,), probably annual because environmental sporadic cyanobacteria variables have blooms different are effects;affected water by meteorological quality factors, factors especially to a large nutrient extent, loadings e.g., wind (TN speed, and windTP)) were direction, positively and temperature, correlated withbut frequent annual cyanobacteria and large-scale blooms cyanobacteria (Figure blooms17), probably from the because interannual sporadic perspective cyanobacteria are dominated blooms are by affectednutrient distribution by meteorological and abundance factors [7 to,26 a,62 large,63]. Meteorological extent, e.g., wind factors speed, were thought wind direction, to play a more and temperature,important role but in frequent cyanobacteria and large bloom-scale formation, cyanobacteria especially blooms when from Taihu the Lake’s interannual nutrient perspective loadings were are dominatedsufficient [ 64by,65 nutrient]. Intensive distribution precipitation and abundance carries nutrient [7,26,62,63]. loadings Meteorological from adjacent factors catchments were thought to the tolake play [59 ], a therefore, more important precipitation role was in cyanobacteria also positively bloom correlated formation, with monthly especially and annual when Taihucyanobacteria Lake’s nutrientblooms as loadings what we were got (Figures sufficient 16 and [64,65]. 17). Intensive precipitation carries nutrient loadings from adjacent catchments to the lake [59], therefore, precipitation was also positively correlated with monthly and annual cyanobacteria blooms as what we got (Figures 16 and 17).

Remote Sens. 2019, 11, 2269 23 of 32

Table 4. Redundancy analysis (RDA) input variables and related statistics.

Monthly Permutation Annual permutation Individual Type Item Abbreviation Dimension Test Test Explanation P-Value 1 F-Value P-Value F-Value Monthly Annual

Monthly mean of daily mean temperature T ◦C 0.002 ** 104.364 0.374 0.831 0.273 0.168 Monthly mean of daily maximum T C 0.002 ** 101.183 0.838 0.101 0.265 0.010 temperature max ◦ Monthly mean of daily minimum T C 0.002 ** 104.251 0.406 0.791 0.283 0.212 temperature min ◦ Monthly mean of daily mean relative RH % 0.140 2.525 0.644 0.256 0.015 0.117 humidity Meteorological Monthly cumulative precipitation from P mm 0.046 * 3.928 0.292 0.784 0.029 0.167 factors 20:00 to 08:00 20-8 Monthly cumulative precipitation from P mm 0.040 * 4.436 0.228 1.567 0.010 0.125 08:00 to 20:00 8-20 Monthly cumulative precipitation from P mm 0.014 * 5.725 0.178 1.531 0.024 0.186 20:00 to 20:00 20-20 Monthly amount of total radiation TR MJ/m2 0.002 ** 29.516 0.752 0.161 0.078 0.010 Monthly amount of sunshine hours SH h 0.002 ** 15.073 0.778 0.129 0.034 0.008 Monthly number of days with sunshine SD days 0.028 * 4.239 0.848 0.098 0.006 0.002 time greater than 60% in a day 60 Monthly number of days with sunshine SD days 0.014 * 6.696 0.694 0.199 0.012 0.037 time less than 20% in a day 20 Monthly number of days with sunshine time SD days 0.112 2.317 0.448 0.421 0.010 0.007 greater than 20% but less than 60% in a day 20-60 Monthly maximum wind speed WSmax m/s 0.028 * 5.566 0.524 0.380 0.018 0.079 Monthly mean wind speed at 08:00 WS8 m/s 0.058 3.078 0.210 1.672 0.008 0.292 Monthly mean wind speed at 14:00 WS14 m/s 0.196 1.690 0.226 1.426 0.001 0.245 Monthly mean wind speed at 20:00 WS20 m/s 0.216 1.533 0.136 2.059 0.006 0.286 Total nitrogen TN mg/L 0.002 ** 24.194 0.388 0.730 0.146 0.135 Water quality Total phosphorus TP mg/L 0.838 0.077 0.658 0.217 0.001 0.135 factors Ratio of total nitrogen to total phosphorus TN/TP - 0.008 ** 8.432 0.634 0.318 0.079 0.021 Chemical oxygen demand CODMn mg/L 0.010 ** 8.247 0.748 0.168 0.066 0.110 Remote Sens. 2019, 11, 2269 24 of 32

Table 4. Cont.

