1 INTRODUCTION

Land-use change is one of the leading causes of global biodiversity change (Jung et al., 2019). Urbanization is one of the most common and dramatic forms of land-use change, and is characterized by increased human population density, increased artificial impermeable surface (e.g., roads, parking lots, and sidewalks), increased pollution, elevated temperature, and habitat fragmentation (Kondratyeva et al., 2020; Walsh et al., 2005). These changes can have profound impacts on all ecosystems experiencing urbanization. Land-use change can be used as a proxy for numerous finer-scale aspects of urban change as several environmental variables that drive ecological change are often correlated with land-use change. For example, transitioning from forest to urban land cover is expected to impact hydrological dynamics and lead to more runoff, thus increasing the concentration of and types of contaminants flowing into water bodies (Arnold & Gibbons, 1996; Walsh et al., 2005). Potential ecological stressors include the concentration of the contaminant, and knowledge of both requires intensive sampling which is difficult to conduct at large spatial scales. Calculating land-use change can be done consistently over large spatial extents through remote sensing and can reveal broader patterns of change across a watershed (Turner et al., 2007). The downside of using land-use change as a proxy for ecological change is the potential loss of fine-scale variation; however, the benefit of increased power to detect change and draw inferences over large spatial extents is appealing.

Currently, 56% of the global human population (83% in North America) lives in urban regions, and this is expected to grow to 68% by 2050 (89% in North America), making further urbanization imminent (Khor, 2022). Understanding how organisms respond to urban pressure is a fundamental ecological goal in the ongoing United Nations “Decade on Restoration” (2021–2030). This is particularly crucial for freshwater organisms, which are estimated to have declined by 84% since the 1970s (WWF, 2020). The interaction between further urbanization and restoration goals is important to consider, especially due to the increased valuation of urban green and blue spaces, which are naturalized vegetated spaces such as parks and forests (green), and aquatic spaces like ponds, wetlands, and rivers (blue). The value of urban blue space was recently recognized within the 2023 Kunming-Montreal Global Biodiversity Framework (GBF), demonstrating the high level of global concern for the ecological integrity of urban water bodies (CBD, 2022). Target 12 of the GBF calls for more biodiversity-inclusive planning and greater access to higher-quality blue and green spaces in urban regions (CBD, 2022). Target 19 calls for more effective resource mobilization, including domestic resource mobilization for conservation, and Target 21 reiterates the importance of biodiversity monitoring (CBD, 2022). Regional specific biodiversity monitoring data provide opportunities to study how urbanization causes ecological changes, which in turn informs how resources can be effectively mobilized for conservation research and better biodiversity-inclusive urban planning into the future.

We leveraged regional monitoring data to further the understanding of how freshwater ecological communities have changed in urban regions. We analyzed trends in alpha and beta diversity in a large urban region using long-term fish community monitoring data from a local watershed management agency. Fish species distributions are closely tied to water quality and stream condition and thus change in fish communities can be indicative of broader environmental changes (Trautwein et al., 2012). Further, long-term datasets collected through standardized sampling methodology provide invaluable opportunities to study ecological change (Hughes et al., 2017). The standardized methodology of long-term monitoring datasets provides critical data which can be compared across time facilitating confident conclusions on biodiversity change (Turner et al., 2015). Long-term ecological datasets allow for comparisons of within-site change, instead of relying on space-for-time substitutions which are common in urban ecological studies when long-term datasets are unavailable (Damgaard, 2019). Temporal datasets better control for aspects of site uniqueness, as many urban sites are impacted differently through time due to historic patterns of land-use (i.e., legacy effects), which may lead to artifacts in space-for-time substitution approaches (Foster et al., 2003).

Using several measures of biodiversity change better encapsulates ecological change as each provides a different interpretation or estimate of the same underlying data patterns (Lyashevska & Farnsworth, 2012). Alpha diversity measures community diversity at a particular site, with different metrics placing emphasis on the number of species or the abundance of individual species (Jost, 2007). To capture community change, measures of alpha diversity can be compared among sites or at the same site through time. However, when alpha diversity is considered alone, information on species identity can be lost. Beta diversity measures community variation among sites (Koleff et al., 2003), or among time periods at the same site, with comparisons across space being more common, in part due to a lack of long-term within-site data. Importantly, measures of beta diversity can retain information on species identity, providing information on the loss and gain of individual species at sites (Baselga, 2008, 2010). Long-term datasets with repeated measures at multiple sites allow for comparisons across both space and time (Hughes et al., 2017). Spatial–temporal datasets like the one we use are becoming more common as long-term local monitoring datasets mature. Used in tandem, alpha and beta diversity indices are useful to understand broad patterns of biodiversity change through both number of species and composition of communities (Magurran, 2021).

