1. Introduction
Coastal lagoons are often formed when a sandbar encloses a former bay [
1]. These wetland ecosystems, like other lentic water bodies, are highly sensitive to anthropogenic pressures and impacts. These can alter their ecological functionality and compromise their uses for human activities [
2]. The ecological status of water bodies is the central concept of the European Water Framework Directive (WFD), relating the actual status of a given water body with a set of reference conditions that represent its ecological optimum [
3]. The status is assessed based on a set of biophysical variables obtained from conventional in situ monitoring. Despite not being included in the initial formulation of the WFD, remote sensing techniques are increasingly being used to assess, in a systematic and synoptic manner, some of those biophysical variables. Optical satellite sensors can retrieve some key indicators of a water body’s ecological status, including chlorophyll-a (Chl-
a) concentration, the total suspended matter (TSM), the absorption coefficient of dissolved organic colored matter, and the water transparency, measured as the Secchi disk depth (SDD) [
4].
The Chl-
a concentration is commonly used as an indicator of the phytoplankton biomass since almost all species contain the key photosynthetic pigment. Chl-
a concentration is also an indicator of the eutrophication process, caused by increased nutrients (N, P and other elements), which is often associated with human activities. Therefore, Chl-
a acts as a connector between nutrient concentration and primary production [
5]. This biophysical parameter has two distinct absorption peaks, centered in vivo at wavelengths of 440 nm and 675 nm [
6].
Particles present in the water column absorb and scatter light. A higher concentration of particles leads to increased scattering and therefore an increase in the turbidity of water bodies [
7]. The TSM determined by filtration of water samples consists of inorganic particles (minerals, sand, etc.) as well as other organic particles (phytoplankton, detritus, etc. [
8]. An increase in the TSM increases scattering and therefore the reflectance of a water body. This effect is especially noticeable in spectral regions where the absorption of other optically active constituents of the water body is low. There is a direct positive correlation between the reflectance in the red spectral region and the concentration of suspended solids in water due to their particle dispersion properties [
9]. The Secchi disk depth (SDD) is used in limnological studies to estimate the extinction of light in water [
10]. The SDD is inversely related to the amount of suspended matter in the water [
11]. As water clarity decreases, brightness in the red spectral region usually increases [
9].
The majority of water quality studies are conducted through in situ measurements and subsequent chemical analysis in laboratories. These methods allow researchers to gain insights into aquatic ecosystems, forming a basis for long-term monitoring to assess the ecological situation and identify trends in water bodies. The water reflectance is measured to analyze its optical properties and derive concentrations of optically active components present in the water column. The use of remote sensing as a tool for monitoring the ecological quality of water masses has great potential as a complementary strategy in an indirect, remote, frequent, and continuous manner over time. Remote sensing methods are often based on semi-empirical or empirical algorithms derived from the statistical regression, between the reflectance values of the water in different spectral ranges and the concentration of the water constituents [
12]. Many authors have successfully used remote sensing technologies to map different water quality variables: TSM in the Gironde estuary [
13]; TSM in the Frisian Lakes [
14]; Chl-
a, SDD and TSM in the Albufera of Valencia and natural ponds [
15]; and Chl-
a and SDD in the Albufera of Valencia and different Spanish reservoirs [
16]. The small size and complex shapes of most inland waters require the use of sensors with high spatial resolution. Among them, Landsat’s sensor is the go-to for these studies because it has a relatively high spatial resolution (30 m) and a long period of archived data. Some of these studies have even been done in the Albufera of València [
17,
18]. On the other hand, Landsat’s sensor has a limited spectral resolution and low temporal resolution, which makes it difficult to accurately determine some of the water quality variables. ESA’s Sentinel-2 (S2) mission has outstanding characteristics for measuring inland waters’ biophysical variables. The spatial resolutions of 10, 20, and 60 m, coupled with a revisit frequency of 5 days, significantly enhance the capabilities of the Landsat mission. Recent studies have successfully used S2 images to measure phycocyanin [
19] and turbidity and TSM [
20]. SPOT satellites have spatial resolutions of less than 10 m. Despite having considerable bandwidth, studies have developed satisfactory results measuring Chl-
a and TSM in lakes [
21] or Chl-
a, SDD, and total phosphorus in reservoirs [
22].