Monthly Permutation Annual permutation Individual Type Item Abbreviation Dimension Test Test Explanation P-Value 1 F-Value P-Value F-Value Monthly Annual Monthly/annual mean area of floating FA km2 cyanobacteria mean Cyanobacteria Monthly maximum area of floating 2 bloom FAmax km cyanobacteria characteristics Initial date of significant floating FA day cyanobacteria si End date of significant floating FA day cyanobacteria se Duration of significant floating FA days cyanobacteria sd 1 * P < 0.05, ** P < 0.01. Remote Sens. 2019, 11, x FOR PEER REVIEW 24 of 32 RemoteRemote Sens. Sens. 20192019, ,1111, ,x 2269 FOR PEER REVIEW 2425 of of 32 32

FigureFigure 16. RedundancyRedundancy analysis analysis (RDA) (RDA) ordination ordination diagram diagram of of monthly monthly cyanobacteria bloom Figurecharacteristicscharacteristics 16. Redundancy and and environmental analysis variables. (RDA) ordination diagram of monthly cyanobacteria bloom characteristics and environmental variables.

FigureFigure 17. 17. RDARDA ordination ordination diagram diagram of of annual annual cyanobacteria cyanobacteria bloom bloom characteristics characteristics and and environmental environmental Figurevariables. 17. RDA ordination diagram of annual cyanobacteria bloom characteristics and environmental variables. 4. Discussion 4. Discussion 4.1. Accuracy Deviation Sources in Our Workflow 4.1. Accuracy Deviation Sources in Our Workflow

Remote Sens. 2019, 11, 2269 26 of 32

4. Discussion

4.1. Accuracy Deviation Sources in Our Workflow The validation results show that for rapid and automatic monitoring purposes, our workflow provides acceptable accuracy; however, some deviation sources remained. Firstly, compared with Rayleigh-corrected reflectance data, the MODIS land surface reflectance data (MOD09) used in our study are sensitive to ambient disturbance, which leads to some inevitable noisy, negative, or even a lack of pixel data, as well as patches [50,66]. FAI > 0.004 was used to distinguish cyanobacteria − bloom pixels from non-cyanobacteria blooms [15], but no field investigation was performed to validate the accuracy using this critical threshold [15]. Liu et al. recommended FAI > 0.025 as a cyanobacteria − bloom extraction threshold in Taihu Lake according to their field investigation [41]. Since most of the previous studies used FAI > 0.004 to distinguish cyanobacteria pixels in Taihu Lake, we also adopted − this threshold to allow for comparison of results. FAI could just distinguish dense cyanobacteria blooms that have formed floating scums and have not mixed with the water column [15,16,52], so it is more suitable for monitoring high Chla concentration cyanobacteria blooms [13]. The FAI threshold can extract some other types of aquatic vegetation and contaminate the final statistics, including emerged and floating-leaved macrophytes, which would lead to an overestimation of cyanobacteria blooms, especially for East Lake and Gong Bay, but this deviation should not affect the long-term temporal and large-scale spatial patterns. Regardless, macrophytes distributed in the above two segments can be deduced according to previous studies and the frequency-based approach [41,58]. Another accuracy uncertainty was the land surface adjacent effect. Stray light from brighter land pixels that are adjacent to a dark water target would lead to increased radiance of NIR bands [67]; thus, the water pixels’ FAI values would be elevated according to the FAI formula, which would cause erroneous segmentation. The land surface adjacent effect may be influential for up to many kilometers offshore [67,68], but the land mask cannot be shrunk too much toward the water, under the consideration of comprehensive statistics. Cloud contamination was another deviation source. When processing floating cyanobacteria pixels covered by thin cloud, visual inspection could be used to identify them and classify those pixels to the final floating cyanobacteria statistics, but the single-band threshold would mask thin cloud, so an underestimation of cyanobacteria blooms may have occurred in some cloudy images. The cloud mask would also remove some pixels that were contaminated by the adjacent effect, insufficient atmospheric correction, satellite sensor mechanical fault, and other accidental factors.

4.2. Environmental Driving Factors of Taihu Lake’s Severe Cyanobacteria Blooms around 2017 Massive nutrient loading input to Taihu Lake came from sewerage, agriculture fertilizer, and livestock drainage. These factors were generally considered fundamental in creating the 2007 cyanobacteria bloom crisis [5,8], which affected water safety and environment of millions of citizens. Other related environmental variables also helped to create a suitable environment for cyanobacteria bloom outbreak, e.g., temperature and wind [5,69]. After that crisis, the local government has made considerable efforts to control the nutrient loadings and pollutants enter Taihu Lake over the whole catchment. After over one decade, opposite trends were observed among some key water quality indices according to monitoring data for the entire lake: TN, NH3-N, and CODMn declined rapidly after the 2007 peak, then remained at a moderate level; TP declined after the 2007 peak, then remained moderate between 2008 and 2014, rose rapidly after 2015, and reached a new peak around 2017 [36]. During the past decade, measures to control catchment pollutants have apparently been effective. However, the huge floods in 2015 and 2016 carried a huge amount of nitrogen and phosphorus to Taihu Lake. Most of the nitrogen with soluble form was discharged after the floods, but most of the phosphorus was deposited in the lake in particulate form. The reserved phosphorus that accumulated in lake sediment was gradually released by means of Microcystis growth, acidity, and oxidation changes [35]. Taihu Lake’s severe cyanobacteria blooms in 2017 were attributed to high Remote Sens. 2019, 11, 2269 27 of 32

