Floating objects in the open ocean: Unveiling modifications of the pelagic habitat induced by forest cover change and climate variations

Natural floating objects (NLOGs) are a major component of the habitat of pelagic species. Since the 1990s, the number of floating objects in the open ocean has increased greatly as a result of the introduction of drifting fish aggregating devices (DFADs) by the industrial tropical tuna purse seine vessels. These changes, and their potential impacts on the species that associate with floating objects, remain poorly understood. If the habitat modifications induced by DFADs have been recently characterized and quantified, the impact of other human activities on the number of floating objects is poorly studied. Relying on lagrangian simulations at the scale of the whole Indian Ocean, from 2000 to 2019, we assess the potential modifications of the pelagic surface habitat that could originate from forest cover change and climate variations. We develop several scenarios, based on coastal and river forest cover, precipitations and river discharge, to simulate densities of NLOGs. Our results suggest no significant increase in average NLOG densities in the ocean and highlight important regional and seasonal variations of these densities driven by both forest cover change and precipitations. These preliminary findings underscore the limited understanding of this critical element of pelagic species habitat. Therefore, there is pressing need to intensify monitoring efforts for pelagic species habitat and raise awareness about potential impacts of habitat modifications on tuna and other pelagic species.


Introduction
In the context of global change, ocean biodiversity is facing degradation due to several direct and indirect drivers (IPBES, 2019).Among these, the exploitation of fish and seafood by fisheries stands as the foremost direct driver, imposing largest impacts (IPBES, 2019).The consequences of increased fishing activities extend beyond mere depletion of marine resources; they also trigger indirect impacts by altering natural habitats (Neumann et al., 2016;IPBES, 2018).For example, research has shown that fishing activities influence the functional composition of benthic species communities (Neumann et al., 2016;Dupaix et al., 2021b) and modify pelagic species habitat (Dupaix et al., 2021a).These habitat modifications, in turn, can substantially affect marine species in terms of distribution, reproduction, behavior and ultimately their fitness (Macura et al., 2019).Consequently, it becomes crucial to understand and assess the scale of these habitat modifications, which result either from global change or direct exploitation of animals.
Numerous pelagic species, including tropical tuna, manifest a wellknown association with floating objects (FOBs, Fréon and Dagorn, 2000;Castro et al., 2002).Originally, FOBs were primarily natural debris like parts of trees floating out in the oceans (designated as NLOGs).Fishers have exploited the associative behavior of pelagic species for centuries to aid in their search and catch.Oppian, a Greek poet, detailed in his poem Halieutica written in the 2nd century, how fishers use ship wrecks to facilitate the catch of dolphinfish (Coryphaena hippurus; Oppian, 1963).Historical evidence also points to the use of anchored fish aggregating devices (AFADs) -i.e.man-made objects, moored in coastal areas, to attract targeted fish species.Artisanal fishers in the Mediterranean were deploying AFADs as early as in the 14th century (Taquet, 2013), while a similar practice was observed in Japan during the 17th century (Nakamae, 1991).In more recent times, industrial tuna purse seine vessels introduced drifting fish aggregating devices (DFADs; Davies et al., 2014) in the 1980s.The deployment of DFADs has surged over the years, with most recent global estimates ranging from 81,000 to 121,000 deployments yearly (Gershman et al., 2015).Specifically, in the Indian Ocean (IO), Maufroy et al. (2017) reported a fourfold increase in the use of DFADs from 2007 to 2013.
The rising number of FOBs in the open ocean related to the increasing use of DFADs has led to various ecological impacts on tropical tuna populations (Dagorn et al., 2013b;Capello et al., 2023;Pons et al., 2023).Firstly, the presence of DFADs reduces the search effort required by purse seiners, significantly increasing tuna catchability and consequently their fishing mortality (Wain et al., 2021;Song and Shen, 2022).Secondly, fishing on DFADs results in the catch of smaller yellowfin (YFT, Thunnus albacares) and bigeye (BET, T. obsesus) tuna compared to targeting them in free-swimming schools (not associated with FOBs; IOTC, 2022d; Merino et al., 2020).Within the IO, YFT and BET, two out of the three tropical tuna targeted by purse seiners, are currently facing challenges of overfishing (IOTC, 2022b;IOTC, 2022a).This means that their spawning biomass is below the spawning biomass reference point for maximum sustainable yield.Purse seine vessels account for 34 % of the yearly catch of YFT and 42 % of BET in the IO (IOTC, 2022b;IOTC, 2022a).Consequently, the management of these fisheries necessitates careful consideration of the two aforementioned impacts caused by drifting fish aggregating devices (DFADs).
Beyond the direct effects of increasing fishing mortality, DFADs have the potential to induce indirect ecological impacts on tunas by modifying tuna surface habitat.Evidence indicates that tuna caught in association with FOBs exhibit lower body condition compared to those caught in free-swimming schools (Hallier and Gaertner, 2008;Marsac et al., 2000;Robert et al., 2014).Furthermore, changes in FOBs density may disturb the large-scale movements of tuna, leading to individuals being retained or transported to ecologically unsuitable areas (Marsac et al., 2000;Fonteneau et al., 2013;Schaefer et al., 2015), and potentially influencing tuna schooling behaviour (Capello et al., 2022;Sempo et al., 2013).While scientific consensus on the extent of indirect impacts of DFADs on tuna behaviour and life-history parameters (e.g.physiological condition, reproduction; Dupaix et al., 2024) is yet to be reached, these potential impacts emphasize the need to characterize pelagic surface habitat modifications induced by human activities.A recent study (Dupaix et al., 2021a) focused on the Western IO (WIO) and provided insights into the modifications of the pelagic habitat induced by DFADs.By analyzing data from observers onboard purse seine vessels, they revealed a strong increase in the total number of FOBs from 2006 to 2018.The entire WIO was impacted: in recent years, DFADs represent over 85 % of the total number of FOBs and the number of FOB observations per day increased fivefold between 2006 and 2018.To assess these modifications, Dupaix et al. (2021a) took the number of observed NLOGs in the WIO as a reference state, representing conditions unaltered by human activities.
The number of NLOGs, mainly composed of tree parts transported by surface currents, could also be significantly affected by human activities, such as forest cover change, and climate variations which can impact oceanic currents (Krajick, 2001;Thiel and Gutow, 2005;Russell et al., 2014).Deforestation, involving the human-induced conversion of forested land into non-forested land, is mainly driven by agriculture and logging (IPBES, 2018).Since 2000, global logging, for materials, construction and energy production, has seen a substantial increase, with 20% of tropical forests experiencing selective logging (IPBES, 2022).In tropical areas, forested land has decreased in recent decades with tropical forest loss estimated to be approximately 10 million hectares per year, with 60 to 90 % attributed to agriculture (Pendrill et al., 2022).The number of NLOGs can also exhibit strong seasonal and inter-annual variations, influenced by factors like precipitations, modifications in surface currents, and extreme climatic events impacting the release and circulation of NLOGs from sources like rivers and coastal forests (Caddy and Majkowski, 1996;Hinojosa et al., 2011).For example, in Taiwan, the Morakot typhoon in 2009 led to the release of over three million trees into the sea Doong et al. (2011).
In the absence of trajectory and/or position data of NLOGs across the entire ocean, employing Lagrangian simulation models with virtual NLOGs becomes valuable in estimating their distribution.Lagrangian simulations are increasingly being used at ocean basin and global scales to characterize the transport of debris, and the dispersal of fish larvae and plant propagules (Andrello et al., 2017;Van Sebille et al., 2018;Van der Stocken et al., 2019;Fontoura et al., 2022).NLOGs are most likely sourced from river basins or coastal forests (Thiel and Gutow, 2005;Doong et al., 2011), and previous studies have indicated that FOBs drift in a manner similar to oceanic drifters (Imzilen et al., 2019).NLOG's drift can thus predominantly be attributed to surface oceanic currents.This study focuses on characterizing the natural surface habitat of tropical tuna in the IO, specifically determining how the density and distributions of NLOGs has changed between 2000 and 2019, potentially revealing the effects of human activities, such as deforestation, and changes in oceanic circulation linked to climate variations.Given the absence of precise estimations of NLOG distribution in the IO, we aim to (1) describe the distribution of NLOGs and their seasonal variations at the scale of the entire oceanic basin and (2) assess any potential trends in NLOG numbers, which could be linked to deforestation or surface currents' variations.

