3.1. Field Collected Chl-a Data
Considering the sample analysis results obtained by CETESB regarding Chl-a at the sampling point in Salto Grande Reservoir (Fig. 3), we observe a variation of Chl-a concentrations, which ranged from 10 µg/L to 130 µg/L in the historical series analyzed.
In 2017, Chl-a concentration reached its highest values, with results higher than 60 µg/L, while the following period, from 2018 to the first semester of 2019, showed lower Chl-a values.
The months of June and August 2019 registered the greatest differences in Chl-a concentration, from 11.49 µg/L, the lowest value sampled, in June to 123.63 µg/L in August. This period coincides with the winter school vacations in July, which may have influenced the number of residents or frequenters of summer houses located near the reservoir.
Quantitative and qualitative water variations depend on the type of ecosystem (ZAKEYUDDIN et al, 2016), the dynamics of aquatic systems, and the concentrations and interactions between their components, such as Chl-a concentrations, which are influenced by biotic and abiotic factors.
A natural abiotic factor influencing Chl-a concentrations is rainfall. Periodic precipitation can lead to concentration or dilution of dissolved nutrients and particulates in the aquatic system (BOUVY et al, 2003).
We also compiled the remote sensing reflectances (Rrs) of the eight selected images, at the CETESB sampling point (Fig. 4), at different wavelengths, according to the images of the spectral bands obtained by the Sentinel-2 MSI sensor.
Figure 4 points to the presence of Chl-a, such as high absorption in the blue and red regions, high reflectance in the green region (550 nm), and especially the reflectance peak in the region near 700 nm.
But some of these peaks are influenced by other water components, such as the region near 560 nm, where contributions from CDOM and suspended solids can occur, and in the region near 700 nm, where we observe the influence of suspended solids. (SANTOS et al., 2019).
The spectrum for 28/08/2019 shows well-defined peaks, corresponding to Chl-a, in the visible region and an increase in reflectance starting at 700 nm, and is the only spectrum to present higher reflectances in this wavelength than in the visible region. These peaks suggest high suspended solids content and high phytoplankton biomass.
To verify the presence of suspended solids in the reservoir, we compiled turbidity, total dissolved solids, and total solids data available at the InfoÁguas platform (Table 2), on the dates corresponding to the images. We calculated the suspended solids values by subtracting the total dissolved solids from the total solid’s values.
Results show that the sampling date 29/08/2019, whose corresponding image was taken in 28/08/2019, presents not only the highest Chl-a index, but also the highest suspended solids content, as well as the highest turbidity, of all the samples. Turbidity is directly proportional to the presence of suspended solids, since it measures the difficulty for a beam of light to pass through water due to the presence of suspended solid matter.
Table 2
Content of total dissolved solids, total solids, turbidity, Chl-a, and suspended solids at the sampling point.
Sampling date (dd/mm/yy) | Imaging date (dd/mm/yy) | Chl-a (µg/L) | Total Dissolved Solids (mg/L) | Total Solids (mg/L) | Suspended Solids (mg/L) | Turbidity (UNT) |
22/02/18 | 24/02/18 | 20.58 | 128 | 132 | 4 | 6.6 |
07/06/18 | 09/06/18 | 43.44 | 208 | 218 | 10 | 5.7 |
16/08/18 | 18/08/18 | 27.00 | 196 | 202 | 6 | 4.5 |
06/12/18 | 06/12/18 | 12.43 | 126 | 128 | 2 | 15.00 |
21/02/19 | 24/02/19 | 27.45 | 148 | 162 | 14 | 4.34 |
26/06/19 | 24/06/19 | 11.49 | 148 | 164 | 16 | 2.33 |
29/08/19 | 28/08/19 | 123.63 | 188 | 208 | 20 | 34.2 |
12/02/20 | 14/02/20 | 33.86 | 184 | 192 | 8 | 11.2 |
08/02/21 | 08/02/21 | 36.98 | 100 | 112 | 12 | 9.54 |
3.2. Application of algorithms
By applying the algorithms and adjusting several linear and nonlinear trend lines, we verified the degree of correlation (R²) between the algorithm-estimated values and the CETESB data. Figure 5 shows the regressions that showed the highest correlation with the lowest error rates for each algorithm.
Most points are concentrated in a range from 20 to 40 mg/m³, with one high Chl-a point above 120 mg/m³ and two points below 20 mg/m³. In all algorithms, the 2nd degree polynomial and linear regression trend lines showed the highest correlations.
In the DG2B algorithm, the linear and second-degree polynomial regressions showed similar correlation and error values, however, we chose the linear regression as the most indicated because it presented the lowest MAPE value and the lowest bias.
The SLOPE algorithm had the highest correlation value and the lowest errors, except for the MAPE values, which present the lowest error with NDCI linear regression.
The remaining algorithms showed similar values, with a 0.95 correlation, RMSE around 7 mg/m³, NRMSE at 6%, and MAPE ranging from 9 to 11%. DG2B can be considered the second-best algorithm, showing the best RMSE and NRMSE correlation values, followed by NDCI and, finally, DG3B, with the highest error values and lowest correlation value.
Some authors have obtained similar results, such as Neil et al. (2019), who evaluated the performance of 48 different Chl-a estimation algorithms based on data collected from 185 continental and coastal aquatic systems, which covered 13 optically different water types. Of these, four algorithms stood out as most suitable and accurate, including DG2B and NDCI.
