ABSTRACT
Sustainable drainage systems (SuDSs) have gained popularity, however, guidance for monitoring and maintaining SuDS components remains limited, especially considering their long-term performance in changing environmental conditions. This study begins to address this gap by developing a proof-of-concept model for monitoring infiltration trench (IT) ‘health’. In this study, ‘health’ refers to the physical condition, performance, and overall well-being of the IT. A physical model, constructed following UK SuDS manual guidelines, serves as a testing ground to evaluate IT performance under various maintenance scenarios. The physical model was instrumented to deliver a system for health evaluation. By identifying distinct maintenance parameters, we assess the IT health in terms of infiltration and attenuation/storage volume. The two parameters of ‘leaf build-up’ (surface condition) and ‘sediment build-up’ (subsurface condition) were used as indicative health parameters. The IT instrumentation was able to quantify the adverse effect of sediment build-up on both storage volume and infiltration time. After the sediment was added, the average peak attenuation volume decreased by 40% under a low flow rate and by 20% under a high flow rate. Additionally, the average infiltration time dropped by 26% for both high and low flow rates. Results suggest soil moisture measurements can indicate when IT maintenance is needed.
HIGHLIGHTS
Evidence-based research is essential for understanding long-term SuDS maintenance.
SuDS maintenance level determines SuDS ‘health’.
Effective SuDS health monitoring requires simultaneous consideration of stressors.
Soil moisture measurement can serve as an indicator for sediment build-up in SuDS.
Coupling of instrumentation and imagery enhances SuDS health monitoring.
INTRODUCTION
To ensure the long-term functionality of sustainable urban drainage systems (SuDSs), local authorities are urged to include SuDS maintenance provisions in planning agreements and enforce compliance through site inspections. Despite the growing popularity of SuDSs and their recognised benefits, there remains a lack of comprehensive guidance for monitoring and maintaining diverse components, especially considering their long-term performance in changing environmental conditions. A notable knowledge gap exists in understanding the impacts of various maintenance strategies on SuDS functionality when not adequately implemented. Hence, evidence-based research is essential to gain a clear understanding of stressors and maintenance factors and their interactions for more effective inspection and monitoring, in order to determine spatial and temporal maintenance needs. In urban environments, infiltration trenches serve as effective solutions due to their adaptability to available spaces, particularly when runoff is directed through underground pipes (Burton & Pitt 2001). Experimental studies by Ohnuma et al. (2015) and Lopes-Bezerra et al. (2022) have highlighted the positive contributions of infiltration trenches in water storage, water budget, and as compensatory measures for urban drainage.
Despite their benefits, infiltration structures, such as trenches, face a drawback in their susceptibility to sediment accumulation, diminishing their efficiency (Furumai et al. 2005; Hatt et al. 2007; Freni et al. 2010). Clogging and sediment deposition are commonly cited reasons for the failure of these structures (Erickson et al. 2010; Minnesota-PCA 2024). The clogging of SuDS due to sedimentation is a recognised maintenance challenge for porous surfaces/subsurfaces (Coleri et al. 2013; Yang et al. 2019; Chen et al. 2020; Xu et al. 2022; Garcia-Haba et al. 2023). Legacy infiltration trenches without pretreatment are particularly prone to sediment accumulation, leading to stressful urban conditions (Emerson et al. 2010; Lewellyn et al. 2015). Hence, there remains a demand for a standardised method to evaluate long-term maintenance challenges and their integration into modelling exercises (Conley et al. 2020). To address effective maintenance, understanding diverse design factors and their impacts on infiltration trench performance is pivotal. However, few studies look into quantification of the impacts of clogging on SuDS performance (Siriwardene et al. 2007b; Kandra et al. 2014; Tu & Traver 2018) or dependence of clogging on best management practice design parameters.
