One of the least disturbed marine coastal ecosystems on Earth: Spatial and temporal persistence of Darwin’s sub‐Antarctic giant kelp forests

Marine habitats and their dynamics are difficult to systematically monitor, particularly those in remote locations. This is the case with the sub‐Antarctic ecosystem of the giant kelp Macrocystis pyrifera, which was already noted by Charles Darwin in his accounts on the Voyage of the Beagle and recorded on the nautical charts made during that expedition. We combined these and other nautical charts from the 19th and early 20th centuries with surveys conducted in the 1970s and 1980s and satellite detection algorithms from 1984 to 2019, to analyse kelp distribution through time and the factors that correlate with it.


| INTRODUC TI ON
There is an increasing awareness of the need to protect the last remaining terrestrial forest ecosystems with limited anthropogenic impact (Watson et al., 2018). Intact forests on continents are periodically monitored by remote sensing systems to better understand changes in the quantity and quality of these ecosystems Potapov et al., 2008Potapov et al., , 2017. In contrast, there has been less attention to the status of intact or pristine underwater ecosystems, particularly kelp forests. Kelps are considered 'foundation' species, as their presence increases the number and diversity of species by the control of the physical environment in the water column, also known as physical engineering (Lamy et al., 2020;Miller et al., 2018). In many regions of the planet, kelp forests are degrading under multiple stressors such as ocean warming, pollution or overharvesting (Krumhansl et al., 2016;Wernberg et al., 2019).
A missing area in the recent account of the global trends of kelp forests is the sub-Antarctic region (Krumhansl et al., 2016), which paradoxically was one of the first kelp ecosystems mentioned in natural history studies. In Voyage of the Beagle, Charles Darwin commented on the conspicuous forests of the giant kelp (Macrocystis pyrifera) surrounding Tierra del Fuego, saying 'I can only compare these great aquatic forests of the southern hemisphere, with the terrestrial ones in the intertropical regions. Yet if in any country a forest was destroyed, I do not believe nearly so many species of animals would perish as would here, from the destruction of the kelp' (Darwin, 1845).
These tenets forged the pillars of kelp ecology (Miller et al., 2018).
Darwin was not the first to record the giant kelp: the first written description of M. pyrifera was made by Juan Ladrilleros during his expedition through the Strait of Magellan in 1557-1558, which illustrates how prevalent this alga was in the region (Martinic Beros, 1982).
The combined extent of the giant kelp canopies of M. pyrifera on the sea surface (kelp forests hereafter) detected with satellite imagery of kelp forests at Isla de los Estados (Argentina, 54°47′S 64°15′W) experienced little changes over the last four decades (Friedlander et al., 2020). Furthermore, kelp forests in a nearby fjord (Yendegaia, Beagle Channel, Chile, 54°52′S 68°44′W) showed physiological tolerance to the indirect effects of glacier melting such as reduced light availability and increased turbidity (Huovinen et al., 2019;Palacios et al., 2021). Contrary to other kelp regions suffering thermal stress, the sub-Antarctic currents at present do not show signs of tropicalisation (Vergés et al., 2014).
The sub-Antarctic kelp forests in this region are found along a high gradient of geographical coastal diversity, from the Fjordlands of Southern Chile, to the extensive coastal shelf of the Falkland Islands and the rugged and heavily glaciated landscapes of South Georgia.
These coastlines have been subject to direct and indirect influences of humans for thousands of years. Indigenous peoples in the Chilean Fjordland region have utilised M. pyrifera and other local seaweed species for food and medicine over at least 14,000 years BP (Dillehay et al., 2008). The Falkland Islands and South Georgia, without human occupation before European arrival, have seen successive waves of economic activity since the mid-18th century, such as livestock grazing and seal and whale hunting (Alonso Marchante, 2014;Bridges, 1988;Palomares & Pauly, 2015), but the impact of these activities on kelp forests is unknown. At present, the spread of the salmon farming industry in Chile has had a devastating impact in the northern Patagonia coastal region, and is now threatening the coastal and fjord ecosystems in the southernmost regions of the country (Friedlander et al., 2018;Quiñones et al., 2019). Still, the nature and extent of the impacts of this activity on kelp forests are poorly understood.
In the present study, we aim to characterise key aspects of the biophysical characteristics of sub-Antarctic giant kelp forests, and to observe the long-term patterns in their distribution as inferred from observational records. By addressing these questions, we aim to develop a broad biogeographical conceptual model of the environmental adaptations and long-term trends in these unique ecosystems.