TP concentration, high temperature, and frequent weak wind, all of which boost cyanobacteria bloom formation [70], which is consistent with our RDA analysis. The role of nitrogen is still ambiguous to some extent, some studies indicated that reducing phosphorus is more effective at causing a decline in cyanobacteria abundance [8,11,69,71,72]. We also believe that reducing TP by the means of controlling external and endogenous sources would be a feasible long-term management strategy, especially under the condition that explanation of TP was 0.135 in annual perspective, much higher than its monthly explanation (0.001) (Table4). Long-term climate warming and wind speed decreasing over the past several decades also create a suitable environment for cyanobacteria blooms [37,59,64]. Therefore, establishing a cyanobacteria bloom monitoring system is vital towards taking preventative action in protecting water environments and safety. As has been mentioned in above sections, FAI is not sensitive enough to distinguish between cyanobacteria and macrophyte [58], so our obtained cyanobacteria bloom spatiotemporal patterns were contaminated by macrophytes more or less, especially in East Lake and Gong Bay. Several studies have indicated that large areas of macrophytes disappeared in East Lake and Gong Bay after 2015, primarily due to mechanized salvage by the local government [42,62,73]. Taihu Lake has changed from a macrophyte-dominated state to a phytoplankton-dominated state under anthropogenic pressure [37]. The existence of an aquatic macrophyte ecosystem would help improve the underwater light environment [8], stabilize phosphorus in the sediment [73], and suppress cyanobacteria growth [74]. Local government and scientists have made great efforts to recover macrophytes; however, artificial planting and restoration in the early 2000s was unsuccessful, largely due to the water environment being unsuitable for macrophyte growth. The artificial restoration of macrophytes can only succeed when the level of nutrient loading is reduced to an appropriate level [75].

4.3. Broader Applications of Our Workflow Inland water has more complicated inherent optical properties (IOPs) than ocean water [76]. Atmospheric correction, the adjacent effect, and other complicated inland water remote sensing problems hinder the data mining and extension of the current methodology to different research locations [77]. We embedded and integrated related processing procedures into a cloud platform (GEE), achieved an acceptable accuracy, and obtained the cyanobacteria bloom spatiotemporal patterns of Taihu Lake. We believe our workflow has the potential to be applied to other lakes and coastal waters for similar purposes. Researchers should attempt to embed different atmospheric correction algorithms and cyanobacteria bloom extraction methods for different inland waters when using our workflow. The floating cyanobacteria coverage we extracted can only represent the horizontal distribution, but not the cyanobacteria biomass, which must consider the vertical distribution of cyanobacteria biomass in the water column. Thus, related remote sensing inversion algorithms may be embedded into our workflow in future. The FAI threshold method in our workflow could potentially be improved by the algae pixel-growing algorithm (APA) [57] or the histogram adaptive threshold, etc. Overall, further efforts are urgently required to embed many types of conventional inversion algorithms and operational processes into cloud platforms, which will help us find new information by combining remote sensing big data and specific natural or social scenarios.

5. Conclusions In this study, we developed an operational cyanobacteria bloom monitoring workflow based on Google Earth Engine, processed nearly 7000 MODIS daily images, and retrieved spatial and temporal patterns of cyanobacteria blooms in China’s Taihu Lake, over 19 years (2000 to 2018). Our workflow provides a basis for automatic remote sensing monitoring of cyanobacteria blooms. Several conclusions were reached in our study: (1) Some sources of error in our workflow were reported, such as the FAI threshold, the land surface adjacent effect, and cloud contamination. However, inspected by the validation using in-situ data and remote sensing observations, the workflow still achieved an acceptable accuracy for rapid Remote Sens. 2019, 11, 2269 28 of 32 and automatic monitoring purposes. Our workflow can improve the automation and efficiency of routine environmental management of Taihu Lake and has the potential to be applied to other inland and coastal water areas. (2) From the perspective of temporal coverage distribution, cyanobacteria blooms in Taihu Lake were moderate before 2006, peaked in 2007, declined rapidly after 2008, remained moderate and stable until 2015, and then reached another peak around 2017. Several lake segments displayed similar temporal trends to the entire lake, especially for those open lake segments. From the perspective of interannual spatial distribution, bay areas and the northwest lake area had more severe cyanobacteria blooms than open lake segments. From the perspective of monthly spatial distribution, most cyanobacteria blooms primarily occurred in April, became severe in July and August, then improved after October. We also found that macrophytes distributed in East Lake and Gong Bay disappeared after 2015 and have not yet been restored. (3) According to the redundancy analysis, meteorological factors (temperature, wind speed, precipitation, etc.) were positively correlated with cyanobacteria bloom characteristics, from both monthly and interannual perspectives; however, nutrient loadings were only positively correlated with cyanobacteria bloom characteristics from an annual perspective. Reducing total phosphorous (TP) by controlling external and endogenous sources, together with restoring macrophyte ecosystem, would be the feasible long-term management strategies for Taihu Lake.

