Designing Effective Low-Impact Developments for a Changing Climate: A HYDRUS-Based Vadose Zone Modeling Approach

Abstract

Low-Impact Developments (LIDs), like green roofs and bioretention cells, are vital for managing stormwater and reducing pollution. Amidst climate change, assessing both current and future LID systems is crucial. This study utilizes variably saturated flow modeling with the HYDRUS software (version 4.17) to analyze ten locations in Ontario, Canada, focusing on Toronto. Historical and projected climate data are used in flow modeling to assess long-term impacts. Future predicted storms, representing extreme precipitation events, derived from a regional climate model, were also used in the flow modeling. This enabled a comprehensive evaluation of LID performance under an evolving climate. A robust methodology is developed to analyze LID designs, exploring parameters like water inflow volumes, peak intensity, time delays, runoff dynamics, and ponding patterns. The findings indicate potential declines in LID performance attributed to rising water volumes, resulting in notable changes in infiltration for green roofs (100%) and bioretention facilities (50%) compared to historical conditions. Future climate predicted storms indicate reduced peak reductions and shorter time delays for green roofs, posing risks of flooding and erosion. Anticipated extreme precipitation is projected to increase ponding depths in bioretention facilities, resulting in untreated stormwater overflow and prolonged ponding times exceeding baseline conditions by up to 13 h at numerous Ontario locations.

1. Introduction

According to the Intergovernmental Panel on Climate Change (IPCC), climate is changing around the world, particularly due to economic and population growth rapidly increasing the production of anthropogenic greenhouse gases [1]. The adverse impacts of climate change become evident through the warming of the atmosphere and oceans, the reduction in snow and ice cover, and the progressive rise in sea levels. Canada, like any other nation, is not immune to the effects of a changing climate. Recent studies have shown that Canada is becoming warmer, as the annual mean temperature averaged over the land has increased by 1.7 °C from 1948 to 2016 [2,3]. It is projected that the number of growing days will also increase with this change in the mean temperature across Canada [3]. Moreover, Canada has experienced noticeable trends of rising annual mean precipitation [3]. Under a low-emission scenario, the annual mean precipitation is projected to increase by 7% by the late 21st century, and under a high-emission scenario, the annual mean precipitation is projected to increase by 24% [3]. With the increase in temperature and frost-free days, there will be a shift from snow to rain in the spring and fall seasons.

With the increase in both the intensity and frequency of extreme precipitation events due to climate change, the risk of flooding rises in dense cities such as Toronto, Ontario, Canada [4]. In order to counter the risk of flooding, engineers have developed ingenious stormwater management solutions such as the use of Low-Impact Developments. Low-Impact Developments (LIDs) are defined as a stormwater management strategy that aims to mitigate the impacts of increased runoff and stormwater pollution by managing runoff as close to its source as possible [5]. With LIDs, stormwater can be treated as a resource in helping to preserve and recreate natural landscapes rather than as a waste that needs to be rerouted from its source [5]. LIDs assist in developing sustainable cities and include systems such as permeable pavements, green roofs, bioretention cells, and rain barrels.

The amount of research conducted on the performance of LIDs under a changing climate is limited. Hathaway et al. [6] noted that the evaluation of the performance of urban stormwater control measures under climate change projections have not been thoroughly examined. Using baseline scenario data between 2001 and 2004 and future scenario data between 2055 and 2058 under two climate emission pathways, Hathaway et al. [6] determined that the frequency and magnitude of overflow from a bioretention system in North Carolina, USA, is projected to increase substantially under future climate scenarios. Borris et al. [7] investigated the runoff quality of bioretention in urban and suburban catchments in Ostersund, Sweden, for future scenarios under different greenhouse gas (GHG) emissions and socio-economic pathways. Wang et al. [8] analyzed the performance of bioretention under a changing climate in Guangzhou, China. The performance factors included the reduction in the runoff volume, peak flow, and first flush. The effect of first flush is defined as a high concentration of contamination in the initial portion of the surface runoff [8]. In the highly dense urban city, Wang et al. [8] noted that the performance of bioretention systems decreases with climate change and cannot replace older stormwater management systems. A recent study completed by Weathers et al. [9] demonstrated an overall decrease in the performance of bioretention facilities in the future by modeling 17 locations across the United States using 10 regional climate models and the United States Environmental Protection Agency’s (USEPA) Storm Water Management Model (SWMM). Wang et al. [10] also used the SWMM to demonstrate the reduction in a bioretention system’s long-term performance under Representative Concentration Pathway (RCP) 8.5 in Guangzhou, China.

The SWMM is a dynamic rainfall-runoff simulation model that computes infiltration using either Horton’s method, the Green–Ampt method, or a curve number method [11]; however, it does not consider water movement under unsaturated or partially saturated conditions, which can lead to conservative LID designs. Niazi et al. [11] notes that modeling LID performance using the SWMM is relatively simple, as the program analyzes hydrological fluxes at a catchment scale. A detailed design of LIDs may prove difficult with the current application, as internal processes of an LID cannot be simulated, such as contaminant transport. Studies by Baek et al. [12], Tu et al. [13], and Haowen et al. [14] suggest that the SWMM is a valuable tool for green infrastructure design and performance assessment; however, it often falls short in accurately representing unsaturated (vadose) zone hydrology. This limitation arises because it primarily focuses on catchment hydrology and hydraulics, simplifying the complex processes occurring within the vadose zone. These studies further state that the hydrogeological aspects of LIDs can be better modeled by vadose zone modeling software such as HYDRUS.

