Performance of GPM-IMERG satellite precipitation for rainfall-runoff modeling in Indonesia

Accurate discharge predictions are critically needed for the management of water resources, including irrigation, drinking water supply, hydropower, reservoir management, environmental flow, flood assessment, and other related activities. Analyzing flood hydrographs, particularly in unmeasured basins, can be facilitated by utilizing satellite precipitation data or hydrological models to estimate river discharge. Prediction of runoff can be achieved by employing rainfall-runoff modeling. Hydrologic Engineering Center - Hydrologic Modeling System (HEC-HMS) is a common rain flow modeling tool to used worldwide. This program is easy to use because hydrological modeling can be done with rainfall and discharge data as input and can perform various roles, one of which is to estimate the maximum flow rate of a watershed (Febriyanto et al. 2018). Based on the previous study, HEC-HMS generally produced acceptable calibration and validation results, performing slightly better than Soil and Water Assessment Tool (SWAT) during the calibration phase for the Manafwa catchment, which statistical tests such as the t-test confirmed (Sempewo et al. 2023). However, precise runoff modeling encounters challenges due to the limited availability of rainfall data, as not all watersheds include rain gauges that are uniformly distributed across the whole watershed namely the Mayang Watershed.

A potential way to address the absence of field rainfall data is to utilize satellite precipitation data to substitute for missing or incomplete rain gauge data. Satellite precipitation data is capable of capturing rainfall information by remotely measuring or recording rainfall, a task that is challenging for rain gauges deployed in the field. Satellite rain data products provide the benefits of being readily accessible, providing real-time information, and offering extensive coverage in terms of both time and space (Yeditha et al. 2022). The availability of satellite data enables the acquisition of rainfall information at any location and time. The Global Precipitation Measurement – Integrated Multi-Satellite Retrieval for GPM (GPM-IMERG) is one of the satellites that collects rainfall data globally. GPM-IMERG provided adequate spatiotemporal coverage and was capable of monitoring the precipitation rate in near real time, utilizing passive microwave and infrared satellite observations to provide rain rate estimates every 30min in 0.1° × 0.1° fields (Giannaros et al. 2022). A study by Saouabe et al. (2022), found that while the GPM-IMERG data slightly underestimated the amount of rainfall, it had a good correlation with ground observations and showed smaller mean error values, indicating better performance in certain metrics. Based on previous study in Bedadung Watershed, Jember Regency, Indonesia, the uses of GPM-IMERG outperform the Tropical Rainfall Measuring Mission (TRMM) for various hydrological purposes, suggesting that it provides a more accurate representation of rainfall data (Hidayah et al. 2021). Furthermore, based on the study by Ahmed et al. (2020), the GPM-IMERG is considered a suitable replacement for TRMM 3B42 for rainfall-runoff modeling in ungauged watersheds because of higher correlation coefficients, better Nash–Sutcliffe efficiency (NSE) values, and lower percent biases, offering better accuracy and reliability in Chenab River, Pakistan.

The main technologies used in the collection of rainfall data by the GPM and TRMM satellites involve advanced instruments such as the precipitation radar and dual-frequency precipitation radar, which emit microwave signals to measure the intensity and vertical distribution of rainfall (Ryu et al. 2021; Gorooh et al. 2022). Additionally, the TRMM Microwave Imager and GPM Microwave Imager detect natural microwave radiation to estimate rainfall intensity, while infrared sensors measure cloud temperature to identify potential precipitation (Sekaranom & Masunaga 2019; Milani et al. 2022). These satellites regularly orbit the Earth, collecting data from various regions globally, and the data collected is processed and validated with ground-based observations to ensure accuracy (Seela et al. 2024). This process enables continuous monitoring of rainfall worldwide, including in areas that are difficult to reach by conventional measurement methods (Ryu et al. 2021; Fusco et al. 2022; Li et al. 2023).

The Mayang Watershed is characterized by a small size and an unequal distribution of rain gauges, as well as variations in topography. Consequently, it is necessary to evaluate the average amount of precipitation across a certain area. Therefore, this study will examine the distribution of rainfall using the Arithmetic, Polygon Thiessen, and Isohyet methodologies due to the small size and highland location of the studied area. This study aims to predict discharge data based on satellite precipitation data GPM-IMERG using the HEC-HMS program in the Mayang Watershed, Jember Regency. Talang station serves as the designated outflow point for the calibration and model validation of the Mayang Watershed, namely for observation discharge control. The sensitivity test is used to identify the most influential parameters affecting the model outputs. The model results will be reassessed with various temporal variations of rainfall data and sensitivity analyses on the parameters will be conducted.

