A decision support system for indirect potable reuse based on integrated modeling and futurecasting

Optimal operation of water reclamation facilities (WRFs) is critical for an indirect potable reuse (IPR) system, especially when the reclaimed water constitutes a major portion of the safe yield, as in the case of the Occoquan Reservoir located in Northern Virginia. This paper presents how a reservoir model is used for predicting future reservoir conditions based on the weather and stream ﬂ ow forecasts obtained from the Climate Forecast System and the National Water Model. The resulting model predictions provide valuable feedback to the operators for correctly targeting the ef ﬂ uent nitrates using plant operations and optimization model called IViewOps (Intelligent View of Operations). The integrated models are run through URUNME, a newly developed integrated modeling software, and form a decision support system (DSS). The system captures the dynamic transformations in the nutrient loadings in the streams, withdrawals by the water treatment plant, WRF ef ﬂ uent ﬂ ows, and the plant operations to manage nutrient levels based on the nitrate assimilative capacity of the reservoir. The DSS can provide multiple stakeholders with a holistic view for design, planning, risk assessments, and potential improvements in various components of the water supply chain, not just in the Occoquan but also in any reservoir augmentation-type IPR system.


INTRODUCTION
Reliable and optimal operation of a water reclamation facility (WRF) is a fundamental risk management component for an indirect potable reuse (IPR) system. It requires timely and informed decision-making in response to fluctuating operational conditions, such as weather patterns, plant performance, and water demand. Futurecasting of IPR systems consists of modeling different near-term future scenarios supplemented by medium-to long-range weather forecasts, which provides useful feedback to decision-makers. Such predictive analyses can aid in alleviating future risks associated with water availability and quality without the need to install and maintain large standby capacities.
Over the years, the interest in integrated water resource management (IWRM) has increased significantly. IWRM takes a comprehensive approach to water management by viewing the water supply, drainage, and sanitation systems holistically. To develop design and management tools for a better understanding of these integrated systems and their complex behaviors, an integrated modeling approach is required. According to Rauch et al. (), the concept of integrated modeling was proposed as far back as 1970.
The first integrated urban drainage model was applied in 1980, though the concept did not become widely adopted until the 1990s (Mitchell et al. ). Since then, there have been many applications of IWRM modeling focused on the simulation of entire urban water systems (Schütze et al. ; Coombes & Kuczera ). Similar approaches have also been used in integrated watershed management. There are several approaches to integrated modeling.
Some software applications provide built-in integration to simulate a coupled system, where the models for different sub-systems are either contained within the software or are linked automatically. Coupled models are based on integrating different standalone models that simulate various components of the target system individually (e.g. watershed, reservoir and treatment plants). Integration of such models is carried out using different levels of sophistication: from manual coupling to fully automated systems.
Coupled models are commonly operated sequentially (Schütze et al. ; Rauch et al. ; Xu et al. ). This loosely coupled modeling approach is unidirectional and flow paths are configured as a tree-like structure. Although this approach seems relatively simple, there is voluminous data input, output, and transfer between the models (Azmi & Heidarzadeh ). In addition, some of these models are not very user-friendly, take a significant amount of time to run, and require a significant level of familiarity with the model structure to operate. A more advanced coupling approach called iterative coupling, on the other hand, relies on the tight coupling of sub-systems to create model synchronization based on back-and-forth data transfer for each timestep using either a standard or custom protocol.
An example of one such standard framework is OpenMI (Open Modeling Interface) (Moore & Tindall ). forecast data (7-45 days) to study water and agriculture resource management in India using the GEFS (Global Ensemble Forecast System) and CFS v2 (Climate Forecast System). By using forecast data as an input to simulate forecasted runoff and soil moisture, they were able to show that their methodology could provide timely information in decision-making for farmers and water managers. A near real-time drought monitor was also developed to estimate the severity and extent of agricultural and hydrologic droughts (Shah & Mishra ).
The overarching goal of this research is to develop a futurecasting application based on the integration of various in-house and weather forecast models and historical data sources. The outputs can provide useful feedback for plant operators to identify and analyze strategies to manage the WRF performance dynamically in response to future reservoir conditions. This paper will discuss the development and implementation of the integrated modeling application and will demonstrate its effectiveness as a decision support system (DSS) to inform and improve the reliable operation of the IPR system.