| Characterising the abiotic habitat
Abiotic variables such as depth, complexity of the rocky substrate and wave stress are essential in determining kelp presence and persistence (Young et al., 2016). However, detailed bathymetric data of the sub-Antarctic are sparse, which precludes a comprehensive overview of the ecosystem. For this reason, this paper will analyse the abiotic niche of kelp forests based on the geomorphology of the coastline, exposure to the ocean, and sea surface temperatures.
Geomorphology informs on the general patterns of rocky nearshore ecosystems. In glacial and paraglacial regions with parallel glacial history such as in the Chilean Fjordland and South Georgia (Hodgson et al., 2014), coastal and submarine geoforms like lateral or terminal moraines, erratic rocks and sills, may provide available rocky substrates at the borders of trough valleys or U-shaped fjords. In contrast, the Falkland Islands experienced limited glacial presence (Clapperton, 1990) with a large shelf, which provided an ecological refugium during the Pleistocene for some marine species (González-Wevar et al., 2012;Leese et al., 2008). Following this premise, similar geomorphological processes may be linked to similar coastal configurations and spatial patterns of kelp forests.
Consequently, if the geomorphological origin of a rocky substrate has been under a slow geological pace of marine sedimentation, the substratum for kelp forests should remain persistent within modern records, unless direct disturbance led to local extinctions.
Oceanographic variables such as exposure to currents can also support habitat persistence or shifts. Kelp generally grows best under high hydrodynamic regimes due to the increased flow of nutrients and inorganic carbon (Parnell et al., 2010;Wernberg et al., 2019). On exposed coastlines, kelp attains higher biomass and growth rates in open waters with more water movement, as observed by Van Tussenbroek (1989b) in the Falkland Islands and by Dayton (1985b) at Isla de Los Estados and southern Tierra del Fuego.
However, we have limited evidence of kelp forest sizes in exposed or sheltered areas in sub-Antarctic kelp ecoregions, or whether strong storms are able to remove kelp forests in the most exposed areas.
In addition, sea surface temperature (SST) records can help to confirm temperature ranges for kelp persistence. At latitude 42°S, north of the Chilean Fjordland, SST >15-17℃ was observed to increase the mortality of annual populations of kelp sporophytes (Buschmann et al., 2014). It is not yet clear whether sub-Antarctic SST records for the last decades are fully within the optimal thermal tolerance of kelp.

| Long-term persistence
Estimating long-time resilience of kelp forest ecosystems in these three different ecoregions is challenging, considering the limited temporal range of ecological studies in this region (Barnes et al., 2006;Castilla & Moreno, 1982;Dayton, 1985b;Van Tussenbroek, 1989c). Instead, the concept of persistence (sensu Connell & Sousa, 1983) refers to a population that did not go extinct, or if it did go extinct locally or regionally, it recolonised during a given period of time in a given area.
To gain perspective on the long-term persistence of sub-Antarctic kelp forests, we compared the first detailed cartographical records made during the voyages of the HMS Beagle and Adventure in 1826-1836 (King et al., 1836) and similar early charts from South Georgia made in 1882-1931 with surveys and remote sensing imagery from the second half of the 20th century up to the present.

