Mapping and Quantification of the Dwarf Eelgrass Zostera noltei Using a Random Forest Algorithm on a SPOT 7 Satellite Image

: The dwarf eelgrass Zostera noltei Hornemann ( Z. noltei ) is the most dominant seagrass in semi-enclosed coastal systems of the Atlantic coast of Morocco. The species is experiencing a worldwide decline and monitoring the extent of its meadows would be a useful approach to estimate the impacts of natural and anthropogenic stressors. Here, we aimed to map the Z. noltei meadows in the Merja Zerga coastal lagoon (Atlantic coast of Morocco) using remote sensing. We used a random forest algorithm combined with field data to classify a SPOT 7 satellite image. Despite the difficulties related to the non-synchronization of the satellite images with the high tide coefficient, our results revealed, with an accuracy of 95%, that dwarf eelgrass beds can be discriminated successfully from other habitats in the lagoon. The estimated area was 160.76 ha when considering mixed beds ( Z. noltei -associated macroalgae). The use of SPOT 7 satellite images seems to be satisfactory for long-term monitoring of Z. noltei meadows in the Merja Zerga lagoon and for biomass estimation using an NDVI–biomass quantitative relationship. Nevertheless, using this method of biomass estimation for dwarf eelgrass meadows could be unsuccessful when it comes to areas where the NDVI is saturated due to the stacking of many layers.


Introduction
Seagrass meadows exist in most shallow, sheltered, soft-bottomed marine coastlines and estuaries throughout the world and rank among the most productive systems in the ocean [1][2][3]. They fulfill important trophic and structural functions in the coastal ecosystems [4][5][6] and rank among the priority habitats classified by the Convention on Biological Diversity (1992). They have been listed as one of the five biological quality elements to be included in ecological quality assessments in marine waters by the European Water Framework Directive (2000/60/EC) [7]. Therefore, due to the widespread decline of seagrass beds on a global scale [1], monitoring these ecosystems is a fundamental step for any program to manage these key habitats. Many techniques are used for monitoring seagrass beds, however the most used indices worldwide relate to their extent, fragmentation, bathymetric limits (colonization of the lower depth), abundance, recovery rate, biomass, density, canopy height and leaf biometry, specific composition and coverage, associated macroalgae, and associated benthic invertebrates [8].
Monitoring of the extent of seagrass meadows is a useful approach to estimate the impacts of natural and anthropogenic stressors. Indeed, this index is the simplest way to study these habitats at both small and large scales [9]; it has a minimal impact on the environment and can prove signs of degradation and suggest the presence of anthropogenic disturbances such as eutrophication, changes in the use of space, and increasing aquatic activities. In addition, the presence or absence of seagrass beds or their regional distribution defines the limits for the sampling of the cover, biomass, or density along depth gradients, which are supplementary indicators of health of seagrass beds [8].
The dwarf eelgrass Z. noltei Hornemann, 1832, is one of the predominant global species in intertidal zones, thus representing the land-sea interface [10]. The distribution of Z. noltei extends from the eastern Atlantic shores starting from Mauritania to southern Norway and the Kattegat Sea, and throughout the Mediterranean, Black, Azov, Caspian, and Aral Seas and the Canary Islands [11][12][13]. Z. noltei beds represent a habitat which allows the spatial structuring of many biological communities and are considered the basis of exceptional trophic compartments for many primary consumers [14]. Moreover, Z. noltei beds can contribute quantitatively to the function of tidal flats, as they are an extended juvenile habitat for some of the extremely important species in the food web [15]. Indeed, Bououarour et al. [16] showed that macrofaunal biomass has a significant positive relationship with the presence of dwarf eelgrass. They demonstrated that variations in the macrobenthic faunal structure are mainly due to the existence of Z. noltei. In semi-enclosed coastal areas, this species is particularly vulnerable to climate-changederived effects and to anthropogenic pressures [17]. Despite the fact that Z. noltei is listed as a species of minor concern by the International Union for the Conservation of Nature (IUCN), the overall population is reported to be declining. Z. noltei forms meadows mainly within the intertidal zone, which make it likely to be vulnerable, in particular to climate-change-derived effects, mainly the sea water temperature and increasing sea level rise (SLR), which will affect the state, composition, and evolution of occupied areas, ranging from homogeneous sectors to fragmented patches scattered on the foreshores [14,[18][19][20].
Studies over time have shown that remote sensing applications for seagrass ecosystems represent good tools for monitoring. This can be achieved at a primary level (detection of seagrasses, spatial coverage, species-level discrimination, biomass detection, growth patterns, and degradation) or at a secondary level (environmental variables influencing seagrasses, such as sea surface temperature (SST), salinity, sea level rise, pollution, detection of epiphytes, etc.) [21][22][23][24][25]. The combination of remote sensing techniques with field surveys is essential when monitoring seagrass statuses and changes over long time scales in large regions [26][27][28]. Z. noltei meadows are the most dominant seagrass areas in semi-enclosed coastal systems (SECS) of the Atlantic coast of Morocco [29,30]. Aspects related to mapping and long-term monitoring of dwarf eelgrass are very incomplete or even absent.
The present study aimed to map Z. noltei meadows in the Merja Zerga coastal lagoon using a SPOT 7 image provided by the Royal Center for Remote Sensing, along with field data. This is a first step towards future monitoring and an important tool for the management of Z. noltei meadows along the Atlantic coast of Morocco.

