Landscape-scale responses of freshwater biodiversity to connectivity and stressors

Aim
There is compelling evidence that drivers and patterns of biodiversity and ecosystem functioning vary across multiple spatial scales, from global to regional, landscape and patch. However, macroecological processes impacting freshwater biodiversity are poorly understood compared to marine and terrestrial ecosystems. Despite step changes in data availability, we have a fragmented view beyond the local-scale of how hydrological and landscape connectivity interact with ecosystem stressors to shape freshwater biodiversity and functioning. While macroecological patterns can vary substantially among taxonomic groups, previous studies have focussed on individual habitat types, sites, or taxonomic groups within landscapes, hindering direct comparisons. We present a cross-landscape, multi-species analysis of the interactive effects of landscape and hydrological connectivity and stressors on standing freshwater quality and the diversity of several major freshwater taxonomic groups.

Location
Great Britain (United Kingdom).

Time period
2000-2016

Major taxa studied
Phytoplankton chlorophyll-a, macrophytes, molluscs, Coleoptera, Odonata, fish and birds.

Methods
Using random forests and generalised additive modelling, we quantified the interactive effects of landscape and hydrological connectivity and stressors on water quality (phytoplankton chlorophyll-a) and the diversity of selected taxa in standing freshwaters.

Results
We found evidence of connectivity changing from positive to negative relationships with biotic responses with increasing human-induced stress levels. Some species groups showed the inverse, reflecting complexities of modelling at large, cross-landscape scales. Almost all responses were affected by stress or connectivity, often interacting, and with non-linear relationships.

Main conclusions
Patterns in stressor-connectivity interactions differed across taxa, but were important in shaping 6 of 8 biotic responses. This emphasises the need for taxon-specific analyses to resolve freshwater ecological responses to stressors, connectivity and their interactions. Our results also highlight that connectivity effects must be integrated in landscape-scale, evidence-led decision-making, designed to reduce impacts of stressors on water quality and biodiversity.

Methods
Data collation and quality assurance
Chlorophyll-a concentration
Chlorophyll-a (μg/l) data are available for 518 water bodies between 2005 and 2014 across Scotland and England. These data were provided by the Scottish Environment Protection Agency (for Scotland) and the Environment Agency (for England). Data are available for 333 lakes and 185 reservoirs. We calculated the mean chlorophyll-a concentration over the time period, for each location. We removed data for water bodies that did not have associated connectivity and stressor metrics, and we removed reservoirs, resulting in a dataset of chlorophyll-a concentrations for 287 water bodies.

Macrophytes
Data are available for 2607 water bodies in Scotland, England and Wales, for the species described in Table 4.2 of Willby, Pitt and Phillips (2012). Data come from publicly available primary datasets generated from macrophyte surveys undertaken by or on behalf of UK environment and conservation organisations and were collated by the University of Stirling. Data were collected between 1832 and 2005. We removed records with a survey date before 1970. All surveys from 1970 onwards were carried out using a whole-lake or transect method. Some water bodies had multiple surveys, and in this case we either took the mean value of all data points from the most recent survey year (for continuous variables), or we took the maximum value of all data points from the most recent survey year (for count variables), or we took one of the values (for categorical variables, which should not change over time). Some parts of water bodies were identified separately in the dataset and these data were merged using the multiple-survey criteria above. After removing data points that had no waterbody ID attached, or that had IDs that were not in the dataset of connectivity and stressor metrics, and after removing data points where not all variables of interest were available, we retained data for 1654 water bodies.

For each waterbody, we have data for the macrophyte species richness (including INNS), the number of macrophyte functional groups, the Lake Macrophyte Nutrient Index as an Ecological Quality Ratio (i.e. relative to value expected in absence of eutrophication impact), the overall quality score based on UK WFD method, and the morpho-edaphic index value.

Molluscs
32,594 records of presences of freshwater molluscs between the years 2000 and 2016 were obtained from the Conchological Society of Great Britain & Ireland non-marine molluscs dataset (Conchological Society of Great Britain & Ireland, 2018), available under the CC-BY 4.0 licence. We removed records with grid references less precise than 1km, or with locations outwith Great Britain, or which did not match with a waterbody location with connectivity and stressor data, or which did not have macrophyte data available. We retained data for 46 mollusc species across 136 1km grid cells. We calculated the total number of species observed in each 1km grid cell between 2000 and 2016 and used this as our response variable.

Beetles
227,156 records of the presence of 475 species of freshwater beetles between the years 2000 and 2018 were obtained from the Aquatic Coleoptera Conservation Trust. We removed records with grid references less precise than 1km, or with locations outwith Great Britain, or which did not match with a water body location with connectivity and stressor data, or which did not have macrophyte data available. We retained data for 395 beetle species across 435 1km grid cells. We calculated the total number of species observed in each 1km grid cell between 2000 and 2018 and used this as our response variable.