Studies by Siriwardene et al. (2007a), Bergman et al. (2011), Barraud et al. (2014), and Toran & Jedrzejczyk (2017) have explored clogging mechanisms and performance reduction over time in infiltration systems through laboratory investigations and field installations. Infiltration as a primary volume reduction strategy has been noted in various studies, such as Lewellyn et al. (2016) and Tirpak et al. (2021) showcasing the effectiveness of well-designed and maintained infiltration trenches in managing runoff. Understanding the sedimentation potential of basins is crucial at both local and catchment levels, as highlighted by Lassabatere et al. (2010) and De Carlo et al. (2020). Furthermore, an emerging approach involves the use of smart technologies for data collection, combined with evidence-based observations. Siriwardene et al. (2007b), Conley et al. (2020), Markovic & Zelenakova (2017), and Al-Rubaei et al. (2013, 2016) exemplify the integration of laboratory and field data for understanding clogging mechanisms and their impacts on infiltration performance. While there have been experimental studies focusing on infiltration trench maintenance, particularly in addressing clogging challenges, consideration is needed in both laboratory experiments and modelling studies to structure and understand functionality. Various studies, such as Conley et al. (2020), Siriwardene et al. (2007a), and Emerson et al. (2010), emphasise the importance of exploring combinations of maintenance challenges and the potential impact on system performance due to sediment build-up over time.
This research aims to tackle this challenge by developing a proof-of-concept model for an infiltration trench monitoring system. To achieve this objective, a physical model was constructed in the laboratory following the guidelines outlined in the UK SuDS manual. The model serves as a testing ground to systematically evaluate the performance of the Infiltration Trench Model (ITM) under various maintenance scenarios, facilitated by the implementation of a semi-automated monitoring system specifically designed for this purpose. The degree of maintenance required serves as a metric for assessing the overall ‘health’ of the infiltration trench. Distinct maintenance parameters were identified, and through a scenario-based approach, the system's performance was assessed in terms of infiltration capacity and susceptibility to flooding. Within the scope of this study, leaves and sediment build-up were considered as indicators of the infiltration trench's health, necessitating evidence-based inspections for more targeted and efficient maintenance practices. The overarching vision of this type of research is to formulate strategies for enhancing inspection and maintenance protocols, aiming for improved long-term functionality grounded in evidence, with the potential for enhanced performance and reduced costs over time.
METHODS
The UK SuDS manual (Woods-Ballard et al. 2015) defines infiltration trenches as shallow excavations filled with rubble or stone that serve as temporary subsurface storage for stormwater runoff. This approach enhances the natural ground's capacity to store water to support groundwater recharge. In particular, infiltration trench systems exhibit significant potential in reducing both the peak flow and volume of runoff, especially during high-return-period events. They are typically positioned alongside impermeable surfaces like roads and parking lots, with a pretreatment layer such as a vegetated filter strip or equivalent alongside. These systems are relatively straightforward to incorporate into landscape design, but their long-term performance hinges on consistent maintenance particularly sediment removal (Minnesota-PCA 2024). This section provides an overview of the methods employed to monitor an ITM, the primary aim is to monitor its ‘health’ from a maintenance perspective. The goal is to enhance our understanding of how insufficient/inadequate maintenance practices can impact infiltration trench performance and to generate experimental evidence to support understanding.
ITM: model specifications
The design of the filter material is crucial for water storage in the infiltration trench. The storage capacity depends on the void ratio, and the outlet throttle controls the level of storage. For the subsurface geotechnical layers, the SuDS manual design guideline was employed. The model consists of four layers: a 10 cm layer of 10 mm pea gravel at the top, a 20 cm layer of 20 mm coarse gravel, a 10 cm layer of compost and finally a layer of permeable foam at the bottom, see Figure 1(b) The foam layer is necessary for a practical reason to ensure accurate outflow rate measurement, as the ITM uses a single drain point connected to a flowmeter. This single drain point imposes horizontal flow across the bottom of the ITM, horizontal flow would be impeded without a permeable bottom layer. The foam underwent infiltration testing to ensure it does not significantly affect the flow rate at the outlet. Within the model, a vertical space of 10 cm with an overflow outlet (8 cm above the top layer) was reserved at the top to accommodate potential inundation, see Figure 1. The outflow and overflow are collected in a plywood spill tray and are pumped back into the hydraulic bench to be reused.
ITM: hydraulic design
In the UK SuDS design manual, the infiltration trench design process prioritises three key considerations: peak flow, storage/attenuation volume, and exceedance flow design. This study specifically concentrates on the first two factors. The analysis revolves around the infiltration rate in soil, influenced by natural soil characteristics such as texture, compaction, and structure, as well as external factors like the inflow rate into the ITM. Conceptually, this ITM represents an infiltration trench built in dense clay soil with limited lateral infiltration. Therefore, we primarily focus on the model's vertical infiltration and retention capacity to maintain simplicity.