Supplementary Materials: (1) The following is available online at http://www.mdpi.com/2072-4292/11/19/2269/ s1, Figure S1: monthly and annual average of environmental factors (including meteorological and water quality factors) in Taihu Lake. (2) Link of GEE scripts of our study: https://code.earthengine.google.com/ e6c4d627ec7f4f0dbc1a4f77fdeb3bb3. Author Contributions: T.J. designed and coded the workflow on GEE and analyzed the data, T.J., X.Z. and R.D. drafted and revised the paper. Funding: This research was supported by the National Key Research and Development Program (2017YFC0506006 and 2016YFC0503605). Acknowledgments: The authors thank the anonymous reviewers for their valuable comments and suggestions. The authors acknowledge the data support from Lake-Watershed Science Data Center, National Earth System Science Data Sharing Infrastructure, National Science & Technology Infrastructure of China (http://lake.geodata.cn), National Ecosystem Research Network of China (www.cnern.org), and Taihu Basin Authority of Ministry of Water Resources (www.tba.gov.cn). Conflicts of Interest: The authors declare no conflict of interest.

References

1. Sellner, K.G.; Doucette, G.J.; Kirkpatrick, G.J. Harmful algal blooms: Causes, impacts and detection. J. Ind. Microbiol. Biotechnol. 2003, 30, 383–406. [CrossRef][PubMed] 2. Wang, M.; Shi, W.; Tang, J. Water property monitoring and assessment for China’s inland Lake Taihu from MODIS-Aqua measurements. Remote Sens. Environ. 2011, 115, 841–854. [CrossRef] 3. Kudela, R.M.; Palacios, S.L.; Austerberry, D.C.; Accorsi, E.K.; Guild, L.S.; Torres-Perez, J. Application of hyperspectral Remote Sensing to cyanobacterial blooms in inland waters. Remote Sens. Environ. 2015, 167, 196–205. [CrossRef] 4. Blondeau-Patissier, D.; Gower, J.F.R.; Dekker, A.G.; Phinn, S.R.; Brando, V.E. A review of ocean color Remote Sensing methods and statistical techniques for the detection, mapping and analysis of phytoplankton blooms in coastal and open oceans. Progr. Oceanogr. 2014, 123, 123–144. [CrossRef] 5. Duan, H.; Ma, R.; Xu, X.; Kong, F.; Zhang, S.; Kong, W.; Hao, J.; Shang, L. Two-Decade Reconstruction of Algal Blooms in China’s Lake Taihu. Environ. Sci. Technol. 2009, 43, 3522–3528. [CrossRef] 6. Zhang, Y.; Ma, R.; Duan, H.; Loiselle, S.; Xu, J. Satellite analysis to identify changes and drivers of CyanoHABs dynamics in Lake Taihu. Water Sci. Technol. Water Suppl. 2016, 16, 1451–1466. [CrossRef] 7. Chen, Y.; Qin, B.; Teubner, K.; Dokulil, M.T. Long-term dynamics of phytoplankton assemblages: Microcystis-domination in Lake Taihu, a large shallow lake in China. J. Plankton Res. 2003, 25, 445–453. Remote Sens. 2019, 11, 2269 29 of 32

