A MACHINE LEARNING CLUSTERING ALGORITHM TO IDENTIFY GAPING BEHAVIOUR IN Mytilus spp. UNDER CONTRASTING ENVIRONMENTAL CONDITIONS

Introduction

Lagoons and deltas, are highly heterogenous transitional systems, subject to multiple pressures. Species inhabiting these areas have adapted to cope with the natural heterogeneity

 but local and global anthropogenic pressures, including climate change, may increase stress and in some cases lead to mortality. Studying behavioural responses can be the key to identify sub-lethal  stress, as behaviour can have physiological links. Mussel gaping, for example, is a highly dynamic process, linked to key functions of metabolism, such as feeding and respiring: changes  in  its temporal pattern  can  affect  energy  intake and allocation,  ultimately influencing growth and reproduction. This can be important in the context of  sustainable cultivation of these species, where resource utilisation should be optimised by maximising the return from the input of seed, thus avoiding periods of metabolic suppression or energetically costly processes. The aim of this study was therefore to (1) test the use of a machine learning algorithm to identify key behaviours and (2) understand whether  consistent patterns of behaviour  could  be linked to  specific environmental conditions .

Method s

Biophys sensors

 were deployed at multiple  aquaculture sites of Mytilus galloprovicnialis within the Venice Lagoon ( VL: 6 sensors, 3 sites, 1 year) and  M. edulis in the Wadden Sea ( WS: 30 sensors, 2 sites, 1 month) . With the use of these  electro-magnetic sensors the valve gaping amplitude ( between 0: close and 1:open) can be obtained every minute, coupled with the temperature of the surrounding water.

Other environmental parameters likely to influence behaviour were measured in continuum (VL, 1 station: Dissolved Oxygen, Chlorophyll, Turbidity ,  sampling frequency 12 minutes ; WS, 2 stations: Chlorophyll, Turbidity ,  sampling frequency 5 minutes ).   A clustering algorithm (fuzzy k-means) was applied to  daily  patterns  of gaping and  to  the environmental parameters measured .  Within the days corresponding to a specified “environmental cluster”, the occurrences of the different “gaping clusters” were counted in order to understand the prevalence of certain behaviours under specific combinations of environmental conditions.

Results

 The algorithm  led to identify the three distinct gaping clusters shown in Fig. 1, named “narrow” , “mid” , “wide ”  at all sites, except one, where the "mid" cluster was not found.   Environmental variables also resulted in distinctive  environmental  clusters (VL: temperature and  dissolved oxygen:  5, chlorophyll: 4 and turbidity: 3 ; WS: temperature: 3, chlorophyll and turbidity:2).  At all sites in the VL mussels were more frequently widely opened (20 vs 10 days)  at the highest temperature (cluster means 25-27°C) and when dissolved oxygen exhibited a more evident fluctuation. On the other hand, they were more frequently narrow (15 vs 5 days) when temperatures were in the mid-low (cluster means 10-12°C) and oxygen saturation mid-high. No relation could be found in VL with chlorophyll and the number of observations for medium and high turbidity were too low. In the WS there was a separation between the two sites, which exhibited distinct patterns of chlorophyll and turbidity . One site had low levels of chlorophyll and turbidity and mussels were more frequently narrow gaping (narrow: 35, mid:50, wide:18) while in the other site, where both chlorophyll and turbidity were higher they predominantly had medium and wider gaping (narrow: 30, mid: 80, wide: 50).

Discussion

 This study found that, an unsupervised consolidated clustering algorithm, allowed to identify patterns of behaviour and characterise some of  its  potential  drivers.  For example,  at higher temperatures, mussels were consistently  displaying a wider opening angle. This  is consistent with shellfish physiology and formulations used in bioenergetic models, which predict an increase in oxygen demand with temperature. However, this behaviour could also increase the energy intake, which, on the other hand, is assumed to be maximum at the organism optimal temperature. Therefore, a more detailed analysis of these data could lead to improve shellfish bioenergetic models, which are being increasingly used in shellfish farming. Furthermore, mussels maintained a wide gape even in high turbidity conditions . At this site mussels have wider palps and grow less, suggesting adaptation to the conditions coming at a cost. 

Acknowledgements

 Scientific activity performed in the Research Programme Venezia2021, with the contribution of the Provveditorato for the Public Works of Veneto, Trentino Alto Adige and Friuli Venezia Giulia, provided through the concessionary of State Consorzio Venezia Nuova and coordinated by CORILA. The research leading to these results has also received funding from the GAIN project, European Union’s HORIZON 2020 Framework Programme under GRANT AGREEMENT NO. 773330.

References

Bertolini C, Rubinetti S, Umgiesser G, Witbaard R, Bouma TJ, Rubino A, Pastres R (2021) How to cope in heterogeneous coastal environments: Spatio-temporally endogenous circadian rhythm of valve gaping by mussels. Sci Total Environ 768:145085