1 A Novel Algorithm to Estimate Algal Bloom Coverage to Subpixel

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

1

2

A Novel Algorithm to Estimate Algal Bloom Coverage to Subpixel Resolution in Lake Taihu

3

Yuchao Zhang, Ronghua Ma, Hongtao Duan, Steven A. Loiselle, Jinduo Xu, and Mengxiao Ma

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Abstract—Remote sensing has often been used to monitor the distribution and frequency of floating algae in inland aquatic environments. However, due to the spatial resolution of the most common satellite sensors, accurate determination of algae coverage remains a major technical challenge. Here, a novel algorithm to estimate floating algae area to subpixel scales, denominated the algae pixel-growing algorithm (APA), is developed and evaluated for a series of image data from the Moderate Resolution Imaging Spectroradiometer (MODIS). The algorithm utilizes the Rayleighcorrected reflectance (Rrc) and a floating algae index (FAI) derived from Rrc in three spectral bands. Comparison with concurrent Landsat TM=ETMþ data indicate that the APA provides more accurate estimates of both algal bloom area and algal bloom intensity (i.e., floating algae coverage) (RSE ¼ 15:2 km2 and RE ¼ 9:9%), compared to other commonly used methods such as the linear unmixing algorithm (LA). Furthermore, this study confirms that FAI is a better index with respect to normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) for the estimation of algae area coverage, especially when combined with the APA. Finally, the study provides a theoretical basis for the objective assessment of bloom severity in complex inland waterbodies.

26 Index Terms—Algae pixel-growing algorithm (APA), algal 27 blooms, floating algae index (FAI), Landsat, linear unmixing 28 decomposition, Moderate Resolution Imaging Spectroradiometer 29 (MODIS).

30

I. INTRODUCTION

31 N RECENT decades, many major lakes have experienced 32 an increased occurrence of algal blooms, including Lake 33 Victoria in Africa, Lake Okeechobee in the United States, and 34 Lake Taihu in China [1]–[3]. Compared to traditional in situ

I

sampling and laboratory analysis, satellite-derived optical and thermal data provide synoptic and high frequency information, useful for early warning, real-time monitoring, and postbloom evaluations of bloom events [1]. Over the last three decades, there have been many important developments in the use of remotely sensed data to estimate phytoplankton biomass concentrations and the associated bio-optical properties [4], [5]. This has been particularly important where eutrophication and floating algal mats have compromised lake functioning. Information regarding the coverage and thickness of algal mats provides fundamental information necessary for biomass estimates, which are more appropriate than traditional chlorophyll-a concentration measurements for decision-making. Choosing an appropriate algal bloom algorithm is the first step to estimate bloom coverage. Single-bands, band ratios, and band differences have been used in a similar manner to land surface vegetation detection algorithms [6]–[8]. Indices such as the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), the normalized difference algae index (NDAI), maximum chlorophyll index (MCI), red tide index (RI), the floating algae index (FAI), and the ocean surface algal blooms index (OSABI) have been utilized to identify floating algal blooms [9]–[15]. But in practical application, NDVI, EVI, and FAI are much more popular than the other indices. However, it has been found that NDVI and EVI are particularly sensitive to aerosol characteristics (type and thickness), sun glint, and solar viewing geometry with respect to FAI, which has been found to be relative stable under different environmental and observing conditions [13]. Minimizing the atmospheric effects is vital to time-series study and characterization on algal blooms and water optical properties in turbid inland lakes [16], [17]. Thus, FAI appears to be a reasonable choice to detect and quantify bloom coverage. However, algal mats can be smaller than the satellite pixel size, leading to uncertainties in the total bloom coverage estimates. Indeed, algal blooms may present complex temporal and spatial characteristics, which may limit the use of satellite data. Most algorithms use data from the Landsat TM=ETMþ, Moderate Resolution Imaging Spectroradiometer (MODIS), and Medium Resolution Imaging Spectrometer (MERIS), with spatial resolutions of 30 m, 250=500=1000 m, and 300=1200 m with temporal resolutions of 16 days, 1 day, and 3 days, respectively. Among these, MODIS has the best temporal resolution, but its best spatial resolution is 250 m, which may not detect small patches of floating algae and may lead to uncertainties in area coverage estimates. The ability to detect and quantify surface features at subpixel scales would improve the use of satellite systems for bloom monitoring and management.

IE E Pr E oo f

1

Manuscript received September 02, 2013; revised January 22, 2014; accepted May 12, 2014. This work was supported in part by the National High Technology Research and Development Program of China (2014AA06A509), in part by the National Natural Science Foundation of China under Grant 41101316, Grant 41171271, and Grant 41171273, in part by Scientific Data Sharing Platform for Lake and Watershed, in part by Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences (NIGLAS), in part by the “135” Program of NIGLAS (NIGLAS2012135010 and NIGLAS2012135014), and in part by the Dragon 3 program (10561). MODIS data were provided by the U.S. NASA, and the Landsat TM=ETMþ data were provided by the U.S. Geological Survey. (Corresponding author: Ronghua Ma.) Y. Zhang, R. Ma, H. Duan, and J. Xu are with the State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China (e-mail: rhma@niglas. ac.cn). S. A. Loiselle is with the CSGI, University of Siena, Siena 253100, Italy (e-mail: [email protected]). M. Ma is with the State Key Laboratory of Pollution Control & Resources Reuse, School of the Environment, Nanjing University, Nanjing 210023, China (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTARS.2014.2327076

1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76

77 78 79 80 81 82

2

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112

IE E Pr E oo f

F1:1 Fig. 1. Study site was Lake Taihu, China lying in the Yangtze River Delta. The lake is often divided into several lake segments, which include Zhushan Bay (ZB), F1:2 Meiliang Bay (MB), Gong Bay (GB), West Lake (WL), South Lake (SL), Central Lake (CL), East Lake, and East Taihu. The brown areas represent major cities.

Bloom detection and quantification at subpixel scales depend on the feature size (the relative percentage of the algae bloom within a pixel) and the sensitivity of the satellite sensor (signal to noise ratio). The accuracy of algae coverage estimates is strongly influenced by mixed pixels containing both algae and algae free (water) surfaces. When the total lake area coverage is small, mixed pixels may represent a large portion of the total algae pixels. Hu et al. [18] have used a linear unmixing algorithm (LA) to estimate the total algae area coverage in Lake Taihu, an eutrophic lake in East China. Shanmugam et al. [15] have also applied the similar algorithm to detect ocean surface algal blooms in subpixel scales. However, this algorithm needs to be further improved to allow for the determination of algae coverage at subpixel scales and more accurate estimation of threshold values for algae and nonalgae areas of optically complex eutrophic lakes. In China, satellite-based estimations of algal blooms (coverage, distribution, and duration) are used by environmental management agencies at both state and local levels for a variety of purposes, from policy implementation and water management (including the physical removal of the algae mats). Coverage estimates are generated by different agencies using a variety of methods, most of which rely on human interpretation of the manually adjusted satellite images (to account for variable atmospheric and observing conditions as well as algorithm artifacts). The large degree of subjectivity has resulted in incompatible estimates and inconsistent recommendations among agencies. There is a clear need for an objective method, which would be valid for variable conditions, including the partial algae coverage at subpixel scales. An objective approach would allow

different agencies and groups to generate consistent estimates based on the same data. In this study, we developed a new algorithm [algae pixelgrowing algorithm (APA)] to estimate algae coverage at subpixel scales using MODIS data. We evaluated the accuracy of the APA using concurrent higher resolution (30 m) Landsat TM=ETMþ data by comparing their spatial distributions and coverage statistics. In addition, several practical issues in using MODIS for bloom monitoring were examined, including an automatic and objective determination of the pure algae threshold as well as a method to adjust pixels contaminated by striping noise. The ultimate goal is to provide an objective method to determine bloom coverage and size to establish a long-term baseline record and real-time assessment of bloom severity to support management decisions.

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127

II. STUDY AREA AND DATA SOURCES

128

A. Study Area

2

With an extension of 36800 km and encompassing major population centers such as Shanghai, Suzhou, Wuxi, and Changzhou, the Lake Taihu basin is one of the most highly developed areas of China (Fig. 1). At the center of the Basin is Lake Taihu, China’s third-largest freshwater lake (2400 km2 ). The lake provides drinking water for more than 2 million people, and sustains important fisheries including crabs, carp, and eels [2]. With the rapid economic development in the last decades, Lake Taihu has become highly eutrophic with frequent cyanobacteria blooms as a consequence of increased nutrient loads

129

130 131 132 133 134 135 136 137 138 139

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

3

TABLE I

T1:1 ACQUISITION DATE OF TERRA/MODIS AND LANDSAT TM (MARKED WITH ASTERISK)/ T1:2 ETMþ DATA USED IN THIS STUDY

Fig. 2. In the APA, a 3  3 window was chosen randomly from a MODIS image. Gray-scale shows the FAI value of each pixel. The central pixel was divided into two kinds of subpixels which have the same maximum and minimum FAI values of the other pixels present in the 3  3 window.