Throughout history, urban centers have been tied to water (Hein, 2016). Living near freshwater provides access to drinking water, irrigation sources, transportation corridors, and cultural practices (Hein, 2016). Expansion of cities to accommodate increased human population growth, and rapid increases in migration from rural to urban regions in recent decades, has led to increased density of humans on the shores and in the watersheds of freshwater systems (Haidvogl, 2018). As a result, freshwater rivers and lakes have become degraded over time through multiple mechanisms including fragmentation due to small dams and mills, anthropogenic pollution (e.g., sewage, industrial waste, and household waste), and increased prevalence of introduced non-native species (Haidvogl, 2018). The removal of soils and natural vegetation and increases in impermeable surfaces have fundamentally reduced, if not eliminated, the hydrological buffering capacity of urban watersheds leading to flashy freshwater systems prone to flooding (Arnold & Gibbons, 1996). Changes in hydrological dynamics, species dispersal ability due to in-stream barriers, and increased pollutant loading can inevitably alter ecological dynamics in freshwater bodies, often causing declines in species sensitive to environmental change (Walsh et al., 2005).

Hydrological connectivity is important for riverine dynamics as changes in connectivity impact not only species dispersal patterns but also nutrient cycling and productivity (Ward & Stanford, 1983). Reduced ecological connectivity due to historic mills and dams, and contemporary roadway culverts, buried stream sections, and dams impacts a variety of ecological processes such as access to spawning grounds, access to prey, and ability to avoid predators (Choy et al., 2018; Edge et al., 2017). At the metacommunity level, decreased connectivity can reduce species colonization of sites from larger source populations which may supplement smaller populations across a watershed (Brown & Swan, 2010). Therefore, isolated sites may see higher rates of local extinction due to the lack of immigration from other sites (i.e., less dispersal).

To understand how freshwater biodiversity is changing in a highly urbanized region, we examined changes in riverine fish communities and potential drivers of change across 15 years in Toronto, Ontario, Canada's largest urban center. Toronto has experienced increased urban sprawl since the beginning of European settlements in the region in the mid-18th century (Careless, 1984). Urbanization progressed steadily from the northern shore of Lake Ontario northward throughout the region's river valleys (Careless, 1984). Through time, the City buried rivers in underground pipes and dredged or filled wetlands to reduce exposure to polluted water and increase developable lands (Bonnell, 2014). Three large river systems and several smaller systems are found in Toronto, of which each has varying degrees of changes in their land-use history due to uneven patterns of urban sprawl through time. In our study, we compared alpha and beta diversity of fish communities through time and examined drivers of change through analyzing land-use change due to wetland loss, woodland loss, and anthropogenic land intensification over the study time period. Additionally, we examined the impact of connectivity through time by considering dendritic connectivity and change in connectivity as a potential driver of fish community change. The first objective of our study was to examine whether fish communities are experiencing more gains or losses in species or abundance-per-species through time (Question 1). We predicted fish communities would show more losses than gains due to increased urbanization in the study region over time and changing environmental quality ultimately contributing to species decline and local extirpation. Our second objective was to determine whether changes in upstream catchment land use and dendritic connectivity explained variation in fish community indices through time (Question 2). We predicted land-use change, specifically loss of more natural land-use cover (woodland, and wetland), would best explain variation in within-site biodiversity through time.

2 METHODS

2.1 Fish community data sampling

Fish community data were obtained from the Toronto and Region Conservation Authority through their Regional Watershed Monitoring Program (TRCA, 2022). Sites in the same watershed were sampled once every 3 years, and watersheds were sampled in different years. Fish sampling was performed using the Ontario Stream Assessment Protocol and single-pass electrofishing (Jarvie & Jackson, 2022; Stanfield, 2010). Fifty-four sampling locations from five watersheds were included in our analysis: Etobicoke Creek (n = 11), East Don River (n = 13), Highland Creek (n = 5), Mimico Creek (n = 3), and the Rouge River (n = 22) (Figure 1). Individuals were identified to species, and abundances were tallied by the TRCA. Where samples were not identified to species level, species were assigned based on site history and/or expert knowledge or were removed from the dataset (see Table S1 for details). Fish data were summarized at the site level for two different time periods, T1: 2001–2003 (T1) and T2: 2016–2018 (T2). Each time period included one sampling event per site. Site Strahler order ranged from 1 to 6.