Odonata
We took a 2-stage approach to modelling Odonata species richness. Data for 39 non-invasive Odonata species were extracted from the British Dragonfly Society Recording Scheme (British Dragonfly Society Recording Scheme, 2018), available under the CC-BY 4.0 licence, covering the time period 2000 to 2016. Records from each 1km Ordnance Survey (GB) National Grid cell that were matched to a water body were retained, giving 2265 1km cells over England, Wales and Scotland. Since Odonata species are likely to be imperfectly recorded, we first fitted the multispecies Bayesian occupancy model described in Belmont (2021, Section 2.4.1) to the data to obtain predicted Odonata species richness, given Odonata traits data from Powney et al. (2014), available under the CC0 licence. Specifically, we used flight period (i.e. number of months each year in which the species flies), median body size and number of habitat types that each species inhabits (of lowland rivers and canals; bogs, moorland and lowland wet heath; ponds; lakes; and woodlands) in this stage of the modelling. The second stage of the modelling of Odonata species richness used the predicted richness from stage 1 as the response, in contrast to the other species responses where the observed species richness data were used directly.

Fish
We merged three major Great Britain data sets of freshwater fish species records. Firstly, we obtained Biological Records Centre (BRC) Freshwater Fish Recording Scheme data (Freshwater Fish Recording Scheme, 2018), which collates records of fish sightings from recreational anglers and the wider public, and is available under the CC-BY 4.0 licence. At the time of data download (31st October 2018), this source yielded 1,602 species presence records.  Secondly, we obtained 304,899 professional survey and volunteer records from the Database and Atlas of Freshwater Fish (Biological Records Centre, 2018), which was downloaded on 22nd May 2018 and is available under the CC-BY 4.0 licence. Finally, we accessed Environment Agency fish distribution data (Environment Agency, 2018), which was downloaded on 10th September 2018 and is available under the Open Government Licence v3.0, comprising 302,292 records derived from professional surveys across England and Wales. 

Our fish data set included both opportunistic presence-only records, and presence-absence data from professional surveys. Species records were assigned to 1km2 grid cells spanning the land surface of Great Britain, as described in the Data subsection of the main text. Altogether, the fish data included records of 66 species, subspecies and hybrids. Since taxonomic resolution varied among datasets we merged records of subspecies to species-level prior to further analysis. We also excluded hybrid records, chiefly among the Cyprinidae and Salmonidae from further analysis. These hybrids were very rarely recorded, resulting in a minimal loss of data.

Birds
The dataset used to calculate bird species richness in this report was the Wetland Bird Survey (WeBS), available at https://www.bto.org/our-science/projects/wetland-bird-survey/data. Organised by the British Trust for Ornithology (BTO), alongside partners RSPB and JNCC, WeBS is a national-scale citizen science monitoring scheme, where volunteers count species at wetland sites once per month throughout the year. In this work, we considered non-zero species counts from breeding season months (April-June) over the period 2006-2016 at different wetland sites throughout England, Scotland and Wales. There were 36 species considered, chosen on the basis of knowledge of their ecology and behaviour as being most likely to follow hydrological features (on the water or in the air) in moving around the landscape within the spring and summer.

Data
Connectivity measures
We used a suite of connectivity metrics, a selection of which are available from the UK Lakes Portal (https://eip.ceh.ac.uk/apps/lakes/). Spatial data for these metrics are available from the Spatial Inventory of UK Waterbodies at https://catalogue.ceh.ac.uk/documents/b6b92ce3-dcd7-4f0b-8e43-e937ddf1d4eb. Other datasets were collated from the UK digital river network (https://catalogue.ceh.ac.uk/documents/a78c90a2-8da4-4f0a-9c6a-c1d1a4a3c2b0) and the OS MasterMap Water (https://www.ordnancesurvey.co.uk/business-government/products/mastermap-water). All variables considered are listed in Table S1. These include the lengths of rivers (split by Strahler stream order) and canals (representing connected linear waters), the areas, perimeters and counts of lakes and ponds (representing connected standing waters), and the numbers of obstacles to fish migration, in the catchment or within buffers of different distances around each waterbody. Strahler number reflects the branching complexity of each stream segment, with Strahler number 1 indicating a headwater segment, and higher numbers indicating increasing numbers of confluences with other segments upstream. (See Figure S1.) In this study, we categorised segments with Strahler number 4 or greater as “Strahler 4+”. Obstacles to fish migration were obtained from the SEPA Obstacles to Fish Passage dataset (https://marine.gov.scot/maps/1746) for Scotland and the Environment Agency River Obstacles dataset (https://www.data.gov.uk/dataset/0df09ef3-9220-438a-8112-6879b3a51ac5/river-obstacles) for England and Wales, both of which are available under the Open Government Licence. These datasets contain information on natural and man-made barriers such as weirs and waterfalls that act as at least partial barriers to fish movement. To standardise connectivity metrics across catchments and buffers, we divided these values by their respective catchment or buffer areas. We also used a set of human vector connectivity metrics, reflecting the likelihood of visits for fishing, watersports and recreation, as described in Chapman et al. (2019). Since it was not known a priori at which geographical scale each of these processes might affect water quality and richness across our taxonomic groups, these metrics were derived for six geographical scales: catchment (reflecting direct upstream hydrological connectivity), and buffer sizes of 100m, 500m, 1km, 1.5km and 2km around the boundary of each waterbody (reflecting landscape connectivity for sizes 500m and above, and reflecting riparian connectivity for the 100m buffer that is most influenced by land directly bordering the waterbody). These metrics were available for waterbodies across England, Scotland and Wales. The categorical variable levels are given in Table S2.