[19]–[21]. In 2007, a massive algal bloom led to a drinking water shortage for more than 1 million people [2], [22]. Although the Chinese government recognized the importance of managing nutrient loads and monitoring lake conditions, the long-term management of the Lake Taihu Basin is still in its infancy. Since 2007, remote sensing has been used to monitor the occurrence of algal blooms as part of an early warning system with daily communications to national and local decision makers.

between the red band (645 nm) and short-wave infrared band 176 (1240 or 1640 nm) 177 F AIMODIS ¼ Rrc ð859Þ  R0rc ð859Þ R0rc ð859Þ ¼ Rrc ð645Þ þ ½Rrc ð1240Þ  Rrc ð645Þ  ð859  645Þ=ð1240  645Þ

where Rrc is defined as the difference between the calibrated 178 sensor radiance [after adjustment for ozone (Lt  ) and other 179 gaseous absorption] and Rayleigh reflectance (Rr ) [24] 180 Rrc ¼ Lt  =F0 cos 0  Rr

149 B. Satellite Data

(1)

IE E Pr E oo f

140 141 142 143 144 145 146 147 148

150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170

Terra MODIS data at 250 and 500 m resolutions and Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper Plus (ETMþ) data at 30 m resolution were used in this study. Landsat TM=ETMþ Level-1 data were obtained from the Earth Resources Observation and Science Center (http:// glovis.usgs.gov). MODIS data were obtained from the LAADS website of the U.S. NASA Goddard Space Flight Center. Twenty-four pairs of concurrent (i.e., same day within a hour) MODIS and TM=ETMþ data (Table I) with low cloud coverage ( < 10%) were georeferenced (UTM projection) with an error of < 0:5 pixel. The 500-m resolution MODIS data at 1240 nm and 1640 nm were resampled to 250 m resolution (to match the 645-nm data). TM=ETMþ data were atmospherically corrected using the radiative transfer calculations based on the second simulation of the satellite signal in the solar spectrum [23]. MODIS data were corrected by removing the molecular (Rayleigh) scattering effects, and then converted to Rayleigh-corrected reflectance (Rrc ) following Hu et al. [24]. The lake segment of East Lake Taihu was excluded from the bloom analysis as this area is characterized by aquatic macrophytes growing from the bottom [25].

171

III. METHODOLOGY

172 A. Floating Algal Index Algorithm 173 The FAI algorithm for MODIS proposed by Hu [13] utilizes a 174 baseline subtraction, defined as the difference between reflec175 tance at 859 nm (vegetation “red edge”) and a linear baseline

F2:1 F2:2 F2:3 F2:4

(2)

where F0 is the extraterrestrial solar irradiance at data acquisition time and 0 is the solar zenith angle. FAI has been used to map bloom size to derive long-term statistics [18]. However, the estimation of bloom size is subject to large errors if a single threshold is used to determine the bloom presence or absence. Hu et al. [18] used an LA to estimate partial coverage at subpixel scales (pixel ) through a statistical analysis of the 9-year MODIS FAI dataset with FAI thresholds for 100% bloom coverage (F AIalgae ) and 0% bloom coverage (F AInonalgae ). When F AInonalgae < F AIpixel  F AIalgae , LA is expressed as

181 182 183 184 185 186 187 188 189 190 191

pixel ¼ ðF AIpixel  F AInonalgae Þ=ðF AIalgae  F AInonalgae Þ: (3)

B. Algae Pixel-Growing Algorithm (APA)

192

1) Theoretical Considerations: Considering n high-resolution 193 pixels (Rrc Hi ) that make up a low-resolution pixel (Rrc Lo ), where 194 Rrc Lo is the arithmetic mean of all Rrc Hi 195 Rrc Lo ¼ Rrc Hi ðiÞ:

(4)

Similarly, there exists a relationship between the low- 196 resolution pixel and high-resolution pixels (or subpixels 197 within the low-resolution pixel) for FAI 198 F AI pixel ¼ F AI subpixel :

(5)

If we assume that, in a 3  3 pixel window, the central pixel is 199 a function the maximum and minimum FAI values present within 200 the window (Fig. 2), (5) becomes 201 F AIcenter ¼   F AImax þ ð1   ÞF AImin

(6)

4

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

IE E Pr E oo f

Fig. 4. Changes in the MODIS-derived algae coverage area with increasing F4:1 iterations (steps) of the APA. The stars mark the true algal bloom area identified F4:2 from the corresponding Landsat TM=ETMþ data. F4:3

F3:1 Fig. 3. Flowchart of the APA process. 202 203 204 205 206

where  is the decomposition parameter of the 3  3 window that is determined based on the relationship between the known FAI values (center, max, and min). In a mixed pixel, the algae coverage is defined as the proportion () of the pixel covered by floating algae such that thresh thresh F AI ¼   F AIalgae þ ð1  Þ  F AIwater   thresh thresh ¼ F AIalgae  F AIwater  thresh þ F AIwater

(7)

thresh thresh 207 where F AIalgae and F AIwater are the FAI thresholds for pure 208 algae (100%) and nonalgae water (0%) pixels, respectively. thresh thresh 209 Assuming that F AIalgae and F AIwater are constant in a 3  3 210 window, the FAI of central pixel could be expressed as follows:

  thresh thresh  F AIwater F AIcenter ¼ F AIalgae  center þ

thresh F AIIwater :

(8)

211 The FAI of max and min pixels in a 3  3 window could also 212 be similarly expressed in the same way. Based on (6)–(8), FAI 213 has a linear relationship with the floating algae coverage  in the 214 mixed pixel

center ¼   max þ ð1   Þ  min

(9)

215 where max and min are the algae coverage of pixels with 216 maximum and minimum FAI values in a 3  3 window, 217 respectively.

Fig. 5. Estimation of floating algae coverage using the APA (the aquatic macrophytes area marked with gray color), from the initial estimation (step 1), the determination of the growing points (step 2), and the final estimation of algal coverage after 3 iterations (step 3). A and B are the ends of a transect within the lake.

F5:1 F5:2 F5:3 F5:4 F5:5

2) APA Implementation and Data Processing: There are three iterative steps in applying the APA to MODIS Rrc data (Fig. 3). 1) The preprocessing step identifies the location of the maximum and minimum FAI value in a roaming 3  3 window, and then calculates  for the central pixel using (6). 2) The “seed” pixels are then identified in two ways: a) if local pixels are found to have thresh their F AI values > F AIalgae , these pixels are taken as the seed pixels and their algae coverage are defined as 1.0 (100%); thresh b) if no local pixels have F AI values > F AIalgae , then local center pixels with maximum FAI values in the 3  3 window are taken as the seed pixels. The coverage is determined via (3) using thresh a pure-algae endmember threshold of F AIalgae . The coverage of other pixels is assigned 0.0 in the initial estimation. 3) The final algae coverage at each pixel is determined iteratively using (9) whenever the maximum coverage in each 3  3 window is not zero. The iteration is terminated when two consecutive calculations yield similar results. In practice, three iterations were necessary before the termination (see below). Through iterations, the algal bloom coverage expands from the initial pure algae

218 219 220 221 222 223 224 225

226

227 228 229 230 231 232 233 234 235 236

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

5

F6:1 Fig. 6. Comparison of algal bloom coverage derived from several different methods ( coverage interval ¼ 0:005): the full line marks the true bloom coverage using F6:2 30 m TM=ETMþ data resized to 250 m; the dashed lines show the estimate of bloom coverage from the concurrent MODIS image using the APA; the dotted lines F6:3 represent the estimate of bloom coverage from the same MODIS image using an LA.

250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273

274 275 276 277 278 279

pixels or high-coverage pixels to low-coverage pixels, according to the relationship between adjacent pixels in a 3  3 pixel window. For very thin algae layers, the measured reflectance may always be confused with surface waters no algal bloom but with high concentrations of chlorophyll-a. However, by including the constraints related to adjacent pixels with highalgae coverage, this error is reduced unless these conditions are present over a large geographic area. If such conditions are not present, the problematic pixel is assigned as nonalgae pixel. Total algae coverage is determined by the no zero algae coverage. 3) FAI Threshold for Pure Algae Coverage: The APA results thresh are sensitive to the choice of the F AIalgae and to the termination thresh conditions of the iterative calculation. The F AIalgae was determined from concurrent higher resolution TM=ETMþ observations, but could also be estimated using in situ reflectance spectra of the pure algae. Due to the lack of in situ reflectance data covering the spectral range of > 1000 nm, the former method was used. With a higher spatial resolution, Landsat TM=ETMþ data provide a better resolution of the pure algae pixels. We assumed that there were only pure algae pixels and pure nonalgae pixels existing in Landsat TM=ETMþ images. A pixel’s gradient was defined as the FAI difference from the adjacent pixels in a 3  3 window. It was found that the pixels associated with the maximum gradient determined mode could be used to separate floating algae from other nonbloom waters very well [18]. Using the large gradient in FAI values across the algae and nonalgae boundary, the mean value of all pixels associated with the gradient was used to represent the TM=ETMþ threshold value for pure algae. After resizing from 30 to 250 m (to match the MODIS resolution), the partial algae coverage in the new low-resolution pixel could be determined. If all TM=ETMþ pixels within the low-resolution pixel had 100% algae coverage, the low-resolution pixel could be identified as the pure algae pixel. From all these low-resolution pure algae pixels, the FAI value (2  standard deviations) was used to represent the threshold value for pure (100%) algae thresh coverage (F AIalgae ). 4) The Condition to Terminate Iterations: For each MODIS pixel, the algae coverage is a function of the FAI maximum and minimum in each 3  3 window. For each iteration, the estimated coverage increases until it reaches a stable value (Fig. 4). The termination condition allows the calculation to

IE E Pr E oo f

237 238 239 240 241 242 243 244 245 246 247 248 249

Fig. 7. North south profile of estimated algae coverage (0%–100%) from resized ETMþ pixels (all at a 250-m resolution) and MODIS using the APA and an LA, along a hypothetical transect from Meiliang Bay to Tiaoxi River (white line in the inset image). Each point represents an arithmetic mean of 3  3 pixels along the transect.

