MAERL HABITAT COMPLEXITY

Presentation

Welcome to my Professional Practice! I'm Victor and during the past 8 weeks, I've been working on this project with my supervisors Jacques Grall and Olivier Gauthier (also at researchgate and twitter), from the European Institute for Marine Studies (IUEM). Our study was in collaboration with the Laboratoire des sciences de l'Environnement Marin (LEMAR) and in the scope of the Marine Observatory of IUEM (l’Observatoire Marin de L’Institut Universitaire Européen de la Mer), which monitors coastal and offshore marine environments in an interdisciplinary way, working with biological, chemical and geological data to investigate the current state and predict the future of these ecosystems and their responses to global changes. One of their main missions is to monitor benthic communities on several contrasting important habitats such as sandy beaches, intertidal seagrass beds, rocky shores, subtidal sediment and subtidal maerl beds. The time series for maerl beds dates from 1992 and follows 11 sites around Brittany. The Observatory is also concerned with methodological questions, and the main goal of our study was to start exploring means to quantitatively measure maerl fragments structural complexity.

I will guide you through the development of my work, sharing mostly my scientific findings, but letting you know a little bit of my personal experiences during this project. I hope you'll enjoy it!

Introduction

TIP: If you're interested in the topic, all references are linked to the publication's DOI

Since it was first brought to light by MacArthur & MacArthur in 1961, the role of habitat complexity on biodiversity has long been a topic of importance to ecology, with higher diversity usually being associated with more complex structures. Because this is thought to be especially true for the aquatic environment, where such structures can provide shelter from both physical and biological pressures like currents and predators, this theme has been explored extensively in aquatic ecology (Tokeshi & Arakaki, 2012) . However, there’s still a lack of hypothesis-driven quantitative assessments on the role of habitat complexity in biodiversity, with most of the available work being of qualitative nature (Kovalenko, Thomaz, & Warfe, 2012). Moreover, when investigating it quantitatively, most studies either focused on estimating complexity indirectly from specific morphological attributes, such as the density of structures, or only explored one aspect of complexity when measuring it directly (Kovalenko et al., 2012; Tokeshi & Arakaki, 2012)

Crustose coralline algae are important bioengineers that can aggregate and form free-living complex structures called maerl. Maerl beds are ecologically important ecosystems distributed worldwide that can be found from coastal areas to depths over 200 m. The high biodiversity associated with maerl beds is historically mostly attributed to its three-dimensional complexity (Nelson, 2009; Riosmena-Rodríguez, Nelson, & Aguirre, 2017; Schubert et al., 2020). Nevertheless, until very recently, there were only a few projects of qualitative nature that explicitly investigated the role of maerl structural complexity on biodiversity (Gabara et al, 2018). And although some studies have indirectly linked quantitatively assessed morphological attributes to maerl complexity (Villas-Boas, Riosmena-Rodriguez, & Figueiredo, 2014), quantitative studies that address the role of complexity as their main hypothesis and use methods to directly assess it, instead of estimating it through attribute measures, are still lacking.

Due to this need, during my Professional Practice, our team started exploring the application of different methods to assess maerl complexity, aiming to develop a more holistic approach. Such quantitative methods consisted of both inferences of complexity through linear and whole-attribute measurements, as well as direct assessments of complexity through fractal and interstitial space analysis. These last methods have been previously described for other habitats, such as coral reefs, mussel reefs, macrophyte-dominated habitats, and seagrass beds(Kovalenko, Thomaz, & Warfe, 2012), but never applied to maerl before.

