IntroductionMunicipalities that have embraced green stormwater infrastructure (GSI) as a stormwater management approach are discovering that streamlining the maintenance of these systems is needed to sustain performance at a reasonable cost. The need to rethink and refine GSI maintenance was called out in Wadzuk et al. (2021); several recent publications also highlighted the need for maintenance, and offered suggestions for how to perform GSI maintenance (Houle et al. 2013; Blecken et al. 2017; ASCE 2018; Clary et al. 2018; DelGrosso et al. 2019). Wadzuk et al. (2021) recommends specific focus on improving maintenance approaches using data and risk to make informed and financially viable decisions about GSI maintenance for the community. Having an affordable and robust maintenance program is the key to ensuring long-term functionality of GSI and the broader protection of the watershed. Further, with the recent emphasis on the equity of GSI systems and distribution throughout a community (Heckert and Rosan 2016; Baker et al. 2019; Meerow 2020), a dynamic, data-driven, and risk-based maintenance framework will help to safeguard design intentions to provide all GSI, including social cobenefits.Each community has unique regulatory requirements and management goals, as well as unique challenges. However, there is a common opportunity and need for improved data management and communication that leverages existing knowledge and experience on GSI maintenance to improve efficiencies and reduce costs. Experience with frequency and cost of GSI maintenance from the literature (Caltrans 2004; Hunt et al. 2005; Erickson et al. 2013; Houle et al. 2013; Welker et al. 2013; Clary and Piza 2017; Erickson et al. 2018; PWD 2019; BMP and TC 2019) can be combined with performance monitoring and inspection data, an understanding of dynamic hydrologic processes (Traver and Ebrahimian 2017; Ebrahimian et al. 2021), data management tools, and new technologies to improve operational decisions and provide a framework for more cost-effective maintenance activities.A common challenge for GSI maintenance programs is the cost of what is perceived as labor-intensive maintenance, which can inhibit implementation (Meng et al. 2017). Perceptions about cost are particularly problematic, as revealed in a national survey of GSI owners, operators, and maintenance practitioners, who reported that 56% of surveyed asset managers indicated they had no dedicated budget for GSI maintenance (BMP and TC 2019). Another challenge is the scarcity of guidance on how to convert inspection and maintenance reports to usable data (e.g., Clary et al. 2018) to track and create connections among maintenance efforts, cost, and resulting GSI performance, or how to use these data to continually improve operational maintenance plans over the longer time horizon. Further challenges arise when data siloes are created, wherein data are inaccessible to other stakeholder groups; that is, one group may be unaware of similar data collected by other groups, or does not share such data with other groups that can benefit from it, even within the same organization. An obstacle and also a benefit to dynamic maintenance is that data collection is advancing rapidly to provide fine-scale environmental data, but there is a lack of comprehensive data management tools to enable the proper use of these large quantities of GSI data, often from disparate sources, to make data-driven decisions. Benefits of a data-driven framework are to minimize and more clearly set expected maintenance costs, so that funds can be worked into long-term budgets and extend the life of GSI systems, and conduct data collection with an understanding of data processing.The core concept behind a data-driven maintenance framework is that it should be an adaptive approach (Wadzuk et al. 2021) that can adjust as new and more data are obtained, the quality of data improves, and the support technology advances. This approach opens the door for more efficient and directed GSI maintenance in the ever-changing urban environment. Data-driven tools improve the management of maintenance programs and the associated sustainment of GSI system performance, aid regulatory reporting and support public outreach efforts to help watershed protection programs, and enable a more holistic design approach and governance structure, addressing sustainable, resilient, and equitable GSI systems that are attuned to the community perspective. A GSI system that is not maintained properly leads to loss of public support and poses a risk for health and safety. An additional benefit to using a data-driven, dynamic