The study area encompasses six distinct land use categories: residential and public areas, forests, gardens, fields, rice fields, and shrubs as the detailed shown in Table 1. The soil type is utilized in the classification of soil type and land use to ascertain the value of the curve number (CN). Furthermore, there are two types of soil in this area which are andosol and Mediterranean.

This study includes using hydrological and geographical data. The hydrological data includes the daily rainfall data and the daily discharge observed data (2012–2021) was obtained from Binamarga PU and SDA Jember Regency, along with the GPM-IMERG satellite rain data spanning a decade (2012–2021) was acquired from the website https://giovanni.gsfc.nasa.gov/giovanni/. The available observed data with good and consistent quality only covers a period of 10 years, while longer records have poor data quality. Moreover, the geographical data, including the digital elevation model data, were sourced from the website https://tanahair.indonesia.go.id/demnas/#/demnas, and the land use map (2017) was obtained from the website https://tanahair.indonesia.go.id/portal-web/download/perwilayah.

During the data processing stage, a rainfall correlation analysis was conducted, followed by calculating the area average utilizing both observational and satellite rainfall data from GPM-IMERG.

If the correlation coefficient value shows a larger value between the two, it means that the relationship between the two is getting stronger. In other words, the satellite data estimation pattern is getting closer to the actual data pattern at the observation rain station (Ng et al. 2023). The following justification can be drawn from the correlation coefficient value: 0 (no correlation between two variables), 0–0.25 (very weak correlation), 0.25–0.5 (moderate correlation), 0.5–0.75 (strong correlation), 0.75–0.99 (very strong correlation), and 1 (perfect correlation) (Octafiani Hartawan et al. 2021).

The watershed encompasses multiple rainfall stations, necessitating the computation of average rainfall to gauge precipitation across the area. Three methods used to calculate the area's average rainfall for observational and satellite rain data in this study are the algebraic, Thiesen, and Isohyet methods because the watershed area is small. The Arithmetic method offers simplicity by summing rainfall measurements from all gauges and dividing by the total count, but it overlooks spatial variability, assuming uniform distribution (Ribeiro et al. 2021; Zakaria et al. 2023). In contrast, the Thiessen polygon method partitions the watershed into polygons around each gauge, weighting rainfall data based on proximity, thus addressing spatial variability (Ribeiro et al. 2021). However, it assumes uniform rainfall within polygons and may not adequately represent distant areas (Smith et al. 2023). Meanwhile, the Isohyet method delineates lines of equal rainfall intensity across the watershed, considering both spatial and topographic variability but demanding more intricate calculations (Workneh et al. 2024). The Arithmetic, Thiessen, and Isohyet methods, as shown in the following equations (Ribeiro et al. 2021).

HEC-HMS is a program that aims to produce a simulated discharge value based on the value of the rainfall data by considering the observed discharge value. In this study, following spatial data analysis using the Arc-GIS tool, hydrological modeling can be conducted to simulate discharge. The quasi-distributed is used which divides the watershed into multiple sub-basins, allowing for spatial variability in parameters and inputs. This process involves several steps. The basin model is the initial step in establishing a project in the HEC-HMS program. It primarily involves creating important hydrological components such as subbasins, reaches, junctions, and sinks. Subsequently, Terrain data are produced by integrating the Mayang watershed boundary map obtained from Arc-GIS delineation in .tif format. This involves navigating the Geographic Information System (GIS) menu and doing actions such as preprocess sink, preprocess drainage, and delineate element. These tasks lead to the creation of subbasins, flow direction, flow accumulation, reaches, and junctions. In the meteorologic model step, precipitation calculations are chosen. This study specifically uses the specified hyetograph method, which entails using rainfall values that correspond to a 5-year return period for each sub-basin. The control specification establishes the timeframe for the execution of the process, using rainfall data from 1 January 2012, to 31 December 2021, with a daily frequency for both rainfall and discharge. Time series data utilizes different forms of data as inputs for modeling, such as recorded rainfall, GPM-IMERG precipitation data, and recorded daily discharge throughout a 5-year period from 1 January 2012, to 31 December 2016. The rainfall data are obtained by calculating the average amount of rainfall from 12 stations located within the Mayang watershed, measured in millimeters. On the other hand, the discharge data refers to the rate at which water flows into the Talang outlet, measured in cubic meters per second. Finally, the initial condition entails the computation of parameter values prior to the execution of the model. These values are derived through manual calculations using methods supplied in the HEC-HMS Technical Reference Manual. The resulting values are then inserted into the selected runoff calculation technique or model. These methods and parameters selected for HEC-HMS modeling are detailed in Table 2.