F7:1 F7:2 F7:3 F7:4 F7:5

conclude when the absolute difference or relative difference of consecutive steps reaches a minimum value. For variable bloom areas (from several hundreds to > 1000 km2 ) with millions of algae pixels, the absolute difference or relative difference inconsecutive steps can be significantly different between images. Importantly, it was found that the smallest relative difference between MODIS and TM=ETMþ data occurred after three iterations in all 24 concurrent MODIS/TM image pairs (Fig. 4). This is a natural consequence of the fact that mixed pixels are usually found adjacent to pure-algae pixels or high algae-coverage pixels. The APA approach builds algae coverage from pure-algae pixels with respect to nonalgae pixels (or from high-coverage pixels to low-coverage pixels). An example of these calculation steps is presented for a lake transect characterized by pure algae to nonalgae water in Lake Taihu on September 24, 2011 (Fig. 5). The algae coverage of some pixels varied at every iteration. In general, the algae coverage of mixed pixels increased gradually, as the algae coverage of the FAI maximum pixel or FAI minimum pixel changed in each 3  3 pixel. From the algae coverage distribution of each iteration, the algal bloom coverage expanded from the pure algae pixels or high-coverage pixels to low-coverage pixels, based on the condition of adjacent pixels in a 3  3 pixel window. In total, the algae coverage will be ignored if there is no algal bloom existing in adjacent pixels.

280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304

6

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

F8:1 Fig. 8. Comparison of the floating algae area estimated from TM=ETMþ and MODIS using the two algorithms [the APA and an (LA)] when different FAI threshold F8:2 values are used to represent pure (100%) algae coverage. u and δ represent the mean and standard deviation of the threshold values.

IV. RESULTS AND VALIDATION

306 A. Algal Bloom Coverage Estimations

307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345

B. Sensitivity Analysis

346

In this study, the MODIS threshold of pure algae was determined from concurrent TM=ETMþ observations. To examine how sensitive the MODIS results were to the threshold estimation, three MODIS images in different years (April 6, 2007; October 20, 2009; May 5, 2012) with different bloom coverage were compared with TM=ETMþ (Fig. 8). The bloom area was found to increase sharply when the threshold for pure algae coverage was lower than the mean minus the standard deviation. The results confirm that both the APA and LA significantly overestimated the algae coverage if the pure algae threshold was low. A higher threshold (mean plus standard deviation) resulted in a slight under estimation, which was less sensitive to further changes of the threshold. An artificial sensitivity test was also conducted by varying the pure algae threshold by 10% to þ10%. The total algae area from the APA estimates changed þ15:1% and 13:5% with respect to TM=ETMþ values, whereas the area from the LA estimates changed þ9:7% and 8:3%. In general, the change in the total algae area estimates was proportional to changes in the threshold values.

347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366

V. DISCUSSIONS

367

IE E Pr E oo f

F9:1 Fig. 9. Linear relationship between MODIS Rrc ð1240Þ and Rrc ð1650Þ from F9:2 several MODIS images. 305

provided a similar frequency distribution for MODIS data with resized TM=ETMþ images. In order to examine the algae coverage distributions derived from different methods, an example of the MODIS results on September 24, 2011 is shown in Fig. 7. An artificial transect from Meiliang Bay (North Lake) to the mouth of Tiaoxi River (South Lake) was chosen to show a range of algae coverage. In order to minimize the impact of image noise and geometric correction error, the average coverage of the 3  3 window was used to represent the central pixel along the transect. Both APA and LA yielded reasonable results when compared to ETMþ. However, both APA and LA underestimated algae coverage when the algae coverage was > 40% in the mixed pixels, and overestimated coverage when the percent coverage was very low. On average, the difference between MODIS and ETMþ was 13.7% when the APA was used and 15.4% when the LA was used.

Algal bloom coverage identified from the 24 TM=ETMþ images ranged from 14.8 to 505:7 km2 . The relative standard error (RSE) between MODIS estimates and TM=ETMþ estimates was 15.2 and 24:8 km2 , respectively, for the APA and LA methods, with their corresponding relative error (RE) of 9.9% and 17.3%. The results of APA and LA presented here should be interpreted as MODIS pixels with both 100% and partial algae bloom coverage. If we ignored the partial algae bloom coverage and took the mean FAI value of pixel with 10% algae coverage as the threshold of algae pixel and nonalgae pixel, algae coverage area from MODIS images is increased by > 40%, and its corresponding RE between MODIS estimates and TM=ETMþ estimates is more than 30%. Due to the partial coverage provided by APA and LA, their errors are much smaller (as gauged by the highresolution TM=ETMþ data). APA provided more accurate estimates than the LA, with the already small errors nearly halved. A comparison of MODIS algal bloom coverage with paired TM=ETMþ coverage, resized from 30 to 250 m, gave similar results (Fig. 6). When the pure algae pixels of MODIS were set with resized TM=ETMþ, LA over-estimated the number of pixels with partial algae coverage. The over-estimation indicated that the nonalgae threshold was low. The APA determination

A. Lake Specific FAI Thresholds

368

There are two options to determine FAI thresholds for long- 369 term applications: 1) threshold based on multi-image statistics 370

7

IE E Pr E oo f

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

F10:1 Fig. 10. Histogram distributions of Rrc ð1240Þ before and after removal of the Rrc ð1240Þ striping noise. 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387

[18]; and 2) threshold based on the individual images. Because environmental factors such as air temperature, wind speed and direction, and hydrological conditions all influence algal aggregation characteristics (i.e., algal bloom thickness) [26], the pure algae threshold (as well as the pure water threshold) may vary between images. Therefore, an image-specific threshold is more accurate. In this study, this was achieved by comparing concurrent MODIS and TM=ETMþ image pairs. However, for routine monitoring, TM=ETMþ data are not available at the MODIS temporal frequency and the pure algae threshold needs to be determined in alternative ways. Because the red-edge effect would lead to elevated reflectance in the near-IR and shortwave IR, the pure algae pixel was determined when Rrc (1240) was greater than Rrc (645) and FAI was great than 0. Comparison with the thresholds determined from the MODIS and TM=ETMþ pairs showed consistency, allowing for operational use of APA with MODIS data in absence of TM=ETMþ.

388 B. MODIS Striping Noise 389 390 391 392 393 394 395 396

One of the MODIS bands used to derive the FAI, the 1240-nm band on MODIS Terra, suffers from significant striping noise. This would seriously affect the FAI value and APA performance. In this study, the noise-contaminated data were replaced with coincident the 1640-nm data, using the linear relationship between the two bands derived from the high-quality (undisturbed) pixels of the same image. A significant linear relationship between Rrc ð1240Þ and Rrc ð1640Þ, especially in algae-covered

areas was evident for all the images examined (Fig. 9). Although the relationship varies among images, an image-specific relationship to derive Rrc ð1240Þ for noise-contaminated pixels can be used in FAI calculations. The results in indicate that such a correction yields different statistics in algae area coverage (Fig. 10).

397 398 399 400 401 402

C. Comparison of FAI, NDVI, and EVI

403

FAI has been used to calculate algae coverage because of its reduced sensitivity to variable environmental and observational conditions [18]. However, other indexes have also been widely utilized. Among these are NDVI [27] and EVI [28], defined as

404 405 406 407

NDV I ¼ ðRNIR  RRED Þ=ðRNIR þ RRED Þ EV I ¼ G  ðRNIR  RRED Þ=

ðRNIR þ C1  RRED  C2  RBLUE þ C3 Þ

(10)

(11)

where RNIR , RRED , and RBLUE are the reflectance in the nearinfrared (NIR), red and blue bands; G is the gain factor, and C1 , C2 , and C3 are the pixel-independent coefficients to compensate for aerosol effects and vegetation background. For MODIS data, G ¼ 2:5, C1 ¼ 6, C2 ¼ 7:5, and C3 ¼ 1. Given these alternative choices on the vegetation indexes, we examined if FAI is the best choice to derive the algae coverage with the APA. A MODIS Terra image for August 10, 2013 was used to evaluate algae coverage using FAI, NDVI, and EVI, where sun glint influences part of the image (Fig. 11). The FAI image

408 409 410 411 412 413 414 415 416 417

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

IE E Pr E oo f

8

F11:1 Fig. 11. Comparison among MODIS FAI, NDVI, and EVI values for MODIS data obtained on August 10, 2013 and corresponding algae coverage distributions F11:2 (derived from the APA). Areas with aquatic macrophytes are marked in gray. 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433

displayed a near homogeneous water background regardless of the sun glint. In contrast, the NDVI image revealed higher variability in the water background in areas where sun glint was significant. Sun glint led to a significant increase in the NDVI values, making it more difficult to differentiate algae pixels from the water background. The performance of EVI was much better than NDVI, but still showed a higher sensitivity to sun glint contamination with respect to FAI. The algae areas determined from the three images, all using the APA, were 178.04, 353.68, and 236:26 km2 , respectively. Thus, when sun glint was significant, both NDVI and EVI overestimated algae coverage. Because of the low-latitude (31 N) location of Lake Taihu, most images during the summer months contain sun glint. Thus, FAI is a preferred index for its tolerance to this as well as to the interference of thick aerosols (not shown here) for long-term, routine monitoring of the bloom coverage.

434 D. MODIS-Based Data Record and Event-Driven Response 435 436 437 438 439 440 441 442 443

The APA approach, together with the statistically determined FAI thresholds, provides an objective method to estimate the bloom severity. This can be used to estimate bloom spatial extent and effective algae coverage as well as to compare current bloom conditions against historical baselines determined using the same objective method. This will lead to more accurate estimates of the bloom severity in Lake Taihu in near real-time as well as pave the way to obtain consistent answers from various research groups and management agencies.

VI. CONCLUSION

444

Remote sensing has been widely used to assess algal blooms in Lake Taihu, but the methods and parameterization varied substantially between different users, making it difficult to compare results and agree on a common action. The present approach to develop and validate a more objective method to characterize the bloom severity provides a novel to meet this challenge. The results, based on the MODIS FAI data, showed improved performance over other methods or indices. The APA approach serves as an objective and more accurate method to determine bloom severity in both near real-time monitoring and historical analysis, thereby improving the capacity of decision makers to manage Lake Taihu and its basin.