Furthermore, Stovin et al. [15] conducted a 30-year continuous simulation to examine the performance of green roofs for four locations located in the United Kingdom under different climatic regimes. Viola et al. [16] attempted to quantify the retention performance of green roofs worldwide by considering several climate conditions using a simple stochastic weather generator. Both Stovin et al. [15] and Viola et al. [16] did not consider the impact of climate change for these long-term conditions. A study by Herrera-Gomez et al. [17] considered the impact of climate change in Spain. The study was related to the mitigation of the urban heat island effect using green roofs. Berkompas et al. [18], Fassman-Beck et al. [19], Schultz et al. [20], and Voyde et al. [21] have completed studies on the performance of extensive green roofs, particularly their stormwater retention performance. From these studies, it can be noted that the storage, and therefore the runoff performance, is dependent on the size of the storm rather than solely the substrate depth. Overall, there remains a general scarcity regarding the design aspects of LIDs under a changing climate, with particularly limited studies having been completed in Canada. The majority of the available studies that model the performance of LID facilities consider saturated conditions in the soil media, where, in reality, unsaturated flow typically dominates [22,23,24]. For instance, the recent studies by Abduljaleel [25] and Zhang et al. [26] examined the hydrological performance of LID facilities under short-term rainfall events, while the studies by Yoon et al. [27] and Lee et al. [28] examined the performance under long-term conditions. These studies all utilized the SWMM in their analysis of LID performance under changing climate. Additionally, there is a notable absence of a clear, standardized methodology encompassing the compilation, analysis, and classification of historical and future climate data for both long-term and extreme events, as well as the identification of various design elements and systematic modeling procedures to assess critical design considerations through a climate change perspective.

When designing LIDs, the use of Darcy’s equation for saturated flow may lead to inaccurate results such as ponding or overflow within the soil media [22]. Rather, it is more likely that unsaturated flow conditions dominate both green roofs and bioretention systems [22,23,24]. Liu and Fassman-Beck [22], Brunetti et al. [29], Meng et al. [30], Qin et al. [31], and Stewart et al. [32] have demonstrated that the HYDRUS software [33] presents good accuracy in producing the hydraulic response of LID systems. HYDRUS is a modeling software used in the analysis of water and energy flow and solute transport in variably saturated soils.

In light of the anticipated increase in both the quantity and intensity of rainfall due to climate change, it becomes imperative to analyze Low-Impact Development (LID) strategies within this evolving context. This study utilizes the vadose modeling software version 4.17 HYDRUS [33] to analyze variably saturated Low-Impact Development (LID) substrates. It offers a thorough examination of green roof and bioretention substrates under changing climate conditions. This research focuses on the following objectives:

• Conducting one- and two-dimensional analyses of green roof and bioretention facilities using the HYDRUS [33] software. The goal is to determine if simple two-dimensional LID systems can be reasonably approximated using one-dimensional geometry to facilitate long-term, computationally intensive simulations under changing climate conditions.
• Evaluating the hydrological performance of green roof and bioretention facilities under long-term climate conditions for the city of Toronto, Canada. The hydrological performance metrics include the total water quantity (net infiltration), runoff, and plant survival.
• Examining a broad spectrum of design elements for green roof and bioretention facilities across ten locations in Ontario, Canada, ranging from the dry subhumid conditions of Toronto to the humid climate of North Bay. For green roofs, the design parameters include the peak delay (volume and time) and runoff. For bioretention facilities, the performance metrics include the ponding depth, residence times, and runoff.

2. Materials and Methods

Leveraging the HYDRUS family of software, the numerical modeling in this study employed variably saturated flow modeling with a soil–atmosphere boundary to simulate realistic environmental scenarios. As HYDRUS uses Richards’ equation for variably saturated flow, two non-linear hydraulic properties are required to solve the equation, namely, the soil water characteristic curve (SWCC) and the unsaturated hydraulic conductivity. These unsaturated hydraulic properties, along with climate data inputs, geometry, and the initial and boundary conditions, are all required to establish the numerical model in HYDRUS and are elaborated upon in subsequent sections. The model outputs include water balance components at the ground surface such as net infiltration, actual evaporation, ponding height, and runoff. These outputs are used to estimate the performance metrics, which encompass factors such as the total water quantity, peak storm delay, total ponding depth and duration, and overall runoff. This provides a comprehensive assessment of the system’s behavior under varying climatic conditions and design considerations. Additionally, the spatial and temporal distribution of the water content within the substrate is assessed for plant survivability. Figure 1 provides a succinct overview of the tasks undertaken to achieve the research objectives. These tasks include the compilation of extreme precipitation events and long-term climate data, the laboratory measurement of material properties for green roofs and bioretention media, and the development of a numerical model geometry. The subsequent sections provide a comprehensive discussion of each of these critical elements.

2.1. Material Properties

Green roof and bioretention substrates were sourced from local suppliers. These substrates were tested in the laboratory to determine their soil properties. Laboratory testing included measurement of the organic content [34], specific gravity [35], particle size distribution using both the sieve test [36] and hydrometer test [37], and the saturated hydraulic conductivity using the constant head test [38]. The soil water characteristic curve was measured using the HYPROP measurement system [39]. Details regarding the soil hydraulic and geotechnical properties are provided and discussed in Guram [40] and in Guram and Bashir [41].

Table 1. Hydraulic parameters of green roof and bioretention media.

Media	Saturated Hydraulic Conductivity,	Fitted van Genuchten [42] Parameters
	Ks (cm/s)	θs (cm3/cm3)	θr (cm3/cm3)	α (1/cm)	n
Green Roof	0.19	0.47	0	0.09	1.13
Bioretention	0.04	0.43	0.16	0.11	1.73

presents the hydraulic properties used in the numerical analysis for the green roof and bioretention substrate, identified as GR2 and BR2 in Guram [40] and in Guram and Bashir [41]. The green roof’s saturated hydraulic conductivity (K_s) was measured to be one order of magnitude greater than the bioretention media. A smaller K_s for a bioretention medium compared to a green roof is desirable, as it assists in pollutant removal processes, such as filtration. A large K_s for a green roof assists in reducing the risk of surface runoff, leading to a greater roof load.

