3.2.1. MLC Results

The results from the class separability test of image-derived spectra showed that all of the class pairs had values greater than 1.90, indicating globally a good class separation (Figure 6) [72].

**Figure 6.** Average reflectance of the different spectral classes between 450 and 900 nm: (**a**) macroalgal groups, water and substratum mean reflectance; (**b**) detailed spectra of each Fucales.

The MLC classifier, trained using image-derived spectra, revealed a dense cover of intertidal Fucales (26.9% of the site) (Figure 7). The overall classification accuracy for the MLC was 95.1% and the kappa coefficient was 0.93. The four bathymetric/vegetation levels appeared clearly, forming four distinctive bands. The Pc-Fspi level (upper shore) was dominated by a thin band of both *P. canaliculata* and *F. spiralis* (1.3% and 0.1% of total pixels, respectively). The An and Fser levels (mid-shore) were dominated by a large band of *A. nodosum* and of *F. serratus* (5.6% and 6.8% of total pixels, respectively). The He-Ld level (lower shore) was characterized by the important development of *H. elongata* (9.7% of total pixels) and a cover of red macroalgae greater than in higher levels (1.7% for all of the site). Green algae were mainly present in the lower shore (1.7% of total pixels). The 'Substratum' and 'Water' classes represented the majority of the site (51.8% and 21.3% of total pixels for the site, respectively). The macroalgal classes '*A. nodosum*', '*F. serratus*' and '*H. elongata*' showed the highest producer/user accuracies (Table 4). There were some misclassifications between the Fucales '*A. nodosum*' and '*F. serratus*' (1.56%), and between '*A. nodosum*' and '*P. canaliculata*' (7.04%). The lowest producer/user accuracy was for '*F. spiralis,*' with some misclassifications between '*F. spiralis*' and '*P. canaliculata*' (44.13%) and between '*F. spiralis*' and '*A. nodosum*' (14.81%). 'Green' and 'Red' algae were also well classified (96.92% and 90.54%, respectively) but there were some misclassifications between 'Red' algae and the Fucales '*P. canaliculata*' (3.04%) and '*H. elongata*' (2.06%).

**Figure 7.** Maximum likelihood classification (MLC), trained using image-derived spectra resulting from the hyperspectral UAV survey at Porspoder. Seven macroalgal cover classes are displayed over the UAV RGB imagery and an orthophotography (Mégalis Bretagne et collectivités territoriales bretonnes—2015), giving an overview of the site. The 'Substratum' and 'Water' classes are not represented on the map. Class codes 'Green' and 'Red' represent grouped green and red macroalgal species, respectively.


**Table 4.** Maximum likelihood classification (MLC) confusion matrix, calculated, using ENVI 5.6.1, by comparing pixels of known class locations to those predicted by the classification workflow for each of the nine cover classes. Results are displayed as percentages of pixels assigned, correctly or incorrectly, to each class. User/producer accuracies (User Acc. and Prod. Acc., respectively) are

#### 3.2.2. SAM Results

The SAM classifier, trained using image-derived spectra, revealed a similar cover of intertidal Fucales as MLC (27.6% of the site) (Figure 8). The overall classification accuracy for the SAM was 87.9% and the kappa coefficient was 0.82. By contrast with MLC, the four bathymetric levels appeared less distinct. The Pc-Fspi level was dominated by a thin band of both *P. canaliculata* and *F. spiralis* (1.4% and 1.3% of total pixels, respectively) with a better cover of *F. spiralis* than for MLC. The An and Fser levels (mid-shore) were dominated by a large band of *A. nodosum* and of *F. serratus* (5.5% and 3.5% of total pixels, respectively), but the cover of *F. serratus* was less important than for MLC. The He-Ld level (lower shore) was characterized by the important development of *H. elongata* (11.8% of total pixels). The cover of red macroalgae was distributed on all of the site (2.2% of total pixels) and was more present in the Fser level compared to the MLC results. Green algae were mainly present in the lower shore (1.9% of total pixels). The 'Substratum' and 'Water' classes represented the majority of the site (54.8% and 17.5% of the site, respectively). The macroalgal classes '*H. elongata*' and 'Green' showed the highest producer/user accuracies (Table 5). As for MLC, there was some misclassification. First, 18% of '*P. canaliculata*' pixels had been classified as '*F. spiralis*' (9.44%) and '*H. elongata*' (9.49%). The lowest producer/user accuracy was for '*F. spiralis*', with the largest misclassification (37.54%) in '*A. nodosum'* and 14.37% of pixels in '*P. canaliculata*'. Of the '*A. nodosum'* pixels, 18.76% were misclassified as '*F. spiralis',* but also 7.35% and 4.62% of '*A. nodosum*' pixels were misclassified as '*F. serratus*' and '*P. canaliculata*', respectively. Of the '*F. serratus*' pixels, 18.86% were misclassified as 'Red', and 17.37% of pixels should have been classified as '*F. serratus*', when they were in fact classified as '*A. nodosum*'. 'Green' and 'Red' algae were globally well classified, but there was some misclassification between 'Red' algae and some Fucales, such as '*A. nodosum*'. 'Substratum' and 'Water' classes had the highest producer/user accuracy and so were well classified on the entire image.