8. Zhang, M.; Shi, X.; Yang, Z.; Yu, Y.; Shi, L.; Qin, B. Long-term dynamics and drivers of phytoplankton biomass in eutrophic Lake Taihu. Sci. Total Environ. 2018, 645, 876–886. [CrossRef] 9. Qin, B.; Yang, G.; Ma, J.; Wu, T.; Li, W.; Liu, L.; Deng, J.; Zhou, J. Spatiotemporal Changes of Cyanobacterial Bloom in Large Shallow Eutrophic Lake Taihu, China. Front Microbiol. 2018, 9, 451. [CrossRef] 10. Peretyatko, A.; Teissier, S.; Backer, S.D.; Triest, L. Assessment of the risk of cyanobacterial bloom occurrence in urban ponds: Probabilistic approach. Annal. Limnolog. Int. J. Limnol. 2010, 46, 121–133. [CrossRef] 11. Descy, J.P.; Leprieur, F.; Pirlot, S.; Leporcq, B.; Van Wichelen, J.; Peretyatko, A.; Teissier, S.; Codd, G.A.; Triest, L.; Vyverman, W.; et al. Identifying the factors determining blooms of cyanobacteria in a set of shallow lakes. Ecol. Inform. 2016, 34, 129–138. [CrossRef] 12. Rigosi, A.; Hanson, P.; Hamilton, D.P.; Hipsey, M.; Rusak, J.A.; Bois, J.; Sparber, K.; Chorus, I.; Watkinson, A.J.; Qin, B.; et al. Determining the probablity of cyanobacteria blooms: the application of Bayesian networks in multiple lake systems. Ecol. Appl. 2015, 25, 186–199. [CrossRef][PubMed] 13. Oyama, Y.; Fukushima, T.; Matsushita, B.; Matsuzaki, H.; Kamiya, K.; Kobinata, H. Monitoring levels of cyanobacterial blooms using the visual cyanobacteria index (VCI) and floating algae index (FAI). Int. J. Appl. Earth Observ. Geoinform. 2015, 38, 335–348. [CrossRef] 14. Kuster, T. Quantitative detection of chlorophyll in cynaobacteria bloom by satellite. Remote Sens. Limnol. Oceanogr. 2004, 49, 2179–2189. 15. Hu, C.; Lee, Z.; Ma, R.; Yu, K.; Li, D.; Shang, S. Moderate Resolution Imaging Spectroradiometer (MODIS) observations of cyanobacteria blooms in Taihu Lake, China. J. Geophys. Res. Oceans 2010, 115.[CrossRef] 16. Hu, C. A novel ocean color index to detect floating algae in the global oceans. Remote Sens. Environ. 2009, 113, 2118–2129. [CrossRef] 17. Zhang, Y.; Feng, L.; Li, J.; Luo, L.; Yin, Y.; Liu, M.; Li, Y. Seasonal-spatial variation and Remote Sensing of phytoplankton absorption in Lake Taihu, a large eutrophic and shallow lake in China. J. Plankton Res. 2010, 32, 1023–1037. [CrossRef] 18. Duan, H.; Ma, R.; Hu, C. Evaluation of Remote Sensing algorithms for cyanobacterial pigment retrievals during spring bloom formation in several lakes of East China. Remote Sens. Environ. 2012, 126, 126–135. [CrossRef] 19. Hu, C. An empirical approach to derive MODIS ocean color patterns under severe sun glint. Geophys. Res. Lett. 2011, 38.[CrossRef] 20. Feng, L.; Hu, C.; Chen, X.; Cai, X.; Tian, L.; Gan, W. Assessment of inundation changes of Poyang Lake using MODIS observations between 2000 and 2010. Remote Sens. Environ. 2012, 121, 80–92. [CrossRef] 21. Hu, C.; Li, D.; Chen, C.; Ge, J.; Muller-Karger, F.E.; Liu, J.; Yu, F.; He, M.-X. On the recurrent Ulva prolifera blooms in the Yellow Sea and East China Sea. J. Geophys. Res. Oceans 2010, 115.[CrossRef] 22. Huang, C.; Li, Y.; Yang, H.; Sun, D.; Yu, Z.; Zhang, Z.; Chen, X.; Xu, L. Detection of algal bloom and factors influencing its formation in Taihu Lake from 2000 to 2011 by MODIS. Environ. Earth Sci. 2013, 71, 3705–3714. [CrossRef] 23. Zhang, Y.; Ma, R.; Zhang, M.; Duan, H.; Loiselle, S.; Xu, J. Fourteen-Year Record (2000–2013) of the Spatial and Temporal Dynamics of Floating Algae Blooms in Lake Chaohu, Observed from Time Series of MODIS Images. Remote Sens. 2015, 7, 10523–10542. [CrossRef] 24. Kabenge, M.; Wang, H.; Li, F. Urban eutrophication and its spurring conditions in the Murchison Bay of Lake Victoria. Environ. Sci. Pollut. Res. Int. 2016, 23, 234–241. [CrossRef][PubMed] 25. Wang, M.; Hu, C. Mapping and quantifying Sargassum distribution and coverage in the Central West Atlantic using MODIS observations. Remote Sens. Environ. 2016, 183, 350–367. [CrossRef] 26. Shi, K.; Zhang, Y.; Zhou, Y.; Liu, X.; Zhu, G.; Qin, B.; Gao, G. Long-term MODIS observations of cyanobacterial dynamics in Lake Taihu: Responses to nutrient enrichment and meteorological factors. Sci. Rep. 2017, 7, 40326. [CrossRef][PubMed] 27. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary- scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [CrossRef] 28. Murphy, S.; Wright, R.; Rouwet, D. Color and temperature of the crater lakes at Kelimutu volcano through time. Bull. Volcanol. 2017, 80.[CrossRef] 29. Griffin, C.G.; McClelland, J.W.; Frey, K.E.; Fiske, G.; Holmes, R.M. Quantifying CDOM and DOC in major Arctic rivers during ice-free conditions using Landsat TM and ETM+ data. Remote Sens. Environ. 2018, 209, 395–409. [CrossRef] Remote Sens. 2019, 11, 2269 30 of 32

