Aerial Biomass and Nitrogen Estimation of Emerged and Floating Macrophytes using Optical Remote Sensing: A Non-Destructive Method

Freshwater resources faces threats with aquatic plants invasion, considered biological pollution with deep effects on water quality and nutrients cycling due to their rapid growth. Orbital remote sensing has been an effective instrument of monitoring large water bodies. Thus, the aim of this study was to analyze the relation between reectance and eld measurements (biomass and nitrogen concentration) of aquatic plants to develop estimation equations and to test vegetation indices to use in orbital remote sensing. The most common tropical infesting species (Salvinia auriculata, Pistia stratiotes, Eichhornia crassipes and Eichhornia azurea) were collected during a year, measured their spectral response to simulate satellite bands, and the biomass and nitrogen concentration measurements. The bands intervals of Sentinel-2 satellite were choosing to the simulation due to their narrow bands and the RedEdge new band. The obtained eld data were correlated with the reectance obtained from spectroradiometry of each species and the equations showed R² = 0.64 to estimate biomass and R² = 0.60 to estimate nitrogen using the entire spectrum. Several indices described in the literature were tested with different Sentinel-2 bands but with no signicant results. The NDVI index showed a separation among species using RedEdge band and can be used to identify the species, but not to estimate their biomass.


Introduction
Aquatic macrophytes play an important role in the aquatic ecosystem balance (Esteves 2011;Pompêo 2017). However, the misuse of water resources by man, mainly in reservoirs for multiple urban uses, favor rapid plant growth that become aquatic weeds which harmed water quality. This condition needs constant monitoring of biological parameters, which is a fundamental component in freshwater resources management.
The pollutants release, deforestation, urban growth in the surroundings and dams contributes with a high concentration of organic matter in water, creating ideal conditions for aquatic plants spread, becoming invasion aquatic weeds (Jampeetong and Brix 2009;Julien et al. 2002;Sullivan et al. 2011). Some infecting of emerged and oating species is so agglomerated that could be misinterpreted as an island.
The infestation became a biological pollution with signi cant effects on water quality, since disturbs the ability to alter nutrient cycling and the ecological functioning, creating harmed consequences for ecosystem and public health. This is why managers need fast and consistent techniques to monitor infestation condition taking preventive and/or remedial actions (Pompêo 2017).
Optical remote sensing of freshwater resources is increasingly becoming important to monitoring macrophytes infestation, especially the submerged ones (Ni et al. 2020; Rotta et al. 2018). There are few studies with emerged and oating plants, but these species cause infestations that normally occupy enormous areas, becoming similar to "crowded mats" on water surface (Coelho et al. 2005). Groundbased measurements provide the most direct and accurate distribution and quantity of aquatic weeds, but they are time-consuming, labour-intensive and spatially limited. So, the availability of satellite data provides great potential for the spatial and temporal monitoring of aquatic plants in a timely and costeffective approach (Dube et al. 2017).
The spectral response of vegetation enables to separate species and can be transformed into mathematical vegetation estimation models (Ferwerda et al. 2005; Ponzoni et al. 2012; Tian et al. 2011;Ullah et al. 2012). Absorption characteristics of plants can be incorporated in these empirical models using vegetation indices (VIs) to identify stress, biomass, productivity and other biophysical traits (Ghosh et al. 2016; Rotta et al., 2018). Most of the infestations shows one predominant specie and detecting it could be very helpful for the managers. Also, the estimation of biophysical parameters enables large scale of aquatic weeds monitoring, which would not be possible using the more expensive and timeconsuming ground-based measurements. The repeated coverage of satellite sensors provide data for long-term monitoring, which is crucial in identifying the success of controlling measures of aquatic weeds (Penatti et al. 2015).
