Extending the conceptual model of river island development to incorporate different tree species and environmental conditions

Riparian vegetation survival and establishment in gravel‐bed rivers depends on the balance between vegetation growth and flood disturbance. We present four examples of vegetation and landform development in gaps (linear open spaces) between established islands and/or floodplain within a reach of the middle Tagliamento River, Italy. Gaps offer shelter to vegetation, supporting higher colonization success and different vegetation‐landform evolution pathways. Time sequences of aerial images track vegetation development over 30 years in the four gaps. In combination with the flood disturbance time series, we interpret vegetation dynamics and identify the fate of sexual and asexual reproduction strategies by observing vegetation expansion from lines of young plants and shrubs and from uprooted deposited trees and pioneer islands, respectively. Analysis of image sequences reveals common features across the four gaps that are generalized to extend a conceptual model of island development. Growing conditions, disturbance energy and time (window of opportunity) between major floods are the main controls on vegetation colonization. These vary among rivers, among reaches along the same river and locally, as in the investigated gaps, allowing different tree species with different life history traits (e.g., Populus nigra, Alnus incana) to engineer local river landforms in different and complementary ways. Although the conceptual model is inspired by observations on the Tagliamento River, consideration of species life history traits and the joint influences of growing conditions, disturbance energy and windows of opportunity provide a framework that may be applied to other temperate rivers where trees drive landform development.

Research over two decades on the Tagliamento River, N.E. Italy, has concentrated on such plant-physical environment interactions.
The physical roles of species from the Salicaceae family have been investigated, particularly Populus nigra L. (black poplar), which dominates islands and floodplain margins along the middle and lower reaches of the river. This research led to the proposal of a conceptual model of island development, whereby species from the Salicaceae family drive island development (Gurnell et al., 2001;Gurnell & Petts, 2002;Gurnell, Tockner, Edwards, & Petts, 2005). The conceptual model incorporates three pathways along which Salicaceae species may colonize the surfaces of river bars and initiate island development: (1) germination and growth of widely dispersed seeds when they are deposited at suitable germination sites; (2) germination and growth of seeds that accumulate in sheltered locations such as in the lee of wood piles and (3) sprouting of shoots and roots from deposited wood pieces or entire uprooted trees. However, in the high-energy context of the Tagliamento, Pathway (3) has been identified as the most likely to initiate island development with the rapid rooting and sprouting of flood-deposited trees and large wood playing a crucial role in stabilizing bar surfaces and trapping fluvial sediments, wood and plant propagules to construct small "pioneer" islands . Pioneer islands provide shelter for further vegetation development and sediment retention, and they may coalesce to form larger, building islands and extensions of the floodplain. Gurnell and Petts (2006) extended the conceptual model by incorporating the impacts of changes in flood unit stream power and also the depth and variability of the alluvial water table along narrowing and widening sections of a river. Such longitudinal changes in the balance between flood disturbance severity and the growth performance of young trees in response to the local groundwater regime explains observed longitudinal changes in the presence, spatial extent and persistence of islands within different reaches of the Tagliamento. In particular, Gurnell and Petts (2006) emphasized reach-scale variations in the potential of Pathway (3) to support pioneer and building island development following the deposition of uprooted trees and large wood pieces of Salicaceae species.
