Why Model Real Time When You Can Model the Future? The Fort Wayne Experience

Our presentation will summarize the development and implementation of the forecasting tool for Fort Wayne City Utilities' collection system model. The tool uses an embedded Python script in PCSWMM to obtain the NOAA rainfall forecast and NWS river stage forecast to simulate 48 hours into the future for operational insight. In addition to the forecasting tool, a cloud-based Azure Function was developed to run the model in the cloud and reduce wear and tear on City computers. Fort Wayne City Utilities provides water, wastewater and stormwater to the second largest city in Indiana with approximately 72,200 customers serviced by 1,500 miles of sewers over a 68 square mile service area. City Utilities is presently implementing a combined sewer overflow (CSO) consent decree that will complete in 2025 that includes the Three Rivers Protection & Overflow Reduction Tunnel (3RPORT). As the tunnel and other consent decree projects are brought online, City Utilities needs to leverage its long term investment in hydraulic modeling by incorporating near-term forecasting. Figure 1 presents the forecasting process. The forecasting tool uses an embedded Python script in PCSWMM to obtain the following public domain, free forecasts: 1.NOAA High Resolution Rapid Refresh (HRRR). The 2.5-km, 1-hour resolution forecast provides the next 18 hours of rainfall for the wastewater service area. 2.NOAA Global Forecasting System (GFS). The 20-km, 3-hour resolution forecast provides the next 72 hours of temperature. 3.NWS Advanced Hydrologic Prediction Service (AHPS). The 6-hour resolution forecast provides the next 4-5 days of river stage from multiple gauging stations. Located at the confluence of the St. Marys, St. Joseph, and Maumee Rivers, the collection system is influenced by high river stages. The NWS river forecast is obtained by a simple web scraping technique while the NOAA forecasts are obtained using the MetPy library (Unidata, 2021). The three forecasts are then formatted to PCSWMM's structure, automatically added to the model as rainfall, climate, and river stage time series files, with the model simulation time set to the reflect the forecast period. Figure 2 presents City Utilities' model and the embedded script, which can be used to automatically run the model after obtaining the forecast. The forecasting script was completed in October 2022 and has been utilized by City Utilities staff ahead of storm events. For the October 31, 2022 event, the HRRR forecast ranged from approximately a quarter of an inch to two-thirds of an inch. Table 1 presents a comparison of the modeled forecast to actual metered CSO data. As shown in the table, the actual rainfall was consistent with the NOAA forecast, and forecast model is conservative consistent with City Utilities' periodic comparisons of historical model simulations and outfall metering data. Similar to many combined sewer communities with a Consent Decree, City Utilities frequently simulates a typical 5-year period as part of assessing CSO controls. The 5-year duration has provided challenges to staff running the long-term simulation on their laptops. To ease the burden on City Utilities' staff laptops, an Azure Function was developed to run the model in the Microsoft Azure cloud environment as a lower-cost alternative to a virtual machine (VM). The Azure function loads the model input file from an Azure storage container, runs the model using the PySWMM library (McDonnell et. al., 2020), compresses the model output, and saves it to the Azure container for City Utilities staff to download. The Azure function was completed in October 2022 and is presented in Figure 3. We believe our presentation will benefit any wastewater or stormwater utility with a hydraulic model and interest in forecast modeling or running the model in the cloud, applied hands-on by utility staff. We will present our lessons learned in deploying the forecasting tool and the cloud application and benefits to City Utilities.