Lagrangian simulations
Lagrangian simulations were performed from 2000 to 2019 to simulate the distribution of NLOGs in the Indian Ocean (IO) using the Ichthyop computational tool v3.3.12 (Lett et al., 2008;Barrier et al., 2023).The delimitation of IO considered in this study is framed in Fig. 1.This tool allows to account for advection-diffusion processes affecting the dispersion of particles (NLOGs in this case) released in a given location of the ocean based on ocean currents.Surface currents were extracted from an eddy-resolving (1/12 • ) simulation performed using NEMO Ocean General Circulation Model (Madec, 2016) in the framework of the OceaniC Chaos -ImPacts strUcture predicTability (OCCIPUT) project (https://meom-group.github.io/projects/occiput;Penduff et al., 2014;Bessières et al., 2017).The hydrodynamic simulations were driven by the 3-hourly fully-varying atmospheric forcing (the DFS5.2 oceanic forcing product, derived from the ERA-I re-analysis; Dussin et al., 2016).Other simulation settings relevant for this study include release location (Davies et al., 2017;Curnick et al., 2021), release time (Siegel et al., 2003;Curnick et al., 2021) and NLOG lifetime (Pineda et al., 2007;Huret et al., 2010;Van der Stocken et al., 2019).A total of 6,408 cells (1/12 • × 1/12 • ) were identified as coastal cells, defined as cells whose centers lie within 1/12 • distance from the coastline.The coastline data was obtained from the IHO (International Hydrographic Organization) coastline shapefile v3 (Institute, 2018), covering a total area of 67 million km 2 .To have a holistic view of all the potential trajectories, n = 1000 particles were released every week within each coastal cell, corresponding to a total of 6.7 billion particles simulated.Particles were advected using NEMO surface currents.Particles were considered stranded and were removed from the simulation when they entered a land cell.Particle densities were then saved as a 2D array every 7 days for each release location and date.
The lifetime of NLOGs, dependent on wood property and on its susceptibility to destruction by secondary consumers, remains poorly known and can range from a few days to up to several years (Thiel and Gutow, 2005).However, past simulations with lifetimes varying from 30 days to a year, revealed that the results were relatively insensitive to lifetime in the IO (Dupaix et al., 2021a).In the current study, an average NLOG lifetime of one year was considered.To this end, particles were transported for a fixed duration of 500 days.The average drifting time of one year was then derived by multiplying the 2D arrays by the probability mass function of an inverse Poisson distribution with λ = 360 days, i.e. multiplying by 1 − e − λ λ t t! with λ = 360 days and t the number of days since release.This approach assumes that NLOGs will sink on average aftre one year, while allowing for some variability in sinking times.The simulations were initiated 500 days before 2000-01-01, to allow the equilibrium to be reached at the beginning of the study period.