Sensitivity analysis is used to identify the parameters that significantly influence model responses and to analyze their interactions, range of choices, and spatial variability to enhance model outcomes (Devak & Dhanya 2017). This process involves determining the impact of various parameters on the calibration process and their interactions with other parameters (Li et al. 2023). Parameter sensitivity is measured by the magnitude of the increase or decrease in the values of the differences, which is the greater the magnitude of the difference, the higher the sensitivity of the parameter.

On the other hand, the Thiessen Polygon method divides the study region into polygons. The Thiessen Polygon method divides the study area into 12 different areas, each area represented by a single rain gauge, as shown in Figure 3(b). This rain gauge collects rainfall for the entire area contained within its polygon. The rain gauge rainfall value is then considered to represent the entire polygon uniformly. This approach takes rain gauge proximity into account and assumes that rainfall is consistent within each polygon. Every point in each polygon is closer to its enclosing sample point than to any other sampling point.

Meanwhile, by using the Isohyet method, the area is separated into subregions based on reported rainfall values, which are classified as very high, high, moderate, low, and very low, as shown in Figure 3(c). These categories are depicted by Isohyet contour lines. It is believed that each subregion within the Isohyet range will have identical rainfall characteristics. This method considers the spatial variability of rainfall. By using multiple rain gauges, it accounts for variations in rainfall across different locations.

Calibration is used to improve simulation results to align them more precisely with observed values. The calibration parameters used in this study include max storage, CN, time of concentration (T_c), and storage coefficient. The calibration was conducted from 1 January 2012, to 31 December 2016, by considering outflow discharge in Talang station that was generated by daily rainfall. The rain gauge being analyzed is located at Seputih Station, which represents the upstream sub-watershed area that is indicated to have precipitation data (using satellite), but it is not recorded in the field. The parameter values obtained from the calculations were adjusted either by decreasing or increasing them as needed.

The NSE values for observational rainfall data using the Arithmetic, Thiessen Polygon, and Isohyet methods are 0.254, 0.439, and 0.212. However, with precipitation GPM-IMERG satellite data using these three methods, the NSE values were generally better than those for observational data, as shown in Table 4. Specifically, the Thiessen Polygon method showed the highest accuracy for both observational and GPM-IMERG data among the three mean areal precipitation methods. Furthermore, this study also evaluates the TRMM satellite for comparison. The Thiessen Polygon method with GPM-IMERG and TRMM satellite data resulted in NSE values of 0.515 and 0.445, respectively, indicating that GPM-IMERG data performed better. Additionally, observational rainfall data generally showed lower and negative percent bias values compared to GPM-IMERG and TRMM satellite data. This indicates that discharge responses from observational rainfall data tend to underestimate, while satellite precipitation data tend to overestimate, except for the Arithmetic and Isohyet methods.

Based on the previous study by using other satellite data such as TRMM in Bedadung watershed show that hydrological modeling performance using manual rainfall recorder (MRR) rain data input was better than the performance using satellite data TRMM 3B42 (Alhamda et al. 2020). The R² and NSE values for MRR rain data input were 0.72 and 0.71. In contrast, the R² and NSE values for the TRMM B342 satellite rain data input were 0.57 and 0.55, respectively. In other studies in the Sampean Baru, Bedadung, and Mayang watersheds, the MMR provided a better discharge response in all watersheds compared to TRMM and GPM, however, in terms of satellite products, the GPM-3IMERGDF showed slightly better results than the TRMM 3B42 (Hidayah et al. 2021).

The validation process took place between 1 January 2017, and 31 December 2021. It involved using daily input data and calibrated parameter values collected earlier. The validation results indicate that the NSE value for observational rainfall data using the Thiessen Polygon method is 0.481. For GPM-IMERG precipitation data, the NSE values using the Arithmetic, Isohyet, and Thiessen Polygon methods are 0.392, 0.333, and 0.655, respectively. The Thiessen Polygon method using GPM-IMERG data shows a higher NSE value than the observational data, as shown in Table 5. Additionally, the validation results indicate that all treatments show an overestimation, except for GPM-IMERG using the Isohyet method, as shown in Figure 11.