Lake, catchment, and climate features
To account for the effects of lake, catchment and buffer features on biodiversity and ecosystem function, we used a set of lake, climate, catchment and buffer typology variables. These were also sourced from the UK Lakes Portal the Spatial Inventory of UK Waterbodies, the UK digital river network and the OS MasterMap Water. These included waterbody area, perimeter, fetch and altitude, and catchment and buffer mean slope, mean air temperature and mean annual rainfall. We included River Basin District (RBD, available at https://environment.data.gov.uk/dataset/779ada21-d465-11e4-8a4f-f0def148f590 for England and Wales, under the Open Government Licence, and at https://spatialdata.gov.scot/geonetwork/srv/eng/catalog.search#/metadata/354cb13c-edeb-40b0-964d-f4925d3e71e9 for Scotland, under the Open Government Licence) to account for broad spatial variation across Great Britain that would be difficult to model without using several other variables. RBDs are based on hydrological connectivity of river basins, so that they provide a suitable indicator of different land use and geology than simply using longitude and latitude. For the fish model only, we included a binary variable indicating whether the lake was hydrologically connected to the sea (“accessible to fish migration from sea”), indicating the absence of downstream barriers to fish migration. We included categorical variables such as depth type and upland/lowland classification to represent known hydrological groupings of lakes that behave similarly, with distinct relationships that are often not apparent when considering the population as a whole. (For example, lakes are known to behave differently when depth exceeds a narrow threshold, due to mixing and the ability of light to reach the lakebed, while small and large lakes tend to differ in terms of dominant habitats, with open water dominating for large lakes and littoral habitats for small lakes.) Identifying types of lake environments in which responses differ may allow more targeted management of actions to protect or restore freshwater biodiversity. More details of these lake, catchment and climate feature variables are given in Table S1.

Stressors
We used land use metrics as proxies for the relative intensity of human-induced pressures or aggregate stressors on each focal waterbody (henceforth referred to as stressors). As primary indicators of human modifications, we obtained the percentage agricultural land use and the percentage urban land use in the catchment and buffers of each waterbody from the UKCEH Land Cover Map 2007 (https://catalogue.ceh.ac.uk/documents/e02b4228-fdcf-4ab7-8d9d-d3a16441e23d). These stressor metrics were available for the same six geographical scales as the connectivity metrics. (More details are given in Table S1.)

Biodiversity and water quality data
We selected aquatic macrophytes, molluscs, beetles, Odonata, wetland birds that feed or breed by freshwater habitats and freshwater fish, as representative species of different levels of mobility and of different trophic levels (Figure 1 in the main text). We used records from several Great Britain-scale freshwater monitoring schemes (see Supplementary Material Data collation and QA section), including both presence-absence and presence-only data. The nature of species’ absences in distributional data (i.e. whether or not they are formally recorded, and how they can be interpreted) are influential to ecological inference (Lobo, Jiménez-Valverde & Hortal 2010). For presence-only data sets, we adopted a standardised approach to inferring such absences; a focal species was assumed to be absent from a specific location if it was not recorded, but other species were recorded from that same location (Kéry et al., 2010; Elliott, Henrys, Tanguy, Cooper & Maberly, 2015; Huang & Frimpong 2016). Since Odonata species are likely to be detected imperfectly (Van Strien, Termaat, Groenendijk, Mensing & Kéry, 2010; Van Strien, Van Swaay & Termaat, 2013), we use predicted Odonata species richness obtained using the method of Belmont (2021, Section 2.4.1) as the response variable in our analyses. We used phytoplankton chlorophyll-a as a biological indicator of water quality. For all responses except macrophytes, we used data for 2000-2016, providing us with sufficient spatial locations for analysis. We used data from 1970 onwards for macrophytes, since these are sampled infrequently due to their low mobility that means their richness changes slowly over time. We did not see evidence of substantial temporal trends in the data.