Accounting for NLOG release sources and their variability
In order to account for different NLOG sources, as well as potential factors affecting their release rate, we developed different weighting scenarios.For each release location and date, we considered the following factors: (i) associated precipitations, (ii) associated rivers and their discharge and (iii) the surface area covered by forests.Precipitations data at the release location and date were extracted from Adler et al. (2003), which provides global monthly precipitations estimates in mm.days − 1 per 2 • cells.Information about rivers and their discharge were obtained from the HydroAtlas database v1.0 (Linke et al., 2019).Rivers were associated to the nearest release locations considering the position of the river mouth relative to the center of the release cell.We selected rivers with a mouth in the IO and a maximum discharge at the mouth equal to or greater than 100 m 3 .s− 1 , totaling over 195,000 km of river length in the study region.To account for forest cover, we created a 1 km buffer around the rivers and coastline.We used Global Forest Change products (GFC 2020v1.8, Hansen et al., 2013), which provide time series analysis results of Landsat images, to characterize global forest extent in year 2000 and changes in global forest cover from 2000 to 2020 at 30-meter resolution.The forest cover was mapped in the coastal areas of the IO (Fig. 1) using Google Earth Engine (GEE), an application programming interface (API) using image processing algorithms.The GFC 2020v1.8product was available in GEE to import GFC global forest extent data in the previously-defined buffer zone for year 2000.For each year after 2000, the GFC pixels (with a resolution of 30 × 30 m 2 ) were successively compared to identify losses and gains of forest area.Yearly raster maps of forest cover were produced, representing forest presence for each pixel with values ranging from 0 (no forest) to 100 % (fully forested pixel).The resolution was then reduced to cells of 90 × 90 m 2 size by averaging the pixels.
Six different weighing scenarios were considered, detailed in Table 1  and Fig

CL
Coastal Length: in this weighting scenario, particles were uniformly released along the coastline.The weight for each release locationdate was determined by measuring the total surface area of the coastline buffer associated to each release location.

CC
Coastal Cover: this scenario involved releasing particles in proportion to the coastal forest cover.Similar to CL scenario, the weight was calculated based on the coastal forest cover surface associated with each release location and date.

RC
River Cover: in this scenario, particles were released in proportion to the forest cover associated with the rivers and the rivers discharge.
For each release location and date, the weight was determined by multiplying the forest cover surface of the associated rivers with the discharge at the rivers' mouths located within each release cell.

CCp
Coastal Cover and Precipitations: this scenario was similar to CC but introduced seasonality by considering precipitations.The weight of each release location and date was calculated by multiplying the corresponding forest cover surface with precipitations.

RCp
River Cover and Precipitations: the weighting scenario was similar to RC but introduced seasonality by considering precipitations.The weight was determined by multiplying the forest cover surface of the associated rivers with the discharge at the rivers' mouths and the precipitations at the release location and date.

R&CC
River and Coastal Cover: this scenario considered the overall forest cover (both from rivers and coastal areas).For each release location and date, the weight was calculated by summing the coastal forest cover surface and the forest cover surface associated with rivers.
abundance were then obtained as follow: with N i,t the NLOG abundance in cell i at time t, R the number of release locations (R = 6,408), T the duration of a simulation (T = 500 days), w j,t− k the weight associated with release location j at time t − k and n the number of particles in cell i at time t that were released in release location j at time t − k.