30. Lin, S.; Novitski, L.N.; Qi, J.; Stevenson, R.J. Landsat TM/ETM+ and Machine-Learning Algorithms for Limnological Studies and algal Bloom Management of Inland Lakes; SPIE: Bellingham, WA, USA, 2018; Volume 12, pp. 1–17. 31. Traganos, D.; Aggarwal, B.; Poursanidis, D.; Topouzelis, K.; Chrysoulakis, N.; Reinartz, P. Towards Global-Scale Seagrass Mapping and Monitoring Using Sentinel-2 on Google Earth Engine: The Case Study of the Aegean and Ionian Seas. Remote Sens. 2018, 10, 1227. [CrossRef] 32. Traganos, D.; Poursanidis, D.; Aggarwal, B.; Chrysoulakis, N.; Reinartz, P. Estimating Satellite-Derived Bathymetry (SDB) with the Google Earth Engine and Sentinel-2. Remote Sens. 2018, 10, 859. [CrossRef] 33. Chen, J.; Quan, W.; Zhang, M.; Cui, T. A Simple Atmospheric Correction Algorithm for MODIS in Shallow Turbid Waters: A Case Study in Taihu Lake. IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. 2013, 6, 1825–1833. [CrossRef] 34. Duan, H.; Ma, R.; Zhang, Y.; Loiselle, S.A. Are algal blooms occurring later in Lake Taihu? Climate local effects outcompete mitigation prevention. J. Plankton Res. 2014, 36, 866–871. [CrossRef] 35. Zhu, W.; Tan, Y.; Wang, R.; Feng, G.; Chen, H.; Liu, Y.; Li, M. The trend of water quality variation and analysis in typical area Lake Taihu, 2010-2017. J. Lake Sci. 2018, 30, 296–305. [CrossRef] 36. Gu, X.; Zeng, Q.; Mao, Z.; Chen, H.; Li, H. Water environment change over the period 2007 - 2016 and the strategy of fishery improve the water quality of Lake Taihu. J. Lake Sci. 2019, 31, 305–318. [CrossRef] 37. Shi, K.; Zhang, Y.; Zhang, Y.; Li, N.; Qin, B.; Zhu, G.; Zhou, Y. Phenology of Phytoplankton Blooms in a Trophic Lake Observed from Long-Term MODIS Data. Environ. Sci. Technol. 2019, 53, 2324–2331. [CrossRef] [PubMed] 38. Huang, L.; Fang, H.; He, G.; Jiang, H.; Wang, C. Effects of internal loading on phosphorus distribution in the Taihu Lake driven by wind waves and lake currents. Environ. Pollut. 2016, 219, 760–773. [CrossRef] 39. Li, J.; Hu, C.; Shen, Q.; Barnes, B.B.; Murch, B.; Feng, L.; Zhang, M.; Zhang, B. Recovering low quality MODIS-Terra data over highly turbid waters through noise deduction and regional vicarious calibratioin adjustment: A case study in Taihu Lake. Remote Sens. Environ. 2017, 197, 72–84. [CrossRef] 40. Jiang, G.; Ma, R.; Loiselle, S.A.; Duan, H.; Su, W.; Cai, W.; Huang, C.; Yang, J.; Yu, W. Remote Sensing of particulate organic carbon dynamics in a eutrophic lake (Taihu Lake, China). Sci. Total. Environ. 2015, 532, 245–254. [CrossRef] 41. Liu, X.; Zhang, Y.; Shi, K.; Zhou, Y.; Tang, X.; Zhu, G.; Qin, B. Mapping Aquatic Vegetation in a Large, Shallow Eutrophic Lake: A Frequency-Based Approach Using Multiple Years of MODIS Data. Remote Sens. 2015, 7, 10295–10320. [CrossRef] 42. Wang, S.; Gao, Y.; Li, Q.; Gao, J.; Zhai, S.; Zhou, Y.; Cheng, Y. Long-term and inter-monthly dynamics of aquatic vegetation and its relation with environmental factors in Taihu Lake, China. Sci. Total Environ. 2019, 651, 367–380. [CrossRef][PubMed] 43. Vermote, E.W.; Wolfe, R. MOD09GQ MODIS/Terra Surface Reflectance Daily L2G Global 250m SIN Grid V006 [Data set]. NASA EOSDIS LP DAAC 2015.[CrossRef] 44. Vermote, E.W. MODIS/Terra Surface Reflectance Daily L2G Global 1kmand 500m SIN Grid V006 [Data set]. NASA EOSDIS LP DAAC 2015.[CrossRef] 45. Wang, S. Tropical state assessment of global inland waters using a MODIS-derived Forel-Ule index. Remote Sens. Environ. 2018.[CrossRef] 46. Liu, R.; Liu, Y. Generation of new cloud masks from MODIS land surface reflectance products. Remote Sens. Environ. 2013, 133, 21–37. [CrossRef] 47. Chen, P.Y.; Srinivasan, R.; Fedosejevs, G.; Narasimhan, B. An automated cloud detection method for daily NOAA-14 AVHRR data for Texas, USA. Int. J. Remote Sens. 2002, 23, 2939–2950. [CrossRef] 48. Chen, Z.; Hu, C.; Muller-Karger, F. Monitoring turbidity in Tampa Bay using MODIS/Aqua 250-m imagery. Remote Sens. Environ. 2007, 109, 207–220. [CrossRef] 49. Wang, S.; Li, J.; Zhang, B.; Shen, Q.; Zhang, F.; Lu, Z. A simple correction method for the MODIS surface reflectance product over typical inland waters in China. Int. J. Remote Sens. 2016, 37, 6076–6096. [CrossRef] 50. Feng, L.; Hu, C.; Li, J. Can MODIS Land Reflectance Products be Used for Estuarine and Inland Waters? Water Resour. Res. 2018, 54, 3583–3601. [CrossRef] 51. Wang, S.; Yang, M.; Li, J.; Shen, Q.; Zhang, F. MODIS surface reflectance product (MOD09) validation for typical inland waters in China. SPIE Proc. 2014, 9261, 92610F. [CrossRef] 52. Moradi, M. Comparison of the efficacy of MODIS and MERIS data for detecting cyanobacterial blooms in the southern Caspian Sea. Mar. Pollut. Bull. 2014, 87, 311–322. [CrossRef][PubMed] Remote Sens. 2019, 11, 2269 31 of 32

