Introduction
Ponds are discrete aquatic habitats distributed across the terrestrial landscape to form a naturally fragmented network, or 'pondscape' (Boothby
1999) and many pond-dwelling organisms are effective dispersers that have the capacity to move long distances between pond habitats in order to acquire resources, avoid predators, competitors, and disturbance, and seek out conspecifics (Fahrig
2007). As a consequence, local populations in ponds become linked by the movement of individuals to form metapopulations sustained across the wider pond network (e.g. Jeffries
1994; Briers et al.
2004). Therefore, the ability of pond-dwelling organisms to disperse among ponds is especially pivotal in promoting species persistence in a dynamic habitat network (Gibbs
2000; Fortuna et al.
2006) in which individual ponds are gained and/or lost through time via a range of natural and anthropogenic processes (Jeffries
2012).
Land-use change such as urbanisation can limit the natural processes that create ponds such as erosional processes or floodplain dynamics (Indermuehle et al.
2008; Williams et al.
1998b,
2010) and accelerate the destruction of natural ponds (Sukopp
1981) or those used formerly for agriculture or industrial purposes (Wood and Barker
2000). This loss and destruction of pond habitat is common to many countries across the world (Fairchild et al.
2013; Hassall
2014). Across the United Kingdom (UK), 32% of ponds are estimated to have been lost over 120 years between 1880 and 2000: a rate of 0.27% per year (Biggs et al.
2005). Losses have occurred in both rural and urban areas, however, the greatest loss (>80%) has been estimated for urban areas such as London between 1870 and 1984 (Langton
1985) and the city of Cardiff (Rich
1998) or areas of intensive agriculture (Beresford and Wade
1982). These major declines are likely to mask a relatively high turnover of sites as ponds are lost and gained over time (Williams et al.
1998a). Some evidence has emerged to suggest that pond losses may have slowed or reversed recently (Biggs et al.
2005; Williams et al.
2010), potentially as pond creation has become imbedded within amenity developments (Jeffries
2012), or as a result of conservation action (e.g. the Million Ponds Project). Nevertheless, in many regions the number of ponds in the modern landscape is still likely to be the lowest in recorded history, with 80% of remaining ponds in the UK existing in a degraded state (Williams et al.
2010) consistent with other wetland habitats (Defra
2011).
The loss of ponds can threaten the persistence of metacommunities when distances between extant ponds begin to exceed the dispersal abilities of the species they support. Species populations that become isolated by pond loss are at a greater risk of local extinction when faced with environmental disturbances or pollution since they lack nearby habitats from which to source recolonists (Tischendorf and Fahrig
2000; Petersen and Masters
2004; Caquet et al.
2007). Within the network, connectivity to large ponds is important as these often support source populations (Van Geest et al.
2003; Sondergaard et al.
2005; Hill et al.
2015) consistent with source-sink island biogeography (MacArthur and Wilson
1967). Equally, connectivity to small ponds is also important since these are more likely to be fishless and serve as important reservoirs of aquatic invertebrates, amphibians and macrophytes (Oertli et al.
2002; Sondergaard et al.
2005; Scheffer et al.
2006). The overall spatial configuration and topology of the pond network (locations and distances between habitats) is thus a key consideration for freshwater biodiversity conservation (Biggs et al.
1994; Boothby
1997; Lundkvist et al.
2002; Jeffries
2005) and questions remain as to how the loss of ponds affects the metacommunity structure of the wider network. Through spatial analyses such as graph theory (Harary
1969), it is possible to determine the extent to which pond loss has fragmented the pond network, threatening species metapopulations.
Graph theory has recently emerged as a powerful tool to evaluate the connectivity of habitat networks and the movements of wildlife and genes (Garroway et al.
2008), and here we apply it to investigate the possible impacts of urbanisation on pond networks. In graph theory, networks are distilled into graphical form with nodes representing habitat patches, and edges indicating the existence of functioning connections or ‘ecological flux’ between node populations (Urban et al.
2009). Traditional applications of graph theory in the field of ecology have focused on modelling species networks, such as food webs, plant-pollinator mutualistic relationships or host-parasitoid webs (e.g. Proulx et al.
2005; Bascompte et al.
2006). To date, graph theory approaches have focussed on terrestrial habitat networks (e.g. Laita et al.
2010; Gurrutxaga et al.
2011; Decout et al.
2012) and application to aquatic systems has been largely confined to riverscapes (Erős et al.
2011; Segurado et al.
2013; Eros and Campbell Grant
2015), with scant application to lentic systems (Ishiyama et al.
2014).
Percolation theory, the science of clustering or clumping in random networks (Stauffer
1987), can be used to complement graph theoretical analyses in order to identify important network characteristics. Percolation analyses can elucidate network redundancy or robustness where, for example, apparently redundant nodes provide alternative dispersal pathways should any nodes be lost or impacted (Laita et al.
2011). Transposed into analyses of landscape connectivity, percolation theory is the quantitative analysis of connectivity in spatially structured systems (With
2002). Frequently, percolation analyses are undertaken to reflect known dispersal ability of a focal organism or organisms in order to gain an understanding of the relative connectedness of the network (O’Brien et al.