Many indices have been used to better estimate aerial biomass and nitrogen leaves concentration ). All of them use the spectral bands Blue, Green, Red, Red Edge, and Near Infrared (Ferwerda et al. 2005;Ponzoni et al. 2012;Tian et al. 2011;Ullah et al. 2012). Since the Red Edge (RE) vary within species, Aparicio 2007 calculated it for twelve macrophytes using radiometric data, in order to simulate their spectral signatures registered by satellites. However, there are few studies that test vegetation indices (VI) in aquatic macrophytes, estimating biophysical and biochemical parameters in an infestation area using re ectance, that can be faster and less expensive to vegetation measurements. In addition, the launch of satellites with new spectral bands, such as RE and other subdivisions of the near infrared (NIR), brings the need to carry out new tests. through the eld collection, measuring radiometry directly from the plant, doing the opposite way. Also, other studies observe spatial variability of macrophyte cover, but there are no studies estimating biomass or nitrogen concentration in an infestation condition. Rotta et al. (2018) discuss that models with more eld data can enhance the accuracy to periodically map aquatic vegetation height and biomass. The existence of various indices enables to nd relationships between eld, laboratory and satellite data, but they need to be tested on emerged and oating aquatic plants using bands from multispectral satellite that have rarely been used.
For a large-scale modeling of biophysical and biochemical parameters to be successful, it is necessary to: 1) correct selection and application of remote sensing; 2) that is coupled with the eld data for calibration and validation; 3) and that is integrated with an appropriate mathematical modeling (Barbosa et al. 2014). The Sentinel-2 satellite includes a Multi-Spectral Instrument (MSI) and has been used in monitoring the earth surface characteristics. This innovative equipment captures high-resolution images with 13 spectral narrow bands for a new view of the soil and vegetation. Studies have shown that indices calculated from narrow bands improve the estimation of several vegetation parameters (Gong et al. 2003;Lee et al. 2004;Mutanga and Skidmore 2004). In addition, data obtained through Sentinel-2 can be used to evaluate biomass every 5-10 days (Dusseux et al. 2015), improving the di culties of eld sampling and the consistency in collecting data for management. Sentinel-2 satellite was the rst optical Earth observation satellite to have three spectral bands located in the "red edge" band, providing important information about the state of plants, although these bands have not been commonly used, compared to the remaining bands (Ni et al. 2020). So, this satellite can be used to discriminate and classify invasive plants.
In the case of emerged and oating macrophytes, it is important for managers facing infestation issues that new models are made and indices are tested to estimate biomass and nitrogen concentration, in order to collect information for better decisions. It is necessary that this collection is not expensive, labour-intensive (non-destructive) and time consuming as ground-based measurements. Therefore, the aim of this study is to calculate models and to test indices that estimate aerial biomass and nitrogen leaves concentration of emerged and oating macrophytes obtained in the eld, using their re ectance measured in the laboratory, simulating satellite observation. For that, the range spectral used for simulation is from satellite Sentinel-2, with narrow spectral bands and RedEdge band included.