Instrumentation, operation and monitoring: sensors
The model is instrumented using two mini-turbine flowrate sensors and an electromagnetic soil moisture probe. The flowrate sensors have a rated flowrate range of 0.2–10.0 l/min and deliver a stream of digital pulses, while the frequency of the pulses is proportional to the rate at which water is passing through the sensor. One flowrate sensor measures the rate at which inflow water is delivered to the top of the ITM, the other sensor measures the rate at which water flows out of the bottom of the model. A perforated PVC tube has been installed around the edges of the model to spray the inflow water over the surface of the model, which replicates surface runoff flowing into the top of the trench, see Figure 1(a). The single soil moisture probe is installed vertically in the ITM soil/gravel media, the probe has four individual sensing zones (top, upper-middle (UM), lower-middle (LM), and bottom (Bot) – Figure 1(b) along its length which are spaced at approximately 100 mm, see Figure 1(b). The bottom sensor is located in the soil layer, the middle two sensors (LM and UM) are in the gravel layer and the top sensor is close to the top of the pea gravel layer.
The connection between the ITM and the computer system serves two purposes: data from the moisture probe is delivered as a digital signal using the SDI-12 communications protocol. Data from the flowrate and moisture sensors is captured via a data logger that records the results at 5-s intervals. A Windows PC is used to download, store and analyse the logged data.
Instrumentation, operation and monitoring: aerial imagery
This study also investigates the usage of aerial imagery using multispectral cameras for more effective monitoring of ITM health, see Figure 2. In this study, the aerial imagery aspect focuses only on surface conditions captured by an overhead multispectral camera which simulates the use of aerial multispectral cameras. Aerial and multispectral imaging have been extensively used in various similar scenarios ranging from industrial site monitoring and maintenance to precision agriculture with a primary focus on automating, monitoring and maintenance processes and cutting long-term costs. These technologies can streamline data collection, providing detailed insights without physical access. This reduces costs and risks, enables early anomaly detection and informed decision-making, optimises resource allocation, reduces manual intervention, and lowers long-term maintenance costs, which is crucial for the health monitoring of SuDS.
Hence, this study adopts aerial and multispectral imaging technologies and examines their efficacy and potential impact in smart health monitoring of infiltration trenches. Specifically, this research aims to identify how coupled sensing, and imagery can be used for more effective monitoring of the ITM performance. In this study, a MAPIR Survey3 multispectral camera equipped with OCN Filter (Orange + Cyan + NIR) has been used for data collection. The OCN filter provides increased contrast and reduces soil noise which improves the ability in detection and classification of various ITM surface conditions in test scenarios.
Infiltration trench health monitoring: health parameters
In this study, ‘health’ refers to the physical condition, performance, and overall well-being of an Infiltration Trench (can be applied to any SuDS component) over its operational lifespan. Two health parameters are considered for inspection and monitoring in this study: ‘leaf build-up’, representing surface condition, and ‘sediment build-up’, representing subsurface condition. A scenario-based approach was adopted to test and monitor the impact of various health scenarios on ITM performance, aiming to understand the challenges arising from inadequate or lack of maintenance. While a wide range of scenarios were tested in this project, a few that best reflect the key observations were selected for demonstration in this paper.
Surface leaf build-up may pose a challenge to an infiltration trench in the following two ways: (a) by covering the surface and thus delaying infiltration and (b) by causing clogging of the voids after degradation of the leaf material, this study focuses on the former. So, the following leaf litter accumulation scenarios were tested:
‘no leaf build-up’.
‘medium leaf build-up’ which consisted of an approximately 30 mm thick layer of initially dry leaves across the top surface of the ITM.
‘high leaf build-up’ which consisted of an approximately 60 mm thick layer of initially dry leaves.
Two subsurface sediment health scenarios were considered: one with ‘no-sediment build-up’, representing a clean and healthy subsurface, and the other ‘with-sediment build-up’, representing an unhealthy poorly maintained subsurface condition. A simple approach was adopted to replicate sediment build-up in the model. The aggregate layers were mixed with sharp sand (up to 4 mm particle size), 20% by weight of sharp sand was added to the aggregate layers, to replicate a sediment build-up scenario.
Results of sieve analyses of the original coarse and pea gravel materials alongside the sand modified coarse and pea gravel material, used for the ‘with-sediment build-up’ scenarios are provided in Figure 4. For completeness, the plot also shows the sieve analysis of the sharp sand that was mixed with the gravels to replicate sediment build-up in this study. We note that the maximum particle size for all test scenarios is less than 25 mm and that there is only a small proportion of material with a particle size below 150 μm. We also note that the added sediment material is dominated by particle sizes in the range 0.2–4.0 mm.