**Figure 8.** Spectral angle mapper (SAM), trained using image-derived spectra resulting from the hyperspectral UAV survey at Porspoder. Seven macroalgal cover classes are displayed over the UAV RGB imagery and an orthophotography (Mégalis Bretagne et collectivités territoriales bretonnes— 2015), giving an overview of the site. The 'Substratum' and 'Water' classes are not represented on the map. Class codes 'Green' and 'Red' represent grouped green and red macroalgal species, respectively.

**Table 5.** Spectral angle mapper (SAM) confusion matrix, calculated, using ENVI 5.6.1, by comparing pixels of known class locations to those predicted by the classification workflow, for each of the nine cover classes. Results are displayed as percentages of pixels assigned, correctly or incorrectly, to each class. User/producer accuracies (User Acc. and Prod. Acc., respectively) are also presented.


#### *3.3. Comparison of Field Sampling and Hyperspectral Classification*

Covers determined by field sampling are compared here to the classification results by both MLC and SAM. A visual representation of the comparison between in situ sampling, an infrared picture and the two methods is provided in Figure 9 and Appendix A Figures A1–A3.

**Figure 9.** (**a**) Picture of a sampling spot on the Pc-Fspi level at Porspoder taken during field sampling in June 2021. (**b**) NIR-G-B image of the same sampling spot. (**c**) Result of the MLC classification. (**d**) Result of the SAM classification. The red square corresponds to the mobile grid structure used for field sampling. Color code in (**c**,**d**) corresponds to the following classes: '*P. canaliculata*' (brown), '*F. spiralis*' (yellow), '*A. nodosum*' (purple), '*F. serratus*' (coral), '*H. elongata*' (orange), 'Red' algae (red), 'Green' algae (green), 'Substratum' (grey) and 'Water' (blue).

To compare covers estimated in situ to those obtained through hyperspectral classification, in situ erect and crustose red algae on one side, and *H. elongata* and *L. digitata* on the other side, have been mixed up in order to attain classes similar to remote classification (Figures 10 and 11).

**Figure 10.** Db-RDA (scaling 1) performed from the macroalgal and pixels covers data for the in situ and the two classifications algorithms. Sampling spots are indicated by circles (Pc-Fspi = red circles, An = yellow circles, Fser = black circles and He-Ld = blue circles), variables (classes used in situ and for classifications: *Pelvetia canaliculata*, *Fucus spiralis*, *Ascophyllum nodosum*, *Fucus serratus*, *Himanthalia elongata*, Red, Green and Substratum) appear in purple, and the explanatory variables (level: Pc-Fspi, An, Fser and He-Ld; method: in situ, MLC and SAM) in red.

The results of the db-RDA are shown in Figure 10. Only the distance between objects (replicates) is considered; therefore, Scaling 1 has been chosen. The model is significant, with an R2 of 91% (F-value of 117.89, *p*-value < 0.001). Axes 1 and 2 explain a significant portion of the total variability, with 42% and 23%, respectively. The points, corresponding to the replicates, are grouped into four homogeneous clusters, in agreement to the four bathymetric levels, with no significant distinction between the three methods. The explanatory variable 'level' is obviously correlated to the four clusters, with 'Pc-Fspi', 'An', 'Fser' and 'He-Ld' associated with the clusters corresponding to these four levels while the variable 'method' appears to have no significant influence on the distribution of replicates within each level. The results obtained from the three methods thus appear similar.