Materials And Methods
The emerged and oating macrophyte species chosen for this study are the most common infesting in Brazilian reservoirs, especially those surrounded by urban areas: Salvinia auriculata, Pistia stratiotes, Eichhornia crassipes e Eichhornia azurea (Pompêo 2017). Except for S. auriculata, the other four species were already studied and had calculated the RedEdge Position for each species (Aparicio 2007;Aparicio and Bitencourt 2015).
A diversity of samples of each species was collected to estimate biomass and nitrogen concentration for emerged and oating macrophytes. To obtain the eld parameters, the species were collected from different locations and periods, aiming to diverse the environmental conditions which should improve the nal model. However, the lack of enough infestation due to the water treatments using algaecide in the reservoirs led the study to diversify the way of sampling. So, the locations of collection were: three reservoirs in the State of São Paulo, Brazil (Paraitinga, Biritiba and Guarapiranga); Guaratuba river inside the Bertioga Restinga State Park (BRSP); some individuals were purchase at garden stores; and some from cultivation developed in a greenhouse. The greenhouse cultivation consisted of water tanks to each species, with application of a nutrient solution (every 15 days), water circulation and ltration system for no algae growth. The collection of species in all these places occurred from August 2017 to November 2018, totaling 30 samples of S. auriculata, 30 of P. stratiotes, 23 of E. crassipes and 7 of E. azurea ( Table  1).
The plants were sampled using the square method, size 0.25 m x 0.25 m (Pompêo 2017), and carried out in boxes with water to the laboratory in the same day until the radiometric measurement (re ectance) was made. The measure was taken under controlled light conditions provided by a 1000-watt halogen lamp, positioned nearly 1 m from the sample and previously heated. This lamp simulates the natural conditions of the sun. After calibration of the program with white and black plates, the re ectance percentage of each specie was measured using the Ocean Optics® spectroradiometer (USB4000 model), with a sensor, positioned at a height of 25 cm from the sample, that detects and measures the radiation that leaves the plant's surface chasing a view angle of 24,3°. The measurement produces the re ectance values per wavelength (nm) almost continuously, allowing the creation of graphs for each sample called spectral curves. The plants were positioned in glass vats simulating the same position that appear in the eld. Therefore, the radiometric reading using the spectroradiometer correspond to the same reading registered by the satellite in natural conditions: only from the re ectance of the aerial part of the plants.
Under infestation conditions, the plants form "crowded mats" on the water surface, making it impossible to observe the interference of their submerged parts (e.g. stem, root), water or soil in the re ectance observed by the sensor remotely located on the satellite.
After the radiometric measurements (re ectance data), the plants were separated into aerial and submerged fractions and the fresh and dry weights (60°C) were obtained. With the high and the angle of the radiometric sensor from the sample, the area and the fresh biomass of the aerial part "read" by the spectroradiometer was determined to calculate the plant density (g/m²) for each sample. The same was done for nitrogen concentration to determine the sensor's area of view in g/m². After drying, the leaves were ground with a Wiley mill to pass through a 1-mm mesh screen. Dried and ground leaves were analyzed for chemical composition, prepared for nitrogen analysis of foliar N concentration following the method of Kjeldahl (1883).
The re ectance data, obtained through radiometric measurement, were segmented according to the band's width (range) of the satellite Sentinel-2 (10m pixel -bands 2, 3, 4 and 8; and 20m pixel -bands 5, 6, 7 and 8b), ranging from 400 to 900 nm. Regions smaller and larger than these values (150 -300, 1800 -1950 and 2400 -2500 nm) were discarded due to the strong absorption and noise of water and atmosphere ). This satellite has two spectral bands in the Red edge region and three in the near infrared (NIR), used in vegetation studies and now tested for aquatic macrophytes (Table 2).
From the average re ectance values collected from each sample based on the bands range from Sentinel-2, equations of biomass and nitrogen concentration using re ectance obtained in laboratory were calculated with all spectral bands using simple linear regression. First, the aim is to obtain a general mathematical model to estimate biomass and nitrogen concentration in future orbital images for the main species of emerged and oating invasive aquatic plants. Second, using the re ectance of each specie, based on Sentinel-2 bands range, the aim is to test know vegetation indices in aquatic plants, checking the possibility to estimate biomass and nitrogen concentration in infestation conditions from orbital images without the need of collection eld. Therefore, four VI and two NI, reviewed by Wang et al. (2016), were used to calculated these parameters in aquatic plants through regression analysis. All VIs used RedEdge, Red, Green and NIR bands. One NI used Blue and Red Edge and the other NI used Blue, Red Edge and NIR ( Table 3).
Analysis of variance (ANOVA) were used to identify (statistically signi cant -α = 0.05) windows of spectral reparability from re ectance between the aquatic plant species. Also, between the VI and NI indices of each species based on Sentinel-2 bands. Aparicio (2007) calculated the Red Edge Position of twelve macrophytes using radiometric data obtained in laboratory through the interpolation and derivative methods, in order to simulate their spectral signatures registered by satellites. The Red Edge values species speci c were used in this study to improve the mathematical models and the indices tests, and are presented in Table 4. According to the author, all species studied here can be observed from the same Sentinel-2 RedEdge band (RE1). The analysis of variance (ANOVA) test was also used to show the differences between the RedEdge of each species. Aparicio (2007) calculated the RedEdge for Salvinia molesta, and in this study it was used Salvinia auriculata. Despite being different species, they present very similar spectral curves, differentiating them from other species of oating aquatic plants.