Maerl bed shot by Erwan Amice, Centre National de la Recherche Scientifique (CNRS)—Laboratoire des sciences de l'Environnement Marin (LEMAR)

If you want to know a little bit more about maerl beds and habitat complexity, you can listen to this podcast session of the series The Yound Men and the Sea, hosted by Remco Lameijer and in which I talked about my project:

TIP: scroll down the preview page below and hit play so you can listen to it while checking the website!

mETHODS

Study design (& practicalities)

Samples were collected in the context of the REBENT (Réseau Benthique) monitoring program (http://www.rebent.org) from four subtidal maerl beds along the coast of Brittany in February of 2019. The sites sampled were the Glénan Islands maerl bed, the Trévignon maerl bed, the Rozegat maerl bed situated at the Bay of Brest and the Molène maerl bed, located at the Quéménès island. The locations were chosen so that maerl fragments with different apparent complexity levels would be compared, since those habitats are under different physical and human pressures, resulting in contrasting conservation states. Samples were taken using a 0.1 m² Smith McIntyre grab.

Map of the sampling sites, part of the REBENT monitoring program. Kindly made by Alice d'Hurlaborde.

Since I’m Brazilian and don’t have a fix address in Europe, my friend was kind enough to shelter me during the pandemic, and I was working from her family’s home during the confinement. Since I was still in France and my supervisors were very invested in the project, they trusted me enough to send me the samples by post. I had to subsample and sort them in a way that the analysis was suitable given my working conditions. Due to material availability and other practical limitations, such as camera resolution, only the 15 biggest maerl fragments (>10 mm) of each sample were selected for analysis.

Whole-plant measurements

Three measures of sphericity and branching density, typically used whole-plant attributes to describe maerl morphology that have been linked to complexity (Figueiredo et al., 2007; Villas-Boas, Riosmena-Rodriguez, & Figueiredo, 2014; Warfe, Barmuta, & Wotherspoon, 2008) were taken for each maerl bit. Sphericity measurements were estimated as described by Sneed & Folk (1958) using the longest diameters of the three main axes of the fragment, which were measured to the closest millimetre using a Vernier caliper. The maximum projection sphericity, shown in Eq. 1, where L is the longest diameter of the rhodolith, I is the longest diameter of the intermediate size axis, perpendicular to L, and S is the longest dimension of the shorter axis, perpendicular to L and I; and two diameter ratios, hereby mentioned as DR1 (Eq. 2) and DR2 (Eq. 3), were calculated.

Equation 1

Equation 2

Equation 3

Animated scheme of the method used for estimating branching density.

Branching density was estimated by manually placing a quadrat of 1 cm² over each maerl fragment five times in a random manner and counting the number of apical branches entirely placed inside it, as illustrated. The average number of apical branches per sample was taken as the final branching density.

The number of apical branches in each quadrat was counted manually.

Image Acquisition and Editing

For both fractal analysis and interstitial space index measurements, an image-based analysis was performed according to methods by Warfe et al. (2008) and Thomaz et al. (2008). Due to the Covid-19 pandemic, a photography set was improvised with the materials available. A camera was mounted in a tripod in front of each maerl thalli, which were positioned over a black background and were placed under light from two directions to decrease shadow noise. Two to five photographs of each fragment were taken, each with a different shallow depth of field. The first was taken focusing on the parts closest to the camera, and each successive photograph was taken focusing back, until all structures were completely blurred. The number of pictures varied due to the difference in size and morphology of each fragment. Due to the limitations of the camera used, a Nikon D90 with a 18-105mm zoom lens (not macro appropriate) all pictures were taken at the same distance and magnification (6x) and finally cropped to the same size (1200 px²), so that all would be at the same scale.

Final improvised photography set for specimen photography (yes, the black background was an old t-shirt stuck to a cardboard box).