The "recruitment box" model (Mahoney & Rood, 1998) is relevant to Pathways (1) and (2) of the island development model. It couples (a) the seasonal production of seeds by riparian Salicaceae species with (b) synchronous seed dispersal by floods to suitable germination sites and then (c) rapid seedling growth promoted by post-flood declines in the water table during a period without significant flood disturbance. This model has been widely applied and much research has developed from it to explain the frequent presence of single species, single age cohorts of Salicaceae species bordering many rivers and, more generally, how such riparian systems function and can be conserved (e.g., Braatne, Jamieson, Gill, & Rood, 2007;Cooper, Merritt, Andersen, & Chimner, 1999;Foster & Rood, 2017;Kalischuk, Rood, & Mahoney, 2001;Lytle & Merritt, 2004;Scott, Auble, Dixon, Johnson, & Rabbe, 2013;Scott, Auble, & Friedman, 1997). However, growth from seedlings or small vegetative fragments do not appear to be a major driver of the first stages of tree colonization along the middle and lower Tagliamento main stem. Even in the upper part of the main stem, where sprouting of vegetative fragments deposited on the river bed is relatively rare, accumulations of large wood and/or boulders are crucial for providing shelter within which seedlings (notably of the locally co-dominant species A. incana) survive and initiate pioneer islands (Pathway (2), Gurnell et al., 2001). The relative lack of effectiveness of sexual reproductive Pathways (1) and (2) for initiating vegetation establishment and landform development in the middle and lower Tagliamento has been attributed to (a) slower early growth rates of seedlings in comparison with sprouts from vegetative propagules of varying size, particularly from uprooted, deposited trees (Francis, 2007;Francis & Gurnell, 2006;Gurnell, 2016;Moggridge & Gurnell, 2009), (b) frequent "flashy" flood disturbances capable of uprooting or burying seedlings and (c) the predominant occurrence of the largest floods in autumn and thus well beyond the spring season of seed production by P. nigra and other species from the Salicaceae family that are present along the river (Karrenberg & Suter, 2003). Point (c) is particularly important because seeds of the riparian Salicaceae have an extremely short period of viability (Gosling, 2007;Karrenberg & Suter, 2003). However, this is not such a limiting factor for dispersal and successful establishment of species that produce seeds with a longer period of viability or that reproduce asexually. For these species, flood timing does not strongly limit successful recruitment from propagules that are dispersed to suitable germination or sprouting sites, but the length of time without disturbance or "window of opportunity" (Balke, Herman, & Bouma, 2014) for growth following flood dispersal remains a major control on recruitment success.
Recently, Bertoldi and Gurnell (2020) reported on the potential contribution of A. incana (L.) Moench. (grey alder), a member of the Betulaceae family that produces seeds with a longer period of viability than the riparian Salicaceae, for river bed landform development in the middle reaches of the Tagliamento. They investigated broad spatio-temporal and topographic changes across the river bed associated with woody vegetation in general (dominated by P. nigra) and with A. incana in particular. Over the last two decades within a >7 km long study reach, A. incana has shown an expansion in cover; is mainly distributed in lines that broadly parallel the river's course; is located at lower elevations on the river bed than riparian woodland vegetation in general; and is associated with local aggradation of the river bed.
These observations indicate that in addition to general aggradation and expansion of wooded islands dominated by P. nigra, A. incana also appears to be driving island development locally along the margins of some channels, bars and islands. Additionally, Bertoldi and Gurnell (2020) concluded that an increase in cover of A. incana since 2000 is most likely explained by the species' presence at relatively lower elevations than vegetated areas in general and along landform edges that would be preferentially eroded by major floods. Therefore, the recent apparent increase in the cover of A. incana does not seem to relate to an extension of the species' geographical range but rather recovery of the species following the largest flood in the last 35 years in 2000. This conclusion is supported by Lippert, Müller, Rossel, Schauer, and Vetter (1995) and Karrenberg, Kollmann, Edwards, Gurnell, and Petts (2003), who record significant presence of A. incana in the middle reaches of the Tagliamento prior to the 2000 flood. In short, A. incana appears to be associated with landform building in specific locations where it may complement physical engineering by the dominant species P. nigra, and its importance may vary through time because of its greater susceptibility to removal during large floods.
Following from this previous research, we investigate the potential complementary role of A. incana (Betulaceae) to that of the dominant species, P. nigra (Salicaceae), in physically engineering the development of islands and floodplains along the middle reaches of the Tagliamento.
Like P. nigra, A. incana is a pioneer species that can rapidly colonize areas of bare ground. However unlike P. nigra, A. incana (a) releases seeds in the autumn (Wilson, Mason, Savill, & Jinks, 2018) that are easy to store (Gosling, 2007) with some evidence (Thompson, Bakker, & Bekker, 1997) that the seeds form a short-term persistent seed bank (i.e., can remain viable for at least 1 year), ensuring viable seeds are available throughout the year for redistribution by floods; (b) is less likely to reproduce asexually (as noted on the Tagliamento by Kollmann et al., 1999, andexperimentally by Francis, Gurnell, Petts, &, although flood damage may encourage sucker and root stump shoots in a similar manner to that reported for regrowth following coppicing (Rytter, 1996;Rytter, Sennerby-Forsse, & Alriksson, 2000;Wilson et al., 2018), and very occasionally some uprooted stumps may survive and sprout if deposited at suitable sites (authors' field observations) and (c) prefers moister, more nutrient-rich sites (Rytter, 1996). Thus, the two species show important differences in their life history traits and environmental requirements, which may allow some complementarity in their roles as physical ecosystem engineers.