In 2016, the Peruvian Space Agency launched an operational Earth observation mission, named PerúSAT-1, with the main objective of monitoring and evaluating natural catastrophes. This satellite features a high spatial resolution NAOMI sensor [
23] and operational agility, which together enable enhanced spatial and temporal resolution for diverse surface monitoring. Building on these capabilities, this study aims to determine whether PerúSAT-1 images can be used to generate water inland quality assessment products that accurately describe the ecological state of the Albufera lagoon (Valencia, Spain). For this, we propose the following specific objectives: (1) atmospherically correct the images to synthesize advanced products, (2) develop algorithms that can estimate the optical properties of water through the combined use of satellite images and a database of different sampling in the Albufera lagoon, and (3) analyze the suitability of the PerúSAT-1 satellite to monitor the quality of the lagoon remotely and continuously over time.
4. Discussion
Authors such as [
52,
53] or [
54], using SPOT, Landsat, and Modis satellites, determined that remote sensing data could be used as an effective predictor of the trophic status of inland water bodies. In our study, algorithms were developed for five water quality variables using the multispectral bands of the Peruvian satellite PerúSAT-1. For three variables, Chl-
a, SDD, and PIM, the band combination was normalized difference (ND), and for TSM and POM, the band combination was simple ratio (SR). ND reduces the uncertainties in the estimation by excluding the seasonal solar azimuth differences and atmospheric contributions [
55], and the implementation of spectral band ratios in the retrieval algorithms reduces irradiance, atmospheric, and air-water surface influences in the remotely sensed signal [
56,
57]. In the Albufera lagoon, the trophic state is always hypertrophic, with a large amount of suspended matter. Taking into consideration the low spectral resolution of the bands and the correlation between the variables studied, several algorithms share the same bands in their formulation. This is the case for Chl-
a, SDD, and TSM, the most correlated variables, whose algorithms are based on the blue (B1) and infrared (B4) bands.
Chlorophyll-a concentrations vary between 21 and 331 µg/L, and the best algorithm was developed using the ND with infrared (B4) and blue (B1) (NRMSE = 16%). Similar studies demonstrated that by using the combination of the red and infrared bands in the Landsat TM and SPOT satellites, it is possible to determine the Chl-
a in water bodies. For example, [
58] used the TM3/TM1 ratio (R
2 = 0.67) to estimate Chl-
a in Pensacola Bay with a standard error of estimate of 1.55 µg/L. Similarly, [
59] used in their study the ratio between blue and red for Landsat 5-TM (R
2 = 0.72) in small lakes in Italy, with an RMSE of 1.3 mg/m
3. Ref. [
15] found that the ratio TM2/TM4 (R
2 = 0.66) for Landsat images, also valid for SPOT images, was useful for estimating Chl-
a using data from the Albufera lagoon.
Gitelson et al. [
60,
61] explain the principle behind these ratios as both bands corresponding to Chl-
a absorption; Chl-
a is directly proportional to the magnitude of the reflectance in the red band and inversely proportional to that in the blue or green band. When using the NIR band, results may be consistent, as the reflectance in this band may sometimes be less variable due to strong absorption by water [
62,
63].
With respect to SDD values, they range from 0.16 m to 0.42 m. As a result, the ND between infrared (B4) and blue (B1) (NRMSE = 15%) is the one that best fits the available data for quantifying this parameter. A study carried out by [
64] in Argentina determined that using a combination of the blue band and the blue/NIR, green/NIR ratios for the OLI sensor (R
2 = 0.84) provides a strong predictive relationship with SDD, with an RMSE of 0.56 m. While other authors, such as [
65] with a standard error of estimation (SEE) of 0.28 m (R
2 = 0.73), ref. [
66] with an SEE of 0.292 m (R
2 = 0.78), and [
67] with an RMSE of 1.20 m (R
2 = 0.72), concluded that SDD is strongly correlated with the responses in the blue and red bands of Landsat and MODIS data through the blue/red ratio in different lakes. As SDD decreases, the brightness in the red band tends to increase, which is explained by the positive correlation between red reflectance and particulate matter-inducing scattering. Dividing by the blue band, dominated by the absorbing effects of phytoplankton, normalizes the brightness in the red band. Furthermore, using the NIR band, rough atmospheric effects could be corrected [
9].
Finally, TSM in situ presents a variation between 39 and 136 mg/L, with an organic fraction between 7 and 60 mg/L and an inorganic fraction between 2 and 75 mg/L. For these values, the best results are presented between the band ratios blue (B1) and infrared (B4) for TSM (NRMSE = 11%), red (B3) and infrared (B4) for POM (NRMSE = 9%), and the ND of red (B3) and blue (B1) for PIM (NRMSE = 15%). This may be due to the greater amount of POM in the study area, compared to PIM.