The van Genuchten [42] equation was used to mathematically represent the measured soil water characteristic curve. The van Genuchten [42] equation is written as follows:

$$Se=\frac{\theta -{\theta }_r}{{\theta }_s-{\theta }_r}={\left[1+{\left(\alpha \left|\psi \right|\right)}^n\right]}^{-m}$$

(1)

where S_e is the effective saturation, θ_r and θ_s are the residual and saturated water contents, respectively, α is a fitting parameter related to the inverse of the air-entry pressure head, n and m are fitting parameters, with m = 1 − 1/n.

2.2. Selection of Climate Data

Historical climate data were used as a reference point to demonstrate the changes in the future climate data. Climate data for ten different locations across Ontario were compiled. These locations included Toronto, Niagara, Kingston, Ottawa, Timmins, Kenora, Thunder Bay, North Bay, London, and Windsor. Historical climate data between 1981 and 2010 were collected from the Environment and Climate Change Canada portal [43], whereas future climate data between 2011 and 2100 from the Ontario Climate Data Portal (OCDP) were used [44,45,46]. There are 33 general circulation models (GCMs) for four representative concentration pathways (RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5) contained within the OCDP. GCMs are the most commonly used climate models, and their primary function is to understand the dynamics of the physical components of the climate system (atmosphere, ocean, land, and sea ice) and make projections based on future greenhouse gas emissions and aerosol forcing [47]. By the year 2100, RCP 2.6 is considered to be the best-case scenario, whereas RCP 8.5 is the worst-case scenario, with the greatest concentration of radiative forces [1].

From the 33 GCMs contained within the OCDP, Baninajarian [44], Bashir et al. [45], and Pk [46] determined that the Community Climate System Model 4 (CCSM4), Geophysical Fluid Dynamics Laboratory Earth System Model version 2M (GFDL-ESM2M), Norwegian Earth System Model (NorESM1-M), and the Hadley Centre Global Environment Model version 2 (Had GEM2-ES) are the four GCMs that performed best when predicting the historical data for different cities within Ontario. Therefore, these four GCMs were used in long-term analysis of future climate projections in this study. For each of the four GCMs, 90 years of future climate data (2011–2100) were subdivided into 30-year periods. This resulted in 48 future climate ensembles (CE) for the four RCPs (Figure 2). The historical climate data for the period from 1981 to 2010 were designated as CE#1. The climate data from 2011 to 2024 fell under the first 30-year future climate block and were simulated data, not measured climate data, and are therefore presented under future climate data. Further examination of this climate data can be found in Baninajarian [44], Bashir et al. [45], and Pk [46].

As the climate data for the ten locations for the four GCMs and four emission pathways resulted in 490 thirty-year daily climate datasets, the selection of an appropriate design climate was required. It was decided that for Toronto, all 48 future climate ensembles presented in Figure 2 were used to examine the performance of the LIDs. Considering that Toronto is the biggest urban center in Canada, this was carried out for a comprehensive assessment.

Extreme Precipitation Events

According to the IPCC [1], the frequency and intensity of extreme precipitation events have increased since 1950 in North America. In order to design infrastructure and stormwater management systems for a region, intensity–duration–frequency (IDF) curves are used to determine the probability of a storm event occurring under certain intensities. Environment and Climate Change Canada [48] provides historic IDF curves for the province of Ontario, whereas future IDF curves are published by various sources. Baninajarian [44] determined that the future IDF curves published by the Ontario Climate Change Data Portal [49] using the regional climate model called RegCM predicted higher intensities compared to other published sources and were therefore used for this research. RegCM uses the outputs from HadGEM2-ES GCM and has two emissions scenarios available, the RCP 4.5 and RCP 8.5 emissions scenarios. The RCP 8.5 emission scenario was chosen as it was more aggressive, predicting higher intensities and thus allowing for the worst-case scenario investigation of extreme events on the LIDs.

Baninajarian [44] compiled and compared IDF curves for ten different locations in Ontario. These locations include Windsor, London, Niagara, Toronto, Kingston, Ottawa, North Bay, Timmins, Thunder Bay, and Kenora. According to the analysis conducted by Baninajarian [44], there was a greater percent change for most of the locations observed for the 6 and 24 h durations as compared to the 1 h storm duration. In addition, all the locations projected an increase in the intensity of extreme events, with the exception of the city of Kenora.

Figure 3 illustrates a comparison of the percentage change between the baseline and future precipitation levels for 1, 6, and 24 h storm durations for the ten locations in Ontario. For the 24 h, 100-year storm events, the city of Niagara was observed to have the greatest percentage increase of 158%. The city of Kingston also had a large percentage increase of 150%. There was a similar trend observed for most locations with the exception of the cities of Toronto, Timmins, and Ottawa for the 24 h storm events. For Toronto, Timmins, and Ottawa, there was a greater increase in the storms with a 2-year return period compared to the 100-year return period. This means that in the future, Toronto, Timmins, and Ottawa will have an increase in smaller-intensity storms that have a greater probability of occurring within a given year compared to the other locations. Additionally, both Toronto and North Bay presented minimal changes in precipitation for the 6 h storm event compared to the other locations. It can also be noted that the cities located in the northern part of Ontario with 100-year return periods had a smaller change in precipitation compared to the locations in the southern part of Ontario.

2.3. Initial and Boundary Conditions

Soil water pressure conditions, reflecting the average over a 30-year span in the Toronto historical data simulation, were extracted and utilized as the baseline for the predicted storm simulations. For consistency across all the locations evaluated in Ontario, identical initial conditions were adopted for each respective site.