When looking in detail at the classes, the methods appeared broadly similar (Figure 11).

In Pc-Fspi (Figure 11a), the cover of the dominating Fucales '*P. canaliculata*' estimated by the in situ sampling was 32.5%, showing no significant difference to those obtained from MLC (39.9%) and SAM (20.9%) (Kruskal–Wallis, *p* > 0.05). On the contrary, for the other dominating Fucales '*F. spiralis*', the test showed no significant difference between in situ sampling (5.0%) and both methods, but a significant difference was observed between the two models (0.6% for MLC and 5.5% for SAM, Kruskal–Wallis, *p* < 0.04). In contrast, the cover of 'Red' macroalgae was 5.0% in situ, showing a single significant difference with the method MLC (0.2%, Kruskal–Wallis, *p* < 0.001). The estimation performed by SAM (1.4%) did not show any significant difference with the other two methods. Bare rock represented about half of the surface of sampling spots in this level, whatever the method considered, showing no significant differences (Kruskal–Wallis, *p* > 0.05).

**Figure 11.** (**a**) Comparison of in situ (violet boxes), MLC (green boxes) and SAM (yellow boxes) determined covers of main Fucales species, macroalgal classes and bare rock found in Pc-Fspi (**a**), An (**b**), Fser (**c**) and He-Ld (**d**) levels. Covers are given in percentages. In the boxplots, only the classes showing covers for all three methods are represented. Letters refer to statistical differences (Kruskal–Wallis).

In An (Figure 11b), the cover of '*A. nodosum*' for SAM and MLC was higher (74.7% and 84.1%, respectively) than the in situ cover (60%). However, there was no significant difference in its covers between in situ and SAM, but a significant difference appeared between in situ and MLC (Kruskal–Wallis, *p* < 0.05). The '*F. serratus*' cover showed no significant difference between the three methods, i.e., circa 5% (Kruskal–Wallis, *p* > 0.05). The percentage of 'Substratum' was also close and did not show significant differences, with 6% in situ and ca. 3% for MLC/SAM (Kruskal–Wallis, *p* > 0.05). However, covers of 'Red' were significantly higher in the in situ field sampling (25%) than in the MLC method (Kruskal–Wallis, *p* < 0.005). The SAM method was not different for either of the two.

In the Fser level (Figure 11c), the cover of '*F. serratus*' was significantly greater in situ*,* with 94.2%, than in SAM with an estimated cover of 40.5% (Kruskal–Wallis, *p* < 0.05). MLC method did not show significant differences with either method (87.6%).

In He-Ld (Figure 11d), the '*H. elongata*' covers were 56.7% in situ, 61.3% with MLC and 65.5% with SAM and were statistically the same (Kruskal–Wallis, *p* > 0.05). 'Red' covers showed nearly similar values (39.2% and 34.9%, respectively). Even though SAM 'Red' covers were lower (16.8%), no significant difference was observed between the three methods (Kruskal–Wallis, *p* > 0.05). 'Green' covers were also not significantly different between in situ (4.2%), MLC (2.3%) and SAM (1.3%) (Kruskal–Wallis, *p* > 0.05).

To summarize these results, no generalization can be applied, both methods showed contrasted results according to the level and/or class considered. For instance, the cover of 'Red' is better estimated by the SAM in Pc-Fspi, and, on the contrary, better estimated by the MLC in An. The same observation can be made for the dominating Fucales, which are better represented by the SAM method in An, and by the MLC in Fser.

#### **4. Discussion**

Seaweeds are ecosystem engineers and key habitat-formers in temperate marine coastal ecosystems [73,74]. The use of satellite sensors to monitor populations of large temperate macroalgal species is well documented [54,75,76]. Such techniques have been used mainly to characterize the extension of macroalgal communities and related habitats [37]. However, the accurate classification of closely related macroalgal species (e.g., different species within the same Genus) and groups of species in remote sensing analysis remain a key point that still needs to be investigated. Using a high spatial and spectral resolution technology could be part of the solution [44].