All images were then edited using Adobe Lightroom Classic 2020 and Adobe Photoshop 2020, and focus stacks were created in order to have a better 3-D representation of the thallus structure. For fractal analysis, pictures were also converted into grayscale and further edited using ImageJ in order to get only the outline of each maerl fragment for binary analysis. Both colored, grayscale and binary images were used in fractal analysis. If you're interested in having a glimpse of how extensive the image treatment process for a single maerl fragment is , you can take a look at this simplified and sped-up video:

Image treatment.mp4

Fractal analysis

Fractal analysis is a means of directly estimating complexity through the fractal dimension D, which is a measure of the change in detail in function of the change in scale (Sugihara & M. May, 1990). In ecology, it’s been widely applied as a tool to quantify the degree of complexity at different scales (Tokeshi & Arakaki, 2012), with most of the studies relying on the Box-Counting Method. To put in a simple way, this method consists in systematically laying a series of grids of decreasing scale (which are the boxes) over an image and counting the boxes that contain relevant information for each different scale. There’re many ways to apply this method, and “relevant information” differs in literature according to the nature of the analysis and of the image – most studies use binary images with only the outline shape of the object of study, and the information counted is the number of pixels containing the outline, while others use original images and manually count the number of boxes in which the object is completely in focus, for example.

Example of how the Box-Counting method works, image taken from FracLac's help page (below).

Most of the marine ecology studies using the box-count method for fractal analysis have been either done by time-consuming protocols involving manual counting or have been done through outdated, non-accessible software developed by small research teams for internal use and require contacting them to have an answer (Martin-Garin et al. 2007; Thomaz et al. 2008). Aiming to develop a more reproducible protocol, I decided to use an extremely user-friendly available plugin for the free-access software ImageJ named FracLac. FracLac has been available for over two decades and has been widely applied in the fields of biomedicine, mainly for brain tissue histology studies, but has rarely been applied to ecology, even though it has been shown to adequately describe epiphyte complexity (Stanton & Horn, 2013). By using this software I was also able to compare the results of fractal analysis using binary and grayscale images. As mentioned before, most studies in ecology only work with binary images, and that comes with the cost of losing a lot of information that could be ecologically relevant (e.g. available living space), that is overcome partially by using methods that allow grayscale images, because they contain important textural information that are a part of the object complexity (Sarkar & Chaudhuri, 1992). There are 4 different methods available in FracLac for Box-Counting fractal analysis, one based on binary images and three others that use the difference in pixel intensity to estimate the fractal dimension of greyscale images. In this study, we applied all four methods and compared them to investigate which would be the most promising for describing maerl fragments.

If you’re captivated by this theme and want to learn more about fractal analysis and it’s different applications in science, you can explore FracLac’s website, where they explain everything about fractal dimension and the types of analysis that this plugin runs:

INTERSTITIAL SPACE INDEX

Definition

The Interstitial Space Index (ISI), as first described by Dibble et al. (1996), measures the frequency and size of interstitial spaces between structural elements along vertical and horizontal axes, such that a greater number of smaller gaps indicates a more structurally complex plant shape. This index has been shown to be relevant for the associated fauna since it’s a good representation of available-habitat, and thus, is thought to be a biologically-relevant measure of habitat complexity (Kovalenko, Thomaz, & Warfe, 2012; Warfe et al., 2008).

To estimate the ISI, a line-intercept method was used to count and measure the gaps in each axis. The measurements were taken in ImageJ: once the scale was calibrated, a grid was positioned over the focus-stacked picture (in original colors) and the gaps were manually counted and their distance calculated with the line tool. The sampling of gap lengths was systematic, with a fixed distance of 2.3 mm between each transect that was defined accordingly to the smallest maerl samples in order to get at least 5 transects for each thallus. This way, bigger fragments would have a higher number of transects instead of a fixed number of transects with different distances between each maerl bit, avoiding an autocorrelation problem that would arise from this. These measurements provided the mean frequency of gaps along the horizontal axes (f_h), the mean length (height) of those gaps (l_h), the mean frequency of gaps along the vertical axes (f_v), and the mean length (width) of those gaps (l_v). These values were then used to calculate the ISI using the formula:

Unfortunately, since this counting process was made manually and was very time-consuming (as can be seen below), only a few samples had been measured so far and this method didn’t integrate our preliminary statistical analysis. But, because it’s a promising method, our team will keep working on acquiring this data prior to publishing our study.