For the lower boundary condition, both LID substrates were simulated with a free drainage boundary condition. The upper boundary condition was set to atmospheric with a surface layer. Daily records of the precipitation and potential evaporation values constituted the atmospheric boundary. The allowable ponding was taken as zero for the green roof. Bioretention facilities cater to a greater catchment area in addition to the precipitation that directly infiltrates the system. According to the Credit Valley Conservation and Toronto and Region Conservation Authority (CVC and TRCA) [50], the maximum ponding depth should be between 15–25 cm. Therefore, for the bioretention facilities, the allowable surface ponding was set to 20 cm.

The water balance at the ground surface describes the water that moves across the soil–atmosphere boundary. The components of the water balance at the ground surface include the precipitation (P), potential evaporation (PE), actual evaporation (AE), surface runoff (RO), and net infiltration (NI). The net infiltration refers to the amount of water that enters the soil and can be described as follows:

$$NI=P-AE-RO$$

(2)

AE is dependent on the prevailing water quantity in the soil near the ground surface and is therefore, in most cases, less than the potential evaporation. To compute the actual evaporation, HYDRUS employs a unique boundary condition known as a system-dependent soil–atmosphere boundary. This boundary condition differs from traditional fixed-value controls (such as precipitation or potential evaporation) by allowing the water flow across the soil surface to be influenced by the dynamic interplay between the soil and the atmosphere. Consequently, the actual water flow is determined by a combination of external factors (such as precipitation and evaporation demand) and internal factors (including soil moisture conditions and hydraulic properties).

2.4. Geometry

For this research, both 1D and 2D analyses of the LID geometry were examined. For the 1D analysis, HYDRUS 1D version 4.17 was used [51]. HYDRUS (2D/2D) version 3.01 was used for the 2D analysis [52]. A triangular mesh was used for the 2D analysis, where a finer mesh was applied to the top of both the green roof and bioretention systems.

The green roof was simulated with a 15 cm soil profile. Li [53] simulated a green roof configuration using HYDRUS 2D with depressed water storages and raised drainage openings. As this configuration is a generic geometry of a green roof, it was used for the 2D analysis. Figure 4a illustrates the geometry used for the 1D green roof configurations. For the 2D geometry, depressed water storages and raised drainage openings can be observed in Figure 5.

For bioretention systems, the CVC and TRCA [50] recommend that the substrate depth should be between 1 and 1.25 m. In addition, bioretention facilities can be located above any soil type; however, for soils with an infiltration rate of less than 15 mm/h (hydraulic conductivity of less than 10^–6 cm/s), an underdrain is required [50]. Therefore, the bioretention system was simulated with a 100 cm soil profile above 300 cm of loamy sand (Figure 4b). The hydraulic properties of the loamy sand were predicted using Rosetta Lite DLL (Dynamically Linked Library), which is included in the HYDRUS software [33], to predict the van Genuchten [42] parameters and saturated hydraulic conductivity [54]. The hydraulic properties of the loamy sand media are presented in

Table 2. Hydraulic properties of loamy sand.

Saturated Hydraulic Conductivity,	Fitted van Genuchten [42] Parameters
Ks (cm/h)	θs (cm3/cm3)	θr (cm3/cm3)	α (1/cm)	n
4.38	0.3904	0.0485	0.0347	1.7466

.

For the 2D analysis, a typical bioretention configuration was constructed and is presented in Figure 6. The main difference between the 1D and 2D geometries in terms of bioretention is that for a 2D geometry there, will also be lateral movement of the water. A side slope of 4 horizontal to 1 vertical (4H:1V) in a 2D geometry is noted to be best suited for urban areas, particularly for pedestrian comfort [55]. To reach the required side slope, defined as the ratio of the horizontal distance to the vertical distance, the slope requires either cutting or filling to satisfy the prescribed ratio.

2.5. Leaf Area Index

In order to simulate plant survivability in HYDRUS, the model set-up needs to be updated to consider root water uptake. The soil profile is updated with the addition of roots. The amount of water extracted is dependent on the distribution and depth of the roots [56]. A triangular distribution would extract less water with depth, whereas a rectangular distribution would extract a constant amount with depth [57]. As the green roof substrate had a depth of only 15 cm, neither a constant nor a triangular root distribution resulting in a large difference.

To partition the potential evapotranspiration into the potential evaporation and transpiration, the leaf area index (LAI) was used. The LAI is a dimensionless quantity that describes the amount of vegetation cover over an area of land. A large amount of plant cover indicates a higher LAI, as there is a higher transpiration rate from plants compared to evaporation from soil. The LAI can be approximated as follows [51]:

$$LAI=0.24\times CropHeight$$

(3)

The LAI was approximated to be 1.2 for this study, as it was assumed that the vegetation on the green roof was grass with a height of 5 cm. Furthermore, Brunetti et al. [29] assumed an LAI value of 2.29 for a Sedum mix for a green roof located in Italy. Therefore, an approximated LAI of 1.2 seems reasonable. However, it is noted that a recent study by Hörnschemeyer et al. [58] observed an influence of plant-specific parameters, such as the LAI, on the water balance and peak runoff using the SWMM, and the plant parameters should be validated.

The LAI is applied when there is vegetation cover; therefore, the growing season needed to be defined. The growing season is not equal to the active period; rather, it is the period during which the weather conditions are preferable for plant growth. The growing season length is calculated as the number of days between the last occurrence of 0 °C in the spring and the first occurrence of 0 °C in the fall [59]. In Toronto, the baseline climate data has 275 active days, from 26 March to 26 December, as illustrated in Figure 7. The length, start, and end of the growing season is contained within the 4th Edition [60] of the National Atlas of Canada. The length of the growing season is 200 days, starting on 15 April and ending on 31 October. The LAI value was applied during the growing season.

Thirty years of Toronto historic climate data were initially simulated using the root water uptake function in HYDRUS. From the unsaturated hydraulic properties, the field capacity and permanent wilting point were defined. The field capacity is the amount of water held in the soil substrate after gravitational drainage has occurred. Typically, the field capacity is the water content at a soil water pressure of 33 kPa, whereas the wilting point is the water content at a soil water pressure of 1500 kPa. If the soil water pressure exceeds this point, plants will not recover and will wilt. From the green roof’s unsaturated hydraulic properties, it was determined that the volumetric water contents at 33 kPa and 1500 kPa were 16.1% and 4.89%, respectively.