approach is that additional markets can evolve—for example, software and hardware companies may advance development of data management and analysis tools or sensors that are geared toward stormwater management applications and maintenance. Newly emerging tools can help build better data sensing, collection, storage, and management systems that dissolve data silos and improve communication and modeling.Components of Dynamic and Data-Driven MaintenanceIt is posited that effective inspection programs and maintenance planning must use data for driving identification of high-priority issues and to communicate uncertainty to the user or owner of the GSI system based on variable processes and inputs. This can assist with identifying the factors that cause failure in GSI systems and evaluating how maintenance activities can reduce the risk of failure while minimizing associated GSI costs. Several innovative data strategies could be employed to better understand and inform maintenance practices, monitoring and inspection approaches, and data management technologies, which can be taken together to form a comprehensive approach to operation (Fig. 1).Data CollectionInspection programs offer key data on the physical site conditions that can be used in a data-driven approach, such as inspection frequency, maintenance effort outcomes, system longevity, seasonal effect, and presence of any elements in a GSI system that routinely cause issues. There are some strategies to increase the reach of limited inspection resources. For example, an inspection program based on random and unannounced spot checking rather than full coverage could encourage routine maintenance by private GSI owners while reducing the number of field inspections required (Debo and Reese 2002). To increase data collection, municipalities could shift inspection responsibilities to the property owners and certified inspectors, and require reporting structures to provide data to be used for dynamic maintenance analysis. Municipalities could implement self-reporting requirements, with a system for verification, whereby property owners are educated to perform their own GSI inspections or hire a certified inspector, and a municipal inspection is triggered by failures to submit required self-inspection results—e.g., as successfully seen in the Clean Air Act (Salzman and Thompson 2014).Inspection programs coupled with new tools for monitoring and GSI site data collection (Fig. 1) provide new means for viewing projects in time and space. Monitoring capabilities can offer a level of ongoing inspection between physical site visits. The current explosion of technology allows for collecting information about GSI systems at ever-increasing rates and resolutions. The tools available to collect information are constantly changing and becoming more affordable. However, the variables critical to understanding GSI function and performance remain largely the same. Care should be taken during data collection that data types are standard and durable over time. All data, continuous or discrete and stored as a vector or raster, should be categorized in a manner that allows for relational mapping.Database Architecture for Stormwater ManagementBecause maintenance and inspection programs generate high volumes of data, including contributions from multiple groups, it becomes increasingly important to store data in an integrated, relational structure (Horsburgh et al. 2009). Cities and private contractors tend to use their own software platforms to store and process data, which allows program managers to select the software that best matches their preferences. However, this approach can create data silos that are unable to integrate with other data types or are inaccessible to potential collaborators, neighboring departments (K. Flynn, personal communication, 2019; L. Sherman, personal communication, 2019; T. Saldutti, personal communication, 2019; K. Vacca, personal communication, 2019), or larger agencies. Many municipalities have work order management systems that integrate and relate GSI structures linked to maintenance data, which aids in scheduling and managing work and associated field data (e.g., trash litter loads, effort data). The ability to leverage data streams throughout and across organizations can continue to improve and enrich the way data can be used and manipulated.Horsburgh et al. (2009) proposed an observations data model (ODM), a relational data model that accommodates different types of data series while facilitating data sharing through a controlled vocabulary and format, focused on hydrologic environmental