445 446 447 448 449 450 451 452 453 454 455 456

ACKNOWLEDGMENT

457

The constructive comments from two anonymous reviewers 458 are greatly appreciated. 459 REFERENCES

460

[1] A. Reinart and T. Kutser, “Comparison of different satellite sensors in detecting cyanobacterial bloom events in the Baltic Sea,” Remote Sens. Environ., vol. 102, no. 1–2, pp. 74–85, 2006. [2] L. Guo, “Ecology-doing battle with the green monster of Taihu Lake,” Science, vol. 317, no. 5842, p. 1166, 2007. [3] S. Q. Zhao et al., “The 7-decade degradation of a large freshwater lake in central Yangtze River, China,” Environ. Sci. Technol., vol. 39, no. 2, pp. 431–436, 2005.

461 462 463 464 465 466 467 468

9

[4] S. Sathyendranath, “Remote sensing of ocean colour in coastal, and other optically-complex waters,” International Ocean-Colour Coordinating Group, pp. 47–76, 2000. [5] Z. P. Lee, “Remote sensing of inherent optical properties: Fundamentals, tests of algorithms, and applications,” International Ocean-Colour Coordinating Group, pp. 43–93, 2006. [6] P. S. Richard and C. T. Michelle, “Remote sensing of harmful algal blooms,” Remote Sens. Coastal Aquat. Environ., vol. 7, pp. 277–296, 2005. [7] H. T. Duan, S. X. Mang, and Y. Z. Mang, “Cyanobacteria bloom monitoring with remote sensing in Lake Taihu (in Chinese with English abstract),” J. Lake Sci., vol. 20, no. 2, pp. 145–152, 2008. [8] R. H. Ma et al., “Spatio-temporal distribution of cyanobacterial blooms based on satellite imageries in Lake Taihu, China (in Chinese with English abstract),” J. Lake Sci., vol. 20, no. 6, pp. 687–694, 2008. [9] J. F. R. Gower, “Red tide monitoring using AVHRR HRPT imagery from a local receiver,” Remote Sens. Environ., vol. 48, no. 3, pp. 309–318, 1994. [10] J. Gower, S. King, G. Borstad, and L. Brown, “Detection of intense plankton blooms using the 709 nm band of the MERIS imaging spectrometer,” Int. J. Remote Sens., vol. 26, no. 9, pp. 2005–2012, 2005. [11] Y. H. Ahn and P. Shanmugam, “Detecting the red tide algal blooms from satellite ocean color observations in optically complex Northeast-Asia Coastal waters,” Remote Sens. Environ., vol. 103, no. 4, pp. 419–437, 2006. [12] C. M. Hu and M. X. He, “Origin and offshore extent of floating algae in olympic sailing area,” EOS, Trans. Amer. Geophys. Union, vol. 89, no. 33, pp. 302–303, 2008. [13] C. M. Hu, “A novel ocean color index to detect floating algae in the global oceans,” Remote Sens. Environ., vol. 113, no. 10, pp. 2118–2129, 2009. [14] W. Shi and M. Wang, “Green macroalgae blooms in the Yellow Sea during the spring and summer of 2008,” J. Geophys. Res., vol. 114, no. C12, pp. 1–10, 2009. [15] P. Shanmugam, M. Suresh, and B. Sundarabalan, “OSABT: An innovative algorithm to detect and characterize ocean surface algal blooms,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 4, pp. 1879–1892, Aug. 2013. [16] J. Chen, W. T. Quan, M. W. Zhang, and T. W. Cui, “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., vol. 6, no. 4, pp. 1825–1833, Aug. 2013. [17] M. Wang, S. Son, Y. Zhang, and W. Shi, “Remote sensing of water optical property for China’s inland Lake Taihu using the SWIR atmospheric correction with 1640 and 2130 nm bands,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 6, pp. 2505–2516, Dec. 2013. [18] C. M. Hu et al., “Moderate Resolution Imaging Spectroradiometer (MODIS) observations of cyanobacteria blooms in Taihu Lake, China,” J. Geophys. Res.-Oceans, vol. 115, C04002, 2010, doi: 10.1029/ 2009JC005511. [19] Q. M. Cai, X. Y. Gao, Y. W. Chen, S. W. Ma, and M. Dokulil, “Dynamic variations of water quality in Lake Taihu and multivariate analysis of its influential factors,” Chin. Geogr. Sci., vol. 6, no. 4, pp. 364–374, 1996. [20] Y. W. Chen, B. Q. Qin, K. Teubner, and M. T. Dokulil, “Long-term dynamics of phytoplankton assemblages: Microcystis-domination in Lake Taihu, a large shallow lake in China,” J. Plankton Res., vol. 25, no. 4, pp. 445–453, 2003. [21] J. Chen and W. T. Quan, “Using Landsat/TM imagery to estimate nitrogen and phosphorus concentration in Taihu Lake, China,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 1, pp. 273–280, Feb. 2012. [22] H. T. Duan et al., “Two-decade reconstruction of algal blooms in China’s Lake Taihu,” Environ. Sci. Technol., vol. 43, no. 10, pp. 3522–3528, 2009. [23] E. F. Vermote et al., “Atmospheric correction of visible to middle-infrared EOS-MODIS data over land surfaces: Background, operational algorithm and validation,” J. Geophys. Res.-Atmos., vol. 102, no. D14, pp. 17131–17141, 1997. [24] C. Hu et al., “Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: Initial results from Tampa Bay, Florida,” Remote Sens. Environ., vol. 93, no. 3, pp. 423–441, 2004. [25] R. Ma, G. Jiang, H. Duan, L. Bracchini, and S. A. Loiselle, “Effective upwelling irradiance depths in turbid waters: A spectral analysis of origins and fate,” Opt. Express, vol. 19, no. 8, pp. 7127–7138, 2011. [26] F. X. Kong and G. Gao, “Hypothesis on cyanobacteria bloom-forming mechanism in large shallow eutrophic lakes,” Acta Ecol. Sin., vol. 25, no. 3, pp. 589–595, 2005. [27] J. W. Rouse, R. H. Haas, J. A. Schell, and D. W. Deering, “Monitoring vegetation systems in the Great Plains with ERTS,” in Proc. 3rd ERTS Symp. NASA SP-3511, 1973, pp. 309–317. [28] A. R. Huete and C. Justice, MODIS Vegetation Index (MOD13) Algorithm Theoretical Basis Document Version 3, 1999.

Yuchao Zhang received the Ph.D. degree in environmental science from Nanjing University, Nanjing, China, in 2008. She has been a Researcher in Nanjing University for 12 years. Since 2012, she has been an Associate Professor with Nanjing Institute of Geography and Limnology (NIGLAS), Chinese Academy of Sciences (CAS), Nanjing, China. Her research interests include remotely sensed monitoring of algal blooms and water color remote sensing in inland lakes.

546 547 548 549 550 551 552 553 554 555

Ronghua Ma received the M.S. degree in surveying and mapping from Xi’an University of Science and Technology, Xi’an, China, in 1999, and the Ph.D. degree in cartography and GIS from Nanjing University, Nanjing, China, in 2002. He has been with Nanjing Institute of Geography and Limnology (NIGLAS), Chinese Academy of Sciences (CAS) since 2002, where he is currently a Professor of Lake Remote Sensing. His research interests include water environment remote sensing of inland lakes and the application of remote sensing and GIS technique for the dynamic monitoring of land use/land cover and its effect on environment.

556 557 558 559 560 561 562 563 564 565 566 567 568

Hongtao Duan received the Ph.D. degree in optical remote sensing in inland waters from the Graduate School of Chinese Academy of Sciences, Beijing, China, in 2007. He is currently an Associate Professor with the State Key Laboratory of Lake Science and Environments, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing, China. His research interest includes water color remote sensing in inland lakes.

569 570 571 572 573 574 575 576 577 578

Steven A. Loiselle received the degree in engineering from Rensselaer Polytechnic Institute, NY, USA, and the Ph.D. degree from the University of Siena, Siena, Italy, in 1986 and 2000, respectively. Since 2002, he has been a Research Professor at the University of Siena. He is a Visiting Scholar at the Nanjing Institute of Geography and Limnology where he has studied the shallow lakes of the Yangtze Valley, China. He has coordinated research in the African Great Lakes, the large lakes of the Paraná delta (Argentina, Paraguay), and the coastal lakes of the Mediterranean basin (Italy, France, Morocco) and is leading an urban aquatic ecosystems project involving citizen scientists. His research interests are in the study of local, regional and global environmental drivers and their impact on freshwater ecosystems, using remote sensing and ecological modeling approaches.

579 580 581 582 583 584 585 586 587 588 589 590 591 592 593

Jinduo Xu received the M.S. degree in geographic information systems from Nanjing Normal University, Nanjing, China, in 2008. She is currently engaged in the data sharing applications and services. Her research interests include the development of geographic information system and geographic information database technology.

594 595 596 597 598 599 600 601

Mengxiao Ma received the B.Sc. degree in environmental science from Shandong Normal University, Jinan, China, and is currently pursuing the Master degree in environmental science from Nanjing University, Nanjing, China. Her research interest includes environmental remote sensing for lake applications.