3. Results

3.1. Comparison of One- and Two-Dimensional Analyses

To compare the 1D and 2D geometries, 30 years of Toronto’s historical climate data were used in the simulations for both the geometries. The cumulative net infiltration, actual evaporation, and bottom flux were analyzed and compared.

Figure 8 demonstrates a comparison of the cumulative NI, AE, and bottom flux (BF) values from the 1D and 2D analyses for the green roof system. For the 2D analysis, drainage openings with 1, 2, and 3 cm diameters were simulated. With the increasing diameter sizes, the cumulative water balance components for the 1D and 2D geometries resulted in similar values. The percentage difference between the 1D and 2D cumulative values for NI, AE, and BF were minimal, demonstrating the similarity in the results.

For the bioretention system, Figure 9 presents the cumulative AE and NI values simulated for both the 1D and 2D analyses. Similar values of AE and NI can be observed for both the 1D and 2D bioretention analyses. The minimal differences observed between the 1D and 2D model results for both the green roof and bioretention systems suggest that 1D models can be effectively employed to simulate the performance of LIDs. This finding holds particular value for long-term simulations, where computational efficiency is critical.

3.2. Performance of LIDs under Long-Term Analysis

To effectively design and implement stormwater management systems capable of addressing the challenges posed by future climatic conditions, extensive long-term simulations were conducted. To assess the behavior of the green roof and bioretention substrates under the various climate ensembles, the annual net infiltration values from the different simulations are presented in the form of box-and-whisker plots. Box-and-whisker plots provide meaningful statistical estimates to study the possible trends in the future climate data for different GCMs and RCPs. This also assists in statistical quantification in terms of which climate ensemble would likely produce unfavourable conditions, such as a large increase in the quantity of water in the future compared to the baseline climatic conditions.

The annual NI values for the 48 future climate ensembles for Toronto were compared to the value from the baseline climate ensemble. Figure 10a,b present the box-and-whisker plots for the green roof and bioretention substrates, respectively. The lower and upper ends of the box show the first and third quartiles, whereas the whiskers represent the maximum and minimum values of the annual net infiltration. The dotted line in the center represents the median of the baseline climate ensemble, noting that the baseline climate ensemble represents the historic climate data between 1981 and 2010. This allows for the reader to observe the gradual or drastic change in the climate data between the baseline and future climate ensembles.

The baseline climate ensemble for the green roof substrate had a minimum and maximum annual NI of 19 and 45 cm, respectively. Climate ensemble #3 had a minimum annual NI of 7.6 cm, and CE#45 had a maximum annual NI of 89 cm. The percentage differences between the maximum and minimum baseline and future climate ensembles were 98 and −59%, respectively. For the bioretention system, the baseline climate ensemble had a minimum and maximum annual NI of 325 and 593 cm, respectively. Climate ensemble #14 had a minimum annual NI of 218 cm, and CE#49 had a maximum annual NI of 887 cm. The percentage differences between the maximum and minimum baseline and future climate ensembles were 50% and −33%, respectively. This large increase in the annual NI compared to the historic climate conditions could lead to a reduced hydrological performance of the LID system.

In general, it can be observed that the GCM HadGEM2-ES predicted higher estimates of the annual NI compared to the other three GCMs. Additionally, higher estimates of the annual NI were projected to occur under the RCP 8.5 emission scenario. With the increase in the time and RCP number, an increase in the annual NI was observed. This means that the annual NI was projected to significantly increase later in the 21st century (2071–2100) under the RCP 8.5 emission scenario. Generally, there was also no substantial change in the median, first, and third quartile values in comparison to the baseline, with the exception of the GCMs HadGEM2-ES and NorESM1-M under the RCP 8.5 emission scenario. There was, however, a large increase in the maximum and minimum values projected in the future compared to the baseline predictions.

When comparing the percentage change between the baseline and future ensembles for the green roof and bioretention substrate, the bioretention substrate had a smaller percentage change in the maximum annual NI compared to the green roof medium. The percentage increase in the maximum annual NI compared to the baseline for the green roof was 98%, whereas for the bioretention substrate, it was 50%. Although both of these percentage differences between the baseline and future ensemble were quite large, the percentage change for the bioretention medium was smaller, even with a greater quantity of water entering the system. This means that the annual AE plays a key role in terms of the quantity of water entering the system. As there was no runoff calculated for these simulations, the NI was directly impacted by the AE in addition to the precipitation.

Under the RCP 8.5 scenario, the temperature is projected to substantially increase in the future due to the increase in GHG emissions [1,44]. This means that the potential evaporation (PE) is projected to increase with increasing temperatures. Although the PE is projected to increase, the AE is also dependent on the prevailing water quantity in the soil near the ground surface and is therefore, in most cases, less than the PE. Even though the bioretention medium had a greater quantity of annual NI compared to the green roof, the percentage difference between the baseline and maximum annual NI was noted to be less, by as much as 50%. This means that the soil hydraulic properties have a significant effect on the total annual NI. It is important to note that the AE is a system-dependent boundary and is therefore dependent on the external conditions as well as the soil moisture conditions. Thus, although the total quantity of the annual NI for the bioretention system far exceeded that of the green roof, the percentage increase in the future maximum annual NI compared to the baseline was smaller for the bioretention system due to the system-dependent actual evaporation.