observations data. The Villanova Center for Resilient Water Systems recently launched a revised structured query language (SQL) data model based on the ODM specific to GSI monitoring, called the stormwater infrastructure data model (SIDM; Strauss and Wadzuk, forthcoming). Like the ODM, the IDM allows for analysis of a wide range of continuous and discrete GSI hydrologic and water quality data. The multidimensional structure of the IDM allows for data analysis spanning projects, locations, time, and users, while maintaining specific metadata for each value within the database. The metadata organization provides a relational structure that facilitates findability, accessibility, interoperability, and reusability (FAIR) data principles. For management of physical assets, a similar relational data model could be developed as a shared data management platform for holistic asset management, allowing for communication and collaboration across operations, maintenance, and other asset management issues.Data AnalysisGeospatial AnalysisGeographical Information Systems (GIS) software is a powerful tool for analyzing GSI spatiotemporally. GIS aids systems analysis in the context of their geometry, location, land use, basin, and loading characteristics, along with improved data visualization and communication. Many cities employ GIS on some level for GSI management; however, how these tools are used is highly variable (L. Sherman, personal communication, 2019; T. Saldutti, personal communication, 2019; K. Vacca, personal communication, 2019). Esri (2020) recently launched a set of GIS resources dedicated to GSI. One innovative software tool for geospatial stormwater management is the Esri 2Nform software (2Nform 2019), which was launched in 2017 and is now being used by over 30 California municipalities (Esri 2018). The 2Nform platform cross-references GSI sites with factors including historic maintenance data and litter maps to identify vulnerable sites and prioritize maintenance accordingly. The 2Nform software also includes automated workflow calendars, a regulatory reporting portal, and a mobile application for direct entry of inspection data (2Nform 2017, 2019). These functions integrate with GIS and other data management tools to work with existing data management structures.Artificial IntelligenceIn recent decades, artificial intelligence (AI) has proven to be a powerful tool for computational analysis and data mining (Witten and Frank 2002). Although AI has been shown to be useful in a range of water resources projects (Yitian and Gu 2003; Shamseldin 2010; Ebtehaj and Bonakdari 2014; Yan et al. 2018; Emamgholizadeh and Demneh 2019; Hosseiny et al. 2020), it is yet to be fully realized for GSI, in part due to a lack of large datasets.Within AI, artificial neural network (ANN) modeling is one possible data-driven approach for dynamic maintenance. This modeling approach learns from training datasets to make predictions based on weighted input factors (Tran et al. 2007). As in other industrial and structural health monitoring, system age, maintenance frequency, and sensor data are used for predictive maintenance and in machine learning application. Moving to GSI applications, a model could be trained with past failure data to predict future failures based on input factors such as system age, maintenance frequency, and loading ratio; some stormwater applications are already starting to use such a model (e.g., Strauss and Wadzuk, forthcoming). ANN modeling is well suited to handle partial or poorly understood data sets because the model relies on empirical patterns rather than precise mathematical relationships (Flintsch and Chen 2004), although this approach is most effective with a large efficient set of training data. GSI ANN model reliability is expected to increase as additional data are available, including previously siloed data made accessible through an integrated database structure.Additionally, AI can facilitate a Monte Carlo (MC) simulation approach, which can assess the risk and uncertainty associated with a particular decision (Menendez and Gharaibeh 2017). An MC simulation accepts ranges of values for each input so that the output is a probability density function describing possible outcomes. In GSI applications, MC simulations can inform risk-based decisions that incorporate uncertainty in contributing factors such as weather or soil characteristics (e.g., rain garden clogging and associated risk threshold; William et al. 2019). One limitation is that MC simulations require defined relationships between factors and outcomes, which can be challenging if these