| Study area
Our area of study encompasses the western side of South America from Peninsula Tres Montes at latitude 47°S to Cape Horn at latitude 56°S; and from longitudes 35°W to 76°W along the Chilean coastline to South Georgia (Figure 1a). This area includes the marine ecoregions of Channels and Fjords of Southern Chile, the Falkland Islands (Malvinas), and the island of South Georgia (excluding the South Sandwich Islands; Spalding et al., 2007), and is oceanographically connected with the Pacific and the Atlantic oceans by the Cape Horn Current, the Falkland Current and the Circumpolar Antarctic Current around the South Georgia Shelf (Brandon et al., 2000). The Channels and Fjords ecoregion is connected by a network of channels (called Channels hereafter, described in González et al., 2013) and the Strait of Magellan that separates the South American continent from Tierra del Fuego. The Falkland Islands ecoregion is bisected by Falkland Sound, which separates the West and East Falkland Islands. In this study, we consider the Patagonian Shelf ecoregion that includes Puerto Deseado (Argentina) and the Chiloense ecoregion (Chile, latitude 41-46°S) for historical analysis only because the kelp forests in the former occur within a region of large tidal ranges, which may render them invisible to satellite sensors (Mora-Soto et al., 2020), while in the latter kelp forests are less stable, with weaker attachments to the substrate (Darwin, 1845;Dayton, 1985b). March. KF highlights the presence of kelp canopies through the difference of the red-edge (Band 6, 740 nm) and red reflectance (Band 4, 665 nm) of the Sentinel-2 spectral bands at 20 m resolution. The algorithm further masks values to discard other land or ocean elements, except for Ulvophyceae (a class of green algae) in intertidal areas, due to their similar reflectance. We applied this algorithm in seasonally-averaged images to increase the detection of kelp forests, as they often otherwise remain invisible due to heavy cloud cover, potentially persistent swell, or very small canopies over the study area. Subsequently, all seasons were averaged to calculate a representative extent of kelp canopy for the total range of time.

|
We aimed to obtain a representative sample of kelp forests per ecoregion. In QGIS (version 3.14.15), a random multipoint layer was created within the limits of each ecoregion. If any of those random points were in a range of 500 m near a kelp forest as identified by the KF algorithm, the forest was selected as a sample, and the process was repeated until the number of samples was close to 100 per ecoregion. The final number of sites were 108, 114 and 87 for Falkland Islands, Channels and South Georgia, respectively. For each of these kelp forests, a polygon was delineated covering the maximum extent of the observed canopy, including an inland range of 500 m if the forest was next to a coastline. Sites were doublechecked with visible canopies on very high-resolution Google Earth true colour images to avoid false positives (Mora-Soto et al., 2020).
If the Google Earth image had low resolution, or was obstructed by cloud cover, we used KF pixels that showed a permanent overlay of pixels identified as kelp in concentrated areas, avoiding estuarine or beach areas. The polygon drawing scale ranged from 1:5,000 to 1:20,000.

| Statistical analyses
Kelp forests were characterised by their coastal geospatial attributes (Table 1) A Kruskal-Wallis (1952) rank sum test was applied to determine whether kelp forest size was significantly different along the categorical variables exposure to the ocean, geoform, ocean current and aspect per each ecoregion. Significant p-values were adjusted and compared with the Benjamini and Hochberg (1995) method.
Additionally, Kaiser-Meyer-Olkin (KMO) and Bartlett's Tests (R version 4.0.2, package: 'psych'; Revelle & Revelle, 2015) were used to assess the adequacy of each factor. Spearman's rank-order correlation (Spearman, 1987) was used to identify whether kelp forest size was correlated with the continuous variables of adjacent land slope, longitude and latitude. Statistically significant categorical variables were included in a conditional inference tree (Hothorn et al., 2006; R package: 'partykit'; Hothorn & Zeileis, 2015) to investigate the association between forest size and geospatial variables, using 80% of all forests of the three bioregions combined (N = 254) for calibration and the remaining 20% (N = 55) as an evaluation dataset.