The annual AE for the green roof and bioretention media, simulated using the 49 climate ensembles, are presented in Figure 11a and Figure 11b, respectively. A clear difference in the quantity and trend was observed between the bioretention medium and the green roof substrate. For the green roof medium, there was a slight increase in AE with both the RCP and time, but this increase was less pronounced compared to the bioretention medium, as shown in Figure 11b. For the bioretention medium, the majority of the future climate ensembles predicted annual AE values that exceeded the maximum annual AE of the historical climate ensemble. This is because the bioretention medium had available water at the ground surface due to ponding, which resulted in evaporation rates close to the potential evaporation rate. Notably, the GCM Nor ESM1-M projected higher estimates of the annual AE for the latter part of the 21st century.

Not only is the annual NI directly impacted by the AE, but Guram [40] also demonstrated the significance of active days, water availability, and soil moisture conditions on LID systems. The active period represents the time when the ground is thawed, thus allowing for water to infiltrate into the soil subsurface. With shifting precipitation patterns, critical future scenarios for LID facilities project an increase in the number of active days and water availability, leading to large quantities of water entering the LID systems.

3.2.1. Long-Term Plant Survivability

To understand plant resilience under adverse climate conditions, plant survivability was evaluated in this study. As green roof systems have a large saturated hydraulic conductivity in order to avoid ponding conditions, they are more likely to experience unsaturated conditions compared to bioretention media. Thus, plant survivability was evaluated for the long-term green roof future climate analysis to understand the potential vulnerabilities.

Figure 12a demonstrates the volumetric water content for the baseline climate ensemble. From the analysis, the volumetric water content did not fall below the permanent wilting point. Still, the volumetric water content did reach a minimum value of 5.69% and may require irrigation to avoid wilting of certain plant species. The future CE#40 was examined as well, since it has the lowest annual moisture index value on average. The active period for CE#40 starts on 22 February and ends 28 December. The growing season would also need to be adjusted for the future. An approximation of the growing season was completed using the average temperatures for this climate ensemble. The start of the growing season was taken as 15 April, and the end was taken as 19 November. Figure 12b demonstrates that the volumetric water content from the 30 years of CE#40 also did not pass the permanent wilting point. The volumetric water content reached a minimum value of 5.6%. Therefore, according to this analysis, there was no appreciable change between the baseline and future volumetric water contents in relation to breaching the wilting point threshold. If the volumetric water content did pass the permanent wilting point, a greater substrate depth would be required to increase the storage capacity.

3.2.2. Annual Runoff Reduction for Green Roof Media

In 2006, Toronto approved the Wet Weather Flow Master Plan (WWFMP). This plan is a long-term goal with the aim of reducing and ultimately eliminating the adverse impacts of wet weather flow in Toronto’s environment [61]. Within the WWFMP, a water balance target states that the maximum allowable annual runoff volume from any development site should be 50% of the total average annual rainfall depth. As it is expected that climate change will likely result in an increase in the annual rainfall depth, the green roof substrate was examined to determine whether it can accomplish this water target.

The percentage difference in the average annual precipitation and runoff from the green roof substrate for all the 49 climate ensembles is plotted in Figure 13. The 50% threshold line is represented by a black dotted line. If the value is below 50%, the runoff from the green roof is greater than 50% of the average annual rainfall depth. Furthermore, Figure 13 is separated by future emissions scenarios such as RCP 2.6, RCP 4.5, RCP 6.0, and RCP 8.5. Overall, it can be seen that in the future scenarios, the water balance target was not always met. In particular, the percentage difference in the average annual precipitation and runoff under the RCP 8.5 emissions scenario was the smallest. For instance, in the year 2094 for CE #46, the total annual precipitation was projected to be 101 cm, whereas the total annual runoff was simulated to be 74.8 cm. The percentage difference was 26%, which is far below the water balance target. This demonstrates that due to climate change in the future, the site may not be able to meet the maximum allowable runoff.

To further illustrate the impact of climate change on current water balance targets, Figure 14 shows the number of instances where the annual runoff exceeded 50% of the annual precipitation. In Figure 14, the climate ensembles are organized by GCM source rather than RCP emission scenarios. The bar chart indicates that the HadGEM2-ES model resulted in the highest number of instances where the annual runoff surpassed the allowable site runoff. Compared to the baseline, there was a general increase in such events, projecting a decline in site performance relative to water balance targets. With increasing time and RCP emissions, the frequency of events where annual runoff exceeded 50% of the annual precipitation rose. Consequently, developers may need to consider alternative Low-Impact Development (LID) measures to address the failure to meet water balance targets due to climate change. Excessive runoff can lead to flooding, erosion, and degraded water quality in receiving waters, negatively impacting property owners, downstream ecosystems, and water users [53]. Thus, it is crucial to design for climate change in order to reduce the stormwater runoff generated during rainfall events and to meet the necessary criteria.

3.3. Performance of LIDs under Extreme Precipitation Events

Storm events have a greater chance of presenting runoff or ponding conditions due to the large intensity occurring under short durations. With climate change, the intensity of storm events is projected to increase, leading to a greater risk of flooding. Thus, in this section, the performance of the green roof and bioretention media are examined under extreme precipitation events. The design elements examined for the green roof include the peak reduction and peak time delay, whereas for the bioretention facility, the design elements include the runoff dynamics, ponding, and storage capacities.

3.3.1. Hydrological Performance of Green Roof Systems under Extreme Precipitation Events

Green roofs assist in reducing the peak flow of a storm event as compared to a conventional roof [62]. The peak reduction can be determined by taking the percentage difference between the peak of the storm event and the peak of the runoff. Figure 15 illustrates a hydrograph for a 48 h 2-year storm using Toronto’s baseline and future climate data. Due to the high permeability of the green roof substrate, there was no surface runoff computed, and therefore, the bottom flux presented in Figure 15 represents the amount of water exiting the bottom of the green roof medium as another form of runoff. From this figure, one can see that the green roof system did in fact assist in reducing the peak intensity of the storm event. This helps in not overwhelming the older storm water systems, such as combined sewer overflows, during a storm event. It can also reduce the erosion of nearby streams or reduce flooding during the time of the storm. Nevertheless, it is projected that the peak reduction will be smaller in the future. This will lead to a greater intensity in the peak runoff compared to baseline storm events. Figure 15 demonstrates the decrease in the peak reduction for a future 48 h 2-year storm event in Toronto. The peak reduction for the future event dropped 22% when compared to the historical event. The 48 h 2-year historic and future events had a peak reduction of 91% and 69%, respectively. This demonstrates the large decrease in the peak reduction, and this could lead to a greater risk of flooding around the site that was not initially expected when designing the green roof using historical climate data.