602 603 604 605 606 607 608

IE E Pr E oo f

469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

1

2

A Novel Algorithm to Estimate Algal Bloom Coverage to Subpixel Resolution in Lake Taihu

3

Yuchao Zhang, Ronghua Ma, Hongtao Duan, Steven A. Loiselle, Jinduo Xu, and Mengxiao Ma

4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

Abstract—Remote sensing has often been used to monitor the distribution and frequency of floating algae in inland aquatic environments. However, due to the spatial resolution of the most common satellite sensors, accurate determination of algae coverage remains a major technical challenge. Here, a novel algorithm to estimate floating algae area to subpixel scales, denominated the algae pixel-growing algorithm (APA), is developed and evaluated for a series of image data from the Moderate Resolution Imaging Spectroradiometer (MODIS). The algorithm utilizes the Rayleighcorrected reflectance (Rrc) and a floating algae index (FAI) derived from Rrc in three spectral bands. Comparison with concurrent Landsat TM=ETMþ data indicate that the APA provides more accurate estimates of both algal bloom area and algal bloom intensity (i.e., floating algae coverage) (RSE ¼ 15:2 km2 and RE ¼ 9:9%), compared to other commonly used methods such as the linear unmixing algorithm (LA). Furthermore, this study confirms that FAI is a better index with respect to normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) for the estimation of algae area coverage, especially when combined with the APA. Finally, the study provides a theoretical basis for the objective assessment of bloom severity in complex inland waterbodies.

26 Index Terms—Algae pixel-growing algorithm (APA), algal 27 blooms, floating algae index (FAI), Landsat, linear unmixing 28 decomposition, Moderate Resolution Imaging Spectroradiometer 29 (MODIS).

30

I. INTRODUCTION

31 N RECENT decades, many major lakes have experienced 32 an increased occurrence of algal blooms, including Lake 33 Victoria in Africa, Lake Okeechobee in the United States, and 34 Lake Taihu in China [1]–[3]. Compared to traditional in situ

I

sampling and laboratory analysis, satellite-derived optical and thermal data provide synoptic and high frequency information, useful for early warning, real-time monitoring, and postbloom evaluations of bloom events [1]. Over the last three decades, there have been many important developments in the use of remotely sensed data to estimate phytoplankton biomass concentrations and the associated bio-optical properties [4], [5]. This has been particularly important where eutrophication and floating algal mats have compromised lake functioning. Information regarding the coverage and thickness of algal mats provides fundamental information necessary for biomass estimates, which are more appropriate than traditional chlorophyll-a concentration measurements for decision-making. Choosing an appropriate algal bloom algorithm is the first step to estimate bloom coverage. Single-bands, band ratios, and band differences have been used in a similar manner to land surface vegetation detection algorithms [6]–[8]. Indices such as the normalized difference vegetation index (NDVI), the enhanced vegetation index (EVI), the normalized difference algae index (NDAI), maximum chlorophyll index (MCI), red tide index (RI), the floating algae index (FAI), and the ocean surface algal blooms index (OSABI) have been utilized to identify floating algal blooms [9]–[15]. But in practical application, NDVI, EVI, and FAI are much more popular than the other indices. However, it has been found that NDVI and EVI are particularly sensitive to aerosol characteristics (type and thickness), sun glint, and solar viewing geometry with respect to FAI, which has been found to be relative stable under different environmental and observing conditions [13]. Minimizing the atmospheric effects is vital to time-series study and characterization on algal blooms and water optical properties in turbid inland lakes [16], [17]. Thus, FAI appears to be a reasonable choice to detect and quantify bloom coverage. However, algal mats can be smaller than the satellite pixel size, leading to uncertainties in the total bloom coverage estimates. Indeed, algal blooms may present complex temporal and spatial characteristics, which may limit the use of satellite data. Most algorithms use data from the Landsat TM=ETMþ, Moderate Resolution Imaging Spectroradiometer (MODIS), and Medium Resolution Imaging Spectrometer (MERIS), with spatial resolutions of 30 m, 250=500=1000 m, and 300=1200 m with temporal resolutions of 16 days, 1 day, and 3 days, respectively. Among these, MODIS has the best temporal resolution, but its best spatial resolution is 250 m, which may not detect small patches of floating algae and may lead to uncertainties in area coverage estimates. The ability to detect and quantify surface features at subpixel scales would improve the use of satellite systems for bloom monitoring and management.

IE E Pr E oo f

1

Manuscript received September 02, 2013; revised January 22, 2014; accepted May 12, 2014. This work was supported in part by the National High Technology Research and Development Program of China (2014AA06A509), in part by the National Natural Science Foundation of China under Grant 41101316, Grant 41171271, and Grant 41171273, in part by Scientific Data Sharing Platform for Lake and Watershed, in part by Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences (NIGLAS), in part by the “135” Program of NIGLAS (NIGLAS2012135010 and NIGLAS2012135014), and in part by the Dragon 3 program (10561). MODIS data were provided by the U.S. NASA, and the Landsat TM=ETMþ data were provided by the U.S. Geological Survey. (Corresponding author: Ronghua Ma.) Y. Zhang, R. Ma, H. Duan, and J. Xu are with the State Key Laboratory of Lake Science and Environment, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing 210008, China (e-mail: rhma@niglas. ac.cn). S. A. Loiselle is with the CSGI, University of Siena, Siena 253100, Italy (e-mail: [email protected]). M. Ma is with the State Key Laboratory of Pollution Control & Resources Reuse, School of the Environment, Nanjing University, Nanjing 210023, China (e-mail: [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/JSTARS.2014.2327076

1939-1404 © 2014 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission. See http://www.ieee.org/publications_standards/publications/rights/index.html for more information.

35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76

77 78 79 80 81 82

2

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112

IE E Pr E oo f

F1:1 Fig. 1. Study site was Lake Taihu, China lying in the Yangtze River Delta. The lake is often divided into several lake segments, which include Zhushan Bay (ZB), F1:2 Meiliang Bay (MB), Gong Bay (GB), West Lake (WL), South Lake (SL), Central Lake (CL), East Lake, and East Taihu. The brown areas represent major cities.

Bloom detection and quantification at subpixel scales depend on the feature size (the relative percentage of the algae bloom within a pixel) and the sensitivity of the satellite sensor (signal to noise ratio). The accuracy of algae coverage estimates is strongly influenced by mixed pixels containing both algae and algae free (water) surfaces. When the total lake area coverage is small, mixed pixels may represent a large portion of the total algae pixels. Hu et al. [18] have used a linear unmixing algorithm (LA) to estimate the total algae area coverage in Lake Taihu, an eutrophic lake in East China. Shanmugam et al. [15] have also applied the similar algorithm to detect ocean surface algal blooms in subpixel scales. However, this algorithm needs to be further improved to allow for the determination of algae coverage at subpixel scales and more accurate estimation of threshold values for algae and nonalgae areas of optically complex eutrophic lakes. In China, satellite-based estimations of algal blooms (coverage, distribution, and duration) are used by environmental management agencies at both state and local levels for a variety of purposes, from policy implementation and water management (including the physical removal of the algae mats). Coverage estimates are generated by different agencies using a variety of methods, most of which rely on human interpretation of the manually adjusted satellite images (to account for variable atmospheric and observing conditions as well as algorithm artifacts). The large degree of subjectivity has resulted in incompatible estimates and inconsistent recommendations among agencies. There is a clear need for an objective method, which would be valid for variable conditions, including the partial algae coverage at subpixel scales. An objective approach would allow

different agencies and groups to generate consistent estimates based on the same data. In this study, we developed a new algorithm [algae pixelgrowing algorithm (APA)] to estimate algae coverage at subpixel scales using MODIS data. We evaluated the accuracy of the APA using concurrent higher resolution (30 m) Landsat TM=ETMþ data by comparing their spatial distributions and coverage statistics. In addition, several practical issues in using MODIS for bloom monitoring were examined, including an automatic and objective determination of the pure algae threshold as well as a method to adjust pixels contaminated by striping noise. The ultimate goal is to provide an objective method to determine bloom coverage and size to establish a long-term baseline record and real-time assessment of bloom severity to support management decisions.

113 114 115 116 117 118 119 120 121 122 123 124 125 126 127

II. STUDY AREA AND DATA SOURCES

128

A. Study Area

2

With an extension of 36800 km and encompassing major population centers such as Shanghai, Suzhou, Wuxi, and Changzhou, the Lake Taihu basin is one of the most highly developed areas of China (Fig. 1). At the center of the Basin is Lake Taihu, China’s third-largest freshwater lake (2400 km2 ). The lake provides drinking water for more than 2 million people, and sustains important fisheries including crabs, carp, and eels [2]. With the rapid economic development in the last decades, Lake Taihu has become highly eutrophic with frequent cyanobacteria blooms as a consequence of increased nutrient loads

129

130 131 132 133 134 135 136 137 138 139

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

3

TABLE I

T1:1 ACQUISITION DATE OF TERRA/MODIS AND LANDSAT TM (MARKED WITH ASTERISK)/ T1:2 ETMþ DATA USED IN THIS STUDY

Fig. 2. In the APA, a 3  3 window was chosen randomly from a MODIS image. Gray-scale shows the FAI value of each pixel. The central pixel was divided into two kinds of subpixels which have the same maximum and minimum FAI values of the other pixels present in the 3  3 window.