Spectral curves
The radiometric measure of each sample was performed to describe the spectral curve using samples from several places and dates, showing a great diversi cation of data for the estimate equations. The average of the collected curves and the percentage of re ectance for each species are shown in Figure 1.
The analysis of variance tests show that the curves are different among species (F = 277.07; p = 0.01). The P. stratiotes and E. crassipes species presented the higher values in the RedEdge and NIR region. In addition to E. crassipes showing such prominence, it is also the species with the highest peak in the green region, separating it from the other plants. S. auriculata and E. azurea showed similar spectral behavior in the visible bands (blue, green and red), separating only in the RedEdge (RE1 e RE2) and NIR bands.

Biomass equation
The predictive model for each species separately did not show signi cant results, therefore, the biomass equation brings all species to create a model that is unique for emerged and oating aquatic plants. Thus, to estimate the fresh biomass of emerged and oating macrophytes from radiometric measurements the combination of all spectral bands was used and equation 1 was de ned as the estimation model. The regression coe cient was R² = 0.64 (p<0.05), showing that it is possible to estimate biomass using the plant re ectance ( Figure 2): The spectral bands were tested separately to check which ones had the best result and regression coe cients to biomass estimation ( Table 5). The bands of Red (R² = 0.41) and RedEdge (R² = 0.43) presented signi cant values, but only for E. crassipes (Figure 3 a,b). The results for the other species are not signi cant.

Biomass indices
The biomass indices (VI) from literature ( Table 2) were applied for each species, and the values range of the best indices are shown in Figure 4 (a,b). These values can be used in image classi cation to recognize these species infestation in water bodies using orbital image. The relation between indices and biomass values of each specie separately is described in Table 6. The indices that showed better regression values for all species were NDVI (R² = 0.24) and RVI (R² = 0.31) and the separation of the groups are show in Figure 5 (a,b). However, these values are not signi cant, so it is not possible to use in measuring biomass of emerged and oating macrophytes. The species with the best coe cient was E. crassipes, using the NDVI index (R² = 0.22), but also with a low value of coe cient.
The Red Edge re ectance of each specie based on Aparicio (2007) interpolation method was used to test the relation with green biomass using linear regression. The average was also used do verify the species difference using ANOVA. Although the results for regression analysis were not signi cant, the results from ANOVA showed that Red Edge separated the species and can be use in the vegetation indices to identify these aquatic plants species (F = 20.96; p < 0.05) ( Figure 6).
Using these new narrow band, the vegetation indices (VI) were also calculated for each species and then compared using ANOVA to check if they can be used with emerged and oating aquatic plants. The results show that the NDVI index, using the RE1 band, is different among the species and can be used to identify them in a future orbital image (F = 4.3; p < 0.05) ( Table 7).

Nitrogen equation
To estimate the N concentration of emerged and oating macrophytes using radiometric measurements, the combination of all spectral bands was used to de ne equation 2, with regression coe cient R² = 0.60 (p < 0.05), which suggests that it is possible to estimate the N concentration of aquatic plants from re ectance (Figure 7).
The spectral bands of the Blue, Near Infrared (NIR) and RE1 were tested separately to check which had the best regression coe cient (Table 8). There were no signi cant results, except for E. azurea in the Blue region (R² = 0.53; p < 0.05). Only with the intersection of all spectral bands is it possible to estimate the aquatic plants N concentration from re ectance.

Nitrogen indices
The nitrogen indices described in the literature (Table 2) were tested for all species and then separately ( Table 9). The index with the best regression coe cient for all species was NI_Wang (R² = 0.24). However, this value is not signi cant to estimate the N concentration of emerged and oating macrophytes from NI_Wang index. The differences among the nitrogen indices (NI) were not calculated using ANOVA since they were not signi cant to estimate nitrogen concentration in emerged and oating aquatic plants.
Nevertheless, the values range for each nitrogen index is shown in Figure 8 (a,b). These values can be use in a future image classi cation to recognize the species in an infestation condition in water bodies based on nitrogen using orbital image, but not for estimate it concentration.

Spectral curves
The curves obtained in this study showed that there is a re ectance difference among species, as already suggested by Aparicio and Bitencourt (2015). The mainly differences were in the regions of Green (525 nm to 595 nm), Red (635 nm to 695 nm), Near Infrared -NIR (727 nm to 957 nm) and Red Edge (680 -750 nm), proving the spectral signature species speci c of emerged and oating macrophytes.
When it comes to vegetation, the NIR region presents small absorption and higher re ectance, because it is the result of the incident energy interaction with the structure of the mesophyll that is the internal structure of the leaf. Maximum absorption occurs in the red region, in which chlorophyll absorbs electromagnetic radiation (Ponzoni et al. 2012(Ponzoni et al. , 2015. The differences in the leaf structure of each species, as well as in the concentration of photosynthetic pigments in the leaf, create differences in the re ectance that will be read by the satellite sensor, as observed by the spectroradiometer in this study. Therefore, it is possible to use this information to recognize an aquatic plant species using radiometric data obtained from an orbital image. Peñuelas et al. (1993) observed that species of emerged and oating aquatic plants show greater distinction in the limits of Red and NIR than submerged plants, facilitating this differentiation.
Absorption by all pigments occurs in the Blue region (425 nm to 500 nm), so it has low re ectance values and it is di cult to discriminate species. In the green region, there is a peak of re ectance at 550 nm due to the lower absorption in this region, which con gures the green color of the leaves. The Red Edge region also shows high re ectance values in emerged and oating, showing that this new and narrow band is important in the recognition and collection of vegetation data, and can also be used for aquatic plants (Cho and Skidmore, 2006).