In order to explore the potential physical ecosystem engineering We generalize the outcomes of the above investigations to extend the conceptual "island development model" described by Gurnell et al., 2001 andPetts, 2006. 2 | METHODS 2.1 | Investigative design Bertoldi and Gurnell (2020) present a spatio-temporal and topographic analysis of vegetation development across the three most heavily vegetated areas of the braid plain, totalling almost 1 km 2 area, of the river bed within an island-braided reach of the middle Tagliamento River (Figure 1a). To explore our three working hypotheses, we searched this surveyed area for gaps (linear spaces between established islands and or floodplain) that were oriented parallel to the braid plain (i.e., similar orientation to flood flows). We identified nine suitable gaps from which we selected three of contrasting width We assembled secondary data sources (river stage records, historical aerial images and airborne lidar data sets) and field observations for each gap. These supported definition of gap boundaries and underpinned reconstruction of vegetation and topographic development of the river bed (see Section 2.2).
The above analyses allow us to revisit our three working hypotheses and incorporate them into an extended conceptual model of island development (Gurnell et al., 2001;Gurnell & Petts, 2006). Only one of these flood events (June 1996) did not occur in Autumn but this was followed by another of similar peak magnitude in November 1996. The four later bankfull events (2008,2012,2017 2018) also all occurred in autumn or winter. The 4 year "window of opportunity" without any significant floods between the October 2004 and October 2008 floods is also relevant to the following analyses. Later "windows" (2008)(2009)(2010)(2011)(2012)(2012)(2013)(2014)(2015)(2016)(2017) were not investigated because, following the above criteria, they are too recent to be investigated using the available secondary data sources, and, Rectified aerial images (1986,1993,1997,2003,2005,2010,2011,2013  This provided a detailed pre-flood record of bed morphology, the distribution of woody vegetation and the main tree species, and also the locations of a sample of alder individuals taller than 4 m, allowing their changing height to be extracted from the three lidar surveys. Bertoldi and Gurnell (2020) detail the analysis of lidar data to establish a typical annual vertical growth increment of approximately 0.6 m for A. incana within the study reach and its verification using field measurements.

| Gap boundaries and widths
The boundaries of Gaps 1, 2, 3 and 4 (Figures 2-7) were established from aerial images. Each gap boundary was interpreted from the image which showed it at its widest extent and free of vegetation.
Boundaries of Gaps 2, 3 and 4 were defined from the 2003 images.
Although these gaps were present in earlier images, the largest F I G U R E 2 Maps of the distribution of cover types and vegetation features in and around Gap 1 interpreted from aerial images captured in 1986, 1993, 1997, 2003 and 2005    pioneer islands are P. nigra. Occasionally other Salicaceae species sprout and drive pioneer island development, but Pathway (3) is extremely rarely related to A. incana within the study reach.
Lines of seedlings and young shrubs are rarely observed and, when present, rarely survive for long in the earliest stages of gap evolution or in the more exposed areas outside of the gaps. Lines appear to form preferentially in the central and downstream parts of gaps.
This indicates that lines of seedlings form in the study reach where there is some shelter provided by a gap and/or (pioneer) islands, supporting hypothesis (a) that a reduction in exposure to flood disturbance enhances the relative importance of sexual reproduction in the initiation of vegetated areas. The lines appear to correspond to flow "trash lines" along channel and bar margins and sometimes in the lee of deposited trees and pioneer islands. They are an expression of Pathway (2) of the island model (Figure 9a), since trash lines are usually composed of dead wood, other dead organic material and seeds, but they may also incorporate viable woody fragments. The lines are most likely to develop from seedlings that benefit from macro-shelter by the vegetated sides of the gap and within-gap pioneer islands and also micro-shelter from dead and living-sprouting wood within the trash line. These suppositions are supported by the following facts: A. incana individuals are almost completely confined to these lines; this species rarely reproduces vegetatively in its early years; unlike species of the Salicaceae family, A. incana produces a seed bank and so has an extended time period (possibly more than a year, Thompson et al., 1997) within which its seeds may germinate.