3 RESULTS

3.2 Temporal beta diversity index

Across all sites sampled, there were more gains than losses for species or abundances-per-species on average. Averaged gains (0.59) were larger than losses (0.41) (i.e., species gains averaged 59% and losses averaged 41%; Figure 2). We found seven sites with significant change in species or abundances-per-species (p < .05) prior to Holm correction (Figure 2). These included two sites on the Don River (DN011WM, DN010WMb), one site on the Rouge River (RG019WM), one site on Highland Creek (HL005WM), one site on Mimico Creek (MM002WM), and two sites on Etobicoke Creek (EC001WM, EC010WM) (Figure 3; Table S4). When the data were re-analyzed using occurrence rather than abundance data, two sites were found to have significant change with one gain site (EC010WM) and one loss site (EC001WM) prior to Holm correction (Table S5). We present p-values both pre- and post-Holm correction for multiple testing, as both values may be of interest to ecologists. The TBI values themselves can be used in modeling without significance testing (see Kuczynski et al., 2018; Robichaud & Rooney, 2022); however, we chose to include significance testing for consideration. The 7 exceptional sites pre-correction are subsequently referred to as high TBI sites, as the pre-correction significant sites represent the seven highest TBI sites, all reporting a TBI value >0.89. These seven sites represent the sites with the largest compositional differences through time (Legendre & Condit, 2019).

3.4 Model results

The two averaged top models (Table 2) for richness indicated a positive relationship between change in richness and wetland at T1, wetland loss, and DCI at T1, although estimate 95% confidence intervals overlapped with zero except for wetland loss (Figure 4). There was a negative relationship between change in richness and DCI change, although the estimate 95% confidence intervals overlapped with zero (Figure 4).

TABLE 2. Model selection results for richness, Shannon diversity, Pielou's evenness, and the temporal beta diversity index (TBI).

Response	Model	K	AICc	DeltaAICc	AICcWt	Cum.Wt	LL	$R_{\mathrm{m}}^2$	$R_{\mathrm{c}}^2$
Richness	Wetland	5	272.45	0.00	0.61	0.61	−130.60	.08	.13
	DCI	5	274.36	1.90	0.24	0.85	−131.55	.05	.07
	Woodland	5	276.55	4.09	0.08	0.93	−132.65
	Anthro.	5	276.81	4.35	0.07	1.00	−132.78
	Land cover	9	282.86	10.41	0.00	1.00	−130.39
	Global	11	286.33	13.88	0.00	1.00	−129.02
Shannon diversity	Wetland	5	67.30	0.00	0.33	0.33	−28.02	.03	.04
	DCI	5	67.58	0.28	0.28	0.61	−28.17	.02	.13
	Anthro.	5	68.28	0.98	0.20	0.81	−28.52	.01	.07
	Woodland	5	68.36	1.06	0.19	1.00	−28.55	.01	.09
	Land cover	9	77.73	10.43	0.00	1.00	−27.82
	Global	11	82.58	15.29	0.00	1.00	−27.15
Pielou's evenness	Wetland	5	9.72	0.00	0.39	0.39	0.76	.04	.11
	Woodland	5	10.55	0.83	0.26	0.65	0.35	.03	.10
	Anthro.	5	11.00	1.28	0.21	0.86	0.13	.02	.12
	DCI	5	11.77	2.05	0.14	1.00	−0.26
	Land cover	9	20.02	10.30	0.00	1.00	1.04
	Global	11	25.46	15.74	0.00	1.00	1.41
TBI	Woodland	5	−19.44	0.00	0.40	0.40	15.34	.04	.04
	Wetland	5	−18.68	0.76	0.28	0.68	14.97	.03	.03
	Anthro.	5	−17.94	1.50	0.19	0.87	14.60	.02	.02
	DCI	5	−17.17	2.27	0.13	1.00	14.21
	Land cover	9	−10.46	8.98	0.00	1.00	16.27
	Global	11	−4.43	15.01	0.00	1.00	16.36

• Note: Anthro. represents anthropogenic model. Models with <2 delta AIC include R² values. Models <2 delta AIC were included in model averaging and data visualization.