[19]–[21]. In 2007, a massive algal bloom led to a drinking water shortage for more than 1 million people [2], [22]. Although the Chinese government recognized the importance of managing nutrient loads and monitoring lake conditions, the long-term management of the Lake Taihu Basin is still in its infancy. Since 2007, remote sensing has been used to monitor the occurrence of algal blooms as part of an early warning system with daily communications to national and local decision makers.

between the red band (645 nm) and short-wave infrared band 176 (1240 or 1640 nm) 177 F AIMODIS ¼ Rrc ð859Þ  R0rc ð859Þ R0rc ð859Þ ¼ Rrc ð645Þ þ ½Rrc ð1240Þ  Rrc ð645Þ  ð859  645Þ=ð1240  645Þ

where Rrc is defined as the difference between the calibrated 178 sensor radiance [after adjustment for ozone (Lt  ) and other 179 gaseous absorption] and Rayleigh reflectance (Rr ) [24] 180 Rrc ¼ Lt  =F0 cos 0  Rr

149 B. Satellite Data

(1)

IE E Pr E oo f

140 141 142 143 144 145 146 147 148

150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170

Terra MODIS data at 250 and 500 m resolutions and Landsat 5 Thematic Mapper (TM) and Landsat 7 Enhanced Thematic Mapper Plus (ETMþ) data at 30 m resolution were used in this study. Landsat TM=ETMþ Level-1 data were obtained from the Earth Resources Observation and Science Center (http:// glovis.usgs.gov). MODIS data were obtained from the LAADS website of the U.S. NASA Goddard Space Flight Center. Twenty-four pairs of concurrent (i.e., same day within a hour) MODIS and TM=ETMþ data (Table I) with low cloud coverage ( < 10%) were georeferenced (UTM projection) with an error of < 0:5 pixel. The 500-m resolution MODIS data at 1240 nm and 1640 nm were resampled to 250 m resolution (to match the 645-nm data). TM=ETMþ data were atmospherically corrected using the radiative transfer calculations based on the second simulation of the satellite signal in the solar spectrum [23]. MODIS data were corrected by removing the molecular (Rayleigh) scattering effects, and then converted to Rayleigh-corrected reflectance (Rrc ) following Hu et al. [24]. The lake segment of East Lake Taihu was excluded from the bloom analysis as this area is characterized by aquatic macrophytes growing from the bottom [25].

171

III. METHODOLOGY

172 A. Floating Algal Index Algorithm 173 The FAI algorithm for MODIS proposed by Hu [13] utilizes a 174 baseline subtraction, defined as the difference between reflec175 tance at 859 nm (vegetation “red edge”) and a linear baseline

F2:1 F2:2 F2:3 F2:4

(2)

where F0 is the extraterrestrial solar irradiance at data acquisition time and 0 is the solar zenith angle. FAI has been used to map bloom size to derive long-term statistics [18]. However, the estimation of bloom size is subject to large errors if a single threshold is used to determine the bloom presence or absence. Hu et al. [18] used an LA to estimate partial coverage at subpixel scales (pixel ) through a statistical analysis of the 9-year MODIS FAI dataset with FAI thresholds for 100% bloom coverage (F AIalgae ) and 0% bloom coverage (F AInonalgae ). When F AInonalgae < F AIpixel  F AIalgae , LA is expressed as

181 182 183 184 185 186 187 188 189 190 191

pixel ¼ ðF AIpixel  F AInonalgae Þ=ðF AIalgae  F AInonalgae Þ: (3)

B. Algae Pixel-Growing Algorithm (APA)

192

1) Theoretical Considerations: Considering n high-resolution 193 pixels (Rrc Hi ) that make up a low-resolution pixel (Rrc Lo ), where 194 Rrc Lo is the arithmetic mean of all Rrc Hi 195 Rrc Lo ¼ Rrc Hi ðiÞ:

(4)

Similarly, there exists a relationship between the low- 196 resolution pixel and high-resolution pixels (or subpixels 197 within the low-resolution pixel) for FAI 198 F AI pixel ¼ F AI subpixel :

(5)

If we assume that, in a 3  3 pixel window, the central pixel is 199 a function the maximum and minimum FAI values present within 200 the window (Fig. 2), (5) becomes 201 F AIcenter ¼   F AImax þ ð1   ÞF AImin

(6)

4

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

IE E Pr E oo f

Fig. 4. Changes in the MODIS-derived algae coverage area with increasing F4:1 iterations (steps) of the APA. The stars mark the true algal bloom area identified F4:2 from the corresponding Landsat TM=ETMþ data. F4:3

F3:1 Fig. 3. Flowchart of the APA process. 202 203 204 205 206

where  is the decomposition parameter of the 3  3 window that is determined based on the relationship between the known FAI values (center, max, and min). In a mixed pixel, the algae coverage is defined as the proportion () of the pixel covered by floating algae such that thresh thresh F AI ¼   F AIalgae þ ð1  Þ  F AIwater   thresh thresh ¼ F AIalgae  F AIwater  thresh þ F AIwater

(7)

thresh thresh 207 where F AIalgae and F AIwater are the FAI thresholds for pure 208 algae (100%) and nonalgae water (0%) pixels, respectively. thresh thresh 209 Assuming that F AIalgae and F AIwater are constant in a 3  3 210 window, the FAI of central pixel could be expressed as follows:

  thresh thresh  F AIwater F AIcenter ¼ F AIalgae  center þ

thresh F AIIwater :

(8)

211 The FAI of max and min pixels in a 3  3 window could also 212 be similarly expressed in the same way. Based on (6)–(8), FAI 213 has a linear relationship with the floating algae coverage  in the 214 mixed pixel

center ¼   max þ ð1   Þ  min

(9)

215 where max and min are the algae coverage of pixels with 216 maximum and minimum FAI values in a 3  3 window, 217 respectively.

Fig. 5. Estimation of floating algae coverage using the APA (the aquatic macrophytes area marked with gray color), from the initial estimation (step 1), the determination of the growing points (step 2), and the final estimation of algal coverage after 3 iterations (step 3). A and B are the ends of a transect within the lake.

F5:1 F5:2 F5:3 F5:4 F5:5

2) APA Implementation and Data Processing: There are three iterative steps in applying the APA to MODIS Rrc data (Fig. 3). 1) The preprocessing step identifies the location of the maximum and minimum FAI value in a roaming 3  3 window, and then calculates  for the central pixel using (6). 2) The “seed” pixels are then identified in two ways: a) if local pixels are found to have thresh their F AI values > F AIalgae , these pixels are taken as the seed pixels and their algae coverage are defined as 1.0 (100%); thresh b) if no local pixels have F AI values > F AIalgae , then local center pixels with maximum FAI values in the 3  3 window are taken as the seed pixels. The coverage is determined via (3) using thresh a pure-algae endmember threshold of F AIalgae . The coverage of other pixels is assigned 0.0 in the initial estimation. 3) The final algae coverage at each pixel is determined iteratively using (9) whenever the maximum coverage in each 3  3 window is not zero. The iteration is terminated when two consecutive calculations yield similar results. In practice, three iterations were necessary before the termination (see below). Through iterations, the algal bloom coverage expands from the initial pure algae

218 219 220 221 222 223 224 225

226

227 228 229 230 231 232 233 234 235 236

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

5

F6:1 Fig. 6. Comparison of algal bloom coverage derived from several different methods ( coverage interval ¼ 0:005): the full line marks the true bloom coverage using F6:2 30 m TM=ETMþ data resized to 250 m; the dashed lines show the estimate of bloom coverage from the concurrent MODIS image using the APA; the dotted lines F6:3 represent the estimate of bloom coverage from the same MODIS image using an LA.

250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273

274 275 276 277 278 279

pixels or high-coverage pixels to low-coverage pixels, according to the relationship between adjacent pixels in a 3  3 pixel window. For very thin algae layers, the measured reflectance may always be confused with surface waters no algal bloom but with high concentrations of chlorophyll-a. However, by including the constraints related to adjacent pixels with highalgae coverage, this error is reduced unless these conditions are present over a large geographic area. If such conditions are not present, the problematic pixel is assigned as nonalgae pixel. Total algae coverage is determined by the no zero algae coverage. 3) FAI Threshold for Pure Algae Coverage: The APA results thresh are sensitive to the choice of the F AIalgae and to the termination thresh conditions of the iterative calculation. The F AIalgae was determined from concurrent higher resolution TM=ETMþ observations, but could also be estimated using in situ reflectance spectra of the pure algae. Due to the lack of in situ reflectance data covering the spectral range of > 1000 nm, the former method was used. With a higher spatial resolution, Landsat TM=ETMþ data provide a better resolution of the pure algae pixels. We assumed that there were only pure algae pixels and pure nonalgae pixels existing in Landsat TM=ETMþ images. A pixel’s gradient was defined as the FAI difference from the adjacent pixels in a 3  3 window. It was found that the pixels associated with the maximum gradient determined mode could be used to separate floating algae from other nonbloom waters very well [18]. Using the large gradient in FAI values across the algae and nonalgae boundary, the mean value of all pixels associated with the gradient was used to represent the TM=ETMþ threshold value for pure algae. After resizing from 30 to 250 m (to match the MODIS resolution), the partial algae coverage in the new low-resolution pixel could be determined. If all TM=ETMþ pixels within the low-resolution pixel had 100% algae coverage, the low-resolution pixel could be identified as the pure algae pixel. From all these low-resolution pure algae pixels, the FAI value (2  standard deviations) was used to represent the threshold value for pure (100%) algae thresh coverage (F AIalgae ). 4) The Condition to Terminate Iterations: For each MODIS pixel, the algae coverage is a function of the FAI maximum and minimum in each 3  3 window. For each iteration, the estimated coverage increases until it reaches a stable value (Fig. 4). The termination condition allows the calculation to

IE E Pr E oo f

237 238 239 240 241 242 243 244 245 246 247 248 249

Fig. 7. North south profile of estimated algae coverage (0%–100%) from resized ETMþ pixels (all at a 250-m resolution) and MODIS using the APA and an LA, along a hypothetical transect from Meiliang Bay to Tiaoxi River (white line in the inset image). Each point represents an arithmetic mean of 3  3 pixels along the transect.

