*2.1. Study Area*

Poyang Lake (115◦49.7~116◦46.7 E, 28◦24~29◦46.7 N) is located in the lower Yangtze River Basin. Tributaries of five rivers, namely the Ganjiang, Fuhe, Xinjiang, Raohe, and Xiushui rivers, feed the lake (Figure 1). Seasonal variation of precipitation leads to significant changes in the lake's surface area throughout the year [30]. The water surface area exceeds 3,000 km<sup>2</sup> during the wet season (April–September) and then drops below 1,000 km<sup>2</sup> during the dry season (October–March). Poyang Lake is also one of the world's most ecologically important wetlands, with millions of migratory birds, including about 98% of the world's population of Siberian cranes, inhabiting the lake area for wintering.

**Figure 1.** Map of Poyang Lake, China.

#### *2.2. Field Sampling and Measurement*

Field data were obtained from Poyang Lake in August 2015 (August 1, 3, and 5, 2015), October 2015 (October 23 and 24, 2015), and January 2016 (January 24 and 25, 2016), from 43, 33, and 26 sampling sites, respectively. Figure 2 shows the locations of the sampling sites. For each sampling exercise, water was collected from a water depth of between 0 and 50 cm. All of the samples were held on ice and stored for subsequent measurements of the concentrations of Chl-a and TSM in the laboratory. Two 500ml portions of each sample were used to filter for collecting Chl-a and TSM, respectively. Chl-a was collected while using Whatman GF/F filters (0.7 μm pore size) and extracted with hot ethanol [31]. The concentrations of Chl-a were determined by spectrophotometry using a UV-2600PC UV–vis spectrophotometer (Shimadzu, Inc., Koyto, Japan). Water samples for TSM measurement were filtered while using Whatman GF/C filters (1.2 μm pore size) under vacuum. Subsequently, the filters were weighed gravimetrically to determine the concentration of TSM. Water temperature and turbidity were measured at the sampling sites using a YSI6600 portable multi-parameter water meter (Yellow Springs Instruments, Inc., Yellow Springs, Ohio, USA).

**Figure 2.** Pseudocolor renderings of images from the Gaofen-1 (GF-1) satellite (**a**: 770–890 nm, **b**: 630-690 nm, **c**: 520–590 nm) showing the landscape of the study area at approximately the time that the field samplings were performed.

Remote-sensing-reflectance spectra were measured above the water surface at wavelengths between 350 and 2500 nm (1 nm interval) while using a Fieldspec 4 spectroradiometer (Analytical Spectral Devices, Inc., Boulder, Colorado, USA), following standard protocols [32]. Measurements were performed between 10:00 and 14:00 on sunny windless days. A total of 66 sampling sites (35 for August 2015, 17 for October 2015, and 14 for January 2016) were measured for remote-sensing-reflectance spectra due to the limitation of measurement time. The radiances from water, sky, and a reference panel were measured at each water sampling site.

Remote sensing reflectance ( Rrs) was determined as the ratio of water-leaving radiance ( L w) to the total downwelling irradiance [Ed(0+)].

$$\mathbf{R\_{rs}} = \frac{\mathbf{L\_w}}{\mathbf{E\_d}(0^+)} = \frac{\mathbf{L\_{SW}} - \delta \mathbf{L\_{sky}}}{\mathbf{L\_{\mathcal{P}}} \star \pi / \rho\_{\mathcal{P}}} \tag{1}$$

where LSW is the total upwelling radiance from water, Lsky is the skylight radiance, δ is a proportionality coe fficient that relates Lsky to the reflected sky radiance determined when the detector viewed the

water surface [33], Lp is the radiance from the reference panel, and ρp is the irradiance reflectance of the reference panel.