The four averaged top models (Table 2) for Shannon diversity indicated positive relationship between change in Shannon diversity and woodland T1, wetland T1, wetland loss, woodland loss, anthropogenic T1, anthropogenic intensification, and DCI T1, although estimate 95% confidence intervals overlapped with zero (Figure 4). There was a slight negative relationship between change in Shannon diversity and DCI change, although estimate 95% confidence intervals overlapped with 0 (Figure 4).

The three averaged top models for Pielou's evenness indicate a positive relationship between change in Pielou's evenness and woodland T1, wetland T1, anthropogenic T1, although estimate confidence intervals overlapped with zero (Figure 4). There was a negative relationship with woodland loss, wetland loss, and anthropogenic intensification, although all confidence intervals overlapped with zero (Figure 4).

The three averaged top models (Table 2) for TBI indicate a negative relationship between TBI and woodland T1, woodland loss, wetland T1, and anthropogenic T1, although estimate confidence intervals overlapped with zero (Figure 4). There was a positive relationship between TBI and wetland loss and anthropogenic intensification, although estimate confidence intervals overlapped with zero (Figure 4).

4 DISCUSSION

Our TBI analysis revealed a set of seven sites with substantial change in community composition over time. Further exploratory analysis using occurrence data rather than abundance data revealed a drop in the number of sites signaling significant change pre-Holm correction for multiple testing. This result may indicate species have already been lost prior to the fish surveys in 2001, making fluctuations in abundance of remaining species more common and influential in our TBI analysis. The different results in using abundance over occurrence data also indicate some sites may show signs of local extinction debt, where poor conditions reduce abundances, but the species continues to persist at the site for the time being (Figueiredo et al., 2019). The TBI is a useful metric for identifying sites with large community change and can subsequently be used in further modeling to understand differences in drivers between gains and losses of species of abundance-per-species (Legendre, 2019; Legendre & Condit, 2019). Further analysis of abundance trends related specifically to gains and losses may reveal underlying community dynamics related to extinction and colonization patterns (Carvalho et al., 2012).

The type of change differed between the high TBI sites, with three sites showing losses and four sites showing gains overall, one high TBI site gaining the invasive round goby, and two high TBI sites gaining green sunfish. The observed 4:3 ratio of gains to losses is the opposite direction of our prediction of more losses than gains. The TBI is determined by change in species presence and/or species abundance; we used abundance-per-species due to availability of count data. While sites indicate overall gains or losses, investigating individual species change is important to identify whether gains are gains in species or abundance, and whether gains are native or non-native species. Gains in native species may be reflective of recovery due to restoration efforts but gains in non-native species be indicative of broader invasion dynamics ongoing in urban regions. Urban regions are known to have higher prevalence of invasive species (Padayachee et al., 2017); thus, we may expect TBI to increase through time if invasive species presence and abundance increase, which can have an immediate impact on the interpretation of biodiversity metrics, and a longer-term impact on local biodiversity through changing community dynamics (Gaertner et al., 2017). For example, round goby invasion of tributaries in the Great Lakes region is of concern, as round goby have been found to impact native species through competition for resources and predation on early life stages of native species (Poos et al., 2010). Such competition and predation by non-native round goby may further impact already declining species by further reducing the abundance of native species, a theory supported by the results of Kindree et al. (2023).

Comparing change in individual species abundances across the study region, each individual watershed, and each stream order revealed little significant change in individual species abundances when corrected for multiple comparison testing. This result may be due to diverse spatial patterns of urbanization across the region causing site-specific legacy effects (Perring et al., 2016; Wohl, 2019), and the inherent dendritic nature of rivers creating heterogeneous habitat conditions between sites (Benda et al., 2004). Spatial patterns of urbanization may differ among sites due to time since land-use change, proximity to direct land-use change, or type of land-use change occurring within a watershed. For example, some sites further downstream see little change in upstream catchment land use as they have been urbanized for over 100 years, whereas some sites closer to headwaters see higher relative levels of land-use change as headwaters are relatively less urbanized areas than downstream sites in the study region (Figure S2). Such site specificity may obscure site-level change when aggregated to a broader spatial extent. While inconvenient for making broader-scale conclusions on drivers of change in species abundances, this difference suggests metrics like the TBI are useful for identifying individual exceptional change sites within a broader urban matrix which may become of management interest.