| Decadal trends of SST (1981-2020)
We obtained long-term daily SST records surrounding sub-Antarctic kelp forests. SST trends were analysed using the NOAA CDR OISST: Optimum Interpolation Sea Surface Temperature product (Reynolds, Banzon, & NOAA, 2008) employing the Google Earth Engine API.
As this imagery dataset has a coarser scale (0.25 arc degrees) than the one used for delimitating kelp forests, we selected two repre- (3) South Georgia ecoregion: Annenkov Island at the western side of the island; Stromness Bay at its eastern side.

| Long-term kelp canopy extent monitoring
To examine the variability of kelp forest extent over the longest observational time frame possible, we made use of historical nautical charts, airborne (from aircraft and unmanned aerial vehicles [UAVs]) and satellite imagery, spanning the period 1829-2020. (c) Historical kelp surveys (1972)(1973)(1974)(1975)(1976)(1977)(1978)(1979)(1980)(1981)(1982)(1983)(1984)(1985)(1986)(1987) The first ecological and ecophysiological studies on sub-Antarctic kelp were conducted in the 1970s and 1980s (Dayton, 1985b;Moreno & Jara, 1984;Van Tussenbroek, 1989a, 1989b, 1989c, 1989d, 1993. We located the surveys on the map and verified whether satellitedetected kelp forests exist in the vicinities of the locations indicated in these studies (either in maps or registered coordinates), using the same layered procedure as mentioned above. Finally, we compared the charts with data on human impact using geographical records of historical and current land use from Geographical Information  resolution (Nijland et al., 2019); and the KF derived from Sentinel-2 (note that KF cannot be computed from Landsat due to lack of adequate reflectance bands). As the available Landsat 5 TM images from 1984 to 2011 were few in comparison to Sentinel-2 images from 2016 to 2019, all images were averaged annually. Only Sentinel-2 images were available for South Georgia.
The following protocol was applied to the annually-averaged Kelp forest size is negatively correlated with land slope, although this relationship is not as strong as with latitude or longitude (ρ = −0.14; The significant variables are used in a conditional inference tree to predict the size of the kelp forests ( Figure 3)

| Multi-decadal SST trends
The long daily record (1981-2020) of NOAA SST data shows stationary trends for all the representative pixels (Dickey-Fuller test = stationary; p = 0.01 for all the sites; Figure S1.2). Minimum SST record is -1.74℃ at Stromness, South Georgia, while the maximum recorded temperature is 12.77℃ at Pebble Island in the Falkland Islands (Table   S1.2). No temperatures above the thermal range of kelp survival are recorded between 1981 and 2020. This record demonstrates that SSTs have been stable during the last three decades in the study area.

| Long-term persistence
The complete coverage of UKHO Nautical Charts and satellitedetected kelp forests can be found in the following online resource:

| DISCUSS ION
In this study, we demonstrate that in addition to bathymetry used in other studies to locate potential sites of kelp forests (Costa et al., 2020), currents, coastal geomorphology and exposure can help to successfully predict broad spatial kelp patterns, in line with studies on sub-Antarctic benthic fauna (Barnes et al., 2006). In the words of Dayton et al. (1994, p. 90), 'geological history and oceanographic processes are the warp and woof of the biological understanding of any marine habitat'. Although the predictive value of each explanatory variable was low (KMO > 0.5), their combination generated a robust model that successfully predicted expected kelp forest sizes in the region. Relatively small patches of kelp are abundant in bays and rocky islets in South Georgia, in contrast to the large extent of kelp forests in the Falkland Islands. This may reflect paraglacial patterns (Hodgson et al., 2014) of fjords and moraines and restricted glacial history and extensive shallow shelf (Clapperton, 1990) (Beaton et al., 2020;Dayton, 1985b;Van Tussenbroek, 1989d). We illustrate this general conceptual model of kelp forest habitats in Figure 4. The patchy response of M. pyrifera to a variety of abiotic (geomorphology, SST) and biotic (competitive interactions) between species of kelps presents analytical challenges;