F7:1 F7:2 F7:3 F7:4 F7:5

conclude when the absolute difference or relative difference of consecutive steps reaches a minimum value. For variable bloom areas (from several hundreds to > 1000 km2 ) with millions of algae pixels, the absolute difference or relative difference inconsecutive steps can be significantly different between images. Importantly, it was found that the smallest relative difference between MODIS and TM=ETMþ data occurred after three iterations in all 24 concurrent MODIS/TM image pairs (Fig. 4). This is a natural consequence of the fact that mixed pixels are usually found adjacent to pure-algae pixels or high algae-coverage pixels. The APA approach builds algae coverage from pure-algae pixels with respect to nonalgae pixels (or from high-coverage pixels to low-coverage pixels). An example of these calculation steps is presented for a lake transect characterized by pure algae to nonalgae water in Lake Taihu on September 24, 2011 (Fig. 5). The algae coverage of some pixels varied at every iteration. In general, the algae coverage of mixed pixels increased gradually, as the algae coverage of the FAI maximum pixel or FAI minimum pixel changed in each 3  3 pixel. From the algae coverage distribution of each iteration, the algal bloom coverage expanded from the pure algae pixels or high-coverage pixels to low-coverage pixels, based on the condition of adjacent pixels in a 3  3 pixel window. In total, the algae coverage will be ignored if there is no algal bloom existing in adjacent pixels.

280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304

6

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

F8:1 Fig. 8. Comparison of the floating algae area estimated from TM=ETMþ and MODIS using the two algorithms [the APA and an (LA)] when different FAI threshold F8:2 values are used to represent pure (100%) algae coverage. u and δ represent the mean and standard deviation of the threshold values.

IV. RESULTS AND VALIDATION

306 A. Algal Bloom Coverage Estimations

307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329

330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345

B. Sensitivity Analysis

346

In this study, the MODIS threshold of pure algae was determined from concurrent TM=ETMþ observations. To examine how sensitive the MODIS results were to the threshold estimation, three MODIS images in different years (April 6, 2007; October 20, 2009; May 5, 2012) with different bloom coverage were compared with TM=ETMþ (Fig. 8). The bloom area was found to increase sharply when the threshold for pure algae coverage was lower than the mean minus the standard deviation. The results confirm that both the APA and LA significantly overestimated the algae coverage if the pure algae threshold was low. A higher threshold (mean plus standard deviation) resulted in a slight under estimation, which was less sensitive to further changes of the threshold. An artificial sensitivity test was also conducted by varying the pure algae threshold by 10% to þ10%. The total algae area from the APA estimates changed þ15:1% and 13:5% with respect to TM=ETMþ values, whereas the area from the LA estimates changed þ9:7% and 8:3%. In general, the change in the total algae area estimates was proportional to changes in the threshold values.

347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366

V. DISCUSSIONS

367

IE E Pr E oo f

F9:1 Fig. 9. Linear relationship between MODIS Rrc ð1240Þ and Rrc ð1650Þ from F9:2 several MODIS images. 305

provided a similar frequency distribution for MODIS data with resized TM=ETMþ images. In order to examine the algae coverage distributions derived from different methods, an example of the MODIS results on September 24, 2011 is shown in Fig. 7. An artificial transect from Meiliang Bay (North Lake) to the mouth of Tiaoxi River (South Lake) was chosen to show a range of algae coverage. In order to minimize the impact of image noise and geometric correction error, the average coverage of the 3  3 window was used to represent the central pixel along the transect. Both APA and LA yielded reasonable results when compared to ETMþ. However, both APA and LA underestimated algae coverage when the algae coverage was > 40% in the mixed pixels, and overestimated coverage when the percent coverage was very low. On average, the difference between MODIS and ETMþ was 13.7% when the APA was used and 15.4% when the LA was used.

Algal bloom coverage identified from the 24 TM=ETMþ images ranged from 14.8 to 505:7 km2 . The relative standard error (RSE) between MODIS estimates and TM=ETMþ estimates was 15.2 and 24:8 km2 , respectively, for the APA and LA methods, with their corresponding relative error (RE) of 9.9% and 17.3%. The results of APA and LA presented here should be interpreted as MODIS pixels with both 100% and partial algae bloom coverage. If we ignored the partial algae bloom coverage and took the mean FAI value of pixel with 10% algae coverage as the threshold of algae pixel and nonalgae pixel, algae coverage area from MODIS images is increased by > 40%, and its corresponding RE between MODIS estimates and TM=ETMþ estimates is more than 30%. Due to the partial coverage provided by APA and LA, their errors are much smaller (as gauged by the highresolution TM=ETMþ data). APA provided more accurate estimates than the LA, with the already small errors nearly halved. A comparison of MODIS algal bloom coverage with paired TM=ETMþ coverage, resized from 30 to 250 m, gave similar results (Fig. 6). When the pure algae pixels of MODIS were set with resized TM=ETMþ, LA over-estimated the number of pixels with partial algae coverage. The over-estimation indicated that the nonalgae threshold was low. The APA determination

A. Lake Specific FAI Thresholds

368

There are two options to determine FAI thresholds for long- 369 term applications: 1) threshold based on multi-image statistics 370

7

IE E Pr E oo f

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE

F10:1 Fig. 10. Histogram distributions of Rrc ð1240Þ before and after removal of the Rrc ð1240Þ striping noise. 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387

[18]; and 2) threshold based on the individual images. Because environmental factors such as air temperature, wind speed and direction, and hydrological conditions all influence algal aggregation characteristics (i.e., algal bloom thickness) [26], the pure algae threshold (as well as the pure water threshold) may vary between images. Therefore, an image-specific threshold is more accurate. In this study, this was achieved by comparing concurrent MODIS and TM=ETMþ image pairs. However, for routine monitoring, TM=ETMþ data are not available at the MODIS temporal frequency and the pure algae threshold needs to be determined in alternative ways. Because the red-edge effect would lead to elevated reflectance in the near-IR and shortwave IR, the pure algae pixel was determined when Rrc (1240) was greater than Rrc (645) and FAI was great than 0. Comparison with the thresholds determined from the MODIS and TM=ETMþ pairs showed consistency, allowing for operational use of APA with MODIS data in absence of TM=ETMþ.

388 B. MODIS Striping Noise 389 390 391 392 393 394 395 396

One of the MODIS bands used to derive the FAI, the 1240-nm band on MODIS Terra, suffers from significant striping noise. This would seriously affect the FAI value and APA performance. In this study, the noise-contaminated data were replaced with coincident the 1640-nm data, using the linear relationship between the two bands derived from the high-quality (undisturbed) pixels of the same image. A significant linear relationship between Rrc ð1240Þ and Rrc ð1640Þ, especially in algae-covered

areas was evident for all the images examined (Fig. 9). Although the relationship varies among images, an image-specific relationship to derive Rrc ð1240Þ for noise-contaminated pixels can be used in FAI calculations. The results in indicate that such a correction yields different statistics in algae area coverage (Fig. 10).

397 398 399 400 401 402

C. Comparison of FAI, NDVI, and EVI

403

FAI has been used to calculate algae coverage because of its reduced sensitivity to variable environmental and observational conditions [18]. However, other indexes have also been widely utilized. Among these are NDVI [27] and EVI [28], defined as

404 405 406 407

NDV I ¼ ðRNIR  RRED Þ=ðRNIR þ RRED Þ EV I ¼ G  ðRNIR  RRED Þ=

ðRNIR þ C1  RRED  C2  RBLUE þ C3 Þ

(10)

(11)

where RNIR , RRED , and RBLUE are the reflectance in the nearinfrared (NIR), red and blue bands; G is the gain factor, and C1 , C2 , and C3 are the pixel-independent coefficients to compensate for aerosol effects and vegetation background. For MODIS data, G ¼ 2:5, C1 ¼ 6, C2 ¼ 7:5, and C3 ¼ 1. Given these alternative choices on the vegetation indexes, we examined if FAI is the best choice to derive the algae coverage with the APA. A MODIS Terra image for August 10, 2013 was used to evaluate algae coverage using FAI, NDVI, and EVI, where sun glint influences part of the image (Fig. 11). The FAI image

408 409 410 411 412 413 414 415 416 417

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING

IE E Pr E oo f

8

F11:1 Fig. 11. Comparison among MODIS FAI, NDVI, and EVI values for MODIS data obtained on August 10, 2013 and corresponding algae coverage distributions F11:2 (derived from the APA). Areas with aquatic macrophytes are marked in gray. 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433

displayed a near homogeneous water background regardless of the sun glint. In contrast, the NDVI image revealed higher variability in the water background in areas where sun glint was significant. Sun glint led to a significant increase in the NDVI values, making it more difficult to differentiate algae pixels from the water background. The performance of EVI was much better than NDVI, but still showed a higher sensitivity to sun glint contamination with respect to FAI. The algae areas determined from the three images, all using the APA, were 178.04, 353.68, and 236:26 km2 , respectively. Thus, when sun glint was significant, both NDVI and EVI overestimated algae coverage. Because of the low-latitude (31 N) location of Lake Taihu, most images during the summer months contain sun glint. Thus, FAI is a preferred index for its tolerance to this as well as to the interference of thick aerosols (not shown here) for long-term, routine monitoring of the bloom coverage.

434 D. MODIS-Based Data Record and Event-Driven Response 435 436 437 438 439 440 441 442 443

The APA approach, together with the statistically determined FAI thresholds, provides an objective method to estimate the bloom severity. This can be used to estimate bloom spatial extent and effective algae coverage as well as to compare current bloom conditions against historical baselines determined using the same objective method. This will lead to more accurate estimates of the bloom severity in Lake Taihu in near real-time as well as pave the way to obtain consistent answers from various research groups and management agencies.

VI. CONCLUSION

444

Remote sensing has been widely used to assess algal blooms in Lake Taihu, but the methods and parameterization varied substantially between different users, making it difficult to compare results and agree on a common action. The present approach to develop and validate a more objective method to characterize the bloom severity provides a novel to meet this challenge. The results, based on the MODIS FAI data, showed improved performance over other methods or indices. The APA approach serves as an objective and more accurate method to determine bloom severity in both near real-time monitoring and historical analysis, thereby improving the capacity of decision makers to manage Lake Taihu and its basin.