53. Legendre, P.; Anderson, M.J. Distance-based redundancy analysis: Testing multispecies responses in multifactorial ecological experiments. Ecol. Monogr. 1999, 69, 1–24. [CrossRef] 54. Legendre, P.; Gallagher, E.D. Ecologically meaningful transformations for ordination of species data. Oecologia 2001, 129, 271–280. [CrossRef][PubMed] 55. Chen, S.; Zhang, W.; Zhang, J.; Jeppesen, E.; Liu, Z.; Kociolek, J.P.; Xu, X.; Wang, L. Local habitat heterogeneity determines the differences in benthic diatom metacommunities between different urban river types. Sci. Total. Environ. 2019, 669, 711–720. [CrossRef] 56. Huang, Y.; Huang, J. Coupled effects of land use pattern and hydrological regime on composition and diversity of riverine eukaryotic community in a coastal watershed of Southeast China. Sci. Total. Environ. 2019, 660, 787–798. [CrossRef][PubMed] 57. Zhang, Y.; Ma, R.; Duan, H.; Loiselle, S.A.; Xu, J.; Ma, M. A novel algorithm to estimate algae bloom coverage to subpixel resolution in Taihu Lake. IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens. 2014, 7, 3060–3068. [CrossRef] 58. Liang, Q.; Zhang, Y.; Ma, R.; Loiselle, S.; Li, J.; Hu, M. A MODIS-Based Novel Method to Distinguish Surface Cyanobacterial Scums and Aquatic Macrophytes in Lake Taihu. Remote Sens. 2017, 9, 133. [CrossRef] 59. Paerl, H.W.; Huisman, J. Bloom Like It Hot. Science 2008, 320, 57–58. [CrossRef][PubMed] 60. Visser, P.M.; Verspagen, J.M.H.; Sandrini, G.; Stal, L.J.; Matthijs, H.C.P.; Davis, T.W.; Paerl, H.W.; Huisman, J.