In this study, heterogeneous seaweed habitats vertically distributed were successfully differentiated using both field and remote techniques. The 'undisturbed' sampling method [59] was used in situ to describe the structure of macroalgal communities and was then directly compared with the remote sensing imagery.

#### *4.1. Habitats Characterization through Remote Sensing and Field Sampling*

Orthophotos are often sufficient to remotely describe a habitat dominated by a single species, forming homogeneous populations, such as mussels [77,78] or polychaete reefs [79] that can be identified on a large scale. Eventually, species groups can be discriminated (brown, red and green seaweeds) [43], but this differentiation is quickly overtaken when studying complex ecosystems, showing an intricate microtopography, such as the European rocky shores. In the present study, using hyperspectral imagery, seaweed habitats were successfully differentiated (1) between them, (2) from substratum and (3) from seawater; the spectral signature allowed a clear differentiation between them (Figure 6a) as already reported in previous studies [13,42,75]. Moreover, the spectral signatures of five Fucales species were also differentiated, allowing an accurate mapping of the study site and its habitats. Here, two species of Fucus (*F. serratus* and *F. spiralis*), previously often gathered as a single class [35,44,58], were also discriminated. Indeed, the rocky shore surveyed during the present study showed a succession of several dominant fucoid species, with a conspicuous increase of red seaweeds abundance in the low intertidal zone, as already reported from previous studies in the area, using only field sampling [80,81]. However, technical limitations were spotted: limpets and barnacles were impossible to discriminate (although large limpets can be guessed) because of their heterogeneous distribution on rocky shore and close spectral signature with substratum. The same issue was observed for distinguishing red and green seaweed species, which were heterogeneously distributed and can present similar spectra, or variations of spectra according to their health conditions (e.g., pigment degradation, grazing, occurrence of epi/endophytes) [42,82,83].

In this study, we assigned several subclasses in the 'Red' one because of the dominance in He-Ld by an assemblage of *Mastocarpus stellatus/Chondrus crispus*. That assemblage masked the occurrence of several filamentous or turf red algae which were not identified on hyperspectral images, but with the use of existing spectral libraries [82], it may be an option [44]. In the same way, crustose and erect red macroalgae were also assigned in the 'Red' class for the same reasons, but with distinguishable patches, it could be possible to create two different classes [42]. For further analysis, it would be interesting to refine classes and to include more identified red species to the classification. The kelp *Laminaria digitata* was not added to the classification due to residual seawater on the images. However, it would be useful to separate that species as a spectral class due to its dominancy in the upper sublittoral zone [44,58]. It would also be interesting to transfer the spectral library created for other study sites to check if the method is interoperable.

#### *4.2. Comparison of the Two Classifiers*

Our results showed that remote classification data were in agreement with covers calculated from field sampling inside sampling spots, in spite of an approximation of 5% for in situ estimations, and whatever spectral properties of brown macroalgae, which are very similar [82,84] (Figure 6b).

MLC provided the best producer/user accuracies for dominating algae of three levels (An, Fser and He-Ld) (Table 4) and SAM for two levels (An and He-Ld) (Table 5). Pc-Fspi provided the lowest producer/user accuracy, with the lowest values for '*F. spiralis*' for the two classifications. This could be a site effect due to the lower extension of these two species compare to the others in Porspoder, especially *F. spiralis*, which results in a reduced pixels selection for the algorithms. Moreover, the fact that *F. spiralis* is found in more shaded environments/crevices on this site could affect the data since shadows cause serious difficulties for remote shooting [85]. There is also a clear misclassification between *P. canaliculata* and *F. spiralis,* especially for MLC, which might be explained because of forming confused communities with small size species, and similar colors, in this site.

In An, MLC was less accurate than SAM for *A. nodosum*. Due to the presence of the red alga *Vertebrata lanosa*, which was not taken into account for the classifications, the '*A. nodosum*' class was not considered as a 'pure class' and SAM had confusion overestimating 'Red' seaweed in An and Fser levels. There was also a little confusion in '*A. nodosum*' with *P. canaliculata,* as in Rossiter et al. (2020) [58]. This confusion appeared in the upper limit of *A*. *nodosum* distribution, where *A. nodosum* appeared brighter due to a stronger light stress [86], with a color close to that of *P. canaliculata*. Not surprisingly, *P. canaliculata* was also classified as *A. nodosum* when it was darker than usual.