The TBI can be scaled up to a higher level of biological organization than species (Legendre, 2019). While the focus of the current paper was species-level biodiversity change, higher levels of biological organization such as family, or other classifications based on thermal guild or functional traits can be used in future studies (Austen et al., 1994; Benoit et al., 2021). Guild level analyses may reveal broader patterns of change which can be obscured by species-level investigation. Our results suggest thermal guilds may be an important categorization to consider, as we found one high TBI site shifted from a cooler-water fish dominated community to a warm-water fish dominated community (EC010WM). It is possible land-use change upstream of this site contributed to changes in water temperature, as the site saw anthropogenic land intensification directly upstream, along with small pockets of woodland and wetland loss (Figure 3).

While our temporal beta diversity analysis revealed gains and losses in fish biodiversity across the study sites, our chosen potential drivers of fish community gains and losses had low explanatory power. Model-averaged predictor-estimate confidence intervals overlapped with zero for almost all estimates and reported relatively low R² values (Figure 4; Table 2). While calculating land-use change can be done consistently over large spatial extents through remote sensing, fine-scale variation can be obscured since land use is a proxy encapsulating many urban changes, and this may have contributed to our results. Finer geographic scale like local upstream catchment at smaller distance buffers or riparian zone land use may reveal stronger relationships to biodiversity indices; however, we were interested in cumulative upstream land-use impacts at the entire upstream catchment scale. Degree of land-use change may have fallen below thresholds previously found to explain ecological change. For example, while thresholds for land-use change tend to be geographically specific, they were found to range from 1% to 12% for urbanization and 2% to 37% for agriculture in a recent global study (Chen & Olden, 2020). The effects of land use may also be statistically stronger when considering factors such as habitat composition. While data to this scale are often not available due to collection limitations through time, it is possible finer-scale changes in habitat may be stronger drivers of biodiversity change (Chase et al., 2018). Additionally, it is possible physical variables like change in water temperature, may explain some variation (Jarvie & Jackson, 2022), as one site of interest appears to show a change in thermal community composition. As suggested by Kuczynski et al. (2018), including predictors such as specific pollutants or flow may also improve explanation of variation within sites through time. As temporal datasets mature and ecological monitoring continues, analyzing trends over time periods longer than the 15 years analyzed in the current study may reveal stronger relationships between drivers of biodiversity change if time lags are present (Watts et al., 2020). Ultimately, while curating temporal biotic datasets is of the utmost importance, collecting physical and chemical datasets alongside biological data will aid in future explanatory modeling efforts.

The value of open-source data gathered through standardized sampling methods in long-term monitoring programs should not be underestimated. Such long-term datasets provide the basis to answer numerous ecological and conservation-oriented questions. Long-term datasets are particularly valuable in regions experiencing ongoing land-use change and because we cannot logistically (nor ethically) plan and implement the corresponding large-scale experimental studies that would allow testing of these ecological responses (i.e., experiments to test the effects of environmental stressors common to urbanized regions at the scale of urban centers). The evolution of the “sustainable city” from the “industrial” and “sanitary” city makes for an interesting comparison of ecological change over time (Kaushal et al., 2015). The approach in this paper can be used to examine ecological change, whether due to further urbanization or restoration. Future studies can include more land-use categories where land is restored to more natural land cover, unlike the current study where only land use change in the direction of further anthropogenic intensification was included. However, such studies will only be possible with continued monitoring of ecosystems through time using standardized approaches such as those used in long-term monitoring programs. Considering the ongoing “Decade on Restoration” and the recent 2023 Kunming-Montreal GBF, the collection of data to inform biodiversity-inclusive planning and the funding of research to analyze pre-existing long-term datasets must be a regional and national priority to meet international biodiversity goals.

AUTHOR CONTRIBUTIONS

Lauren Lawson: Conceptualization (lead); data curation (lead); formal analysis (lead); methodology (lead); visualization (lead); writing – original draft (lead); writing – review and editing (lead). Christopher B. Edge: Conceptualization (equal); data curation (supporting); methodology (equal); resources (equal); supervision (lead); writing – original draft (supporting); writing – review and editing (equal). Marie-Josée Fortin: Conceptualization (equal); methodology (supporting); writing – review and editing (supporting). Donald A. Jackson: Conceptualization (equal); methodology (supporting); supervision (lead); writing – review and editing (supporting).

CONFLICT OF INTEREST STATEMENT

The authors declare no conflicts of interest.