Study regions
Simulated NLOG abundances were examined in eight oceanic regions defined on the basis of industrial purse seine fleet behavior in the Western IO (WIO) and oceanographic specificity (Fig. 1; Dagorn et al., 2013a;Schott et al., 2009).
• Mozambique: Encompassing the Mozambique Channel, this region is characterized by the presence of numerous oceanic eddies and a dominant southward flowing current throughout the year.Notably, it is known to be rich in NLOGs from March to May, when industrial purse-seine vessels primarily target NLOG-associated tuna schools in this area (Chassot et al., 2019).Ocean Dipole (IOD) events, this region is characterized by important negative Sea Surface Temperature (SST) anomalies and a shallower thermocline from July to October (Saji et al., 1999).
• Indonesia: This region includes the coastal waters offshore Indonesia, where an important artisanal and semi-industrial tuna fishery operates (IOTC, 2022c).Surface currents flow southward along the coast of Sumatra and are stronger during the winter monsoon.Additionally, a westward current flows in the South of the region, directed towards the Eastern and Southern IO regions (Schott et al., 2009).• Southern IO: Encompassing the entire IO south of 10 • S and the region south of Madagascar in the western part of the region, this region is characterized by westward surface currents and minor tropical tuna fishing effort (IOTC, 2022d;Schott et al., 2009).

Comparison with observers data
NLOG positions were derived from data collected by observers onboard French purse seine vessels operating in the WIO from 2014 to 2019.These observer data included the date, time and location of the vessel's main activities.For each activity taking place on a floating object, the object type was recorded, enabling the discrimination of NLOGs from other FOBs like DFADs.Using this dataset, we computed an index of NLOG abundance per 2 • cell per month, noted N obs i,m for cell i in month m, employing the methodology established in Dupaix et al. (2021a).To account for the data coverage bias stemming from the use of fisheries data, this index is standardized using observation effort.It can be expressed as follows: where D m is the number of days in month m, S i,d the number of NLOGs observed in cell i on day d and O i,d the number of vessels with an observer in cell i on day d.To ensure statistical robustness, cells with a low observation effort ( ∑ Dm d=1 O i,d < 10) were discarded.For each of the retained monthly 2 • cells, we extracted the corresponding simulated NLOG abundance from the different weighting scenarios.Observed and simulated NLOG abundances were then averaged over the years for each quarter and oceanic region where the French purse seine fleet operates, namely the Arabian Sea, Mozambique, SCTR, Somalia and Southern IO.For each scenario, a linear regression was performed between the observed and simulated NLOG abundances to determine which scenario performed the best and a Student's t-test was performed to determine if the slope coefficient of the linear regression significantly differed from zero.

Assessing NLOG density in the IO
Using the outcome of the different scenarios described in Section 2.2, we calculated NLOG densities on a weekly basis, by averaging the estimated NLOG abundances per unit surface over each oceanic region.To standardize density estimations, these values were divided by the NLOG density averaged over the whole study period in each region.Next, we used these standardized weekly density estimates to calculate the mean weekly values over the whole study period for each region and for the entire oceanic basin, giving access to seasonal variations.
We calculated a weekly NLOG density anomaly (noted A t , where t is the time) by removing the seasonal variations from the standardized weekly density estimations.To assess the presence or absence of any global increasing or decreasing trend in NLOG density across the oceanic basin, linear regressions were performed on A t , with time as an explanatory variable (A t = at + b).To evaluate temporal autocorrelation, we calculated the Pearson's correlation coefficient (ρ) between A t and A t− Δt , ∀Δt ∈ {0, 1, …, 52}, with Δt in weeks.We then determined the maximum time lag (Δt m ) for which ρ⩾0.2, i.e.Δt m = max(Δt/ρ(A t , A t− Δt )⩾0.2).A Student t-test was then performed to evaluate if the slope coefficient (a) significantly differed from zero.The test uses degrees of freedom, with s representing the number of values in the time series.

Best performing scenarios
Fig. 3 displays the comparison between observed and simulated NLOG abundances within the oceanic regions and quarters sampled by scientific observers on-board purse seiners (see Section 2.4).The Coastline Length (CL) scenario notably overestimates NLOG density in the Arabian Sea and Somalia in the 3rd quarter, and underestimates the NLOG abundance in the Mozambique Channel, resulting in a nonsignificant relationship between modelled and observed NLOG abundances between regions and seasons (p-value = 0.97).Simulated NLOG abundances obtained with other scenarios display a significant linear relationship with the observed NLOG abundance (linear regressions, pvalue of 1.67 × 10 − 4 , 9.98 × 10 − 5 , 4.06 × 10 − 4 , 1.36 × 10 − 4 and 1.61 × 10 − 4 for CC, RC, CCp, RCp and R&CC scenarios respectively).
In summary, CL is the scenario performing the worst, highlighting the need to take forest cover into account.However, this comparison with data does not allow to determine a scenario that would perform better than others, with R 2 values ranging from 0.73 to 0.79.Most scenarios tend to overestimate NLOG abundance in some regions at given quarters (e.g.Arabian Sea and Somalia regions in the third quarter), and the significant linear relationships are mostly driven by the values observed and simulated in the Mozambique Channel.