Infiltration trench health monitoring: SuDS health scenarios
As shown in Table 1, all combinations of the health parameters mentioned above were chosen for maintenance scenario development in this paper. Each scenario was tested at two flow rates. The left-hand column in Table 1 provides the adopted scenario naming convention. Each scenario was replicated three times to discern potential sources of error and to improve confidence in the observations. Inflow, in this study, refers to the water pumped into the PVC tube from the Hydraulics Bench, inflow rate is controlled via the flowrate control valve on the bench. Outflow denotes the water drained from the model through the bottom outlet, see Figure 1(a), outflow rate is not under direct control, it is the result of ITM behaviour. For this paper, two representative inflow rates were chosen (see Table 1) to replicate both low- and high-rate surface runoff conditions. These rates were determined based on the desired stress level for system testing and the pumping capability of the hydraulic bench. The low inflow rate ranged between 1.0 and 1.3 l/min and the high inflow rate ranged between 3.5 and 4.0 l/min.
Scenarios . | Subsurface condition/Sediment build-up . | Surface condition/Leaf build-up . | Inflow rate . | Statistics on the multispectral image dataset . | |
---|---|---|---|---|---|
SLxxa-LTFyyb-MPzc-Fkd . | No. of RAW images 12 MP . | No. of processed images (224 × 224) . | |||
SL00-LTF00-MPR-FL | No sediment | No leaf | Low flow | 76 | 671 |
SL00-LTF00-MPR-FH | No sediment | No leaf | High flow | 76 | 524 |
SL00-LTFMM-MPR-FL | No sediment | Medium leaf | Low flow | 75 | 507 |
SL00-LTFMM-MPR-FH | No sediment | Medium leaf | High flow | 50 | 295 |
SL00-LTFHH-MPR-FL | No sediment | High leaf | Low flow | 75 | 479 |
SL00-LTFHH-MPR-FH | No sediment | High leaf | High flow | 75 | 482 |
SLHH-LTF00-MPR-FL | High sediment | No leaf | Low flow | 75 | 489 |
SLHH-LTF00-MPR-FH | High sediment | No leaf | High flow | 75 | 479 |
SLHH-LTFMM-MPR-FL | High sediment | Medium leaf | Low flow | 75 | 510 |
SLHH-LTFMM-MPR-FH | High sediment | Medium leaf | High flow | 59 | 460 |
SLHH-LTFHH-MPR-FL | High sediment | High leaf | Low flow | 75 | 482 |
SLHH-LTFHH-MPR-FH | High sediment | High leaf | High flow | 75 | 620 |
Total | 786 | 5,516 |
Scenarios . | Subsurface condition/Sediment build-up . | Surface condition/Leaf build-up . | Inflow rate . | Statistics on the multispectral image dataset . | |
---|---|---|---|---|---|
SLxxa-LTFyyb-MPzc-Fkd . | No. of RAW images 12 MP . | No. of processed images (224 × 224) . | |||
SL00-LTF00-MPR-FL | No sediment | No leaf | Low flow | 76 | 671 |
SL00-LTF00-MPR-FH | No sediment | No leaf | High flow | 76 | 524 |
SL00-LTFMM-MPR-FL | No sediment | Medium leaf | Low flow | 75 | 507 |
SL00-LTFMM-MPR-FH | No sediment | Medium leaf | High flow | 50 | 295 |
SL00-LTFHH-MPR-FL | No sediment | High leaf | Low flow | 75 | 479 |
SL00-LTFHH-MPR-FH | No sediment | High leaf | High flow | 75 | 482 |
SLHH-LTF00-MPR-FL | High sediment | No leaf | Low flow | 75 | 489 |
SLHH-LTF00-MPR-FH | High sediment | No leaf | High flow | 75 | 479 |
SLHH-LTFMM-MPR-FL | High sediment | Medium leaf | Low flow | 75 | 510 |
SLHH-LTFMM-MPR-FH | High sediment | Medium leaf | High flow | 59 | 460 |
SLHH-LTFHH-MPR-FL | High sediment | High leaf | Low flow | 75 | 482 |
SLHH-LTFHH-MPR-FH | High sediment | High leaf | High flow | 75 | 620 |
Total | 786 | 5,516 |
aSLxx: No sediment: SL00; High sediment: SLHH.
bLTFyy: No leaf: LTF00; Medium leaf: LTFMM; High leaf: LTFHH.
cMPz: The positioning of the moisture probe is either on the left or right side of the ITM, to investigate the spatial aspects of ITM health. This study focuses on a single probe positioned on the right-hand side, labelled as ‘MPR’ in Table 1.
dFk: FL, low flow; FH, high flow.