Study Area
The Merja Zerga lagoon, also called Moulay Bousselham lagoon, is a RAMSAR wetland. It is the most northern coastal lagoon along the Atlantic Moroccan coast and is located approximately 120 km north of Rabat, between 34°47′ and 34°53′ north and 6°13′ and 6°19′ west. This lagoon receives daily influence from the Atlantic tides, which are considered the main source of its volume water. It also receives fresh water from two main inflows-the Nador Canal, which is of artificial origin and includes drainage water, as well as the Drader River. Moreover, the lagoon undergoes significant anthropogenic changes [31]. Indeed, the population bordering this lagoon has increased by 150% in the last 25 years [32]. This lagoon's waters are characterized by temperatures ranging from 27 to 28 °C in summer and from 13 to 15 °C in winter [33]. The depth of the lagoon fluctuates between 0 and 2 m but does not exceed 50 cm on average according to the tidal cycle and the precipitation [34,35]. The tide has a mean tidal range of nearly 1.4 m, which increases to 3.60 m during the spring tide, while during the neap tide it decreases to about 0.80 m [36].
The Merja Zerga lagoon provides important socioeconomic support to the local communities through many activities such as tourism and fisheries. Furthermore, it represents a significant ecological resource that has attracted the interest of many scientific researchers. This lagoon has been part of the CASSARINA and MELMARINA projects during the last two decades [37]; thus, several studies have been based on observations and fieldwork in the Merja Zerga, mainly treating hydrological characteristics, environmental changes, contaminants in sediments, plankton, and fish fauna [34,35,[38][39][40][41][42].

Field Data
Seagrass samples were taken from seven dwarf eelgrass stations on 04-05 November 2019, and on 11 November 2019 they were chosen randomly depending on the tidal conditions and scattered on all of the discernible aquatic vegetation areas (pass, east, west and center of the Merja Zerga lagoon) from field investigations and satellite image observations (Table 1, Figure 1).  In order to define the spectral separability of the species lagoon and the mixtures, spectral measurements were performed in situ using a JAZA0478 spectroradiometer with a 348.95-1039.85 nm wavelength range, provided by the LETG laboratory of the University of Nantes. In each station, the reflectance of different types of habitats, in particular the dwarf eelgrass Z. noltei, macroalgae (chlorophyte and Rhodophyta), sand, and mud sediments, were recorded at 1 m above ground. For each type of habitat, an average of three to four spectral measurement were sampled in each station and their corresponding references were also taken. The white and dark references were calculated by measuring the reflectance of the white reference (Spectralon ® WS-1-SL) and dark reference (performed by covering the collimating lens with its cap).
In order to estimate the Z. noltei biomass in g/m² and its cover percentage, seven stations were sampled on 16 September 2018 in the intertidal area located in the downstream, upstream, and center of the lagoon during low tide. Biomass was assessed using hand cores (30 cm in length and 12 cm in diameter), with ten replicates for each station. The samples were randomly collected from Z. noltei meadows, were washed in situ and taken to the laboratory, then frozen at −20 °C before biomass measurements. The samples before being weighed were oven-dried at 60 °C for two days. As for estimating the cover percentage of the Z. noltei, in each station, five replicates of quadrats (0.5 × 0.5 m²) were photographed and their geographic coordinates were saved for mapping validation. The quadrats photos were processed by ImageJ software according to Stationary Bed Monitoring Protocols for the Water Directive (WFD) Zostera marina (L.)-Z. noltei Version 3 [43]. When the use of quadrats was not possible, the cover percentage was estimated visually by assigning it to one of the following classes: 0-25%, 25-50%, 50-75%, 75-90%, 100%.