dynamics are not well understood or vary over time (such as fall leaf drop).Toward Improved Dynamic and Data-Driven MaintenanceWe are presently in a data revolution, and by tapping into the data collection, modeling, analysis, and use capabilities, we can exponentially grow our ability to implement GSI and ensure its longevity. Key to this goal is to understand GSI’s dynamic performance, and more importantly, what performance failure is and how we can appropriately set risk benchmarks.Dynamic GSI systems and performance data, along with data tools and technologies, can enhance operational decisions and cost-effectiveness of maintenance activities (Wadzuk et al. 2021). Within the municipal stormwater management community, there is a desire for improved data management and communication. Each of the data-driven tools presented here represents an exciting area for future research and potential incorporation into municipal management. These tools employ new technologies that are promising but generally not well documented for GSI asset management. The new GSI IDM, for example, constitutes a novel data management platform explicitly tailored to an environmental observations data format that would easily link with outside databases and applications. A relational data model opens the door for a wide range of other analysis techniques (e.g., GIS, ANN, MC) that can be implemented to perform risk assessments for GSI systems under existing and proposed maintenance programs. To move toward a relational data model, it is recommended to begin merging hard infrastructure data (e.g., siting and hydrologic performance) with management data (e.g., litter index and maintenance records), environmental data (e.g., climate and weather application programming interfaces) and potentially other social data (e.g., demographics, school and other community institutions) to identify existing maintenance areas of concern (Christman et al. 2018; Hosseiny et al. 2020). These areas can be assessed to determine why they require more maintenance and to find trends with nonperforming GSI, as well as to develop targeted lower-cost maintenance approaches. A simple example would be targeting fall leaf litter cleanup in areas with trees based on weather conditions, which may reduce other inspection visits and inform the required skillset of the maintenance team. Outcomes from this type of investigation can help with more efficient maintenance of systems, and can be used in planning and siting, to improve design, and provide other benefits. If a municipality chooses to develop a multidimensional relational structure that can enable different working groups to contribute data and analyze data in new ways, skilled personnel will be needed to develop and maintain this platform.To direct data collection efforts, future work needs to be done on understanding which system variables most reliably predict failure of a GSI system. This type of investigation could link GSI hydrologic performance in response to maintenance and could help inform decisions about allocating limited maintenance resources. A key future need—in general and specific to GSI maintenance—is the need to define failure and determine the risk we are willing to accept. Collectively, there is an abundance of GSI performance data over time and space. With advancements in data management technology, all GSI stakeholders can move to better understand the reliability of their systems, which will ultimately lead to increased GSI sustainability and resiliency.References Baker, A., E. Brenneman, H. Chang, L. McPhillips, and M. Matsler. 2019. “Spatial analysis of landscape and sociodemographic factors associated with green stormwater infrastructure distribution in Baltimore, Maryland and Portland, Oregon.” Sci. Total Environ. 664: 461–473. https://doi.org/10.1016/j.scitotenv.2019.01.417. Blecken, G.-T., W. F. Hunt, A. M. Al-Rubaei, M. Viklander, and W. G. Lord. 2017. “Stormwater control measure (SCM) maintenance considerations to ensure designed functionality.” Urban Water J. 14 (3): 278–290. https://doi.org/10.1080/1573062X.2015.1111913. BMP and TC (Stormwater Best Management Practices and Maintenance Task Committee). 2019. Inspection and maintenance of stormwater control measures. Reston, VA: ASCE. Caltrans (California DOT). 2004. BMP retrofit pilot program final report. Rep. No. ID CTSW–RT–01–050. Sacramento, CA: CALTRANS Division of Environmental Analysis. Christman, Z., M. Meenar, L. Mandarano, and K. Hearing. 