As predicted storm data for ten other locations in Ontario were also available, each location was also simulated for a 48 h storm with six different return periods (2-, 5-, 10-, 25-, 50-, and 100-year). The peak reduction for each location was computed for the baseline and future predicted storms. It was determined that the peak reduction for the baseline ranged between 20% (100-year storm) and 96% (2-year storm), whereas the peak reduction for the future ranged between 11% (100-year storm) and 82% (2-year storm). The difference between the future and baseline peak reduction was then calculated and is presented in Figure 16. A negative difference implies that the future has a smaller peak reduction compared to the baseline climactic condition. For example, the Niagara 48 h 2-year baseline and future events had peak reductions of 88% and 61%, respectively. The difference between the future and the baseline was equal to −27%, signifying that there was a smaller predicted peak reduction in the future. From Figure 16, it can also be observed that, overall, there was a smaller peak reduction predicted for the future compared to the baseline. This means that the runoff peak intensity will be higher for future events.

Subsequently, Toronto, with return periods of 25, 50, and 100 years, did not follow a similar trend as the other locations, rather, there was a greater peak reduction predicted for the future for these return periods. Still, the difference in the peak reduction for these return periods was fairly small. Due to the storm distribution, the historic storm for Toronto had a smaller peak runoff reduction compared to the future storm. This highlights the fact that in addition to the intensity and duration of the storm, the distribution of the storm is also very important. In this research, only one method of storm distribution (the Chicago method) was investigated. Therefore, it is recommended that for such analyses, more than one method of storm distribution should be considered.

In addition to the peak reduction, the peak time delay is significant in delaying the stormwater runoff quantity during a storm event. This can assist in reducing erosion, flooding, or overwhelming nearby stormwater management systems in comparison to conventional roofs. The time delay was determined by subtracting the time of the peak runoff from the peak of the storm event. For example, for the Toronto 48 h 2-year event, the peak time delays for the baseline and future storm events were 19.6 min and 7.2 min, respectively. The difference was 12.4 min, which is presented in Figure 17. The differences between the time delay from the baseline and future are presented in Figure 17 for the ten locations in Ontario.

With the exception of Kenora and North Bay, eight locations in Ontario followed a similar trend. Interestingly enough, there was a greater difference between the baseline and future for the 2-year storm events. This large difference in time delay between the baseline and future for 2-year storms implies that frequent storm events, or storms with a greater probability of occurrence, will have a smaller time delay between the storm peak and runoff peak in the future compared to 100-year storms. Nevertheless, there was still a reduced time delay predicted for the future, leading to adverse effects of stormwater runoff. For instance, Kingston and Timmins had the greatest time difference of 26 min. This means that the peak will occur 26 min earlier in the future.

For the larger return periods, the difference between the baseline and future was small. However, the time delay between the peaks for the baseline was not large to begin with. For example, the 100-year baseline and future time delays for Timmins were 4.6 min and 2.4 min, respectively. This may result in a smaller time difference of 2.2 min compared to the maximum difference of 26 min. However, this does imply a percentage difference of 47% between the baseline and future events for Timmins. Figure 18 presents the percentage difference between the baseline and future time delay. The greatest percentage difference was computed for the London 100-year storm (88%), demonstrating a large change between the baseline and future storm events.

Kenora and North Bay were once again the exception, demonstrating a large peak delay in the future compared to the base. Kenora had a reduction in storm quantity in the future compared to the baseline, leading to a reduced time difference. For North Bay, however, the total storm quantity was projected to be greater in the future compared to the baseline. Examining the hydrograph, it is noted that the peak intensity of the future storm was quite low in comparison to the historic data, as shown in Figure 19. Thus, the development of the storm distribution impacts the assessment of the hydrological performance of LIDs.

3.3.2. Hydrological Performance of Bioretention Systems under Extreme Precipitation Events

A key design aspect for bioretention facilities includes the ability to allow for surface ponding, as these systems collect a large quantity of stormwater runoff from a large catchment area. Additionally, surface ponding assists in stormwater pollutant removal processes, particularly via sedimentation. Still, there are regulations to follow when designing for ponding within bioretention systems, such as the maximum ponding depth and allowable ponding time. The CVC and TRCA [41] require a maximum ponding depth of between 15 and 25 cm and a maximum allowable surface ponding time of 24 h after a storm event, since this is less than the time required for one mosquito breeding cycle. With these constraints in mind, the surface ponding depth and time were investigated.

Figure 20 presents the ponding depth difference between the baseline and future events for the ten locations in Ontario. A positive value in the ponding depth difference indicates that the future has a greater ponding depth. For example, the historic 10-year event for Kingston had a maximum ponding depth of 3.8 cm, whereas the future 10-year event for Kingston reached the maximum ponding depth of 20 cm. The difference of 16.2 cm can be observed in Figure 20.

Overall, the majority of the locations in Ontario had an increased ponding depth projected for the future, with the exception of Kenora and North Bay. There was little-to-no ponding difference for the 2-year storms for all ten locations. Toronto had a minimal difference in the ponding depth between the baseline and future. The 5-year storm event for London had the greatest increase of 18 cm in the ponding depth.

Furthermore, it can be observed that the 5-year and 10-year storm events had a large increase in comparison to the storms with a greater return period. This was due to the fact that there was no difference in the ponding for the storms with a greater return period, as both the historic and future storm events reached the maximum 20 cm ponding depth. Thus, the stormwater runoff from the bioretention system was also examined.