Performance indicators: infiltration time
Two parameters are employed as key performance indicators (PIs) in this study to assess the ITM's health and performance under the different health scenarios shown in Table 1. These parameters are generally aligned with the discussions under the Infiltration Trench Health Monitoring section.
Infiltration time is the time between sensing that outflow from the model has started to sensing that outflow has stopped. This time serves as an important metric for assessing the health of the ITM, as it is influenced by factors such as the void ratio. The void ratio, in turn, affects the attenuation/storage capacity of the ITM and consequently its infiltration capability.
PIs: attenuation/stored volume
where Qin is the inflow rate and Qout is the outflow rate.
Image data: multispectral image collection
Extensive multispectral image data collection was undertaken for each scenario (and the three repeats of each) as outlined in Table 1. Multispectral images were captured under uncontrolled lighting conditions to more closely replicate real-world outdoor settings. Additionally, they were acquired from a nearly top-down, perpendicular camera angle to accurately simulate the point-of-view of real-world aerial images. A scaffolding-like fixture has been installed over the model, to aid with camera angle, fitting and positioning (see Figure 2). Images were captured using the RAW format to better exploit the camera sensor's dynamic range. They were then converted to TIFF format to eliminate possible compression artefacts and data loss which usually appears in lossy file formats. The images were then rescaled, the histogram equalised, and the white balance adjusted to improve consistency and homogeneity, and to mitigate possible bias during the data collection process. Undesirable samples such as blurred or poorly exposed (highly overexposed or underexposed) images have been removed from the dataset manually. While the unprocessed raw images were captured at 12 megapixels on the camera, they were sliced into smaller non-overlapping windows of 224 × 224 pixels. This enables the elimination of unwanted areas and objects across the frame and allows us to increase the number of training samples from 786 to 5,516 images across all scenarios. Besides slicing, this study employs several image augmentation techniques (crop, flip, rotation, zoom) to further increase the number of training samples to mitigate overfitting and to improve the overall accuracy of the predictive model. Table 1 shows a summary of statistics on the multispectral image dataset across various scenarios in this research. As can be observed, there is no significant imbalance in the dataset and data is distributed almost equally across various scenarios in this study. A sample of sliced pre-processed images for each scenario is provided in the supplementary material for this article. This dataset has also been made accessible publicly at: https://www.kaggle.com/datasets/mahdimaktabdar/sustainable-drainage-systems-suds.
Image data: image classification
All 12 scenarios illustrated in Table 1 are classified into three categories based on their surface conditions i.e., high leaf, medium leaf, and no leaf (see Table 1). State-of-the-art deep learning image classification models including fine-tuned InceptionV3, EfficientNET B7, and ResNet50 paired with image processing techniques to classify various scenarios were used in this study. EfficientNET B7 is a convolutional neural network architecture designed for efficient yet accurate image classification tasks. It leverages compound scaling to balance model depth, width, and resolution, optimising computational resources. This study chose the B7 variant of EfficientNET due to its performance superiority. ResNet50 is a deep learning architecture that includes skip residual connections which enables the training of very deep neural networks by mitigating the vanishing gradient problem. These shortcut connections allow the network to learn residual mappings, making it easier to optimise and improving accuracy. InceptionV3 is a convolutional neural network architecture, featuring the use of inception modules, which incorporate parallel convolutional filters of different sizes to capture spatial hierarchies efficiently. All aforementioned deep models have been attached to a group of top-layers including Dense (128, Relu), Batch Normalisation, Dropout (0.45), Dense (256, Relu), Batch Normalisation, Dropout (0.45), and the final Dense layers that includes three neurons representing three different surface conditions (classes). All deep models have been paired with categorical cross entropy loss function along with Adam optimiser with the learning rate of 0.00001 and have been fine-tuned (unfrozen) during the training process of 30 epochs.
