Development of an Integrated Observation Data and Global Numerical Weather Prediction System-Based Rainfall-Runoff Forecasting Model as PJT-I’s Effort in Water Resources Disaster Mitigation

Given the vital role of water resources and their vulnerability to climate-driven disasters, especially floods and droughts, the need for effective disaster mitigation strategies in water resource management is evident. PJT-I, a key player in the industry, grapples with this challenge. To address it, PJT-I has embraced the Smart Water Management System (SWMS) platform and innovative enhancements in rainfall-runoff modeling, showcasing their dedication to bolstering real-time monitoring and predictive capabilities. These advances hold the promise of strengthening operational water resource management, spanning allocation and infrastructure operations, and are particularly relevant in the face of escalating extreme weather events. The study’s focus on the Sengguruh reservoir catchment area serves as an illustration, highlighting the creation of two hydrometric-based models: the AWLR Gadang model and the Sengguruh reservoir model. Their effectiveness is validated through data calibration and integration of GFS forecast data, enabling 120-hour forecasting. Rigorous assessment involving correlation, efficiency, determination, and error metrics provides insights into the models’ predictive capacities and limitations amid the dynamic interplay of rainfall-runoff dynamics.


Introduction
Water resources play a crucial role in supporting various aspects of community life, including agriculture, industry, and domestic needs.However, water resources are vulnerable to disasters such as floods and droughts, which can result in severe impacts on individuals, economies, and ecosystems.The increasing frequency and intensity of extreme weather events due to climate change adds to the urgency of enhancing disaster mitigation strategies in water resources management.
As a leading water resources management company, the State-Owned Enterprise Water Resources Management (PJT-I) Jasa Tirta-I faces challenges in effectively mitigating the impacts of water resources disasters.To address these challenges, besides the need for real-time hydrological parameter observations, a reliable weather forecast information and a robust rainfall-runoff modeling approach are crucial.With the presence of these components, we are able to predict the discharge in rivers and water resource infrastructure, thereby assisting decision-makers in identifying disaster potential earlier and preparing precise flood mitigation measures.
In recent years, PJT-I has implemented the Smart Water Management System (SWMS) platform, which is instrumental in real-time monitoring of water resource data, including discharge, water level elevations, rainfall, and water quality parameters.This system evaluates real-time monitoring of IOP Publishing doi:10.1088/1755-1315/1343/1/012013 2 Reservoir Regulation and River Administration Activities (RTOW and RAAT) compliance, as well as river water quality assessments within PJT-I's working area.With the availability of telemetry equipment owned by PJT-I, technological advancements, and access to global numerical weather predictions, PJT-I has embarked on a revolution in rainfall-runoff modeling to enhance the functionality of the existing SWMS platform.
The integration of real-time observation data from telemetry stations with the utilization of the Global Numerical Weather Prediction System yields valuable insights into rainfall patterns and trends.This advancement drives the development of continuous rainfall-runoff modeling facilitated by computational operating system tools.These efforts are aimed at reinforcing PJT-I's capabilities in operational water resources management, specifically in water allocation and water resource infrastructure operation.This paper outlines the steps taken by PJT-I in formulating the development of the rainfall-runoff modeling.

Study Area
The Brantas River originates from the Arjuna -Anjasmara Mountains at an elevation of 1,547 meters above sea level.Flowing south, then west, and finally east clockwise, it covers a distance of about 320 km.In the upper reaches of the Brantas River, there are several large cascade dam infrastructures, namely the Sengguruh Dam, Sutami (Karangkates) -Lahor Dam, Wlingi Dam, and Lodoyo Dam.The Sengguruh Dam, located at the highest point of this cascade, is designed to trap and reduce sediment entering the Sutami Reservoir.Constructed between 1982-1988, it has a storage capacity of about 21.5 million m³.Administratively located in Sengguruh Village, Kepanjen District, Malang Regency, the Sengguruh Dam is situated approximately 24 km south of Malang city.It lies at the confluence of the downstream Brantas River and Lesti River, both of which are upper components of the Brantas River Basin as shown in Figure 1.
This paper focuses on the Sengguruh Dam's catchment area as the study area for rainfall-runoff modeling across PJT-I's entire working region.