The difference in runoff between the historic and future climate for the ten locations is shown in Figure 21. Similarly, a positive value of the runoff difference indicates that the future has a greater stormwater runoff quantity. The 2-year and 5-year storms did not result in runoff for either the historic or future events. Moreover, there was little-to-no stormwater runoff observed for North Bay, Timmins, and Toronto. Also, Kenora had a negative difference, as there was a reduction in runoff observed for the future.

The 100-year storms resulted in a large increase in runoff for the future. The London and Niagara 100-year storms had the greatest increases in the ponding depth, with an approximate 50 cm increase. This is concerning, as the bioretention system may fail and the configuration would need to be updated, e.g., through the addition of an underdrain, to avoid flooding around the bioretention facility. This overflow would also lead to an increase in untreated stormwater runoff, which would have adverse impacts on nearby streams and waterbodies.

The next bioretention performance criteria that was examined was the total ponding time. The ponding time difference between the baseline and future storm events for the ten locations were examined and are presented in Figure 22. There was a general increase in the ponding time for most locations in the future, with the exception of North Bay and Kenora. Windsor, with a 100-year return period, had the largest increase in the ponding time (13 h) compared to the baseline. There was a large increase in storm events with the 100-year return periods and a minimal time difference for storms with a 2-year return period. Still, none of the locations exceeded the 24 h ponding time limit defined by the CVC and TRCA [50]. The future 100-year Windsor storm had the greatest ponding time of 22 h, which is fairly close to the limit.

4. Conclusions

This study initially conducted an examination between one- and two-dimensional analyses of both green roof and bioretention facilities. The results of the analysis presented minimal percentage differences between the one- and two-dimensional models for the long-term climate analysis for both the LID systems considered in this research. Therefore, it can be concluded that 1D models can be effectively employed to simulate LID performances. This finding holds particular value for long-term simulations, where computational efficiency is critical.

The second objective was to examine the hydrological performance of the LID systems under long-term climate conditions. The long-term analysis of the LID performance for Toronto presented two conclusions. First, assessments of this sort should consider separate assessments for various time periods, as was carried out in this research, so that progressive design changes in the LID systems can be considered. Second, comparisons of the performance for different emission scenarios will enable planners and designers to make changes to the design by considering the risks associated with various climate change scenarios.

The review of the climate data and numerical modeling results indicated that the percentage change in the maximum NI values for green roofs can be as much as 100%, while for bioretention systems, it can be in the order of 50%. The lower increase in the NI values for bioretention facilities was shown to be related to the increase in AE as a result of the increased PE in the future, the hydraulic properties of the materials, and the quantity of water that the bioretention facilities have to handle. One can conclude that green roofs are more prone to increases in precipitation in the future, while the effect of increased precipitation on bioretention facilities will be offset by the expected increase in the PE.

Healthy vegetation is important for the proper functioning of LID systems. The availability of enough water in the root zone to maintain healthy vegetation is an important concern for changing climatic conditions. The simulations of LID facilities with vegetation for the city of Toronto indicated that for the historical climate conditions, the water availability in the root zone did not fall below the wilting point. A similar simulation for the future climate ensemble with the lowest expected moisture index indicated that, in general, the water availability in the root zone might decrease from historical levels but will still not breach the threshold of the wilting point. However, it should be noted that irrigation may be required, depending on the plant species.

The annual runoff was also estimated for a green roof under Toronto’s long-term climatic conditions. This was carried out to determine whether the future climate ensembles are able to meet the water balance target suggested by the Wet Weather Flow Master Plan (WWFMP) for Toronto. The water balance target states that the annual runoff must be less than 50% of the annual precipitation from a particular site. After conducting the analysis, it was concluded that there will be an expected increase in runoff for many future climate ensembles, leading to the water balance target possibly not being met in the future. The results also lead one to conclude that there is potential for an increased risk of flooding, erosion, and degraded water quality in receiving waters due to climate change.

The third objective of this study was to examine the hydrological performance of LID systems under extreme climate conditions. For the extreme precipitation analysis, two key features that green roofs assist with compared to conventional roofs are the storm peak reduction and the storm peak time delay. Ten locations in Ontario were examined for their performance in terms of the peak reduction and peak time delay in the future.

The hydrological performance of the green roof demonstrated that there could be a general decrease in the peak reduction in the future. In many instances, this can lead to a substantial increase in stormwater runoff intensity from green roof substrates. Moreover, there could be a general decrease in the peak time delay between the peak of the storm and the peak of the runoff from green roofs in the future. Future storms that have a smaller return period will have a smaller time delay between the storm peak and runoff peak compared to storms with a greater return period. Overall, this reduction in performance could potentially result in flooding and erosion due to greater storm intensities and quantities in Ontario.

In terms of the hydrological performance of bioretention systems in the future, a general increase in the ponding depth and time is expected for most of the locations considered in Ontario. A surface ponding depth of 50 cm is expected for many locations across Ontario. Considering that the design requirements for bioretention facilities for some locations is a mere 20 cm, this will result in large quantities of untreated stormwater runoff overflowing from bioretention facilities. For most of the locations considered in this research, the ponding times are expected to increase in the future. The largest increase was predicted for the city of Windsor, which was only two hours short of one mosquito breeding cycle. Considering that the assessments presented in this research assumed an average initial soil saturation, it can be concluded that there is a higher probability that for above average initial soil saturations, the ponding times in the future could be a major concern.

Overall, this study examined the performance of LIDs under future long-term and extreme precipitation climate scenarios and demonstrated that current LID systems will require adaptation under a changing climate. Through the use of HYDRUS, this study presented various design elements of both green roofs and bioretention facilities, allowing LID designers to similarly complete performance reviews of current and future LID systems with a changing climate.