Distribution of NLOGs in the Indian Ocean
Fig. 4 displays simulated climatological NLOG density maps for the different scenarios.Among all scenarios, the Mozambique Channel, the Bay of Bengal and the western coast of India exhibited the highest NLOG densities, designating them as probable sources (i.e.regions with an important input) and accumulation zones of NLOGs in the IO.However, the relative importance of these areas strongly varies as a function of the scenario considered.In the CL scenario (Coastal Length, uniform release along the coast), the highest densities are located along the coast of India and in the Bay of Bengal, with relatively lower densities in the Mozambique Channel (Fig. 4A).Considering the coastal cover (scenarios CC and CCp) increased the relative density in the Mozambique Channel and more broadly across the rest of the oceanic basin, specifically in Somalia, Eastern IO and Southern IO regions (Fig. 4B&D).Increased NLOG densities in the Southern IO region for scenarios CC and CCp can be explained through an increased relative NLOG density on the coast of Sumatra (Indonesia region) which suggests that the coastal forest cover in this area could also constitute a potential source of NLOGs.In scenarios incorporating river cover (RC and RCp), the NLOG density in the Bay of Bengal was notably higher than in any other region, with a very high density in the south of Myanmar, likely originating from the Ganges delta (Fig. 4C&E).The R&CC scenario (Fig. 4F) displayed intermediate NLOG distribution, falling between those obtained with coastal cover only (CC and CCp) and with river cover only (RC and RCp).In summary, all scenarios designate the Mozambique Channel, the Bay of Bengal, the western coasts of India and Sumatra as the main sources of NLOGs, which then drift towards the oceanic basin.NLOGs originating from the Mozambique Channel feed the WIO (West of SCTR, Somalia and South of Arabian Sea regions), while the Bay of Bengal and the western coasts of India and Sumatra, feed the Eastern and Southern IO regions.

Seasonal variations of NLOG density
Fig. 5 illustrates the seasonal variations in NLOG densities.In some regions (Arabian Sea, Eastern IO, SCTR and Somalia), these variations were marginally influenced by the type of scenario, indicating that the seasonality was mainly driven by the seasonality of the surface currents rather than by that of NLOG inputs.Density maxima were observed in the Arabian Sea between February and May, in the Eastern IO between August and November, in the SCTR between March and April and in Somalia in August (Fig. 5A,C,F&G).However, small differences between   scenarios were still noticeable.For example, in the Eastern IO, the density maximum occurred from July to November depending on the scenario (Fig. 5C).Similarly, in Somalia, the CL scenario showed an earlier maximum compared to other scenarios (Fig. 5G).On the other hand, in the Bay of Bengal, Indonesia, Mozambique and Southern IO, the seasonality varied depending on the considered scenario (Fig. 5B,D,  E&H).In the Bay of Bengal, scenarios considering the river forest cover displayed a strong seasonality, with a density peak in May (RC and R&CC), while accounting for precipitations partially compensated for this seasonality and resulted in a smaller density peak later in June-July (CCp and RCp, Fig. 5B).Noticeably, in the Bay of Bengal, the Ganges river constitutes by far the main source of NLOGs.The strong seasonality observed in scenarios RC and R&CC could be due to the reverse of currents in the area, which provokes the retention around May and the stranding and/or exit of particles originating from the Ganges in other periods of the year.In the Indonesia region, scenarios accounting for precipitations (CCp and RCp) predicted a maximum density in the first half of the year, which is in agreement with the higher precipitations observed in the area from November to April (Figure S1 in Supplementary Materials).However, considering rivers led a maximum density being observed in the second half of the year in the Indonesia region (RC and R&CC, Fig. 5D).In the Mozambique region, very few seasonal variations were observed except in scenarios considering the precipitations, where the maximum density occurred in February-April and in April-May for the CCp and RCp scenarios, respectively (Fig. 5E).

Trends of NLOG numbers in the Indian Ocean
Fig. 6 depicts the variations in simulated NLOG density anomalies, estimated over the entire oceanic basin, from 2000 to 2019.Among all scenarios, only the CL scenario exhibited a significant positive trend (p.value 4.4 × 10 − 2 , Table 2).For CC and CCp scenarios, there was a nonsignificant increasing trend (p.values 6.3 × 10 − 2 and 2.8 × 10 − 1 respectively).However, when considering the river forest cover (RC, RCp and R&CC), the mean NLOG density showed a non-significant decreasing trend (p.values 6.6 × 10 − 1 , 8.8 × 10 − 1 and 5.2 × 10 − 1 respectively).In the CL scenario, where the total weight applied to released particles remained constant over the years (Figure S2 in Supplementary Materials), the slightly positive trend must originate from changes in currents that could lead to higher particles retention within the oceanic basin.Including a decreasing coastal forest cover over the years did not compensate for this trend due to surface current changes, resulting in the positive but non-significant trends observed in the CC and CCp scenarios.However, when considering the forest cover associated with river basins (RC, RCp), the previously observed increasing trend was reversed, although it remained non-significant.This reversal might be attributed to loss of forest cover over the study period, but also to changes in surface currents as the decrease of the total number of weighted particles released was greater for the CC scenario than for the