Data
2.2.1.Observational Data.PJT-I has implemented rainfall and water level observations through the utilization of Automatic Rainfall Recorders (ARR) and Automatic Water Level Recorders (AWLR).Both are equipped with data loggers and GSM modules that can transmit observation data to a central office server database.Through this system, PJT-I can collect and send observational data periodically (hourly intervals).This observational data enables PJT-I to access real-time information from various monitoring locations without the need for physical visits to each station or deploying personnel for manual recording.The scheme can be seen in Figure 2.
Within the scope of this study, the hydrometric network within the Sengguruh reservoir catchment area consists of 18 ARR stations, 1 AWLR station, and 1 manual observation station operated by personnel at the Sengguruh Dam.List of telemetry stations as shown in Table 1.Personnel monitor the reservoir's water level every hour and record it through an application developed by PJT-I.This process ensures that observational data is directly transmitted to the server database in real-time.This hydrometric station network serves as input data and calibration for the developed rainfall-runoff modeling.The GFS employs numerical models and weather observation data with a horizontal grid system resolution of 0.25° x 0.25° to predict weather variables such as temperature, humidity, atmospheric pressure, wind, and precipitation.This model forecasts weather changes within a time range spanning from a few hours up to several weeks ahead.The GFS undergoes updates four times daily at 00:00, 06:00, 12:00, and 18:00 UTC, providing forecasts for up to 16 days in advance.The initial 120 hours of the forecast offer hourly predictions, while days 5 to 16 provide predictions every three hours [1,2].PJT-I employs the GFS as rainfall forecast data in its discharge forecast modeling.Adapting to the telemetry recording frequency of per hour, the GFS data utilized consists of hourly data for the first 120 hours.The GFS data is updated every 6 hours in accordance with the latest released data, ensuring the acquisition of the most current information.Presented as spatial grid data, the GFS data requires interpolation through the bilinear method to estimate rainfall values aligned with the location points of the ARR stations managed by PJT-I.

Bilinear Interpolation Method.
Figure 3 shows the configuration of a target point on which conventional bilinear interpolation proceeds using the four rectangular points surrounding the target point.The procedure is simply linear interpolations [3], first along the x-axis and then along the y-axis.Centering the target point, weighting for the interpolation is given by the area ratio of the four rectangles to their area sum.This way of weighting is identical to the conventional bilinear interpolation method.

Thiessen Polygon.
The Thiessen Polygon method was developed by Thiessen.According to the location of rain gauges, polygons are formed by the perpendicular bisectors of the lines joining nearby gauges.Thus, each polygon contains only one rain gauge, and the weights of the rain gauges are computed by their relative areas, which are estimated with the Thiessen polygon network.The average precipitation in each sub-basin is calculated in Equation (1).In this study, the mean precipitation generated in each sub-basin was incorporated into the hydrological simulation by creating a virtual rain gauge within the centroid of each sub-basin [4].
where  ( 0 ) = average precipitation in the center of sub-basin  (  ) = measured precipitation at the rain gauge i   = area of thiessen polygon associated with gauge i  = area of the sub-basin

Australian Water Balance Model.
The AWBM works on the basis of three independent surface storages for computation of partial runoff from a catchment [5,6] (Figure 4).The model independently calculates water balance for each surface storage using precipitation and evapotranspiration as inputs.
The whole catchment in this model can be divided into three partial areas using three parameters as A1, A2 and A3.The soil moisture for each partial area is computed by adding precipitation and deducting evapotranspiration.At any time, rainfall is added to each of the partial area or storage and after deducting evapotranspiration, the following water balance equation is used to compute storage and if the moisture becomes greater than the capacity of storage, the moisture in excess of capacity becomes runoff.
n =  n−1 +  n −  n (7) Where  n−1 and  n are the storage on n − 1 and nth days,  n and  n are the precipitation and evapotranspiration on nth day.In this model, friction () of runoff in any partial storage gave base flow and reminder may provide surface runoff from that store using following equations.
Both the surface runoff from three partial storage and base flow are depleted on daily or hourly timescale with the help of surface flow recession coefficient () and baseflow recession coefficient () in a linear manner using the following equations: The routed surface and base flow at the outlet of the basin provide total runoff from the basin.The different parameters of AWBM along with the range are given in Table 2.  5).Observation data forms the foundation of the model.This includes rainfall monitoring data from ARR stations, water level and discharge from AWLR stations, and on-site personnel observations in water resource infrastructure such as dams, weirs, and gates.NGWP provides information about potential rainfall forecasts.This data serves as crucial input for the modeling process and enables the model to forecast runoff and potential flood occurrences.
Hydrologic modeling is the core of the system.Integrated model devices are used to depict the hydrological processes within the system.Observation data and NGWP are analyzed by the model, resulting in model discharge and forecast discharge outputs.The initial phase involves hindcasting model, where observation data is used to calibrate and validate model parameters.By doing so, the model becomes more accurate and can reflect actual hydrological conditions.After validation, a forecast model is generated based on the validation parameters.
Lastly, high-performance computing servers are a vital infrastructure supporting data processing and model computations.The urgency of high-performance computing server needs ensures that data processing and computing systems can be conducted quickly and efficiently, allowing the model to respond to changes and produce output in real-time and provide timely information to stakeholders.
Overall, the integrated system model effectively combines observation data processes, global weather prediction data retrieval, hydrologic modeling, and high-performance computing server to provide valuable information for operational water resources management, including water allocation and water resource infrastructure operation.