Satellite Images
To map the distribution of dwarf eelgrass Z. noltei in the Merja Zerga lagoon, we tried to obtain very good images with high resolution, at low tide and with a good tidal coefficient. This was achieved by making a list table with the dates of low tides, coefficients, and the number of images that contained little cloud. Unfortunately, the results were very limited; we must say that this added to the difficulty of achieving very good images. Consequently, we used one SPOT 7 satellite image acquired on August 19, 2019 at 10:54 a.m local time, matching with the falling tide. Indeed, August 19, 2019 corresponds to the lowest tide, which was 0.7m at 11:46 a.m UTC+1, with a tidal coefficient equal to 72. Images were obtained during the summer, which is the period of increased density of eelgrass beds, culminating in August-September [44]. In fact, among many Sentinel 2 and Pleiades images, the SPOT 7 image presented the best conditions required for mapping dwarf eelgrass in the Merja Zerga lagoon.

Typology
In our approach, we mapped the main habitats that form the Merja Zerga lagoon with reference to the floristic typology outlined by Hammada [46] for this lagoon. This typology showed that there are two gradients that influence the vegetation distribution, namely submersion rhythms and salinity. In fact, the submersion gradient allowed the three types of species to be distinguished, which are: We aimed to discern between the aquatic species of the Merja Zerga lagoon marine environment. Hence, the classes that were used for the classification were shallow water, sand, mud, terrestrial plants, algae, mixture of algae and dwarf eelgrass, submerged dwarf eelgrass, and emerged dwarf eelgrass.

Image Processing
The SPOT 7 image used for this study was delivered in bundle (PAN + XS) format. Atmospheric correction for the four bands was achieved by using the GRASS algorithm in order to convert them to reflectance. Therefore, the spectral and spatial resolutions were suitable for use with the 9 vegetation indices (Table 1). Indeed, mapping of dwarf eelgrass Z. noltei in the Merja Zerga lagoon was achieved by using a supervised classification approach applied with a random forest algorithm (Figure 2), which is a model that constructs and combines a large number of base decision trees.
Before applying the random forest model, 117 polygons were created manually using ArcGIS 10.4.1 software based on personal field knowledge and were distributed over the eight classes, which be used for the classification: shallow water, sand, mud, terrestrial plants, algae, mixture of algae and dwarf eelgrass, submerged dwarf eelgrass, and emerged dwarf eelgrass. For this, we used Breiman and Cutler's random forests for classification and regression algorithm with the R package randomforest v.4.6-14. Known as a robust and consistent model, the random forest algorithm is the most popular ensemble-based model for classification problems, and in particular for seagrass mapping [47]. We empirically tuned the random forests model by testing two parameters (because it seems that there is no theory to guide the choice of the mtry number (number of predictors sampled for spliting at each node) or the number of trees [48]. First, the mtry number was fixed at 3 (with the use of the tuneRF function of the randomForest package). Then, with this mtry number, and in order to find the optimal number of trees that correspond to a stable classifier without excessive computation time, we computed the random forests algorithm with different ntree values and recorded the OOB error rate to assess the number of trees at which the out of bag error rate stabilized and reached the minimum (in our case this was fixed at 500).
First of all, we returned the four raw bands of the SPOT 7 image and used them to calculate nine vegetation indices ( Table 2). The resulting layers were then stacked onto a 13-band image, which was then implemented in the random forests model using the R software. The random forests algorithm was validated by the overall out-of-bag (OOB) model error and the confusion matrix (percentage of falsely and classified training pixels) and seemed to be robust (OOB = 3.14). Then, we used the predict function to compute the RF classification prediction with the use of the clusterR interface to speed up the computation (in the R package "raster"). In addition, a final estimation of the quality of the "accuracy" classification error was made using Erdas software by creating 120 equalized random points spread over the eight classes.
To estimate Z. noltei aboveground biomass in the Merja Zerga part of the Moulay Bousselham lagoon using remote sensing, we chose to use the quantitative experimental NDVI -Z. noltei biomass relationship, whereby biomass = 610.61 (NDVI)^1.88 (n = 31, r2 = 0.97), as created by Barillé et al. (2010). Depending on the presence or absence of algae and the emersion or submersion condition, the precedent equation was separated or combined for the estimated percentage cover (Figure 2).
Hence, a specific biomass estimation map was generated for each of the three classes of Z. noltei: two maps were created by directly applying the equation for both the submerged Z. noltei class and the emerged Z. noltei class, and one map was created by combining the equation with the percentage cover of Z. noltei for the mixed class of algae and Z. noltei.