Study area
The Occoquan Reservoir is located in Northern Virginia and is one of the two major water supply sources (the other being the Potomac River) for several municipalities in the region. The reservoir spreads over an area of 6.9 km 2 and its drainage basin, the Occoquan Watershed (Figure 1), and spans across 1,484 km 2 (573 square miles).
The reservoir has a full pool volume of 3.1 × 10 7 m 3 , an average water depth of 5.1 m, and a maximum water depth of approximately 19 m close to the dam. The reservoir has a full pool elevation of around 37 meters above mean sea level (MAMSL), and mean hydraulic residence time is approximately 20 days (Xu et al.  Figure A1, available with the online version of this paper). Temperature, dissolved oxygen (DO), nitrate, pH, oxidation-reduction potential, and conductivity data are collected in situ at different water depths (usually at 1.5 m intervals, starting at 0.3 m from the surface) at frequencies ranging from once a month to three times a month. Many other constituents (such as nitrogen and phosphorus forms and total organic carbon) are measured, via sample retrieval and transport to the laboratory for analysis, at the top and bottom of the reservoir stations. OWML also operates a weather station located close to the WRF (called OWML weather station at UOSA), which measures different meteorological parameters including rainfall, air temperature, solar radiation, wind speed and direction, and humidity, which are transmitted on an hourly basis to OWML's database servers.

METHODS
Regulation of nitrate in reclaimed water throughout the year is a unique operational challenge at UOSA. Its operating permit limits the total annual nitrogen load to 6.0 × 10 5 kg (1.316 × 10 6 lb). More than 90% of this load is discharged in the form of oxidized nitrogen, mainly as nitrate (NO 3 ).
Although the MCL (maximum contaminant level) for nitrate in drinking water is 10 mg/L as nitrogen, the Occoquan Policy (VSWCB ) specifies a Water Quality Objective (WQO) of 5 mg/L at the dam. The WRF is, therefore, required by permit to reduce its nitrate discharge concentration when the nitrate concentrations at the drinking water intakes rise above 5 mg/L.
It has been observed that during thermal stratification and extremely low hypolimnetic oxygen concentrations in the summer, the supply of nitrate from UOSA actually benefits the reservoir. Under these conditions, microbes in the sediment utilize nitrate as an electron acceptor in the absence of oxygen, advancing the denitrification process. As a result, the release of phosphorus (P) and other less preferential electron acceptors, Mn 2þ and Fe 2þ , into the water column is inhibited. The reservoir overturns at the beginning of fall and remains well-mixed throughout the winter and early spring. The mixing redistributes oxygen across the entire water column and restores the aerobic conditions, thereby diminishing the denitrification capacity of the reservoir.
A nitrate discharge optimization study (Bartlett ) concluded that the total nitrogen load delivered into the reservoir by the WRF can be more effectively distributed each month based on the reservoir's denitrification capacity.
Based on these recommendations, UOSA changed its operational strategy by reducing the degree of denitrification in summer to ensure a higher concentration of nitrate in the effluent and then transitioning back to increased denitrification just before reservoir turnover, generally in early fall.
Using the optimized load distribution provided in that study, average monthly concentrations were calculated using the average monthly flows from the last 5 years  As also noted by Bartlett (), these optimized loads cannot be used as a fixed allocation schedule for the WRF due to the dynamic nature of the system. In winter months, the optimized nitrate load distribution is quite conservative, as it is calculated based on the worst-case scenario using extremely low natural streamflows (winter drought).
The effect of any change in the discharged load from the WRF on the nitrate concentrations at the dam is based on the hydraulic retention time of the reservoir, which may vary from a few days to months, depending on the inflows. Therefore, many factors including the pool elevation, volume and quality (temperature and background nitrate concentration) of stream inflows, withdrawal by Griffith WTP, and weather forecast can all affect the future conditions of the reservoir and consequently its denitrifying capacity. Hence, selecting the desired monthly nitrate concentration for the effluent is still a trial-and-error process, largely based on the operator's judgment.
The growing water demand caused by urbanization complicates this process further. The safe yield of drinking water from the Occoquan Reservoir, including the reclaimed water, is 3.0 × 10 6 m 3 /day (79 mgd). UOSA's contribution to the inflows has been steadily increasing over the years and, based on the current projections, will exceed 50% of the safe yield after 2025. In reference to

Intelligent View of Operations Model
UOSA is currently using a wastewater and reuse process simulation software IViewOps (Intelligent View of Operations) as a day-to-day tool for simulating changes to operation and effect of assets out of service for maintenance (Sen et al. ). IViewOps is a multi-layer model that analyzes and optimizes the plant on three levels: (1) biochemical process modeling, (2) asset's condition CFSv2 reforecast data are available for 29 years , from every January 5th, on the same horizontal

National Water Model
The Since CE-QUAL-W2 cannot simulate complex bubbleplume dynamics, oxygen mass rate (kg/day) must be individually inputted for each layer and mixing must be simulated using a vertical mixing coefficient. This method is an oversimplification of complex bubble-plume dynamics and therefore is usually unable to predict the correct oxygen con-    to run a different condition (e.g. input different meteorological data). Furthermore, different scenarios and processes can be run in a batch mode manually or by using the Scheduler feature, which allows runs to be scheduled at different times or on a periodic basis.