Rainfall-Runoff Model
Rainfall-runoff modeling uses a flexible modeling environment designed for hydrological modeling of rainfall and associated river flow in river catchments as well as general time series processing and simulation.The modeling process uses an integrated scripting language providing flexibility for developing modeling and simulation methods as well as the option to adapt or extend the set of standard modeling elements and functions included with the standard library.
The main elements of models are nodes and links.Links typically represent channels, conduits or river reaches through which water may flow from one location to another.In hydrological models, links configured to hydrological routing to model the flow between an upstream and downstream river location.Nodes are connecting points that links attach to and might represent a particular location or branch within a river system.Both nodes and links may have a rule script that defines how they operate at each time step of the model.The model itself will have timing properties that control the period over which the model runs and the time step that it uses.Data from all time-series inputs will be automatically synchronized to the time step of the model which means that the model time step can be changed without any requirement to do anything with the input data.The rainfall-runoff model considers rainfall obtained through thiessen weighting from various monitoring stations and using the AWBM method, both of which are analyzed at the Global Precipitation node.Then a river basin scheme is created to analyze river flow routing.This research presents two models developed for the Sengguruh Reservoir catchment area, namely the AWLR Gadang and Sengguruh Reservoir models.As an example of the modeling process, for calibration purposes as in Figure 6 and 8, the model utilizes data from March to April 2023, likewise for validation as in Figure 7 and 9, the model integrates data from May to July 2023.
During the calibration phase, the model's parameters are adjusted using information from the initial segment of the dataset.Subsequently, the resulting optimal parameters are employed to replicate the observed conditions in the second segment of the dataset as part of the validation process.After the model demonstrates satisfactory performance during the validation process, the calibrated parameters can then be applied to forecasting scenarios for a 120-hour period from the current time.The proactive approach to this prediction involves the integration of GFS forecast data.The output of the process is as shown in Figures 10 and 11

Goodness of fit criteria
The suitability of the AWBM Model was assessed based on criteria such as the coefficient of correlation (r), Nash-Sutcliffe efficiency (%), coefficient of determination (R 2 ), and root mean square error [9].The AWBM model results for the Gadang AWLR during the calibration and validation periods (March to April 2023 and May to July 2023) as in Table 3 show the following performance metrics: coefficient of determination (R 2 ) of 0.393 and 0.228, coefficient of correlation (r) of 0.627 and 0.477, Nash-Sutcliffe efficiency (%) of 24.4% and -24.7%, and root mean square error of 38.479 and 16.999, respectively.
Similarly, for the Sengguruh Reservoir during the calibration and validation periods, the AWBM model yielded the following results: coefficient of determination (R2) of 0.429 and 0.614, coefficient of correlation (r) of 0.655 and 0.784, Nash-Sutcliffe efficiency (%) of 9.3% and 59.4%, and root mean square error of 28.678 and 16.152, respectively.

Summary and Conclusions
Water resources are essential for supporting various aspects of life, yet they are vulnerable to disasters like floods and droughts.The intensifying extreme weather events due to climate change highlight the need for better disaster mitigation strategies in water resources management.As a leading company in this field, PJT-I faces challenges in effectively addressing water resource disasters.We have implemented the Smart Water Management System (SWMS) platform, crucial for real-time water resource monitoring, and are revolutionizing rainfall-runoff modeling to enhance SWMS functionality.This advancement aims to strengthen our operational water resources management capabilities, including water allocation and infrastructure operations.This paper examines rainfall-runoff modeling in the Sengguruh reservoir catchment area.There are two models developed based on the hydrometric network in the catchment area, namely the AWLR Gadang model and the Sengguruh reservoir model.Calibration using data from March to April 2023, followed by validation with data from May to July 2023, demonstrates the models' effectiveness.The integration of GFS forecast data allows for forecasting over a 120-hour period.
The model performance was evaluated using multiple criteria, including correlation coefficient (r), Nash-Sutcliffe efficiency (%), coefficient of determination (R 2 ), and root mean square error.The model exhibited varying results during calibration and validation periods.These findings provide a comprehensive understanding of the model's predictive capabilities and its strengths and limitations in simulating rainfall-runoff dynamics for the respective locations.

Figure 1 .
Figure 1.Location of Sengguruh Dam at the downstream confluence of Brantas River and Lesti River, and telemetry station's locations.

Figure 2 .
Figure 2. Telemetry equipment aystem using database management system as the data storage at PJT-I.

Table 1 .
Telemetry Station in Sengguruh Reservoir Catchment Area 2.2.2.Numerical Global Weather PredictionSystem.The GFS (Global Forecast System) is a global numerical weather prediction system developed and operated by the National Centers for Environmental Prediction (NCEP) under the United States National Oceanic and Atmospheric Administration (NOAA).

Table 3 .
Goodness-fit criteria of AWBM Model