#### *2.3. Model Development and Assessment*

In this study, semi-analytical and empirical approaches were used for estimating the concentrations of Chl-a and TSM by sensitive indices. The retrieval algorithms were established through regression processes while using linear, quadratic polynomial, exponential, logarithmic, and power-law regression approaches. The goodness of fit was judged by the value of the coe fficient of determination (R2). The water samples were measured for remote-sensing-reflectance spectra and concentrations of Chl-a and TSM and then numbered from 1 to n, with 1 representing the samples with the highest concentration of Chl-a/TSM and n representing the samples with the lowest concentration. Subsequently, one sample was selected every three sample numbers—i.e., samples 1, 4, 7, 10, ... , were selected; the selected samples were then used for model validation, while the remaining samples were used for model calibration. The coe fficient of determination (R2), root-mean-square error (RMSE), and mean relative percentage error (MRPE) between the measured and predicted values of Chl-a or TSM concentration were calculated to assess the fitting and validation accuracy. The RMSE and MRPE were determined while using equations (2) and (3), respectively:

$$\text{RMSE} = \sqrt{\frac{1}{n} \ast \sum\_{i=1}^{n} \left[ \mathbf{x}\_{\text{imea}} - \mathbf{x}\_{\text{ipre}} \right]^2} \tag{2}$$

$$\text{MRPE} = \frac{\sum\_{i=1}^{n} \left| \frac{\chi\_{\text{irma}} - \chi\_{\text{irma}}}{\chi\_{\text{irma}}} \right|}{\mathbf{n}} \ast 100\% \tag{3}$$

where n is the number of samples.

#### *2.4. Image Data and Preprocessing*

The GF-1 satellite images were downloaded through the Remote Sensing Market Service Platform of the Chinese Academy of Sciences (http://www.rscloudmart.com). The GF-1 satellite was launched on 26 April 2013. The satellite has a sun-synchronous orbit with an altitude of 645 km, crossing the equator at 10:30 local time in a descending mode. The satellite carries panchromatic (2 m resolution) and multi-spectral (8 m resolution) sensor systems for high-resolution observation and four wide-field-of-view (WFV) sensors for large-scale observation. The four WFV sensors acquire multi-spectral data with a spatial resolution of 16 m, a revisit cycle of four days, and wide coverage (4 × 200 km). Table 1 shows the spectral bands of GF-1. The predicted service life of this satellite is five to eight years. The images that were employed in this study were acquired August 3 and October 24, 2015 and January 30, 2016, respectively. The images captured the lake wetland transition between the highest and lowest water levels in summer 2015 and the follow up winter. The water level change reflected the variation of aquatic environment, in particular for concentrations of Chl-a and TSM.



The preprocessing of GF-1 images includes geometric correction, radiometric calibration, atmospheric correction, and water body range extraction. In this paper, image data were processed while

using the ENVI 5.3 software (Exelis Visual Information Solutions, Inc., Broomfield, Colorado, USA). Geometric correction was processed using the RPC (Rational Polynomial Coe fficients) Orthorectification module in ENVI 5.3. The Landsat8 OLI panchromatic image covering the Poyang Lake area was used as a reference. The GF-1 images were then radiometrically calibrated to covert DN value to radiance. Atmospheric correction was performed while using the FLAASH (Fast Line-of-Sight Atmospheric Analysis of Spectral Hypercubes) module in ENVI 5.3. FLAASH integrates MODTRAN 5 radiative transfer model with all MODTRAN atmosphere and aerosol styles to provide a unique solution for each image [34]. In this study, the mid-latitude atmosphere and rural aerosol were selected in FLAASH to correct the GF-1 images. The results of atmospheric correction were the remote sensing reflectance above the water surface. Figure 3a–c show the comparison of GF-1 bands before and after atmospheric correction, along with in situ reflectance resampled according to GF-1 band configuration. Figure 3d presents the validation result between the Band-3 reflectance of GF-1 imagery and in situ measured reflectance from August 2015. The results showed that the atmospheric interference to sensor had been effectively removed through FLAASH implementation.

**Figure 3.** Comparison of GF-1 reflectance before and after atmospheric correction with in situ measured spectra, (**a**): mean value of data from August 2015; (**b**): mean value of data from October 2015; and, (**c**): mean value of data from January 2016. (**d**) Validation between the Band-3 reflectance of GF-1 imagery and in situ measured reflectance (August 2015).