RESULTS AND DISCUSSION
ITM inflow–outflow
As previously discussed, two constant inflow rates were selected in this study. Figure 5 illustrates behaviour at the lower inflow rate, while Figure 6 depicts behaviour at the higher inflow rate. The inflow rate traces indicate that precise control was challenging; adjustments were needed at the start of each run, and the pump did not maintain a constant rate throughout each run. However, these inaccuracies have a negligible impact on the primary results and outcomes.
Comparing Figure 5(a) with 5(b) and Figure 6(a) with 6(b), it can be seen that there is no significant difference in the lag times (i.e., the time between the initiation of inflow and the initiation of outflow), after sediment is introduced to the ITM. The lag time is also reported in Table 1. This lack of variation can be attributed to the model's size, the short vertical flow path, and the relatively large minimum particle size of the sediment used. See sieve analysis results in Figure 4. Nonetheless, it is crucial to examine this parameter in any real-life case study.
In the tests reported here, the ITM does not achieve equilibrium inflow–outflow due to the choice of relatively short rain events simulated here. During the inflow phase of each experimental run, the volume of water stored in the ITM increases, which leads to a rise in the water head in the model. The elevated water head causes the observed increase in outflow rate. Once the inflow stops, the stored water volume decreases, resulting in a gradual decline in the outflow rate until the model is fully drained.
ITM attenuation/storage
Typical performance of the ITM attenuation/storage is shown in Figure 5(a) and 5(b) and Figure 6(a) and 6(b). In general, the shape of the stored volume trace is as expected, except that it does not fall to zero at the end of the run. This is primarily due to the mini-turbine flow meters used to measure flow rate. The flow meters have a specified flow range of 0.2–10.0 l/min, once the ITM is nearly empty the flowrate falls to a value below 0.2 l/min but continues to drain at a slow rate for an extended period of time. This final slow drainage is not registered by the outflow flowmeter, so is missing from the outflow volume calculation. As a result, the stored volume trace does not fall to zero at the end of a run. While this is a clear anomaly in the results, it does not significantly influence the outcomes or conclusions of this study.
Effects of sediment build-up
The outflow traces in Figure 5(a) and 5(b) and Figure 6(a) and 6(b) show that the no-sediment scenarios have a longer infiltration time than the with-sediment scenarios. Table 2 summarises numerical results of the scenarios presented in Table 1. The values in Table 2 represent the average of the three repetitions for each scenario. Additionally, the table includes normalised averages, using the no-sediment/no-leaves result as the baseline.
Am . | Scenario . | Peak attenuation volume . | Infiltration time . | Lag time . | Peak outflow rate . | ||||
---|---|---|---|---|---|---|---|---|---|
l . | % . | min . | % . | min . | % . | l/min . | % . | ||
Low flow rate | SL00-LTF00-FL | 30.3 | 100 | 99.9 | 100 | 4.47 | 100 | 0.93 | 100 |
SL00-LTFMM-FL | 29.3 | 97 | 100.2 | 100 | 4.83 | 108 | 0.86 | 93 | |
SL00-LTFHH-FL | 34.2 | 113 | 108.0 | 108 | 4.08 | 91 | 0.93 | 100 | |
SLHH-LTF00-FL | 18.2 | 60 | 76.8 | 77 | 4.44 | 99 | 1.21 | 130 | |
SLHH-LTFMM-FL | 16.0 | 53 | 73.6 | 74 | 4.22 | 94 | 1.16 | 126 | |
SLHH-LTFHH-FL | 21.6 | 71 | 78.0 | 78 | 4.06 | 91 | 1.29 | 139 | |
High flow rate | SL00-LTF00-FH | 95.4 | 100 | 123.0 | 100 | 2.17 | 100 | 1.77 | 100 |
SL00-LTFMM-FH | 98.6 | 103 | 131.4 | 107 | 2.44 | 113 | 1.78 | 101 | |