Sped-up simplified scheme of the sampling of gaps for the Interstitial Space Index (only horizontal gaps measurements are shown).

PRELIMINARY DATA ANALYSIS

The data was checked for univariate and multivariate normality, through Shapiro Test, and for homogeneity of variance assumption, through Levene’s Test. Multicolinearity was also tested with Pearson’s correlation in order to select which response variables would integrate the final model. Because not all multivariate assumptions were met, a PERMDISP was performed prior to the non-parametric Permutational Multivariate Analysis of Variance (PERMANOVA – Adonis). To investigate which of the analyzed variables were driving the variance of our samples, a Principal Component Analysis was performed. One-Way ANOVAs were performed for each response variable to further understand which ones could be promising for describing the maerl fragments. All analysis were run in R (https://www.R-project.org/).

preliminary Results & discussion

I understand that you're probably not familiar with maerl fragments (I can't blame you!), and that you're probably very curious to see what they look like. Our improvised photography set resulted in 64 close-up images, and even though the camera used wasn't macro-appropriate, the final products were very detailed due to the focus stack (thank you, Photoshop!). So, before you move on to the preliminary statistical results, take some time to appreciate such interesting algae, and keep in mind that all pictures are at the same scale.

GROUP A

Maerl Fragments from the Molène maerl bed (Quéménès island)

GROUP B

Maerl Fragments from the Glénan maerl bed (Glénan islands)

GROUP C

Thalli from the Rozegat maerl bed (Bay of Brest)

GROUP D

Thalli from the Trevignon maerl bed (Trégunc)

MULTIVARIATE ANALYSIS

When looking for multicollinearity, we found that out of the 4 methods used for estimating the fractal dimension D, the three methods based on grayscale images were almost perfectly correlated (Figure 1). Due to that, only the Binary method and the Grayscale Differential Volume 1 + 1 method (from now on referred as GRAY3 - check FracLac’s website to understand the difference between the methods), which showed the smallest variance, were selected for the final analysis.

Figure 1. Correlation and distribution plots for Fractal Dimension D estimated through the different methods available in FracLac.

The PERMDISP test was significant (F-statistic = 3.6847, p = 0.012), indicating that there’s certainly a dispersion effect, so that we cannot affirm that there’s a location effect, even though the PERMANOVA was also significant (p < 0.001). The dispersion effect is also clear when observing the Principal Component Analysis (Fig. 2). The first two axis of the PCA explain 67.5% of the variance, and a general trend can be seen. Samples from group A and B being more similar to each other and samples from localities C and D being closer to each other. We expected such trend, considering that samples from groups A and B were the smallest and least apparently complex, while samples from localities C and D were the biggest and more visually complex. The difference in complexity also reflects the health and conservation status of those maerl beds – the Glénan site (group B) was extremely explored during the 60s, with maerl extraction resulting in the death of most of the bed’s thalli and associated fauna, for example (Grall & Hall-Spencer, 2003).

Figure 2. Principal Component Analysis distance biplot (scaling 1). One representative of each locality was added for better comprehension.

The PCA also showed that the diameter S (the biggest diameter of the smallest axis of the thallus), used to calculate Sphericity, was driving most of the variance. However it’s important to keep in mind that even though this specific variable was a good morphological parameter to differentiate the maerl thalli, it’s not necessarily related to habitat complexity and available living space if analyzed isolated. We could also see that the measures of Fractal Dimension D show a contrasting trend when compared to the whole-plant attributes branching density and sphericity measures (Maximum Sphericity and diameter ratios). For several different marine environments, fractal dimension has been shown to be more highly correlated to faunal diversity than single morphological attributes, and thus, is considered a better measure of habitat complexity (Kovalenko, Thomaz, & Warfe, 2012; Tokeshi & Arakaki, 2012; Warfe et al., 2008). Our results might suggest that this would also be true for maerl fragments, considering the different trends. However, this still needs investigation with community data, which will be the next step of our study.