Discussion
This study provides the first-ever estimations of relative densities of natural floating objects (NLOG) at the scale of the Indian Ocean, offering insights into their seasonal and long-term variability.These estimations are an important step towards better understanding the impact of human activities on this important component of marine habitats.All tested scenarios indicate that the areas with highest NLOG densities are the Bay of Bengal, the Mozambique Channel and the Eastern Arabian Sea.Considering either the coastal forest cover or the forest cover associated with rivers did not induce any change in the goodness-of-fit of the different scenarios, when comparing the predicted NLOG abundances with those derived from observers data in the WIO.All scenarios suggest no decreasing trend in NLOG densities across the entire IO, despite the observed loss of forested lands globally (Figure S2 in Supplementary Materials, IPBES, 2019).Instead, our results suggest that the loss of coastal forest was compensated by a change in surface currents.This change resulted in a higher proportion of NLOGs being retained in the IO, whereas they would have drifted to adjacent oceans earlier in the study period.Most areas did not display any significant trend over the study period, except for the Arabian Sea in the CL and CC scenarios (corrected p.values 3.7 × 10 − 2 and 1.2 × 10 − 2 respectively; Figures S3&S4 and Table S1 in Supplementary Materials).However, although non-significant, the slope coefficients in some areas were negative whatever the scenario (e.g.Eastern IO, Somalia and Indonesia), suggesting that potential decreasing trends in these areas might be observed over longer study periods.Overall, these findings highlight the complex and region-specific dynamics influencing NLOG densities in the Indian Ocean.Van der Stocken et al. (2019) conducted simulations on the global dispersion of mangrove propagules and found high densities in the Mozambique Channel, Bay of Bengal and in the Eastern Arabian Sea.The global density patterns obtained in our study also align with Lagrangian simulations performed to determine the global distribution of plastics debris or microplastics (Van Sebille et al., 2015;Viatte et al., 2020).However, these studies on plastic distribution identified an accumulation area in the eastern southern IO, which was not present in our simulations.This discrepancy could be attributed to difference in drift time used in the simulations.In Van Sebille et al. (2018) and Viatte et al. (2020), simulations were carried out with much longer drift times, ranging from 20 to 50 years, emphasizing the importance of carefully considering particles lifetimes (Crochelet et al., 2020;Stelfox et al., 2020).Considering that estimations of driftwood lifetimes range from a few days to 1000 days (Thiel and Gutow, 2005) and previous simulations showed little effect of drift time on our simulations (Dupaix et al., 2021a), we are confident that simulating NLOG drift for 360 days was appropriate.Confirmation of these simulated drift times could be achieved by conducting field-based estimations of NLOG drifting times through the study of fixed organisms and wood degradation.However, such estimations are challenging due to the dependence of wood colonization by fixed organisms on the richness of the crossed zones, and the variability of wood degradation based on species characteristics (Russell et al., 2014).Additionally, limited information exists on the tree species found at sea, further complicating such estimates.Alternative methodologies to determine drift times could involve the use of backward-intime lagrangian simulations using NLOGs observed at sea as release locations.Furthermore, the duration for which echosounder buoys, deployed on both DFADs and NLOGs by purse seine fleets, remain active, could also provide insight into the drifting time of NLOGs.However, echosounder buoys are deployed on NLOGs which are already at sea and can be deactivated remotely, making them suitable only as a lower bound for drift time estimates.
When comparing the results obtained through the simulations with observers data, most scenarios apart from the CL performed well.
However, the goodness-of-fit of estimated NLOG abundances was strongly driven by high NLOG abundance indices recorded in the Mozambique Channel (Fig. 3).Improved validation of the best performing scenarios could be obtained by including more field data, specifically from areas and/or time periods beyond those present in the observers data used in this study.The inspection of seasonal patterns (Fig. 5) allowed to highlight further specific features related to each weighting scenario.Industrial purse seine fleets in the IO primarily fish in the Mozambique Channel from March to May (IOTC, 2022d), mainly targeting tuna associated with natural floating objects since the early days of the fishery (Fonteneau, 2003;IOTC, 2022d).This seasonal behavior of the fleet may indicate that scenarios CCp and RCp can best reproduce the expected seasonal patterns of NLOG density in this area, highlighting the need to take precipitations into account.However, it remains unclear whether the absence of NLOGs in the Mozambique area during the rest of year explains the lack of purse seine fishing activity or if other oceanographic factors, such as the upwelling in the Somalia area during the summer monsoon (July-August; Schott et al., 2009), redirect fishers to different areas.In the Maldives, pole-and-line fishers target NLOG-associated tuna from December to April and in October Jauharee et al. (2021).Our findings with all scenarios are consistent with this seasonal patterns in the Maldivian fishery during the first part of the year (Dec-Apr, Fig. 5A).However, no scenario suggest any increase of NLOG density in the Arabian Sea in October.It is essential to note that the Arabian Sea area used in this study is much larger than the waters surrounding the Maldives.When examining quarterly simulated density maps, the scenario that aligns best with Jauharee et al. ( 2021)'s results would be the RCp scenario (Figure S5 in Supplementary Materials).The simulations presented here can be complemented by more localized studies in areas where fishers use NLOGs, relying on Local Ecological Knowledge to improve the proposed scenarios.