TA B L E 3 Summary of the total of kelp forests drawn on the UKHO nautical charts and satellite-detected kelp
understanding their differentiation/integration in response to contemporary environmental pressures would benefit modelling and monitoring of kelp forests in all regions.
In the ecoregions of Channels and Fjords and Falkland Islands, the dominant orientation of larger kelp forests towards the oceanic currents corroborates previous findings (Dayton, 1985b;Van Tussenbroek, 1989b, 1993, supporting the concept that water motion and turbulence play a primary role on nutrient assimilation for kelp (Dayton, 1985a). This prevalent quality ends when maximum wave energy limits kelp canopy, resulting in a hump-shaped relationship between kelp forest size and wave energy (Young et al., 2016).
On the other hand, in areas of very low energy, that is, the fjords of South Georgia and the internal Channels of Southern Chile, kelp forests are strongly influenced by freshwater coming from glaciers and snow fields and rainfall. Previous work in the Beagle Channel has shown that M. pyrifera has adapted their photosynthetic activity to this paraglacial environment, including overshadowing due to high turbidity  or adopting a loose-lying form on the bottom and small pneumatocysts, undetectable from the surface (Gerard & Kirkman, 1984;Van Tussenbroek, 1989c).
Therefore, kelp forests in fjords can persist despite being considered stressed (Dayton, 1985b;Mora-Soto et al., 2020), even with the increased glacial melting rates of the last three decades (Meier et al., 2018).
In the current global context, kelp forests are severely affected by marine heatwaves, ocean acidification, increased frequency of storms and over-harvesting (Krumhansl et al., 2016;Wernberg et al., 2019). However, kelp persistence seems to follow a neutral relationship with SST at higher latitudes, i.e., bull kelp Nereocystis luetkeana in Oregon (Hamilton et al., 2020). Furthermore, our SST analysis confirms that the currents near our study areas show no trends of tropicalisation, a major cause of thermal stress and shifts on kelp distributions in other regions, i.e., Western Australia (Wernberg et al., 2013), or Tasmania, with a sharp decline of kelp canopies since the year 2000 (Butler et al., 2020). The low SST ranges in South Georgia, located south of the Antarctic Polar Front (Moore et al., 1999), could mean that this ecosystem may cope well with polar temperatures, which might be a signal for potential colonisation of the Antarctic, but the small sizes of their canopies might also mean they are close to their lower temperature limit. On the other hand, acidification could have a negative impact on the development of calcareous organisms that feed on kelp, but not the kelp itself (Brown et al., 2014). An increased frequency of subpolar westerly winds and strong storms is affecting the distribution of some species like the bull kelp Durvillaea antarctica (Fraser et al., 2018). Although SSTs have remained stable and the present study supports the long-term stability observed in this marine ecosystem (Friedlander et al., 2018(Friedlander et al., , 2020, more local long-term studies on persistence after storms are needed to confirm this region as a climatic refugium for kelp. Our for this could be due to local areas of higher photosynthetic activity associated with phytoplankton (Saggiomo et al., 2011) or to floating kelp (Wichmann et al., 2012).
The remote sensing approach used in this study was based on previous studies that used KF (Mora-Soto et al., 2020) and NDVI (Nijland et al., 2019), with high levels of overall accuracy at detecting kelp canopy extent in remote locations. This study, however, does not include ground-truthed estimations of kelp canopy biomass, frond density and kelp density linked with satellite imagery. Those variables could help to explain the environmental dynamics of the forests further, but would need a significant longterm sampling effort through SCUBA diving, as done in previous works developed in the Southern California coastline Cavanaugh et al., 2011Cavanaugh et al., , 2013. The baseline presented in this study could be expanded to a similar kind of environmental research in the future.