Biomass
The vegetation indices described in the literature (NDVI, RVI, TVI and RDVI) could be improved by using the Red Edge band since each species may have different Red Edge Position, as showed by Aparicio and Bitencourt (2015). Using only Red and NIR bands (strong absorption and maximum re ectance respectively) the results was not good as expected. The greatest correlation found here between biomass and re ectance are associated with substituting band Red (R² = 0.41) to band Red Edge (R² = 0.43) for the E. crassipes species.
Song and Park (2020) showed a clearly difference between aquatic plants and water surface using NDVI, concluding that is the most effective vegetation indices for detecting aquatic plants. However, the separation between species is more complicated. Peñuelas et al. (1993) found that NDVI was not able to separate oating from emerging plants, explaining that the species show great variation in density and vigor. Therefore, it was only possible to classify biological groups through multivariate analyzes and new combinations of narrow spectral bands, as was observed in the present study with Red Edge band. Zhou et al. (2018) argue that the NDVI index varies with spatial and temporal changes due to the in uence of vegetation groups. However, in similar types of vegetation, as the studied plants, the index can be invariable and unde ned (Chen et al. 2012), requiring the use of narrower bands.
The NDVI index was derived from the NIR and Red regions. If the total photosynthetically biomass increases, the re ectance in the Red decreases and in the NIR increases (Dusseux et al. 2015). However, some studies show that broad spectral bands of NDVI can be unstable, varying with soil color, canopy structure, optical leaf properties and atmospheric conditions (Middleton, 1991;Qi et al. 1995). In addition, the index reaches a saturation level after reaching high values of biomass or leaf area (Gao et al. 2016). That is why studies have shown a non-linear relationship between NDVI and vegetation properties, due to this saturation over densely vegetated areas (Edirisinghe et al. 2011;Viña et al. 2011). This is because the use of broad spectral bands results in the loss of critical information available in speci c narrow bands (Thenkabail et al. 2000), as Red Edge band. Ullah et al. (2012) also found no signi cant results when correlating biomass indices with grass spectroradiometry (low and non-explanatory regression coe cients). For biomass, they also explain that saturation can di cult the estimation in smaller plants with condensed behavior, as seen in emerged and oating macrophytes. The band depth analysis parameters calculated with narrow band sensors were more accurate predictors of biomass than the NDVI index with Red and NIR bands (Mutanga and Skidmore 2004;Chen et al. 2009). Carvalho et al. (2013) also agree that satellite signal saturation is a problem for a precise relation between indices and forest structure. The explanation for the saturation problem is that, from a leaf area index of 3, the amount of red light around 660-680 nm that can be absorbed by the leaves and reaches a peak. The re ectance of the NIR continues to increase due to the multiple dispersion effects (Kuman et al. 2001). This imbalance between the saturation of the red-light absorption and the high re ectance of the NIR results in a slight change in the NDVI, resulting in poor relations with biomass (Mutanga and Skidmore 2004). A vegetation cover rate above 60% shows that the indices using the NIR region are not perceptible to changes in vegetation (Gitelson et al. 2002;Glenn et al. 2008), which it may have occurred with emerged and oating macrophytes from having a dense vegetation cover.
Obtaining data directly from the eld is a challenge since several environmental conditions can in uence the results. Rotta et al. (2018) found a strong similarity between the curves based on in situ data and those based on image data (using SPOT-6), but with a slight underestimation in the eld model. They discuss that the low number of sampling points (n = 8) used in calibration may not be su cient to build a robust prediction model or greater sensitivity of R². This issue was tested in the present study and better results were founded using a greater number of samples from different places. Despite the di culties in estimating submerged aquatic heights from waterbodies due to their optical complexities, Rotta et al. (2018) discuss that the results provide useful initial evidence that it may be possible to use existing radiative transfer models to map them with adequate accuracy.
The literature indices require only two bands and can be limited to explore the rich information contented in the hyperspectral data. Then, multiple linear regression based on more than two hyperspectral bands has been used to predict vegetation parameters, such as leaf nutrient and biomass (Curran et  However, for emerged and oating macrophytes, the results show that the combination of all bands is more e cient than the indices that use speci c bands. Ferwerda et al. (2005) also founded that the correlation between N concentration and hyperspectral measurements was greater when the indices used all ranges of the spectrum than that using only Red and NIR ranges. The N indices (N_Wang and N_Tian) were not signi cant to estimate the concentration of N in emerged and oating macrophytes, as well as failing to estimate %N for crops ).
The species collected in this study belong to the same biological group ( oating aquatic plant, except for E. azurea that is emerged aquatic plant) (Esteves 2011). Ferwerda et al. (2005) observed that indices commonly used to determine N concentration can be explanatory or not, depending on the type of plant (grasses, willows, olive trees, and others), which demands more speci c models. Wang et al. (2016) also found no signi cant difference in the concentration of N between species that belong to the same functional type. It should be considered that there is a variation in the concentration of N that can occur within the species in response to differences in soil properties and history of environmental variations (Smith et al. 2002). Therefore, the variation of N between biological and functional types can be better explained than between species.
In addition, spectral and eld information from one location may not be representative for another. Other factors can also in uence variations in re ectance, such as age, uniformity and canopy layout (Martin and Aber 1997;Towsend et al. 2003). Therefore, it is important that the calibration of a model is based on data from different locations and periods of eld collection, producing an equation that is predictive for emerged and oating macrophytes in general, regardless of age and conditions in which populations are found. It is important to develop an equation that works in multiple scenes without having to collect data for each speci c location (Martin and Aber, 1997), as was showed in this study.
Nitrogen is not easily detected through vegetation indices. Estimating the biochemical parameters of broad-band sensors is more challenging than biophysical parameters. The chemical composition of a target is masked by the average effect of the width of a wide spectral band (Ullah et al. 2012), in addition to broadband scanners losing absorption by N (Ferwerda et al. 2005), making the biochemistry prediction more di cult. The variation of N concentration is so subtle that it may have no effect on re ectance (Ullah et al. 2012;Yu et al. 2014). However, using a sensor with narrower spectral bands, such as Sentinel-2, and a model with all its spectral bands based on eld measurements, it was possible to found an equation capable to estimate the N concentration for emerged and oating macrophytes, what before on a spatial scale was a challenge.