Nevertheless, trash lines may also promote vegetative reproduction from small as well as large woody fragments. While A. incana has occasionally been observed to produce stump sprouts from deposited trees in the study reach, field experiments on the Tagliamento with smaller A. incana cuttings have universally failed and sprouts have rarely been observed on roots exposed by erosion (authors' personal observations). Furthermore, laboratory experiments employing varied soil moisture regimes and sediment calibres have shown extremely poor survival rates for A. incana cuttings but high survival of two Salicaceae species, P. nigra and Salix eleagnos (Francis et al., 2005). Therefore, any asexual recruitment pathway is likely to be confined to Salicaceae species. Furthermore, as pioneer islands and lines of shrubs develop, areas of sheltered but open, relatively fine, moistureretentive sediments may accumulate along the sides and lee of these vegetated landforms that could support recruitment from springdispersed Salicaceae seeds. Corenblit et al. (2016) observed on the Garonne River, France, how a single Salicaceae species, P. nigra, engineered bar surfaces and supported their aggradation, lateral F I G U R E 9 (a) The conceptual model of island development proposed by Gurnell et al. (2001Gurnell et al. ( , 2005. Three different vegetation development pathways (1, 2 3) are associated with progressive aggradation and reinforcement of vegetated landforms and, potentially, the development of established islands and new areas of floodplain. The likely success of each pathway in driving island development depends upon the length of the window of opportunity in relation to the growth performance of trees whose growth is initiated by three different types and sizes of propagule. (b) Idealized growth curves displayed by three different vegetation development pathways according to the impact of local growing conditions on tree growth performance, the length of the window of opportunity between floods, and the local disturbance energy/ shear stress imposed by floods at the end of the window of opportunity [Colour figure can be viewed at wileyonlinelibrary.com] and downstream extension as chronological sequences of stands grew in the shelter of established stands. On the Allier River, France, Tinscert, Egger, Wendelgaß, Heinze, and Rood (2020) found that the most genetically diverse stands of P. nigra were found on the least disturbed sites, supporting our proposal that where this species develops on bare, highly disturbed sites, it reflects a pathway driven by vegetative reproduction, although recruitment from seeds may occur in less-disturbed, sheltered locations. This is also supported by Barsoum (2002), who observed that early stage recruitment of P. nigra along the Drôme River, France, was predominantly from seed, but poor survival over time in response to flood disturbances induced a shift towards vegetative regeneration. Hortobágyi, Corenblit, Steiger, and Peiry (2018) highlighted the engineering roles of different Salicaceae species on the Allier River, where P. nigra acts as the main engineer species at the bar scale but two other Salicaceae species, Salix purpurea and Salix alba with slightly different traits, respectively, colonized and physically engineered the bar sides (coping with the most exposed positions) and tail (benefitting from the most sheltered locations). Here, we have shown the importance of even stronger contrasts in the traits displayed by two riparian trees species drawn from different families for the physical engineering of islands and floodplain edges on the highly disturbed Tagliamento.
This exemplifies hypothesis (b) that "vegetated area initiation and development displays distinctive spatio-temporal patterns reflecting traits of the colonizing woody species." Not only is initial colonization by P. nigra dominated by vegetative reproduction and the development of pioneer islands, but the presence of pioneer islands, particularly at the head of gaps, produces sufficient shelter for seedlings to survive. Furthermore, because of the predominance of autumn floods, initial colonization of open sites requires seeds that are viable in the autumn, precluding those of Salicaceae species but supporting colonization by A. incana. Thus seedling colonization of bare areas of sediment on the Tagliamento is dependent on some shelter from flood disturbances, which is often facilitated by the presence of P. nigra-driven pioneer islands, but as lines of seedlings and shrubs dominated by A. incana grow, Bertoldi and Gurnell (2020) show that they interact with transported sediments to perform a physical engineering role, building linear landforms that often border developing islands and maintain channel edges. In the case of A. incana, sheltered areas not only provide some protection for the developing seedlings, but also the relatively finer, free-draining but relatively moisture-retentive substrates that experiments suggest the species prefers (Hughes et al., 1997). Thus A. incana and P. nigra adopt central positions in two different but complementary Pathways ( (2) and (3)) that drive island development within the study reach.
If they survive, vegetated gaps may have several possible fates.