Integrating spatial data scales
For chlorophyll-a concentrations and the taxonomic richness of macrophyte, fish and bird communities, response and explanatory variables were assigned to waterbodies. Due to the nature of their recording schemes, beetle, mollusc and Odonata records were instead assigned to 1km grid cells following the Ordnance Survey (Great Britain) National Grid (‘monads’). To reconcile these differences in recording, grid cells with at least one waterbody present had their species records spatially matched to the waterbody with the largest catchment of all waterbodies in that cell. If all waterbodies in the cell were too small to have catchments generated for them (<1ha), species records within that cell were matched to the largest waterbody present. (Figure S2.)

References
Belmont, J. (2021). Bayesian hierarchical methods for species distribution modelling under imperfect detection (PhD thesis). University of Glasgow. https://theses.gla.ac.uk/id/eprint/82621

Biological Records Centre (2018). Database for the Atlas of Freshwater Fishes [Dataset]. doi:10.15468/3wfao2

British Dragonfly Society Recording Scheme (2018). Dragonfly records from the British Dragonfly Society Recording Scheme [Dataset]. doi:10.15468/cuyjyi

Chapman, D. S., Gunn, I. D. M., Pringle, H. E. K., Siriwardena, G. M., Taylor, P., Thackeray, S. J., … Carvalho, L. (2020). Invasion of freshwater ecosystems is promoted by network connectivity to hotspots of human activity. Global Ecology and Biogeography, 29(4), 645-655. doi:10.1111/geb.13051

Conchological Society of Great Britain & Ireland (2018). Conchological Society of Great Britain & Ireland: non-marine mollusc records [Dataset]. doi:10.15468/6dexp9

Elliott, J. A., Henrys, P., Tanguy, M., Cooper, J., & Maberly, S. C. (2015). Predicting the habitat expansion of the invasive roach Rutilus rutilus (Actinopterygii, Cyprinidae), in Great Britain. Hydrobiologia, 751(1), 127-134. doi:10.1007/s10750-015-2181-9

Environment Agency (2018). Freshwater Fish Counts for all Species, all Areas and all Years [Dataset]. Available from: https://data.gov.uk/dataset/f49b8e4b-8673-498e-bead-98e6847831c6/freshwater-fish-counts-for-all-species-all-areas-and-all-years.

Freshwater Fish Recording Scheme (2018). Freshwater fish records via iRecord [Dataset]. doi:10.15468/eeafla

Huang, J., & Frimpong, E. A. (2016). Limited transferability of stream‐fish distribution models among river catchments: Reasons and implications. Freshwater Biology, 61(5), 729-744. doi:10.1111/fwb.12743

Kéry, M., Royle, J. A., Schmid, H., Schaub, M., Volet, B., Häfliger, G., & Zbinden, N. (2010). Site-occupancy distribution modeling to correct population-trend estimates derived from opportunistic observations. Conservation Biology, 24(5), 1388-1397. doi:10.1111/j.1523-1739.2010.01479.x

Lobo, J. M., Jiménez-Valverde, A., & Hortal, J. (2010). The uncertain nature of absences and their importance in species distribution modelling. Ecography, 33(1), 103-114. doi:10.1111/j.1600-0587.2009.06039.x

Powney, G., Brooks, S., Barwell, L., Bowles, P., Fitt, R., Pavitt, A., … Isaac, N. (2014). Morphological and geographical traits of the British Odonata. Biodiversity Data Journal, 2, e1041. doi:10.3897/BDJ.2.e1041

Van Strien, A. J., Termaat, T., Groenendijk, D., Mensing, V., & Kéry, M. (2010). Site-occupancy models may offer new opportunities for dragonfly monitoring based on daily species lists. Basic and Applied Ecology, 11(6), 495-503. doi:10.1016/j.baae.2010.05.003

Van Strien, A. J., Van Swaay, C. A. M., & Termaat, T. (2013). Opportunistic citizen science data of animal species produce reliable estimates of distribution trends if analysed with occupancy models. Journal of Applied Ecology, 50(6), 1450-1458. doi:10.1111/1365-2664.12158

Willby, N., Pitt, J.-A., & Phillips, G. (2012). The ecological classification of UK lakes using aquatic macrophytes. Evidence SC010080/R2. Bristol: Environment Agency. https://assets.publishing.service.gov.uk/media/5a7c58d2e5274a1b00423299/LIT_7377_45a305.pdf