The 'Red' class had a lower producer/user accuracy for the SAM classification, with an overestimation in Fser to the detriment of *F. serratus,* which is underestimated (Table 5). For that level, SAM is not a good descriptor at the site level.

On the processed images (Figures 7 and 8), many pixels were classified as 'Green', despite a reduced cover of green seaweeds in the field data. This is mainly due to the positioning of the spots on the shore, chosen because of a clear dominance by brown macroalgae.

Both SAM and MLC misclassified pixels of *H. elongata,* with some occurrence in Pc-Fspi, An and Fser, whereas this species does not develop in higher intertidal zones [7,87]. On the contrary, the high-level species *F. spiralis* could appear in An with both MLC and SAM. An approach taking into account the bathymetric range (strongly affecting certain species) could be determined using a lidar approach [88] and could solve such a problem, considering the vertical zonation of species [17].

The MLC classification was found to be more accurate than SAM at the site level due to better management of spatial heterogeneity of habitats by the MLC, and because SAM does not consider the magnitude of pixels' vectors. Moreover, groups of macroalgal species, Fucales in particular, have close spectral similarities, which could partially explain the lower accuracy of SAM [47,89]. Nevertheless, both supervised classifications mapped the four bathymetric levels sampled in this study and both classifiers were able to separate brown, red and green algae. At sampling spots scale, they provided similar results, so, they can be used to compare macroalgal covers with accuracy. To refine the results, it could be interesting to test an object-based classification that takes into account not only spectral information but also the shape, size, texture, tone and the compactness [90,91] of objects. In that prospect, macroalgae could be an interesting model due to various morphologies and textures.

#### *4.3. Consistency of Specific Identification and Perspectives*

Automated macroalgal classification applied to shores dominated by a single fucoid species is currently manageable [43]. However, discriminating and mapping *Fucus* spp. remain a challenge, as seen in previous studies, in which *Fucus* spp. have been gathered in a single mixed class [35,44,58]. Overall accuracy, used to estimate the quality of the

classification [62,92], presented values indicating a clear distinction between *F. spiralis* and *F. serratus*. Even though the final classification of the entire site of Porspoder pointed out some pixels which were not correctly attributed, the entire distribution of pixels on the site was consistent. The analysis by db-RDA did not show a significant influence by the method in the distribution of the sampling spots, unlike the level. This may be related to the nature of the site studied, with the Pc-Fspi level having much lower macroalgal covers than the An and Fser levels, and to the spectral properties of each species. Moreover, the comparison between in situ covers with MLC and SAM data showed no significant difference for most classes (Figure 11), therefore, validating the algorithms as good descriptors for intertidal macroalgal covers at sampling spots scale. This also confirm the db-RDA results; efficiency does not depend on the method but on the studied bathymetric level. However, there were more differences (at the site scale) with SAM which resulted in more misclassified pixels on images (Figures 9 and A1–A3). So, the use of common algorithms could be perfectible, but it does answer the problem of the study by classifying correctly macroalgal communities. Other algorithms such as random forests or support vector machines might be considered to estimate entire shores, as for coastal/terrestrial objects [93–97].

Indeed, the routine use of the hyperspectral method could be the subject of a longterm study in an ecosystem monitoring context, particularly in the context of Fucales regression on European coasts [98,99]. Indeed, since the last century, covers of some Fucales species have decreased under the action of various factors such as the intensification of grazing [100]. This trend is also well known and studied in various marine phanerogam species [101], and also in kelp species, submitted to increasing grazing and/or heat waves related to global change [102–104]. Indeed, hyperspectral imagery is already being used in many ecosystems in the context of conservation biology [105,106]. Thus, the promising results obtained in this work could serve as a basis for a conservation/monitoring program of intertidal habitats.

## **5. Conclusions**

In light of the results, MLC seems to be a better classifier for mapping a seaweeddominated rocky shore, with a more realistic achievement. To better assess the impact of global change on coastal ecosystems, there is an increasing interest in remote sensing data to evaluate the ecological state of corresponding habitats [107]. Otherwise, the community approach in ecological surveys gives a good opportunity to better understand functional traits of marine vegetation, including relationships with primary production [108,109]. In that context, this study gives for the first time a comparison of cover data for macroalgal habitats obtained by both in situ sampling on the shore and hyperspectral imagery at a centimeter resolution, and a consistent cartography of a site using well-known algorithms.