How rising CO2 and global warming may stimulate harmful cyanobacterial blooms. Harmful Algae 2016, 54, 145–159. [CrossRef] 61. Cao, H.-S.; Kong, F.-X.; Luo, L.-C.; Shi, X.-L.; Yang, Z.; Zhang, X.-F.; Tao, Y. Eﬀects of Wind and Wind-Induced Waves on Vertical Phytoplankton Distribution and Surface Blooms ofMicrocystis aeruginosain Lake Taihu. J. Freshw. Ecol. 2006, 21, 231–238. [CrossRef] 62. Zhang, M.; Yang, Z.; Shi, X. Expansion and drivers of cyanobacteria blooms in Lake Taihu. J. Lake Sci. 2019, 31, 336–344. [CrossRef] 63. Guo, C.; Zhu, G.; Paerl, H.W.; Zhu, M.; Yu, L.; Zhang, Y.; Liu, M.; Zhang, Y.; Qin, B. Extreme weather event may induce Microcystis blooms in the Qiantang River, Southeast China. Environ. Sci. Pollut. Res. Int. 2018, 25, 22273–22284. [CrossRef] 64. Zhang, Y.; Shi, K.; Liu, J.; Deng, J.; Qin, B.; Zhu, G.; Zhou, Y. Meteorological and hydrological conditions driving the formation and disappearance of black blooms, an ecological disaster phenomena of eutrophication and algal blooms. Sci. Total. Environ. 2016, 569–570, 1517–1529. [CrossRef] 65. Shi, K.; Zhang, Y.; Xu, H.; Zhu, G.; Qin, B.; Huang, C.; Liu, X.; Zhou, Y.; Lv, H. Long-Term Satellite Observations of Microcystin Concentrations in Lake Taihu during Cyanobacterial Bloom Periods. Environ. Sci. Technol. 2015, 49, 6448–6456. [CrossRef] 66. Feng, L.; Hou, X.; Li, J.; Zheng, Y. Exploring the potential of Rayleigh-corrected reﬂectance in coastal and inland water applications: A simple aerosol correction method and its merits. ISPRS J. Photogramm. Remote Sens. 2018, 146, 52–64. [CrossRef] 67. Matthews, M.W.; Odermatt, D. Improved algorithm for routine monitoring of cyanobacteria and eutrophication in inland and near-coastal waters. Remote Sens. Environ. 2015, 156, 374–382. [CrossRef] 68. Santer, R.; Carrere, V.; Dubuisson, P.; Roger, J.C. Atmospheric correction over land for MERIS. Int. J. Remote Sens. 1999, 20, 1819–1840. [CrossRef] 69. Huang, C.; Zhang, Y.; Huang, T.; Yang, H.; Li, Y.; Zhang, Z.; He, M.; Hu, Z.; Song, T.; Zhu, A.X. Long-term variation of phytoplankton biomass and physiology in Taihu lake as observed via MODIS satellite. Water Res. 2019, 153, 187–199. [CrossRef] 70. Zhu, W.; Chen, H.; Wang, R.; Feng, G.; Xue, Z.; Hu, S. Analysis on the reasons for the large bloom area of Lake Taihu in 2017. J. Lake Sci. 2019, 31, 621–632. [CrossRef] 71. Huang, J.; Zhang, Y.; Huang, Q.; Gao, J. When and where to reduce nutrient for controlling harmful algal blooms in large eutrophic lake Chaohu, China? Ecol. Indicat. 2018, 89, 808–817. [CrossRef] 72. Huo, S.; He, Z.; Ma, C.; Zhang, H.; Xi, B.; Xia, X.; Xu, Y.; Wu, F. Stricter nutrient criteria are required to mitigate the impact of climate change on harmful cyanobacterial blooms. J. Hydrol. 2019, 569, 698–704. [CrossRef] 73. Yang, L.; Yang, X.; Ren, L.; Qian, X.; Xiao, L. Mechanism and control strategy of cyanobacteria bloom in Lake Taihu. J. Lake Sci. 2019, 31, 18–27. [CrossRef] Remote Sens. 2019, 11, 2269 32 of 32

74. Pelicice, F.M.; Agostinho, A.A.; Thomaz, S.M. Fish assemblages associated with Egeria in a tropical reservoir: Investigating the eﬀects of plant biomass and diel period. Acta Oecolog. 2005, 27, 9–16. [CrossRef] 75. Zhang, Y.; Yao, X.; Qin, B. A critical review of the development, current hotspots, and future directions of Lake Taihu research from the bibliometrics perspective. Environ. Sci. Pollut Res. Int. 2016, 23, 12811–12821. [CrossRef] 76. Shi, K.; Zhang, Y.; Li, Y.; Li, L.; Lv, H.; Liu, X. Remote estimation of cyanobacteria-dominance in inland waters. Water Res. 2015, 68, 217–226. [CrossRef] 77. Palmer, S.C.J.; Kutser, T.; Hunter, P.D. Remote Sensing of inland waters: Challenges, progress and future directions. Remote Sens. Environ. 2015, 157, 1–8. [CrossRef]

© 2019 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (http://creativecommons.org/licenses/by/4.0/).