2018. “Prioritizing suitable locations for green stormwater infrastructure based on social factors in Philadelphia.” Land 7 (4): 145. https://doi.org/10.3390/land7040145. Clary, J., and H. Piza. 2017. Cost of maintaining green infrastructure. Reston, VA: ASCE. Debo, T. N., and A. Reese. 2002. Municipal stormwater management. 2nd ed. New York: Lewis Publishers. DelGrosso, Z. L., C. C. Hodges, and R. L. Dymond. 2019. “Identifying key factors for implementation and maintenance of green stormwater infrastructure.” J. Sustainable Water Built Environ. 5 (3): 05019002. https://doi.org/10.1061/JSWBAY.0000878. Ebrahimian, A., N. Sokolovskaya, and B. Wadzuk. 2021. “Modeling dynamic performance of urban infiltration trench systems: Methodology and a case study in Philadelphia.” J. Hydrol. 594 (Mar): 125938. https://doi.org/10.1016/j.jhydrol.2020.125938. Ebtehaj, I., and H. Bonakdari. 2014. “Performance evaluation of adaptive neural fuzzy inference system for sediment transport in sewers.” Water Resour. Manage. 28 (13): 4765–4779. https://doi.org/10.1007/s11269-014-0774-0. Emamgholizadeh, S., and R. K. Demneh. 2019. “A comparison of artificial intelligence models for the estimation of daily suspended sediment load: A case study on the Telar and Kasilian rivers in Iran.” Water Supply 19 (1): 165–178. https://doi.org/10.2166/ws.2018.062. Erickson, A. J., V. J. Taguchi, and J. S. Gulliver. 2018. “The challenge of maintaining stormwater control measures: A synthesis of recent research and practitioner experience.” Sustainability 10 (10): 3666. https://doi.org/10.3390/su10103666. Horsburgh, J. S., D. G. Tarboton, M. Piasecki, D. R. Maidment, I. Zaslavsky, D. Valentine, and T. Whitenack. 2009. “Environmental modelling and software an integrated system for publishing environmental observations data.” Environ. Modell. Software 24 (8): 879–888. https://doi.org/10.1016/j.envsoft.2009.01.002. Hosseiny, H., F. Nazari, V. Smith, and C. Nataraj. 2020. “A framework for modeling flood depth using a hybrid of hydraulics and machine learning.” Nat. Sci. Rep. 10 (1): 1–14. https://doi.org/10.1038/s41598-020-65232-5. Houle, J. J., R. M. Roseen, T. P. Ballestero, T. A. Puls, and J. Sherrard. 2013. “Comparison of maintenance cost, labor demands, and system performance for LID and conventional stormwater management.” J. Environ. Eng. 139 (7): 932–938. https://doi.org/10.1061/(ASCE)EE.1943-7870.0000698. Hunt, W. F., W. G. Lord, and J. T. Smith. 2005. “Determining BMP inspection and maintenance costs for structural BMPs in North Carolina.” In Impacts of global climate change, 1–6. Raleigh, NC: Urban Stormwater Consortium of the N.C. Water Resources Research Institute. Meng, T., D. Hsu, and B. Wadzuk. 2017. “Green and smart: Perspectives of city and water agency officials in Pennsylvania toward adopting new infrastructure technologies for stormwater management.” J. Sustainable Water Built Environ. 3 (2): 05017001. https://doi.org/10.1061/JSWBAY.0000824. Salzman, J., and B. H. Thompson, Jr. 2014. Environmental law and policy. 4th ed. St. Paul: Foundation Press. Strauss, M., and B. Wadzuk. Forthcomina. “Predictive maintenance of stromwater infrastructure using internet-of-things technology.” J. Environ. Eng. Tran, D. H., A. W. M. Ng, and B. J. C. Perera. 2007. “Neural networks deterioration models for serviceability condition of buried stormwater pipes.” Eng. Appl. Artif. Intell. 20 (8): 1144–1151. https://doi.org/10.1016/j.engappai.2007.02.005. Wadzuk, B., B. Gile, V. Smith, A. Ebrahimian, and R. Traver. 2021. “Call for a dynamic approach to GSI maintenance.” J. Sustainable Water Built Environ. 7 (2): 02521001. https://doi.org/10.1061/JSWBAY.0000945. Welker, A. L., L. Mandarano, K. Greising, and K. Mastrocola. 2013. “Application of a monitoring plan for storm-water control measures in the Philadelphia region.” J. Environ. Eng. 139 (8): 1108–1118. https://doi.org/10.1061/(ASCE)EE.1943-7870.0000714. William, R., P. Gardoni, and A. S. Stillwell. 2019. “Reliability-based approach to investigating long-term clogging in green stormwater infrastructure.” J. Sustainable Water Built Environ. 5 (1): 04018015. https://doi.org/10.1061/JSWBAY.0000875. Witten, I. H., and E. Frank. 2002. “Data mining: Practical machine learning tools and techniques with Java implementations.” ACM Sigmod Rec. 31 (1): 76–77. https://doi.org/10.1145/507338.507355. Yan, J., J. Jin, F. Chen, G. Yu, H. Yin, and W. Wang. 2018. “Urban flash flood forecast using support vector machine and numerical simulation.” J. Hydroinf. 20 (1): 221–231. https://doi.org/10.2166/hydro.2017.175. Yitian, L. I., and R. R. Gu. 2003. “Modeling flow and sediment transport in a river system using an artificial neural network.” Environ. Manage. 31 (1): 0122–0134. https://doi.org/10.1007/s00267-002-2862-9.