Narrow gaps (e.g., Gap 1) may continue to support relatively gentle flows of water during floods and isolated ponds during low flows, reflecting their small size, high boundary roughness and strong bank reinforcement by tree roots. Bank reinforcement is likely to be particularly effective where the channels are bordered by A. incana, because this species is frequently observed to spread its roots across the surface of bank faces in the study reach. Wider gaps tend to be bordered by channels with one or more central bars that aggrade to form islands. In the second widest gap (Gap 2) one of these channels had

| An extended conceptual model of island development
The original conceptual model of island development, inspired by observations along the Tagliamento River (Gurnell et al., 2001Figure 9a), emphasized that vegetation development Pathway (1) is unlikely to be successful, mainly because it requires longer windows of opportunity than are typically available on this river. However, Pathway (1) may support vegetation development in areas sheltered by vegetation patches created by the other two pathways. Pathway (3) is observed to be the most successful. It drives island development in the middle and lower reaches of the river where P. nigra is the dominant riparian tree species. Pathway (2) is observed quite widely in the river's headwaters, where the dominant riparian tree species are S. eleagnos and A. incana  and large dead wood accumulations are available to shelter seedlings. Pathway (2) complements Pathway (3) in the middle reaches (Gurnell et al., 2000(Gurnell et al., , 2001. In the lower reaches, Pathway (2) may also be active in the tail of scroll bars (elongated pioneer islands that develop around lines of sprouting deposited trees-Pathway (3)) on the inside of meander bends (Zen, Gurnell, Zolezzi, & Surian, 2017;Zen, Zolezzi, Toffolon, & Gurnell, 2016). Gurnell and Petts (2006)   is growing conditions. This control operates at all spatial scales from biogeographical region to river catchment to river reach to patches within reaches. Although many factors influence growing conditions for riparian trees, moisture availability is crucial and is dependent upon substrate calibre (water retentiveness), river water surface and alluvial groundwater levels. Different vegetation development Curves (1), (2) and (3) can be conceptualized for a single tree species under "poor," "moderate" and "good" growing conditions (the three graphs shown in Figure 9b). How far vegetation development can progress along each curve depends upon the two remaining controlling factors.
The second control is the window of opportunity that is available following germination or initial sprouting of vegetative propagules (horizontal axis on all graphs, Figure 9b). This controlling factor is entirely time dependent, remembering that for species that do not support a seedbank, the start of the curve is the season of seed production, whereas for other sexual or asexual propagules, it can be any time in the growing season. The third controlling factor is flood disturbance.
In relation to flood disturbance, the development curves can be interpreted as vegetation resistance, since any disturbance that plots higher than a given curve would remove vegetation following that development curve. The frequency and magnitude of floods varies through time, generating extreme flow events of widely varying total power or energy. Total stream power is distributed across the width of the flow, resulting in longitudinal reach-scale variations in unit stream power according to flow width for any single value of total stream power. Furthermore, within a reach, stream power is distributed unevenly according to three-dimensional patterns of water depth and shelter, so that disturbance energy (vertical axes, Figure 9b), the bed shear stresses that it exerts and the resulting severity of sediment erosion and deposition processes can vary across all spatial scales. This has the potential for plants following vegetation growth Pathway (1) to survive locally and Pathway (2) to survive more widely in "seedling safe zones" (Polzin & Rood, 2006) within reaches where only those following Pathway (3) may be capable of resisting removal in the most exposed sites.
The above-described multi-scale approach to considering the success of the different vegetation development pathways, can be interpreted by referring to three scenarios. In wide unvegetated rivers or wide gaps in the largest rivers subject to frequent, high energy disturbances, only Pathway (3)  of the Tagliamento as an illustration and investigate the fate of two engineer tree species. As the primary engineer, P. nigra initially colonizes exposed sites, so it suffers high energy disturbances (highest disturbance line) and thus may only survive by following Pathway (3) to build pioneer islands that aggrade into established building islands (right graph, Figure 9b). However, if deposited by a sufficiently large flood, trees following Pathway (3) will be floated into relatively elevated positions, and if the window of opportunity is sufficiently long, they may induce aggradation so that the developing pioneer island may only suffer medium disturbance from a later high magnitude event and may continue its trajectory towards island development (middle graph). Thus, as deposited P. nigra individuals grow, they can reduce the intensity of their exposure to disturbance from a flood event of a given size and they may also induce areas of reduced flood disturbance energy for propagules of the same species following other growth pathways. This can be extended to other species that may preferentially follow other development pathways. A. incana tends to follow Pathways (1) and (2) because of its stronger dependence on sexual reproduction in its early years. In the study reach, Pathway (1) is unlikely to be successful, because even in sheltered, elevated positions, where flood energy may be reduced locally to medium levels, good growing conditions and a long window of opportunity are needed to resist even low disturbances (right graph). This explains why this pathway is unlikely to contribute to island development unless it develops from trash lines that are already sheltered within gaps and/or benefit from shelter by pioneer and established islands and within embayments in floodplain edges. However, the species may successfully grow in trash lines under moderate growing conditions in lower energy rivers (middle graph) where disturbance energy is generally lower, particularly if the window of opportunity is sufficiently long.