The significant modifications of pelagic surface habitat caused by DFADs have led scientists to hypothesize that DFADs could indirectly impact tropical tunas and act as an "ecological trap" (Marsac et al., 2000;Hallier and Gaertner, 2008).An ecological trap occurs when individuals select habitat that no longer offer suitable conditions due to anthropic changes, negatively affecting their fitness and potentially impacting the population (Battin, 2004;Gilroy and Sutherland, 2007;Swearer et al., 2021).The indicator-log hypothesis suggests that tuna would associate with NLOGs to find biologically productive areas, as NLOGs originate from rivers and accumulate in rich frontal zones (Fréon and Dagorn, 2000;Castro et al., 2002).The ecological trap hypothesis applied to DFADs and tropical tuna proposes that DFADs could alter the density and distribution of FOBs, potentially leading to the retention or transport of tuna to poorly productive or unsuitable areas Marsac et al. (2000).If NLOGs densities decrease in historically high-density areas, and if these areas are still productive, it could exacerbate this ecological trap.However, the scenarios developed in this study (CC, RC, CCp, RCp and R&CC) predicted no significant trends of NLOG numbers in the Bay of Bengal and in the Mozambique Channel, the areas with highest simulated NLOG densities.
The increase in FOB density induced by DFADs may have other potential indirect ecological impacts on tuna.It could lead to increased time spent by tuna associated with FOBs and disturb their schooling behavior (Fréon and Dagorn, 2000;Pérez et al., 2020;Soria et al., 2009).The meeting-point hypothesis is another hypothesis explaining the associative behavior of tuna, positing that tuna use FOBs to facilitate encounters with conspecifics and promote schooling behaviour (Fréon and Dagorn, 2000).An increase of FOB density could disperse tuna among FOBs and hinder the formation of large schools (Capello et al., 2022;Sempo et al., 2013).An increase in NLOG density despite forest cover loss (as suggested by the CC and CCp scenarios) could potentially compound the impacts of DFADs on schooling behavior.
FOBs play a crucial role in marine species' habitat, benefiting a wide range of species from fixed organisms to large pelagic predators (Fréon and Dagorn, 2000;Thiel and Gutow, 2005).In recent decades, human activities, such as fishing, plastic waste, logging and agriculture, have significantly altered their distribution (Dupaix et al., 2021a;Thiel and Gutow, 2005;Van Sebille et al., 2015).While large plastic debris distribution is well-studied (Mendenhall, 2018), the impacts of DFAD fishing and other human activities on FOB distribution remain relatively unknown.Studies have mainly focused on the WIO and Western and Central Pacific Ocean, considering NLOGs as a reference, and have shown that DFADs strongly increase the number of FOBs at sea (Dagorn et al., 2013a;Dupaix et al., 2021a;Phillips et al., 2019).However, the impact of other human activities like agriculture or logging (causing deforestation) has not been extensively studied.This study, showing limited influence of human activities on NLOG numbers but strong seasonal variations, is an initial step towards better understanding this important component of marine habitats.
This study aims to establish a baseline understanding of the respective contribution of currents variations and deforestation on the distribution of floating objects in the Indian Ocean, focusing on a wellsampled period (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007)(2008)(2009)(2010)(2011)(2012)(2013)(2014)(2015)(2016)(2017)(2018)(2019).However, the relatively short time frame considered in this study does not allow for a clear separation between the influences of climate change and natural climate variability on surface currents trends.The methodology presented here could be adapted to partly address the impact of human-induced climate change on the future distribution of NLOGs.This adaptation would involve using future flow fields from Earth System Models (ESMs) rather than current conditions.However, several challenges complicate this approach, including the lack of reliable future deforestation projections (Teo et al., 2023) and the coarse horizontal resolution of ESMs, which often fail to adequately capture meso-scale eddies.Moreover, ESMs exhibit significant biases in their representation of present-day currents, which could undermine the reliability of any conclusions drawn about future scenarios (Hewitt et al., 2020;Zhang and Wang, 2023).Despite these limitations, evaluating the impact of future flow evolution on NLOG distribution could still be possible at a coarser resolution than what was achieved in this study, and under the assumption that deforestation rates remain similar with current observations.However, such an analysis remains beyond the scope of our current work and is left for future studies.
This study presents first-ever simulations assessing the long-term impacts of terrestrial human activities on a crucial natural component of the pelagic surface habitat.Our results not only provide estimates of NLOG density across the entire IO but also enable more accurate local estimations of NLOG densities and variations than previous studies based on fisheries data (Dupaix et al., 2021a).These local estimates, which can still be further refined, can be invaluable for fisheries management and ecological studies aiming to understand the effects of DFAD fisheries on tropical tuna and other species associating with FOBs.The simulations reveal diverse trends across regions, although many are not statistically significant (see Figures S3&S4 in Supplementary Materials), warranting further exploration through Local Ecological Knowledge in areas where fisheries rely on driftwood to enhance their tuna catches (Western Indian Ocean, Indonesia, Maldives; Moreno et al., 2007;Jauharee et al., 2021).The different scenarios, based on coastal forest cover and/or river associated forest cover, suggest no significant increase in average NLOG densities in the ocean.However, these findings underscore the limited understanding of this critical element of pelagic species habitat and the need to assess habitat modifications on an even longer term to determine climate change impacts.Therefore, there is pressing need to intensify monitoring efforts for pelagic species habitat and raise awareness about potential indirect impacts of habitat modifications on tuna and other pelagic species (Dupaix et al., 2024).These impacts should be further investigated and better considered by regional fisheries management organizations to ensure the sustainable management of marine resources.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.
Fig. 1.Map of the different areas considered in the study.IO: Indian Ocean.SCTR: Seychelles-Chagos Thermocline Ridge.
Fig. 2. Schematic representation of the weighting scenarios used in the study.The release point is shown in red, forest cover (in % of the cell surface) is represented in green, the land is represented in grey and the sea in blue.(A) CL: total surface of the coastline buffer (white area delineated in black); (B) CC: sum of the forest cover contained in the coastline buffer; (C) RC: sum of the forest cover in the river buffer multiplied by the mean river discharge at river mouth; (D) CCp: CC multiplied by the precipitations at the release point; (E) RCp: RC multiplied by the precipitations at the release point; (F) R&CC: sum of the forest cover in the union of the coastal and river buffers.