Our results go beyond the global distribution of macroalgal covers as inferred from indices such as NDVI, VCI or IP [35,110,111], but rather provide information on the fine scale repartition of species/groups of species on the shore.

Since coastal rocky shores integrate various and imbricated habitats, the UAVs approach developed here seems to be an adequate tool to evaluate the distribution of macroalgal communities/habitats at the site to geographical area level. Moreover, hyperspectral imaging at the centimeter scale allows for a precise analysis of the seaweed habitat structure in parallel to field monitoring.

**Author Contributions:** Collected field sampling data, methodology, analyzed and interpreted field and remote sensing data, wrote the article, W.D.; remote sensing methodology, made critical revisions for important intellectual content, A.L.B., T.B. (Touria Bajjouk) and S.R.; data analysis, made critical revisions for important intellectual content, M.H.; collected field sampling data, data analysis, made critical revisions for important intellectual content, T.B. (Thomas Burel); hyperspectral acquisition and pre-treatment, made critical revisions for important intellectual content, M.L. and A.G.; conceived and designed the study, supervision, collected field sampling data, funding acquisition, made critical revisions for important intellectual content, E.A.G. All authors have read and agreed to the published version of the manuscript.

**Funding:** This study was founded by the DREAL (Direction régionale de l'environnement, de l'aménagement et du logement) and the Regional Council of Brittany (Rebent Network), the Office Français de la Biodiversité and the Agence de l'Eau Loire-Bretagne (European Water Framework Directive and Marine Strategy Framework Directive). Wendy Diruit received a fellowship from the Doctoral School of Marine Sciences (Ecole Doctorale des Sciences de la Mer et du Littoral).

**Data Availability Statement:** The data presented in this study are available on request from the corresponding author.

**Acknowledgments:** Authors thank Sara Terrin and Alain Guenneguez for helping with the field sampling, and Marion Jaud for her valuable advice in hyperspectral analysis and the use of ENVI.

**Conflicts of Interest:** The authors declare no conflict of interest.

**Appendix A**

**Figure A1.** (**a**) Picture of a sampling spot on the An level at Porspoder taken during field sampling in May 2021. (**b**) NIR-G-B image of the same sampling spot. (**c**) Result of the MLC classification. (**d**) Result of the SAM classification. The black square corresponds to the mobile grid structure used for field sampling. Color code in (**c**,**d**) corresponds to the following classes: '*P. canaliculata*' (brown), '*F. spiralis*' (yellow), '*A. nodosum*' (purple), '*F. serratus*' (Coral), '*H. elongata*' (orange), 'Red' algae (red), 'Green' algae (green), 'Substratum' (grey) and 'Water' (blue).

**Figure A2.** (**a**) Picture of a sampling spot on the Fser level at Porspoder taken during field sampling in June 2021. (**b**) NIR-G-B image of the same sampling spot. (**c**) Result of the MLC classification. (**d**) Result of the SAM classification. The black square corresponds to the mobile grid structure used for field sampling. Color code in (**c**,**d**) corresponds to the following classes: '*P. canaliculata*' (brown), '*F. spiralis*' (yellow), '*A. nodosum*' (purple), '*F. serratus*' (Coral), '*H. elongata*' (orange), 'Red' algae (red), 'Green' algae (green), 'Substratum' (grey) and 'Water' (blue).

**Figure A3.** (**a**) Picture of a sampling spot on the He-Ld level at Porspoder taken during field sampling in April 2021. (**b**) NIR-G-B image of the same sampling spot. (**c**) Result of the MLC classification. (**d**) Result of the SAM classification. The black square corresponds to the mobile grid structure used for field sampling. Color code in (**c**,**d**) corresponds to the following classes: '*P. canaliculata*' (brown), '*A. nodosum*' (purple), '*F. serratus*' (Coral), '*H. elongata*' (orange), 'Red' algae (red), 'Green' algae (green), 'Substratum' (grey) and 'Water' (blue).