Spectral Results
According to the spectra, separability between classes was possible. Moreover, the equivalent image calibrated to reflectance also made discrimination between classes possible. The reflectance spectra of algae (n = 9) and Z. noltei (n = 13) sampled in the Merja Zerga lagoon presented different overall shapes in the visible and NIR ranges between 400 and 900 nm. Indeed, algae spectra were plotted above the dwarf eelgrass spectra. The distance between algae spectra and emerged dwarf eelgrass spectra was estimated using the spectral angle, which had a value of 0.936 radians. The reflectance spectra of Z. noltei were discernible from chlorophyta in the visible wavelengths. However, they had the same tendency (increasing) in the NIR, except for the submerged dwarf eelgrass spectra (decreasing) (Figure 3). The overall shapes of the spectra from the SPOT 7 image we used were similar to those sampled with the spectroradiometer in the field.

Biomass and Cover Percentage of Z. noltei
Our results showed that the stations S6, S4, and S7 had the highest biomass among the five stations sampled in the Merja Zerga lagoon (Table 3). According to the visual interpretation and to the quadrat analysis, station S7 had the lowest cover percentage of Z. noltei, which was limited to 50% (Table 4). The stations S5 and S4 came in the second position, which were mixed beds (Z. noltei and algae) with cover percentages of Z. noltei limited to 60% and 75%, respectively. On the contrary, the rest of the stations of the lagoon were homogeneous beds with the highest cover percentages of Z. noltei, which reached 96%. In fact, despite their small size, the dwarf eelgrass beds in stations "S3", "S2", and "S1" were homogeneous and denser, plus the presence of macroalgae species there was not very prevalent (Figure 4).

Satellite Image Processing
Our random forests model accurately predicted the distribution of all eight classes for the Merja Zerga lagoon. The model's internal class errors fluctuated between 0.00 and 0.25% and the OOB for the model was 3.14%. It turned out that the raw bands and indices that were most relevant for distinguishing between the eight classes according to the mean decrease Gini index were the NDVI and GNDVI indices, followed by the GVI, CVI, and EVI, then followed by the green and blue bands successively ( Figure 5). Indeed, NDVI and GNDVI indices had the highest mean decrease Gini values, which reached 141.03 and 138.69, respectively. The average curve of the pixels per class extracted from the SPOT 7 image showed that each class had a specific band or index which best discriminated it. For example, the IB index was the most relevant for distinguishing submerged dwarf eelgrass ( Figure 6).

Image Classification
We managed to estimate the surface areas of Z. noltei meadows in the Merja Zerga lagoon, which was approximately equal to 160.76 ha (mixed beds included), of which 65.22 ha was the total area for emerged and submerged Z. noltei.
The classification results indicated that Z. noltei beds are more extensive in the center and in the upstream area of the lagoon than in it downstream area (Figure 7), in particular at station S6 (Table 5), which was in agreement with the field observations. Thus, the total surface areas for Z. noltei beds were 27.31 ha, 95.54 ha, and 37.91 ha for submerged, mixed, and emerged beds, respectively.  26 Dwarf eelgrass biomass results were estimated using remote sensing ranges between 0 and 277.81 g DW/m² for the class "emerged Z. noltei" and between 0 and 156.509 g DW/m² for the class "mixed beds", of which 62% (which is the average of the three mixed station: "S5":0.77, "S4":0.60, "S7":0.50) corresponded to Z. noltei, i.e., was between 0 and 97.03 g DW/m². Lastly, biomass values were between 0 and 160.79 g DW/m² for the class "submerged Z. noltei" (Figure 8).