2006; Reunanen et al.
2012; Ishiyama et al.
2014).
Together, graph and percolation theory can be used to gain a strategic oversight of a habitat network (Galpern et al.
2011) and help identify areas of the network with high ecological flux for management planning or policy formation (Fall et al.
2007; Stewart-Koster et al.
2015). For urban areas this could yield better outcomes for nature conservation effort where resources may be limited. Without consideration of the spatial configuration of habitat there remains a risk that, notwithstanding the potential but largely unknown influence of garden ponds, as urban development continues pond networks could become increasingly fragmented and less resilient to multiple environmental stressors and climate change.
Ponds are a good candidate for graph theory analysis (Moilanen
2011), being discrete habitats linked by dispersal of aquatic biota (e.g. invertebrates), many of which live exclusively within pond networks (Céréghino et al.
2007). In this study, major changes in the structure and connectivity of a pond network within Birmingham, a heavily urbanised region of the UK, were identified for a 105 year period (
ca1904 – 2009). The structure of the pond network was assessed over time by digitising ponds present on historical Ordnance Survey mapping and the resilience of the network determined by analysing changes in pond distribution, area and number in relation to shifts in land-use. We tested three hypotheses: 1) that considerable pond loss would be observed over the 105 year sequence, 2) pond losses would be strongly associated with urbanisation, and 3) the structural robustness of the Birmingham pond network would decline as the number of ponds in the network decreased.
Discussion
This study aimed to assess how Birmingham's network of pond habitats has altered in response to increasing urbanisation over a 105 year period and consider changes in network robustness and implications for biodiversity. A rate of pond loss in Birmingham between
ca1904 and 2009 of 0.78% per annum is comparable to that of London (0.79%) between 1870 and 1984 (Table
1), and an 82% total loss in total pond numbers between
ca1904 and 2009 ranks Birmingham second highest in the UK (accept hypothesis one), behind urban London and comparable to losses in Bedfordshire's intensively agricultural landscape (Beresford and Wade
1982). Whilst there are few records of ponds loss outside of the UK, losses in Birmingham appear greater than nationwide losses observed in Sweden (Bjureke et al.
1976) and comparable to those in the Netherlands (Weinreich and Musters
1994). Since
ca1962, the rate of annual pond loss has declined to 0.1%, consistent with the nationwide disappearance of ponds reported by Biggs et al. (
2005). Therefore, pond-dwelling organisms within Birmingham are likely to rely upon fewer ponds, potentially rendering their metacommunities less resilient to stochastic events such as pollution or deterministic changes such as global climate change. Ponds that have remained throughout the study (171 ponds) have a notably reduced surface area, which may suggest the occurrence of natural successional processes e.g. vegetation encroachment, or development pressures.
Pond resource turnover, loss and creation
As reported by a number of other authors (Williams et al.
1998a; Jeffries
2012), the raw numbers mask a high turnover in pond resource. The turnover in stock is clearly linked to land-use change and strongly driven by the process of urbanisation (accept hypothesis two) not dissimilar to the impact of coastal urbanisation upon an estuarine wetland network (Dou and Cui
2014). Here however, former farmland field ponds were either lost to, or enveloped by, suburban development as others (few by comparison) were built as part of those developments. These findings accord with several studies of Birmingham's demography and changing landscape, where the population of Birmingham has increased from approximately 500,000 in 1900 to 1M by the early 2000s (Haynes
2008; University of Portsmouth
2017), which coincided with the expansion of Birmingham city centre throughout the 20th century as villages and hamlets coalesced into suburbs through industrial and residential development on former agricultural land (Axinte
2015).
The vast majority of ponds lost within Birmingham were probably artificial in nature, however in a highly altered landscape they are likely to act as surrogates for natural habitats and studies have shown artificial ponds to have high conservation value (Vermonden et al.
2009; Hill et al.
2017; Thornhill et al.
2017). Nevertheless, it is apparent that many ponds are more isolated from their neighbouring habitats and there is a large body of evidence in the published literature that this degeneration of the pond network with less connected nodes has large implications for local and regional biodiversity (e.g. Table
1).
A recent reduction in the rate of pond loss within Birmingham may be due to the retention of larger ponds, which are frequently located in public green spaces or used for recreation (e.g. boating and fishing). Such cultural landmarks may receive protection through local authority planning policies or legislation. Applying conventional island biogeography (MacArthur and Wilson
1967), the retention of high quality larger ponds could help to preserve some of the network's source populations and reduce the overall impact of pond loss. However, a number of studies have also shown that a cluster of small ponds are key contributors to regional invertebrate biodiversity (Wood et al.
2001; Scheffer et al.
2006; Boix et al.
2012). Reasons for this are complex, however it may be due to an increased fish and waterfowl presence in larger ponds resulting in the exclusion of some invertebrate species and reduction in vegetation complexity (Oertli et al.
2002; Sondergaard et al.
2005; Schilling et al.
2009) or to an increase in the number of habitat niches available across several small ponds (Williams et al.