SL00-LTFHH-FH | 100.1 | 105 | 127.8 | 104 | 2.25 | 104 | 1.75 | 99 | |
SLHH-LTF00-FH | 80.8 | 85 | 97.4 | 79 | 2.17 | 100 | 2.00 | 113 | |
SLHH-LTFMM-FH | 82.1 | 86 | 96.0 | 78 | 2.31 | 106 | 1.96 | 111 | |
SLHH-LTFHH-FH | 70.9 | 74 | 87.8 | 71 | 2.25 | 104 | 1.63 | 92 |
Am . | Scenario . | Peak attenuation volume . | Infiltration time . | Lag time . | Peak outflow rate . | ||||
---|---|---|---|---|---|---|---|---|---|
l . | % . | min . | % . | min . | % . | l/min . | % . | ||
Low flow rate | SL00-LTF00-FL | 30.3 | 100 | 99.9 | 100 | 4.47 | 100 | 0.93 | 100 |
SL00-LTFMM-FL | 29.3 | 97 | 100.2 | 100 | 4.83 | 108 | 0.86 | 93 | |
SL00-LTFHH-FL | 34.2 | 113 | 108.0 | 108 | 4.08 | 91 | 0.93 | 100 | |
SLHH-LTF00-FL | 18.2 | 60 | 76.8 | 77 | 4.44 | 99 | 1.21 | 130 | |
SLHH-LTFMM-FL | 16.0 | 53 | 73.6 | 74 | 4.22 | 94 | 1.16 | 126 | |
SLHH-LTFHH-FL | 21.6 | 71 | 78.0 | 78 | 4.06 | 91 | 1.29 | 139 | |
High flow rate | SL00-LTF00-FH | 95.4 | 100 | 123.0 | 100 | 2.17 | 100 | 1.77 | 100 |
SL00-LTFMM-FH | 98.6 | 103 | 131.4 | 107 | 2.44 | 113 | 1.78 | 101 | |
SL00-LTFHH-FH | 100.1 | 105 | 127.8 | 104 | 2.25 | 104 | 1.75 | 99 | |
SLHH-LTF00-FH | 80.8 | 85 | 97.4 | 79 | 2.17 | 100 | 2.00 | 113 | |
SLHH-LTFMM-FH | 82.1 | 86 | 96.0 | 78 | 2.31 | 106 | 1.96 | 111 | |
SLHH-LTFHH-FH | 70.9 | 74 | 87.8 | 71 | 2.25 | 104 | 1.63 | 92 |
The average infiltration time for no-sediment scenarios in Table 2 is 115 min, while for with-sediment scenarios it is 85 minutes, resulting in a 26% reduction. The outflow traces indicate that the peak outflow rate is lower without sediment. Additionally, the rise and fall limbs of the outflow traces exhibit slower rise and fall times in the no-sediment scenarios compared to the with-sediment scenarios. For example, the average peak outflow rate for high flow rate no-sediment scenarios is 1.77 l/min, while for high flow rate with-sediment scenarios it is 1.86 l/min, representing a 5% increase.
The volume stored traces in Figure 5(a) and 5(b) and Figure 6(a) and 6(b) show that the peak volume of stored water is greater in the no-sediment scenarios than the with-sediment scenarios. For example, the average peak stored volume for high flow rate no-sediment scenarios was 98.0 l, while for high flow rate with-sediment scenarios it was 77.9 l, representing a 20% reduction. The corresponding reduction in stored volume for low flow rate scenarios was 40%.
Additionally, the soil moisture traces in Figure 5(c) and 5(d), Figure 6(c) and 6(d) show that there are significant changes in void ratio and water level that correlate with-sediment build-up. Comparing Figure 6(c) with 6(d) shows that when the LM layer is saturated, the no-sediment scenario has approximately 50% moisture content, while the with-sediment scenario has about 45%. This indicates a 5% reduction in available storage volume for this layer. The soil moisture traces in Figure 5(c) and 5(d) show that the water level for the no-sediment case is only just starting to reach the LM moisture sensor when inflow is turned off, while for the with-sediment case the water level indicated by the LM sensor reaches a significantly higher level. It should be noted that the reduction in bottom layer moisture content shown in Figures 5 and 6 for the with-sediment scenarios is due to unintended additional compaction of the bottom layer during re-construction for the with-sediment scenario.
Sediment reduces the soil void ratio, decreasing the ITM's storage capacity. This causes the water head to rise faster for a given inflow rate, leading to a quicker increase in outflow rate. Both the increased outflow rate and reduced attenuation capacity indicate a performance decline due to sediment build-up.