445 446 447 448 449 450 451 452 453 454 455 456

ACKNOWLEDGMENT

457

The constructive comments from two anonymous reviewers 458 are greatly appreciated. 459 REFERENCES

460

[1] A. Reinart and T. Kutser, “Comparison of different satellite sensors in detecting cyanobacterial bloom events in the Baltic Sea,” Remote Sens. Environ., vol. 102, no. 1–2, pp. 74–85, 2006. [2] L. Guo, “Ecology-doing battle with the green monster of Taihu Lake,” Science, vol. 317, no. 5842, p. 1166, 2007. [3] S. Q. Zhao et al., “The 7-decade degradation of a large freshwater lake in central Yangtze River, China,” Environ. Sci. Technol., vol. 39, no. 2, pp. 431–436, 2005.

461 462 463 464 465 466 467 468

9

[4] S. Sathyendranath, “Remote sensing of ocean colour in coastal, and other optically-complex waters,” International Ocean-Colour Coordinating Group, pp. 47–76, 2000. [5] Z. P. Lee, “Remote sensing of inherent optical properties: Fundamentals, tests of algorithms, and applications,” International Ocean-Colour Coordinating Group, pp. 43–93, 2006. [6] P. S. Richard and C. T. Michelle, “Remote sensing of harmful algal blooms,” Remote Sens. Coastal Aquat. Environ., vol. 7, pp. 277–296, 2005. [7] H. T. Duan, S. X. Mang, and Y. Z. Mang, “Cyanobacteria bloom monitoring with remote sensing in Lake Taihu (in Chinese with English abstract),” J. Lake Sci., vol. 20, no. 2, pp. 145–152, 2008. [8] R. H. Ma et al., “Spatio-temporal distribution of cyanobacterial blooms based on satellite imageries in Lake Taihu, China (in Chinese with English abstract),” J. Lake Sci., vol. 20, no. 6, pp. 687–694, 2008. [9] J. F. R. Gower, “Red tide monitoring using AVHRR HRPT imagery from a local receiver,” Remote Sens. Environ., vol. 48, no. 3, pp. 309–318, 1994. [10] J. Gower, S. King, G. Borstad, and L. Brown, “Detection of intense plankton blooms using the 709 nm band of the MERIS imaging spectrometer,” Int. J. Remote Sens., vol. 26, no. 9, pp. 2005–2012, 2005. [11] Y. H. Ahn and P. Shanmugam, “Detecting the red tide algal blooms from satellite ocean color observations in optically complex Northeast-Asia Coastal waters,” Remote Sens. Environ., vol. 103, no. 4, pp. 419–437, 2006. [12] C. M. Hu and M. X. He, “Origin and offshore extent of floating algae in olympic sailing area,” EOS, Trans. Amer. Geophys. Union, vol. 89, no. 33, pp. 302–303, 2008. [13] C. M. Hu, “A novel ocean color index to detect floating algae in the global oceans,” Remote Sens. Environ., vol. 113, no. 10, pp. 2118–2129, 2009. [14] W. Shi and M. Wang, “Green macroalgae blooms in the Yellow Sea during the spring and summer of 2008,” J. Geophys. Res., vol. 114, no. C12, pp. 1–10, 2009. [15] P. Shanmugam, M. Suresh, and B. Sundarabalan, “OSABT: An innovative algorithm to detect and characterize ocean surface algal blooms,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 4, pp. 1879–1892, Aug. 2013. [16] J. Chen, W. T. Quan, M. W. Zhang, and T. W. Cui, “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., vol. 6, no. 4, pp. 1825–1833, Aug. 2013. [17] M. Wang, S. Son, Y. Zhang, and W. Shi, “Remote sensing of water optical property for China’s inland Lake Taihu using the SWIR atmospheric correction with 1640 and 2130 nm bands,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 6, no. 6, pp. 2505–2516, Dec. 2013. [18] C. M. Hu et al., “Moderate Resolution Imaging Spectroradiometer (MODIS) observations of cyanobacteria blooms in Taihu Lake, China,” J. Geophys. Res.-Oceans, vol. 115, C04002, 2010, doi: 10.1029/ 2009JC005511. [19] Q. M. Cai, X. Y. Gao, Y. W. Chen, S. W. Ma, and M. Dokulil, “Dynamic variations of water quality in Lake Taihu and multivariate analysis of its influential factors,” Chin. Geogr. Sci., vol. 6, no. 4, pp. 364–374, 1996. [20] Y. W. Chen, B. Q. Qin, K. Teubner, and M. T. Dokulil, “Long-term dynamics of phytoplankton assemblages: Microcystis-domination in Lake Taihu, a large shallow lake in China,” J. Plankton Res., vol. 25, no. 4, pp. 445–453, 2003. [21] J. Chen and W. T. Quan, “Using Landsat/TM imagery to estimate nitrogen and phosphorus concentration in Taihu Lake, China,” IEEE J. Sel. Topics Appl. Earth Observ. Remote Sens., vol. 5, no. 1, pp. 273–280, Feb. 2012. [22] H. T. Duan et al., “Two-decade reconstruction of algal blooms in China’s Lake Taihu,” Environ. Sci. Technol., vol. 43, no. 10, pp. 3522–3528, 2009. [23] E. F. Vermote et al., “Atmospheric correction of visible to middle-infrared EOS-MODIS data over land surfaces: Background, operational algorithm and validation,” J. Geophys. Res.-Atmos., vol. 102, no. D14, pp. 17131–17141, 1997. [24] C. Hu et al., “Assessment of estuarine water-quality indicators using MODIS medium-resolution bands: Initial results from Tampa Bay, Florida,” Remote Sens. Environ., vol. 93, no. 3, pp. 423–441, 2004. [25] R. Ma, G. Jiang, H. Duan, L. Bracchini, and S. A. Loiselle, “Effective upwelling irradiance depths in turbid waters: A spectral analysis of origins and fate,” Opt. Express, vol. 19, no. 8, pp. 7127–7138, 2011. [26] F. X. Kong and G. Gao, “Hypothesis on cyanobacteria bloom-forming mechanism in large shallow eutrophic lakes,” Acta Ecol. Sin., vol. 25, no. 3, pp. 589–595, 2005. [27] J. W. Rouse, R. H. Haas, J. A. Schell, and D. W. Deering, “Monitoring vegetation systems in the Great Plains with ERTS,” in Proc. 3rd ERTS Symp. NASA SP-3511, 1973, pp. 309–317. [28] A. R. Huete and C. Justice, MODIS Vegetation Index (MOD13) Algorithm Theoretical Basis Document Version 3, 1999.

Yuchao Zhang received the Ph.D. degree in environmental science from Nanjing University, Nanjing, China, in 2008. She has been a Researcher in Nanjing University for 12 years. Since 2012, she has been an Associate Professor with Nanjing Institute of Geography and Limnology (NIGLAS), Chinese Academy of Sciences (CAS), Nanjing, China. Her research interests include remotely sensed monitoring of algal blooms and water color remote sensing in inland lakes.

546 547 548 549 550 551 552 553 554 555

Ronghua Ma received the M.S. degree in surveying and mapping from Xi’an University of Science and Technology, Xi’an, China, in 1999, and the Ph.D. degree in cartography and GIS from Nanjing University, Nanjing, China, in 2002. He has been with Nanjing Institute of Geography and Limnology (NIGLAS), Chinese Academy of Sciences (CAS) since 2002, where he is currently a Professor of Lake Remote Sensing. His research interests include water environment remote sensing of inland lakes and the application of remote sensing and GIS technique for the dynamic monitoring of land use/land cover and its effect on environment.

556 557 558 559 560 561 562 563 564 565 566 567 568

Hongtao Duan received the Ph.D. degree in optical remote sensing in inland waters from the Graduate School of Chinese Academy of Sciences, Beijing, China, in 2007. He is currently an Associate Professor with the State Key Laboratory of Lake Science and Environments, Nanjing Institute of Geography and Limnology, Chinese Academy of Sciences, Nanjing, China. His research interest includes water color remote sensing in inland lakes.

569 570 571 572 573 574 575 576 577 578

Steven A. Loiselle received the degree in engineering from Rensselaer Polytechnic Institute, NY, USA, and the Ph.D. degree from the University of Siena, Siena, Italy, in 1986 and 2000, respectively. Since 2002, he has been a Research Professor at the University of Siena. He is a Visiting Scholar at the Nanjing Institute of Geography and Limnology where he has studied the shallow lakes of the Yangtze Valley, China. He has coordinated research in the African Great Lakes, the large lakes of the Paraná delta (Argentina, Paraguay), and the coastal lakes of the Mediterranean basin (Italy, France, Morocco) and is leading an urban aquatic ecosystems project involving citizen scientists. His research interests are in the study of local, regional and global environmental drivers and their impact on freshwater ecosystems, using remote sensing and ecological modeling approaches.

579 580 581 582 583 584 585 586 587 588 589 590 591 592 593

Jinduo Xu received the M.S. degree in geographic information systems from Nanjing Normal University, Nanjing, China, in 2008. She is currently engaged in the data sharing applications and services. Her research interests include the development of geographic information system and geographic information database technology.

594 595 596 597 598 599 600 601

Mengxiao Ma received the B.Sc. degree in environmental science from Shandong Normal University, Jinan, China, and is currently pursuing the Master degree in environmental science from Nanjing University, Nanjing, China. Her research interest includes environmental remote sensing for lake applications.

602 603 604 605 606 607 608

IE E Pr E oo f

469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545

ZHANG et al.: NOVEL ALGORITHM TO ESTIMATE ALGAL BLOOM COVERAGE