UNIVARIATE ANALYSIS

We individually tested each variable through One-Way ANOVAs and post-hoc Tukey's tests to investigate if they were significantly different between localities (Figure 3). Both measures of the fractal dimension D, from Binary and GRAY3 methods, were significantly different between groups (p < 0.001 for both cases). However, while the GRAY3 method only pointed out location B as significantly different from others, the Binary method resulted in differences between groups B & C, B & D, and A & D. Overall, both methods show similar trends, but GRAY3 was able to differ groups A and B in terms of complexity, where the Binary failed, while the Binary method was able to show significant differences between groups A & D, which was not seen for the grayscale method. We expected this methods to have different results due to their different nature - while the binary method only deals with the presence and absence of the outline of the thalli, the grayscale method infers the fractal dimension through the difference between pixel intensity, which is more efficient for texturized images (Sarkar & Chaudhuri, 1992), but it’s still unclear which would be more efficient. Fractal dimension varies with scale (Warfe et al., 2008), and therefore, further investigation with more than a single image magnification is necessary. Similarly a study linking those methods with community data is necessary to understand which of them would be a better representation of habitat complexity.

Branching density was significantly different (p < 0.01) between all localities, except between groups A and B. This morphological measurement has been previously related to an increase in faunal abundance and richness (Figueiredo et al., 2007) and thus could be a good indicator of habitat complexity.. Maximum Sphericity was significantly different between groups A & C, A & D and D & B. This is a measure of the algae three-dimensionality, however, it doesn’t take it’s complex branching into consideration since it was first created to describe pebbles, instead, it shows how similar the three main axes’ lengths are (Sneed & Folk, 1958). Since both variables show contrasting patterns from the fractal dimension, it’s important to test those measurements together with community data to understand which would be more ecologically important.

Figura 3. Boxplots of each main response variable with ANOVA p-values and Tukey's Test significance shown.

FINAL CONSIDERATIONS & FUTURE DIRECTIONS

Our study was the first investigation of its kind, and we were able to apply different methods to asses maerl fragments complexity. We started showing that most of those methods seem promising, and are now ready to apply them to other ecologically important questions. During my thesis, we'll investigate if any of those measurements are linked to alpha and beta-diversity to further understand which of them are indeed accurate measurements of habitat complexity and if they are ecologically representative.

But before that, there are still some unfinished business we need to attend! During this 8 weeks, I wasn't able to finalize all the measurements we wanted! We'll keep on working to get one of the most promising methods for estimating complexity, the Interstitial Space Index. Just looking at some representatives of each group, we expect that this method could bring interesting results:

ACKNOWLEDGEMENTS

Thank you for sticking until the end, I hope you enjoyed it and that you learned something new with my website! Now, I'd like to use this platform to thank some of the people that has made my work possible.

First, I want to thank my supervisors Jacques Grall and Olivier Gauthier, who were always so open to my ideas, trusted me with biological material and offered me all the necessary support during the whole project! I'm looking forward to keep on working with you.

I'd also like to thank the D'Hurlaborde family. Alice, Christine, Emma and Jean Jacques, thank you so much! I have no idea of what would've happened if you hadn't so kindly offered me to stay with you during this process! You've opened your home to me and truly made me feel like part of the family, helping me not only with housing, but making me feel integrated and giving me a very much needed emotional support! Also, thank you for letting me transform your library into a crazy algae laboratory, this study couldn't have been finished without you, and I'll always be grateful!

Finally, I'd like to thank the IMBRSea Coordination for finding solutions for the students at these problematic times we've been through the past months! It'd have been a shame if we didn't have the option to keep on working on our projects like so many others hadn't during this pandemic. It hasn't been easy for any of us, but you sure did a great job of coming up with creative solutions.

THANK YOU!

Page updated

Report abuse