Discussion
The overall shapes of the spectral signature of the algae and the dwarf eelgrass found in the field look like those obtained by Bargain et al. [54] and by Dehouck et al. [57]. In fact, Z. noltei reflectance is discernible from green algae in the visible wavelengths and even more in the NIR wavelength. In addition, emerged dwarf eelgrass and algae spectra increase in the NIR wavelength compared to the submerged dwarf eelgrass spectra, which have a decreasing slope due to the residual water layers in them. This is due to the influence of water. Dehouck et al. [57] indicated that in general the spectra of Z. noltei and algae are similar, however small differences are noticeable in narrow bands. They claimed that the spectra need to be analyzed and separated numerically to find out the best spectral bands.
We chose to use vegetation indices to map Z. noltei in the Merja Zerga lagoon because it exists in the form of monospecific seagrass beds. Indeed, the average curve for pixels per class extracted from the SPOT 7 image showed that each class has a specific band or index which best discriminated it. For example, the IB index is the most relevant for distinguishing submerged dwarf eelgrass ( Figure 6). Our OOB and mean decrease Gini results showed that NDVI, GNDVI, GVI, and CVI are the best indices to use for this purpose ( Figure 5). The mapping of seagrass meadows using multispectral and hyperspectral images has often been achieving by using the NDVI, which is the most widely used index to assess green biomass [58]. Indeed, Barillé et al. [59] calibrated SPOT 7 images to NDVI and tested the influence of Z. noltei aboveground biomass variations on spectral reflectance. Hence, they used field-measured hyperspectral data from seagrass and algae and then applied a quantitative experimental NDVI-biomass relationship to map the biomass distribution of Z. noltei. The latter displayed a characteristic steep slope from 700 to 900 nm, which increased with increasing biomass. Later, Bargain et al. [60] demonstrated that vegetation indices used to map these kinds of beds and their biomass with satellite or airborne imagery must be selected carefully to moderate all sources of variability affecting the index value, such as the pigment variations, or those related to background influences, such as the soil spectral contrast, albedo variations, or the specular reflectance at the leaf surfaces.
Our study is the first study of the mapping of Z. noltei beds using a random forest algorithm. This method showed satisfying results because it has been proven effective to estimate the surface distributions of this species using SPOT 7 satellite imagery. Our mapping results showed that Z. noltei beds in the Merja Zerga lagoon are more extensive than in the older studies [61,62]. Indeed, their surface area reached 65.22 ha, and when including associated seaweed mats, this equaled 160.76 ha (Table 6). Table 6. Surface area for Z. noltei at each station in the Merja Zerga lagoon: S1, S2, S3, S4, S5, S6, S7.