When studying the Ibitinga Reservoir, also in São Paulo, Cairo et al. (2020) classified the Chl-a concentrations in ranges and tested several algorithms for each band, pointing to the SLOPE algorithm as the most suitable for waters with Chl-a concentrations between 19.51 and 87.63 mg/m3, corroborating our study.
However, when adjusted for the 2nd degree polynomial and linear regressions, the SLOPE returned negative values. The DG2B algorithm showed similar results when adjusted for the 2nd degree polynomial regression.
Watanabe et al. (2019), when investigating the trophic gradient of three cascading reservoirs on the Tietê River, in São Paulo, described a similar scenario, which led the authors to choose an algorithm presenting positive values, the NDCI algorithm with linear adjustment, even though they presented less favorable correlation and error values.
Thus, we used the NDCI algorithm, because although it presented the third best result, it did not return negative Chl-a concentration values. According to Mishra and Mishra (2012), NDCI has the advantage of varying between − 1 and + 1, so that qualitative Chl-a mapping and bloom detection using remote sensing is possible even for remote areas where field data are unavailable or unusable.
Despite the correlation indexes obtained in this study, even higher than those obtained by other authors, the present study sought to use data already available to the public, without collecting new data, which resulted in a reduced number when compared to other authors.
Due to the reduced number of data, some Chl-a concentration ranges were not considered, which can cause errors in the estimate. As the water body has only one monitoring point, the water characteristics at this point are understood to be the same for the entire reservoir area—a wrong assumption given the trophic state along the reservoir.
We suggest, therefore, collection of new data for more accurate results regarding different situations.
3.3. Spatio-temporal Chl-a distribution
We applied the NDCI algorithm adjusted for the 2nd degree polynomial function to all images, which resulted in Chl-a concentration values in mg/m³ (Fig. 6). Chl-a concentrations ranged from values lower than 25 mg/m³ up to 150 mg/m³, reaching 152.3 mg/m³, with the lowest value being 8.80 mg/m³.
Overall, the lowest concentrations are found closer to other watercourses. Chl-a concentrations tend to increase throughout the reservoir, with the highest concentrations found in areas closest to the dam and banks.
Areas with cloud cover and macrophyte interference were removed and represented by masks. Although these plants interfere in the results, with data showing its reflectance rather than the reflectance of the water, the presence of this vegetation indicates an environment with enriched concentrations of nutrients, which influence its growth.
On 24/02/2019 and 24/06/2019, we can also note the presence of macrophytes near the sampling point. On 28/082019, the concentration of Chl-a is relatively high on the reservoir banks, especially near the CETESB sampling point.
This location is mainly occupied by houses and also hosts the Yacht Club. The circulation and occupancy rate at the site can lead to an increase in domestic sewage or waste, which influence the amount of nutrients and the eutrophication process in this area.
An objective instrument used to compare the eutrophication state of aquatic systems is the Trophic State Index (TSI) (NOVO et al., 2013). In our study, TSI was verified using the Chl-a concentration values (Fig. 7) obtained from the Modified Carlson Index, the same as used by CETESB, an instrument based on three indicators: Secchi depth, total phosphorus and Chl-a.
The TSI consists of six categories classified according to Chlorophyll-a as follows: ultra-oligotrophic (Chl-a ≤ 1.17 mg/m³), oligotrophic (1.17 < Chl-a ≤ 3.24 mg/m³), mesotrophic (3.24 < Chl-a ≤ 11.03 mg/m³), eutrophic (11.03 < Chl-a ≤ 30.55 mg/m³), supereutrophic (30.55 < Chl-a ≤ 69.05 mg/m³) and hypereutrophic (Chl-a ≥ 69.05 mg/m³).
Disregarding areas covered by clouds or macrophytes, most of the reservoir are can be classified as eutrophic or above. According to the lowest estimated value, 8.8 mg/m3, the lowest TSI classification obtained in the reservoir was mesotrophic.
Of lower incidence, the mesotrophic category occupies 0.41% of the area, with no areas classified in this category on 02/24/2018 and 08/28/2019, whereas the eutrophic category, which occupies 63% of the area, shows the highest incidence.
Hypereutrophic areas are usually located near the reservoir’s dam and banks. On 08/28/2019, we can observe the largest area classified as hypereutrophic, with the remaining area classified as supereutrophic.
Although we did not use data collected from the region closest to the dam, due to the presence of macrophytes, this site can have high rates of nutrients and consequently high eutrophication rates, due to the retention of nutrients and other compounds near this location.
On 06/24/2019, the reservoir source is supereutrophic, becoming eutrophic in the second region, contrasting with the other images, in which the second region is shown to have higher Chl-a concentration. Figure 8 illustrates the percentage of area occupied by each category.
On all analyzed dates, most of the reservoir is classified as hypereutrophic or above. On 02/24/2019 and 06/24/2019, most of the area is classified as supereutrophic. On 06/24/2019, 88% of the area can be classified as eutrophic.
Interestingly, the image from 08/28/2019 shows large hypereutrophic and supereutrophic areas, corresponding to approximately 58% and 41.9% of the total area, respectively, leaving only 0.1% classified as eutrophic.
All these analyses unveil the worrying situation of this reservoir. The high eutrophication rate found can interfere with the biogeochemical cycle of this environment and damage the local biodiversity, the neighborhood, as well as the population who have contact with the water or consume some organism from this environment.