Effects of leaf build-up
Image classification results
Deep model . | Recall . | Avg. precision . | F1-score . |
---|---|---|---|
Fine-tuned InceptionV3 | 96.04 | 96.06 | 96.05 |
Fine-tuned EfficientNET B7 | 92.75 | 92.76 | 92.72 |
Fine-tuned ResNet50 | 97.93 | 97.96 | 97.96 |
Deep model . | Recall . | Avg. precision . | F1-score . |
---|---|---|---|
Fine-tuned InceptionV3 | 96.04 | 96.06 | 96.05 |
Fine-tuned EfficientNET B7 | 92.75 | 92.76 | 92.72 |
Fine-tuned ResNet50 | 97.93 | 97.96 | 97.96 |
CONCLUSIONS
This study examines the impact of various maintenance scenarios on a laboratory-scale ITM to assess its health condition. It evaluates performance based on the key indicators of infiltration time and attenuation/storage volume. Additionally, the paper introduces coupled sensing and imagery data acquisition methods and technologies for monitoring purposes. The investigation primarily examines two key health parameters: subsurface sediment build-up and surface leaf accumulation, under varying inflow rates (both low and high), which represent different runoff rates entering the infiltration trench. The main conclusions can be summarised as follows.
Based on our observations, sediment build-up, as expected, reduces the void ratio in the ITM, thereby diminishing its attenuation and storage capacity. The effects of sediment build-up are more evident in high inflow rate scenarios, increasing the frequency of surface inundation in extreme situations. In this study, we investigated a specific sediment build-up scenario using an arbitrary rate of 20% by weight. However, considering various sediment build-up scenarios and sediment particle sizes in future studies could yield further valuable insights, enhancing our understanding of the impacts of sediment build-up on performance. Overall, this highlights the critical role of maintenance in managing sediment build-up to influence the performance of an infiltration trench, with potential exacerbation when combined with other maintenance factors.
Based on the obtained results, the expected influence of surface leaf build-up did not manifest as expected. This outcome can be attributed to two primary factors: (i) the restricted plan area of the model and (ii) the intact condition of the leaves utilised. The limited plan area of the model hinders the replication of changes in lateral flow conveyance across the model surface. It is envisaged that localised patches of intact leaves on the model surface may induce lateral water flow before entering the model, potentially altering the attenuation/storage performance of the model in the presence of intact leaves. Furthermore, the intact condition of the leaves used in this study does not mimic the natural degradation of leaves which can lead to clogging over time, as observed in practical scenarios in the literature. Overall, it can be concluded that the decomposition of leaves, combined with sediment build-up, is expected to exacerbate attenuation and storage reduction, impede infiltration rates, and adversely impact the overall health of the model. This can potentially increase the risk of surface inundation and overflow. Future phases of this study will incorporate the analysis of the cumulative effects of maintenance factors (sediment build-up and clogging due to leaf degradation), elucidating their influence on the overall performance and health of the model.
Also, insights can be gleaned from the experimental setup in this study. The electromagnetic moisture probe is equipped with four moisture-sensing rings, these sensing rings fulfil two primary functions: (i) monitoring water levels in the model and (ii) gauging the impact of sediment build-up on the gravel layers in the model. The experiments showed that the soil moisture measurements from the probe enable a reasonably accurate assessment of the water level within the model. Notably, saturation detected by the top sensor of the moisture probe indicates that the model is nearing inundation, effectively serving as an inundation sensor. Under saturated conditions, variations in the maximum moisture content at each sensor over time can reveal the extent of sediment build-up in an infiltration trench, signalling the need for maintenance.
Finally, while multispectral aerial images can easily classify infiltration trenches based on their surface condition and may be useful for health monitoring, this study reveals only a weak correlation between the surface conditions investigated and infiltration time at the scale of this model. More extensive investigation using a larger scale model is required for more tangible results.
FURTHER WORK
For the next step, the study aims to build on the lessons learned from this proof-of-concept project by further developing and utilising coupled sensing and imagery data acquisition methods and technologies for SuDS monitoring. This approach will enhance our understanding of the interlinkage between surface and subsurface conditions, informing SuDS maintenance strategies. One of the key objectives is to scale up the approach to outdoor conditions for real-world observations, providing opportunities for more realistic temporal and spatial analysis of SuDS performance and maintenance strategies.
AUTHOR CONTRIBUTIONS
M.I. performed original draft preparation, conceptualisation, methodology, resource procurement, data curation. R.B. performed original draft preparation, conceptualisation, methodology, data curation, formal analysis. M.M.O. performed original draft preparation, data curation, formal analysis.
DATA AVAILABILITY STATEMENT
All relevant data are available from an online repository or repositories: https://www.kaggle.com/datasets/mahdimaktabdar/sustainable-drainage-systems-suds.
CONFLICT OF INTEREST
The authors declare there is no conflict.