Z.noltei Beds Surfaces
(ha) S6 S7 S5 S4 S3 S1 and S2 Usually, mapping Z. noltei meadows by using a handheld GPS instrument or by using transects is not enough when some areas are inaccessible due to the nature of the substrate, to the extent of the meadows, and to the tidal cycles. Generally, estimating seagrass distributions using traditional methods with pre-recorded grid patterns or a combination of transects and sampling points [63] is time-consuming and has high associated costs [10]. This can, therefore, be achieved thanks to remote sensing, either by interpretating aerial photographs from an unmanned aerial vehicle (UAV) combined with ground truthing measurements [4][5][6][7] or by terrestrial, oblique, large-scale photography and satellite image classification [64][65][66][67].
Some studies also proved the efficiency of random forest approaches in mapping the extent of seagrass. Indeed, according to Traganos and Reinartz [68], the random forest machine learning classifiers and support vector machines are gaining more interest for coastal habitat remote sensing. Indeed, they assessed the extent of change for Posidonia oceanica (L.) Delile and Cymodocea nodosa (Ucria) Ascherson following atmospheric and analytical water column correction and using a random forest algorithm for the RapidEye time series. Furthermore, Zhang [69] and Traganos and Reinartz [70] proved that random forest algorithms generated promising results regarding classification of seagrasses, while Zhang et al. [71] exhibited the advantages of al. [57] emphasizes that by carefully selecting multispectral bands or by combining them random forest algorithms over the maximum likelihood classifier using hyperspectral imagery.
In contrast to our study, several previous studies have mentioned the presence of Z. noltei in Merja Zerga lagoon, but without giving its overall distribution. First, Dakki et al. [61] estimated that 4.78% of the lagoon's surface was occupied by dwarf eelgrass meadows, the surface area of which corresponded to 140 ha, including algae species. The presence of Z. noltei in the Merja Zerga lagoon was also reported by Boutahar et al. [30], Gam et al. [72], Grignon-Dubois and Rezzonico [73], Natij et al. [74], Touhami et al. [75], and Touhami et al. [76]. The dwarf eelgrass meadows in this lagoon have developed in muddy and sometimes deep substrates, which makes walking through them very difficult, so mapping these meadows on foot is not feasible. Lately, according to Zarranz Elso et al. [62], dwarf eelgrass has been mapped in the Merja Zerga lagoon through spectral characterization and a maximum likelihood classification of a World View 2 satellite image acquired on November 3, 2010, combined with field samples. Hence, the surface area of Z. noltei meadows in Merja Zerga lagoon was estimated to be 71.508 m², however in this case the satellite image used did not cover all of the intertidal areas of the lagoon. However, this study missed the bigger beds located in the downstream area of the lagoon. When interpreting multispectral optical images, confusion persisted between dwarf eelgrass meadows, algae, and the cordgrass Spartina sp., which is an emergent halophilous species that boarders the lagoon with other saltmarsh species, especially Salicornia sp. (L.) and Juncaceae rushes [62,75]. Dehouck et using a multitemporal database they could limit this confusion, and that algae seem to be discriminable only when using fine spectral bands. According to Proença et al. [77], the mapping of Spartina anglica C.E. Hubb. and Spartina maritima (Curtis) Fernald using supervised classification of Pleiades images showed that these species were often confused with the vast areas vegetated by Z. noltei, in particular in October. In order to achieve better results and to avoid this confusion, it is better to mask the majority of the potential terrestrial vegetation from the satellite image before running the classification algorithm. In our case, this problem did not occur because we created a sufficient number of polygons (117) distributed across the eight classes used for the classification. In fact, based on the accuracy assessment results, the random forest classification generated a satisfactory Z. noltei meadow map, with a kappa index equal to 0.9429 and an overall accuracy equal to 95%.
Our classification results allowed the estimation of the overall extent of Z. noltei in the Merja Zerga lagoon and also allowed the estimation of the aboveground biomass of this species. As with the study by Barillé et al. [60], in our case the NDVI-biomass quantitative relationship was applied to all Z. noltei meadows successfully, because despite the stacking of many layers in certain areas, the NDVI was not saturated. Thus, this method was applied directly for the classes "emerged Z. noltei" and "submerged Z. noltei " and was combined with the cover percentage for the class "mixed Z. noltei and algae". Indeed, we observed during field investigations that Z. noltei was not the main living macrophyte in the mapped intertidal area-filamentous green macroalgae was observed in the mixed beds and a few macroalgae species (Cholorophyta and Rhodophyta) were occasionally present (Figure 9). Indeed, Valle et al. [78] claimed that the NDVI is useful for distinguishing between sparsely vegetated and dense Z. noltei beds, and therefore can be further applied to CASI (hyperspectral sensor) imagery to map the cover percentages of the species living in the area.

Conclusions
In our study, the suitability of using remote sensing to firstly map the distribution of Z. noltei and secondly to estimate its biomass was confirmed. In fact, we classified a SPOT 7 satellite image using a random forest algorithm and field data to produce dwarf eelgrass habitat maps. This approach enabled us to define the distribution of Z. noltei beds and associated macroalgae in terms of surface coverage. Our results proved with an accuracy of 95% that GNDVI, NDVI, and EVI are potentially useful vegetation indices for discriminating dwarf eelgrass beds from other habitats in the lagoon. Older maps involving the spatial characterization of Z. noltei in the Merja Zerga lagoon could not determine the entire extent of this seagrass. Compared to our work, previous studies on Z. noltei mapping underestimated its extent due to the fact that they used images that did not fit all of the Merja Zerga lagoon or were taken at a higher tide. Indeed, the overall surface occupied by dwarf eelgrass in this wetland is 160.76 ha when considering mixed beds. Our approach was based on estimating the extent, which is a global index that can reveal the increase or loss of the area of seagrass beds at its limits; it does not consider the losses inside these limits, which are visible at the level of the fragmentation of these beds. Biomass estimation of dwarf eelgrass using the NDVI-biomass quantitative model applied to SPOT 7 images can be very useful as a non-destructive method for estimating the biomass of this seagrass species. In future studies of Z. noltei and in order to purchase the data obtained, the same methodology should be considered.