Conclusions
This study shows that: Emerged and oating aquatic plant species presents differences in their re ectance, allowing specie speci c spectral curves that can be useful to estimate vegetation measurements.
It is possible to estimate the biomass and nitrogen concentration of emerged and oating aquatic macrophytes using re ectance obtained in the eld and measured in the laboratory with spectroradiometer, through equations developed with all spectral bands.
The mathematical models had a good accuracy and are capable to estimate the parameters biomass (R² = 0.64) and nitrogen concentration (R² = 0.60) as a non-destructive collection method, without using time-consuming, labour-intensive and more expensive ground-based measurements.
The Red Edge narrow band showed to be better for calculate vegetation indices for aquatic plants, proving that more advanced multispectral sensor with narrow bands presents a valuable data-source for the accurate mapping of invasive species.
The equations to estimate biomass and nitrogen concentration obtained from all species may simulate a natural situation in which several species can be mixed when observed by orbital sensor.
The indices presented in the literature did not show satisfactory results for estimating the parameters, perhaps because they only use the selection of some broad bands for measurements that are subtle and di cult to perceive re ectance.
The simulation using the range of spectral bands of satellite Sentinel-2 and the eld data open a good possibility for future studies using orbital images. The orbital images can become quite useful considering that the nowadays satellites can have excellent radiometric resolutions, giving a good subsidy for the use of orbital remote sensing in limnology.
The ndings of this study underscore the relevance of the new generation multispectral sensors in providing primary data-source required for mapping biomass and nitrogen concentration at lower or no cost over time and space through re ectance, providing necessary insight and motivation to the remote sensing community, ecologists and environmentalists.
Collecting biomass samples in the eld is quite challenging. Using satellite-based models could be the only viable way, in terms of cost and temporal frequency, to perform periodic collection of plants in waters which can signi cantly aid in ecosystem management. Although the initial results presented in this study are encouraging, the method needs to be further evaluated across different species and various other waterbodies to test its robustness.