Spatial modeling of hydrology shows the importance of the availability of rain gauges that can represent the entire sub-watershed, which can be demonstrated by the Thiessen Polygon result. The rainfall data used for this runoff model simulation has been tested for correlation at all rainfall stations, resulting in a good match between satellite data and observational data. These patterns determine how rainfall translates into runoff and other hydrological processes. Furthermore, based on Michelon et al. (2021), high-density rainfall observations are crucial for accurately capturing these patterns, which are particularly challenging to observe in mountainous environments due to the complex terrain and variability in rainfall distribution. When the density of rain gauges is low or when gauges are irregularly distributed, the precipitation data collected is less representative of the actual spatial variability of rainfall, leading to significant uncertainties in the hydrological model parameters (Zeng et al. 2018). Inaccurate precipitation inputs cause the model to calibrate parameters incorrectly, resulting in poor simulation of hydrological processes like runoff. Thus, observing high-density rainfall, including the distance of rainfall from the outlet, is key to predicting hydrological responses in terms of runoff coefficients and lag time in small mountainous catchment areas. Increasing the density of stations will improve the performance of hydrological modeling; however, it is important to remember that adequate station density depends on the catchment area, type of event, and distribution of stations (Hohmann et al. 2021). Recent studies highlight the variability in the optimum number of rain gauges, which ranged from 14 km²/gauge to 38 km²/gauge, with an average of 25 km²/gauge. It was found that the current rain gauge density needs to be improved by at least a factor of three to mimic the full-scale level (Gyasi-Agyei 2020).

The research on rainfall modeling in streams using GPM-IMERG precipitation data input with the HEC-HMS program yielded satisfactory modeling performance when compared to observed rainfall input. The simulated discharge pattern, using observational rainfall data and GPM-IMERG satellite precipitation data as input, is compatible with the observed discharge, as indicated by the calibration and validation findings. Under conditions of low flow, the simulated release values align with the observed discharge when using both the observed rainfall data and the GPM-IMERG satellite precipitation data. During the highest flow rate, the simulated discharge value does not match the observed discharge when using observational rainfall data as input. However, when using GPM-IMERG satellite precipitation data as input, the simulated discharge aligns well with the observed discharge. Furthermore, the temporal influence shows that high-flow conditions better represent the hydrological response compared to low flow conditions. This result aligns with a study by Li et al. (2021), which GPM-IMERG data was effective in simulating hydrological responses during high-flow conditions and showed rational flow simulations with notable seasonal and regional variations in performance in Taiwan. The study in hydrological modeling over the Nanliu River Basin that evaluates the performance of the GPM-IMERG products against gauge observations found that when driven by IMERG data, the hydrological model's performance improved for high-flow simulations compared to low flow simulations (Tong et al. 2018). However, the temporal and spatial resolution of satellite data, while suitable for capturing high-flow events that involve significant precipitation over large areas, may not be sufficient for accurately representing the finer details of low flow conditions. Smaller, localized precipitation events can be averaged out or missed in the data processing, leading to inaccuracies (Ahmed et al. 2022). Low-intensity precipitation events, which are characteristic of low flow conditions, can be challenging to detect accurately due to the limited sensitivity of the satellite sensors, which can lead to underestimation or missing detection of light rain events, impacting the overall accuracy of low flow simulations (Zhou et al. 2021).