Most of the studies using SPOT and Landsat TM satellite data (e.g., ref. [
68] with an RMSE of 25.77 mg/L (R
2 = 0.94) for SPOT data and 25.31 mg/L (R
2 = 0.98) for Landsat data) conclude from their results that the combination of green and red bands or blue and red bands presents high correlation coefficient values for the quantification of TSM. On the other hand, [
69] demonstrated that through the ratio between blue and red with Landsat 5-TM, it is possible to determine TSM with an RMSE of 0.295 mg/L (R
2 = 0.67). Likewise, ref. [
13] achieved this detection using the combination of bands 3 (infrared) and 1 (green) of the SPOT satellite with an R
2 of 0.93. Therefore, according to [
70] and more recently [
14], the wavelength between 700 and 800 nm is suitable for estimating TSM. The reason for the correlation, usually with red and NIR bands, is explained in [
71] by the contribution of particulate matter, especially the inorganic component, to scattering in these bands. The algorithms for PIM detection used in [
72] (R
2 = 0.99) and [
73] (R
2 = 0.96) are very similar to those for TSM due to the above explanation. Far fewer studies have investigated PIM and POM fractions compared to TSM, making it difficult to find empirical algorithms that effectively separate these signals.
In our study, band 2 was not useful for the determination of any parameter. The study area has very eutrophic waters, and waters with high Chl-
a (above 3–5 mg/m
3) produce distinguishable spectral characteristics in the red and NIR regions of the reflectance spectrum [
74]. Ref. [
75] explains that inorganic mineral particles, unlike organic particles, have a greater refractive index and scatter light more in the NIR, which is better estimated with algorithms that include visible bands. However, ref. [
20] explains that POM in the Albufera lagoon is better determined with bands close to the NIR, due to the high load of particulate matter present in the lagoon, as this high concentration produces a saturation in the visible bands, in favor of the red-edge and NIR bands.
Regarding the comparison of spatial resolution, the fact that the PerúSAT-1 images appear more homogeneous may be due to the width of their bands. This width, as well as the fact that several algorithms use the same bands, can reduce their sensitivity to detecting small variations in the variables studied. However, the very high spatial resolution can be very useful when avoiding the border effect or the inclusion in a water mask of coastal areas with dense aquatic vegetation (see
Appendix A,
Figure A6). Additionally, when correlating the data obtained through the algorithms developed for the PerúSAT-1 products, the error improves slightly with respect to the data obtained with the S2 products. Nevertheless, S2 offers a better revisit time and has a better band configuration, only having lower spatial resolution than PerúSAT-1. Considering that PerúSAT-1 can change its observation geometry, which increases the revisit time, it would be interesting to combine the products obtained from both satellites to perform studies in specific locations, thus obtaining better results and more robust models. Validation of the atmospheric correction with the in situ Rrs data demonstrates that the method developed in this study can be useful for application to PeruSAT-1 imagery. The hydrodynamic cycle that occurs in the Albufera was once modified by human hands, and thematic maps (
Figure 9,
Figure 10,
Figure 11,
Figure 12 and
Figure 13) show different phases of this cycle. Thus, there is a visible increase in Chl-
a, TSM, and its organic/inorganic fractions during spring. This increase coincides with the time when the fields are flooded for rice planting, possibly due to runoff from the fertilizers used in the crops. It can also be observed that the parameters begin to decrease in summer when a nutrient depletion occurs after the spring peak of primary production. Regarding the spatial distribution, it is observed that in the water inlet areas, where Chl-
a concentration is lower, there are higher SDD values. In these same entrance areas, inorganic solids predominate with respect to the center of the lagoon, where, since there is a higher Chl-
a concentration, organic solids predominate, and, in addition, there are lower SDD values.
5. Conclusions
To sum up, our study has identified that through the ND with infrared (B4) and blue (B1) bands of PerúSat-1, it is possible to determine the amount of Chl-a in eutrophic lakes, with a relative error of 16% (NRMSE), and to calculate the SDD with a relative error of 15% (NRMSE). Regarding TSM and its inorganic plus organic fraction, our results have demonstrated that we could use band ratios blue (B1) and infrared (B4) for TSM, red (B3) and infrared (B4) for POM, and the ND of red (B3) and blue (B1) for PIM, with relative errors of 11%, 15%, and 9% (NRMSE), respectively.
Due to an existing correlation between Chl-a and SDD in the eutrophic waters of the Albufera lagoon, the algorithms to determine both variables through PerúSat-1 images are very similar, as they use the same combination of bands.
Finally, the development of those new algorithms obtained through the combination of in situ data and data extracted from the PerúSat-1 satellite leads to the possibility of having more detailed information to study the spatial dynamics that occur in the Albufera lagoon, especially with the greatest spatial resolution that it offers. PerúSat-1 is a tool with the potential to be used in water quality studies, especially in hypertrophic waters.