Operational configuration
The DSS was set up using two separate projects in URUNME.
The first project ( and performs extensive data analysis and manipulation before feeding it to the model as inputs or storing it as historical trends. All the operations carried out during the entire simulation process are fully automated and use hundreds of functions in various process flow diagrams. The overall data flow path to operate the model is shown in Figure 3.

Calibration scenario
The calibration scenario, called 'base', is used to evaluate the current model calibration using the last 3 years of observed data. Note that for the Occoquan case described in this paper, these choices on data and simulation length are those taken for the Occoquan model application and are not limitations imposed by URUNME. As long as there is sufficient storage space and memory available,

Forecast scenarios
To run the forecast scenarios, the latest CFS operational forecast GRIB2 files are automatically downloaded from the NOAA website for the four selected grid points closest to the UOSA weather station (Table A1,

RESULTS AND DISCUSSION
The results presented in this section are for RE02, extracted from the DSS output generated on October 10, 2018. RE02 is located at around 0.5 km from the dam and is the closest sampling station to the WTP intakes ( Figure 2).
Calibration of the coupled model  Table 2.

Temperature
Due to inadequate hypolimnetic mixing, initial runs of the model underpredicted the bottom water temperatures at RE02. The poorer mixing was caused by a lower value of plume rise (DMPR) computed by the LPM than the observed value of plume rise caused by hypolimnetic aeration.
where f i is a calibration factor, SOD i is the temperaturecorrected SOD, and Q i is the oxygen flow rate in N-m 3 /h for segment i at the current timestep. An f value of 0.07 corresponds to a 6.6 times increase in SOD at a maximum oxygen flow rate of 80 N-m 3 /h for each segment. NSE values shown in Table 2 indicate 'good' to 'very good' performance for dissolved oxygen with the exception of RE30, where NSE for dissolved oxygen is 0.63 ('satisfactory'). The PBIAS for all the reservoir stations remained within ±15% which is considered 'very good'. RMSE values remained between 1.7 and 2.5 mg/L and R 2 between 0.7 and 0.87.

Nitrates
The nitrate calibration has shown mostly 'satisfactory' NSE values, with the exception of RE20 and RE35 which are    October to January (Table 1). N 2 and N 3 were set at 7.5 mg/L and 12.5 mg/L, respectively. Discharge and withdrawal scenarios were set similar to 30-day forecast.
The forecasted bottom-layer water temperature profiles are shown in Figure A4   scenarios. Additionally, the reservoir denitrification capacity remained significantly high during September due to relatively higher water temperatures causing low nitrate concentrations.
For D 1 W 1 (Figure 8(a)), the nitrate concentrations at RE02 reached a maximum of 1.7, 2.3, and 2.9 mg/L for N 1 , N 2 , and N 3 , respectively, at the end the forecast. A drawdown of 0.63 m (2.1 ft) in the pool elevation was simulated by the model due to higher WTP withdrawals compared to the total inflows to the reservoir on average. Although D 2 W 1 (Figure 8( As elaborated in the above discussion, the initial pool elevation and, hence, the HRT of the reservoir play a crucial role in the outcome of the forecast. To understand the effect of how lower pool elevations can affect the nitrate concentrations in the reservoir, a hypothetical condition was considered where an already existing drought condition in the fall was extended into winter (the simulation was only done for this study and is not a part of regular DSS forecast).
All the 12 scenarios were rerun by creating an artificial drawdown of 2 m just before the start of the 90-day forecast, which corresponds to a reduction of 33% in the reservoir volume ( Figure 9). It can be seen that in all the scenarios, the higher effluent nitrate concentration of 12.5 mg/L (N 3 ) will eventually push the nitrates to the WQO limit in the forecasted period. The worst-case scenario remains D1W2 where the nitrate reaches 5 mg/L at RE02 in around 77 days.
The above results presented how the dynamic nature of the system can affect the reservoir based on different boundary conditions. Different meteorological parameters can also have significant effects on the outcome. For instance, colder than normal air temperature in fall or winter can also increase the nitrate concentration at RE02 by reducing the denitrification capacity of the reservoir. On the contrary, any significant rain event can washout the entire reservoir diluting the nitrate concentrations to much lower levels.