Of course, the words "poor," "moderate," "high" applied to growing conditions; "short," "moderate," "long" applied to windows of opportunity, and "low," "medium" and "high" applied to disturbance energy are all qualitative and can be conceptualized across different time and space scales and in relation to different riparian species. changes (e.g., Bornette et al., 2008;McCoy-Sulentic et al., 2017;Merritt, Scott, LeRoy Poff, Auble, & Lytle, 2010;Stromberg & Merritt, 2016) and has started to consider how plant traits may be relevant to fluvial geomorphology (e.g., Hortobágyi et al., 2018;O'Hare, Mountford, Maroto, & Gunn, 2016;Tabacchi et al., 2019). The extended conceptual model of island development builds on such ideas by considering their species-specific and complementary consequences for river landform development.

| CONCLUSIONS
In conclusion, our observations of physical ecosystem engineering by P. nigra and A. incana within the study reach of the Tagliamento, emphasizes that on this high energy river, patches of vegetation that survive within the braid plain are mainly initiated by vegetative reproduction from large propagules (usually whole uprooted trees). Because the dominant species, P. nigra, reproduces freely vegetatively, this is the key engineering species to sprout and build pioneer islands on the braid bar surfaces (vegetation development Pathway (3)), although other Salicaceae species may also initiate pioneer islands. Furthermore, even when extended windows of opportunity arise, Salicaceae species are unlikely to initiate new vegetated patches from seeds (Pathways (1) and (2)) because their seeds are released in Spring; they have a short period of viability; and the largest floods on the Tagliamento occur in Autumn (when there are no viable seeds to disperse and when any young seedlings of the year are too small to survive). In contrast, A. incana does not reproduce readily by vegetative means, particularly when trees are fairly young (i.e., within the age range constrained by island turnover on the Tagliamento, Zanoni, Gurnell, Drake, & Surian, 2008). Therefore, A. incana is very unlikely to drive pioneer island development (Pathway (3)), but this species releases seeds in Autumn that have an extended period of viability and so may benefit from dispersal by floods to suitable sites to initiate Pathways (1) and (2). Even so, seedlings are unlikely to survive in the high energy environment of the Tagliamento unless they receive some macro-(potentially Pathway (1)) or micro-(Pathway (2) Although tree species, hydrological and geomorphological conditions may differ within and between other temperate river environments, we have attempted to generalize the ways in which the traits of engineer plant species constrain physical ecosystem engineering outcomes along three main vegetation development pathways. Our extended conceptual model of island development identifies three interacting aspects of the river system that may constrain the functioning and relative importance of those three pathways: flow disturbance energy, windows of opportunity and local growing conditions. We hope that this conceptualization of how multiple tree species might function in different but complementary ways to drive island development will be helpful in the context of other temperate river environments dominated by the same or other riparian tree species.
In addition, the conceptual model provides a clear framework that could be explored in detail using numerical models (e.g., Bertoldi et al., 2014;Caponi & Siviglia, 2018).
Takemon, University of Kyoto for the 2013 lidar data. Figure 1 includes an extract of an image captured on June 26, 2017 from Google Earth, and from the following supplier: Maxar Technologies. In using an extract from this image, we have conformed to guidelines available from https://www.google.com/permissions/geoguidelines/ attr-guide.html (accessed February 27, 2020) including image attributions in the Figure caption that conform to "the text of your attribution must say the name 'Google' and the relevant data provider(s), such as 'Map data: Google, DigitalGlobe'" and we have not obtained written permission to use these images because the guidelines state that "Due to limited resources and high demand, we're unable to sign any letter or contract specifying that your project or use has our explicit permission."

DATA AVAILABILITY STATEMENT
The data that support the findings of this study are available from the corresponding author upon reasonable request.