2004). Nevertheless, the potential ecological value in retaining larger ponds for biodiversity may be compromised in Birmingham due to the loss of smaller stepping stone habitats that would otherwise connect larger ponds to the network by facilitating species dispersal.
Possible implications for the biota
As a result of pond loss, the 2009 percolation threshold (2.36km) suggests that for many aquatic biota the Birmingham pond network is comprised of a series of sub-components, the number of which can be inferred through statistical thresholding (Fig.
6). A percolation threshold of 811m and average MST
f edge length of 462m indicated that more frequent exchange of biota between ponds was likely in the
ca1904 pond network, which would have historically allowed more rapid recovery of local populations from stochastic events. For invertebrates, the exception may be a small percentile of species populations which make long distance movements (Conrad et al.
1999) which may be sufficient to maintain genetic diversity (Lowe and Allendorf
2010) but not population recovery.
The majority of macroinvertebrate dispersal studies have focused on Odonates, though nearly all suggest that dispersal, particularly of Zygopterans, beyond 1km is rare and due to high levels of philopatry the majority of movements are constrained to less than 100m (Rouquette and Thompson
2007;
Supplementary Material Table T1). Though potentially less severe due to stronger dispersal tendencies, the scenario is likely to be similar for some Hemipterans (Briers
1998) and Diptera (Service
1997). Less clear is the impact that pond loss is likely to have had on non-winged (i.e. passive) invertebrate dispersal (e.g. Gastropods and leeches (Hirudinea)) which are largely incapable of self-dispersal between habitats and rely on vectors (Bilton et al.
2001).
Analysis of the spatial configuration of pond networks alone suggests that the majority of the 2009 Birmingham pond network has too few ponds that are typically too far apart to sustain populations of the European protected amphibian,
Triturus cristatus.
T. cristatus is generally considered to disperse up to 250m (Langton et al.
2001), with few studies reporting movements up to 1km (Kupfer and Kneitz
2000) and optimum pond density for a
T. cristatus metapopulation is considered to be 4km
-2 (Oldham et al.
2000), which is seldom achieved by 2009 in this study (mean pond density 1.3km
-2). These findings may substantiate the suggestion that
T. cristatus populations across Birmingham are generally thought to have experienced a decline, though limited data exist (The Wildlife Trust
2000).
Although the overall robustness of the Birmingham pond network has clearly declined with probable implications for many pond-dwelling organisms, further study is required to understand the relative significance of impacts to biota with different dispersal modes and strengths (partially accept hypothesis 3). In addition, this analysis does not represent probable losses in temporary ponds (Jeffries
2012) and or the occurrence small garden ponds which are estimated to be present within 10% of UK gardens (Davies et al.
2009). However, whilst valuable as temporary refuges the biodiversity supported by garden ponds has been shown to be a nested subset of field ponds (Hill and Wood
2014) and potentially unlikely to offset their loss.
Future directions
The present study provides the first application of graph theory to a pond network. However, research in three areas would improve the ecological grounding of such spatial models. First, dispersal across the urban landscape is highly unlikely to be uniform as it is comprised of many obstacles such as roads (Parris
2006) and artificial lighting (Bilton et al.
2001; Smith et al.
2009). Small aquatic invertebrates in particular are difficult to track and efforts have focused primarily on rare species (e.g. Purse et al.
2003; Hassall and Thompson
2012). Second, within habitat quality could not be assessed, yet national studies have identified a decline in the quality of ponds (Williams et al.
2010), which could suggest that many included here may not be suitable for colonisation for pollution sensitive taxa. To this end, public participation in data generation (i.e. citizen science; Thornhill et al.
2016) and improvements in remote-sensing (Palmer et al.
2015) are promising avenues. Thirdly, we used a 1km threshold as a broad average of the dispersal ability of the community, however, a more in depth analysis could apply shorter and longer thresholds to better represent the varied abilities of the ecological community to disperse (Galpern et al.
2011), whilst being careful not carry out analysis without sufficient evidence base (Moilanen
2011); thus referring back to point one above.
Conclusion
The identification of the backbone of an urban pond network by using graph theory and the concept of minimum spanning trees (MSTs) is the beginning of a landscape-scale strategy for the conservation of pond fauna and flora rather than traditional single site management. An extended analysis should be carried out to include the wider pond network such that study boundaries are reflective of natural boundaries. However, this study highlights important clusters and pathways within the current pond stock as well as evidence of a need to improve the networks spatial resilience.
This study finds that ponds have become increasingly scarce in the urban landscape over more than 100 years as they are lost to the process of urbanisation. The loss of ponds since ca1904 is considerable, but the rate of pond loss has slowed in recent times. This may be reflective of the types of ponds which are being retained as they are often in the public eye and form part of amenity parkland.
The manner in which this study has been carried out is stepwise and intuitive and can be undertaken with widely available software such that it may be readily repeated across regions and adapted for other landscapes. Landscape managers should ensure that the ponds that remain are of good quality and could use the analytical approach presented here to strategically create ponds in order to reduce the vulnerability of the pond network to further habitat loss. However, opportunities remain to further refine the approach by incorporating inter-habitat resistance to dispersal due to unfavourable land-use.