| Anthropogenic changes and kelp persistence
A number of places in our area of study have experienced important landscape changes in the last two centuries caused by direct or indirect human interventions in the environment. Nevertheless, kelp forests have generally persisted in the same areas (Table S1.3). The main landscape alterations are the following: F I G U R E 4 A general model of the sub-Antarctic kelp forest in relation to coastal geomorphology. The area of a kelp forest depends on the available substrate originated by geological orogeny followed by glacial, marine, alluvial or organic weathering and/or sedimentation. Topographic slope classes: (a) Level to sloping; (b) steep; (c) vertical and over-hanging walls. Exposure to the ocean has an inverse correlation with the growth rate and positive correlation with the life span of the canopies (Van Tussenbroek, 1989a, 1989c a.  (Hayward, 1983) followed by a high rate of retreat since the 1990s (Cook et al., 2010) with an area loss of 0.88 ± 0.04 km 2 year −1 between 2003 and 2016 (Farías- Barahona et al., 2020). The kelp forests of Molkte Harbour are still in the same location despite the distance to the ice front (currently <10 km; Figure   S1.  (Royle, 1985), and in 1848 the city of Punta Arenas (Chile) was founded in the Strait of Magellan (Martinic Beros, 2002 (Glasser et al., 2004) that opened coastal spaces for conifers and moorlands (Villagran, 2001), whereas closedcanopy Nothofagus forests in Patagonia have been in place since ∼10,000 cal year BP (Moreno et al., 2019). Shell middens along the Beagle Channel indicate human occupation of the coastline during at least the last 6,500 years, with a stable marine biota (Estévez et al., 2001). A 14,000 year multiproxy record of seabird guano and tussac grass (Poa flabellata) in the Falkland Islands showed an abrupt establishment of P. flabellata and seabird colonies ~5000 year BP (Groff et al., 2020), which, in turn, could have meant massive nutrient inputs for kelp growth. In turn, seabirds like the kelp gull Larus dominicanus can forage on bivalves growing on kelp blades like Gaimardia trapesina (Friedlander et al., 2020;Hockey, 1988). Molecular studies have suggested a northern hemisphere origin of giant kelp, with a later dispersal to the southern hemisphere (Astorga et al., 2012;Coyer et al., 2001;Macaya & Zuccarello, 2010). Specifically, Coyer et al. (2001) indicate that multiple inter-hemisphere crossings occurred between 3.1 Mya and as recent 0.01 Mya. More specific research is needed to complement these convergent lines to estimate an age of sub-Antarctic kelp forests.
The observed persistence does not necessarily imply that sub-Antarctic kelp forests are not threatened by local anthropogenic factors. To the north of our study area, kelp forests are harvested with different levels of protection in central and northern Chile. This has a strong overall influence on kelp forest morphology and biomass (Gouraguine et al., 2021;Krumhansl et al., 2016;Vásquez, 2008).
The effect of marine traffic on kelp canopies in the region, particularly near the Strait of Magellan and the Cape Horn is a question that requires future research. The major impending threat to the region is probably the rapid expansion of salmon farming; however, overexploitation of fisheries stocks, destructive fishing practices and threats associated with climate change will all require proactive management actions to mitigate potential negative impacts . To sustain this ecosystem in the long term, it is imperative to develop specific plans for conservation, like the creation of marine protected areas (Friedlander et al., 2018(Friedlander et al., , 2020Rozzi et al., 2007), regulation of kelp harvesting, intensive salmon farming, destructive fishing practices, and pollution (Hinojosa et al., 2011;Iriarte et al., 2010), evaluation of carbon sequestration and ecosystem services (Bayley et al., 2021) as well as encouraging indigenous rights for the protection of the maritorio (seascape) in Patagonia.

ACK N OWLED G EM ENTS
We thank Nacho Juárez and Walter Huaraca (U. of Oxford), Giles Richardson ( app/view/beagl ekelp charts with the aim of encouraging scientists and the public to explore the layers interactively.
The datasets generated during the current study are available in the Beagle Kelp charts repository: https://doi.org/10.5061/dryad. bcc2f qzck.