Process optimization
Every week, the DSS moves forward in time by predicting the reservoir water quality based on the most recent initial and future boundary conditions. When the forecasted water quality going out 30-60 days is above the 3 mg/L Less denitrification means potential cost savings due to the following: • Reduced amount of carbon required for denitrification and therefore more BOD (biochemical oxygen demand) available for the digesters to increase biogas production for heating and electricity generation.
• Less supplemental carbon (methanol and MicroC) required for denitrification.
• Less anoxic zone mixing requirements and reduce nitrate recycle pumping rates.
• Lower anoxic volumes reduce undesired luxury bio-P uptake resulting in less soluble phosphate release and consequently lower struvite formation in digesters. This reduces the potential cost of descaling of the centrifuge system and struvite control chemical feed.

CONCLUSION
This paper presented the development and implementation of a DSS for the Occoquan IPR system to forecast and regulate the nitrate concentrations in the augmented reservoir.
The entire DSS application is powered by newly developed software, called URUNME, to fully automate the operation of the integrated reservoir and process models, forecasting products, and various data sources. URUNME was also used to develop information-rich and user-friendly interactive dashboards to output the updated historical data and forecasting results for the stakeholders.
The reservoir model is operated once every week, running multiple future scenarios based on different combinations of natural stream inflows, plant effluent flows and nitrate concentrations, and water withdrawal by the WTP.
The future weather conditions are simulated using the forecasts obtained from the CFS. A 30-day forecast predicted the reservoir conditions based on the NWM streamflow forecast going out 30 days. A 90-day forecast, on the other hand, is used as a What-If analysis to simulate the effects of any possible future winter drought.
The forecasted nitrate-N concentrations under different scenarios can be used by the WRF to identify and analyze operational strategies dynamically in response to the reservoir future conditions. In the event that nitrates going out 30-60 days exceeds 3 mg/L, the DSS can be used to run different optimization routines on IViewOps to determine the means to reduce the nutrient discharge. Even when the target nitrates effluent concentrations are difficult to achieve due to operational constraint, the DSS output provides a fair estimate of the time it would take for the reservoir to reach unacceptable nitrate levels. This provides valuable feedback to the plant managers for contingency planning, e.g. rescheduling the maintenance of certain assets, adding supplemental carbon chemicals, and changing plant configuration, to avoid any future emergency conditions. As the DSS progresses forward in time, the optimization of the WRF is revisited every 7 days with updated outputs due to the dynamic nature of the system. For instance, any significant rain event can partially or completely wash out the entire reservoir, diluting the nitrate concentrations to much lower levels.
The file transfer system makes it easier for the WRF to run the simulations based on reservoir model runs done at OWML. Each week, URUNME is run to generate a database file that is sent to the WRF via FTP. The plant uses the URUNME interface to run this together with IViewOps and populate relevant data in a screen that is a sister screen to SCADA. The operations and maintenance staff can then make decisions about how to optimize the plant and what the future conditions would be with optimization.

Future applications of URUNME-based DSS
Integrated modeling using the URUNME system provides a comprehensive approach to environmental management by evaluating different components of a system in a holistic manner. It can be used as an effective tool for design, planning, and risk assessments, and can facilitate informed decision-making, enhancing the transparency and collaboration between different stakeholders. Moreover, it can be used for simulating, understanding, and managing a combination of wastewater, drinking water, and groundwater interaction, such as the following: 1. IPR applications in Southern California (Orange County).
2. Wastewater treatment, discharge to a reservoir, extraction for drinking water supply, and drinking water treatment (IPR, Wichita Falls, Texas).
3. Raw water supply from a reservoir, water treatment, and sludge disposal to the wastewater treatment plant (WSSC (Washington Suburban Sanitary Commission), Maryland).
Although developing a reservoir model from scratch is a major undertaking that requires a significant amount of data for setting up and calibrating the model. However, there are certain components of this DSS which can be reused for similar applications without any major modification: • Downloading the data, reading the binary file formats (GRIB2, NetCDF), and calibrating the forecasts from the CFS and NWM.
• Obtaining data from different data sources (stream stations, reservoir stations, and weather stations), manipulating and transforming data, and writing the boundary conditions and input parameters to the reservoir model.
• Different screens created as part of the DSS dashboard.
The use of a DSS driven by URUNME can be used in various systems in addition to water resources. These can include chemical and biological engineering systems and their models, manufacturing and supply chain management systems and their models. URUNME is available free of cost for academic use (www.urunme.com).