Fig. 3 .
Fig. 3. Comparison between simulated and observed NLOG abundance for each weighting scenario.y-axis: NLOG abundance index obtained from simulations, corresponding to the number of simulated NLOGs per 2 • cell.x-axis: NLOG abundance index obtained from observers data, corresponding to the number of observed NLOGs per day of observation.CL: Coastline Length; CC: Coastal forest Cover; RC: River forest Cover multiplied by river discharge; CCp: Coastal forest Cover multiplied by precipitations; RCp: River forest Cover multiplied by river discharge and by precipitations; R&CC: River plus Coastal forest Cover.SCRT: Seychelles-Chagos Thermocline Ridge.On each panel, r.sq: R 2 of the linear regression; p: p-value of the Student's t-test testing if the slope coefficient of the linear relationship significantly differs from zero.Because values from observers data and from simulations correspond to different indices of NLOG abundance, this figure does not allow for quantitative comparisons.

Fig. 4 .
Fig. 4. Simulated density of NLOGs averaged over 2000-2019 for the different weighting scenarios.Maximum value was set to 1 and values below 10 − 4 were discarded.CL: Coastline Length; CC: Coastal forest Cover; RC: River forest Cover multiplied by river discharge; CCp: Coastal forest Cover multiplied by precipitations; RCp: River forest Cover multiplied by river discharge and by precipitations; R&CC: River plus Coastal forest Cover.

Fig. 5 .
Fig. 5. Average variations of mean simulated density of NLOGs per area.The densities are scaled for each scenario and area.Note that the scales do not allow for direct comparison between different areas.CL: Coastline Length; CC: Coastal forest Cover; RC: River forest Cover multiplied by river discharge; CCp: Coastal forest Cover multiplied by precipitations; RCp: River forest Cover multiplied by river discharge and by precipitations; R&CC: River plus Coastal forest Cover.IO: Indian Ocean; SCTR: Seychelles-Chagos Thermocline Ridge.

Table 2 Results of the linear regressions performed between the NLOG density anomaly (A) and time, represented
on Fig. 6, for the different weighting scenarios.Significant slope coefficients are in bold.Δt: maximum time lag such that Pearson's correlation coefficient ρ(A t , A t− Δt )⩾0.2.CL: Coastline Length; CC: Coastal forest Cover; RC: River forest Cover multiplied by river discharge; CCp: Coastal forest Cover multiplied by precipitations; RCp: River forest Cover multiplied by river discharge and by precipitations; R&CC: River plus Coastal forest Cover.