Despite limitations in detecting low flow conditions, GPM-IMERG outperforms the TRMM satellite in producing more accurate precipitation estimates. In this study, the RMSE, NSE, and %Bias of Thiessen Polygon GPM-IMERG are 0.6, 0.515, and 6.13, respectively, while for Thiessen Polygon TRMM, these values are 0.9, 0.445, and 10.01. Thus, GPM-IMERG demonstrates better performance compared to TRMM, indicating more accurate, reliable, and less biased precipitation estimates. In previous studies, such as in the Bedadung Watershed, Jember Regency, Indonesia, GPM-IMERG also showed a better fit than TRMM for various hydrological models such as Variable Infiltration Capacity (VIC), HEC-HMS, and SWAT, suggesting that it provides a more accurate representation of rainfall data in certain contexts (Hidayah et al. 2021). According to Gaona et al. (2016), GPM offers more comprehensive global coverage than TRMM, allowing GPM-IMERG to capture precipitation data over a larger portion of the Earth's surface, including higher latitude regions that experience significant precipitation events. Additionally, GPM-IMERG enhances the spatial resolution of precipitation estimates to 0.1° × 0.1° and provides data at finer temporal resolutions, such as every 30min. This is a significant improvement over TRMM, which had a spatial resolution of 0.25° and temporal resolution of 3h, allowing for more detailed and accurate precipitation measurements (Chen & Li 2016). In comparison to other satellites such as PERSIANN, GPM-IMERG offers more frequent updates and higher resolution data (Pradhan et al. 2021). While GPM-IMERG provides high-resolution and frequent data, it may still miss localized precipitation events or have inaccuracies due to the complexity of weather systems, such as during the dry season and due to topographical influences – factors that can similarly affect other satellite-based estimates such as TRMM and PERSIANN (Zhou et al. 2023). Therefore, further studies should compare satellite rainfall data with other emerging rainfall data products such as microwave or radar rainfall, Climate Hazards Group Infrared Precipitation with Stations, CHAOS simulation (CHAOS-hydro), and the XPOL weather radar (XPOL-hydro) to evaluate their accuracy and reliability in hydrological modeling (Thorndahl et al. 2017; Varlas et al. 2019).

Based on the sensitivity results, the CN is a crucial factor as it measures the runoff potential in a region. Similar to findings from the Meenachil River basin study, CN and lag time are the most sensitive parameters in the HEC-HMS model for predicting peak discharge volume (George et al. 2022). A higher CN value implies a greater tendency for runoff (Kristanto et al. 2021), resulting in less water infiltrating the soil and more water flowing over the surface, leading to a higher volume of river discharge (Priambodo et al. 2021). The increase in runoff and discharge is often exacerbated by the lack of infiltration zones, such as permeable soils or vegetation, which would otherwise absorb some of the rainfall. Additionally, the time of concentration (T_c) and the storage coefficient (R) are crucial variables impacting the timing of the simulated peak discharge. T_c represents the time required for water to travel from the farthest point in the watershed to the outflow, while the storage coefficient indicates the rate at which storage areas release water. Increasing T_c and R values extends the time to reach peak discharge, affecting the model's ability to accurately predict flood peaks and the timing of runoff events. During the calibration phase, simulated discharge levels are compared to observed data, including values from observational precipitation data, satellite precipitation data, and recorded discharge measurements (Esmali et al. 2021). Even so, satellite precipitation data, especially GPM-IMERG, can be effectively used for hydrological modeling in regions with sparse observational data. Overestimated precipitation data using the Thiessen Polygon method is beneficial for flood management planning, as it ensures preparedness for potential extreme weather events. On the other hand, underestimated precipitation data using the Isohyet method is useful for water resource planning, including irrigation, raw water supply, and hydropower plant management, as it encourages efficient and sustainable water use.

GPM-IMERG satellite precipitation data proves to be a valuable tool for hydrological modeling, especially in areas with sparse observational data. Among the methods evaluated, the Thiessen Polygon method was particularly effective when paired with GPM-IMERG satellite data, providing more reliable precipitation data by accounting for the proximity of rain gauges. Calibration and validation results further confirmed the superiority of GPM-IMERG data over observational data, especially when using the Thiessen Polygon method, enhancing the accuracy of simulated discharge patterns, particularly under high-flow conditions. Despite some limitations in capturing low-intensity precipitation events, GPM-IMERG outperforms other satellite products like TRMM in terms of accuracy. Sensitivity analysis revealed that parameters such as CN and time of concentration significantly influence model outcomes. The integration of GPM-IMERG data in hydrological modeling offers valuable insights for flood management and water resource planning, particularly in regions with sparse observational data, ensuring better preparedness for extreme weather events and promoting sustainable water use practices. Further studies should explore emerging rainfall data products, such as microwave or radar rainfall, to continue improving hydrological modeling accuracy. The next development of this research is to perform more advanced rainfall prediction modeling by leveraging various features from satellite data and other meteorological data using machine learning techniques such as deep neural networks or least mean square training.

The authors want to thank Jember University and IsDB for the research grants and support.

Data cannot be made publicly available; readers should contact the corresponding author for details.

The authors declare there is no conflict.