Comparison of flow simulations with sub-daily and daily GPM IMERG products over a transboundary Chenab River catchment

Hydrological processes of the river basins mainly depend on the spatial and temporal structures of precipitation. Several studies have evaluated the effect of the spatial resolution of precipitation on the runoff quantity (Goldstein et al. 2016; Terink et al. 2018; Kim & Kim 2020). Few studies compared the impact of different or low temporal resolution of precipitation on the runoff quantity, especially for large river catchments. This study focuses on the estimation of discharge from a large transboundary catchment of the Chenab River present in Pakistan and India (Figure 1) by using sub-daily and daily precipitation information. Traditionally, gauge precipitation is considered a reliable source of precipitation for the calculation of hydrological processes. However, the gauge precipitation data is not always available from the transboundary river catchments due to fewer data-sharing practices among the riparian states. As an alternative, satellite precipitation products (SPPs) can overcome this problem by continuously capturing and supplying us with the precipitation data from a transboundary catchment. SPPs are considered a consistent source of precipitation due to their expansive spatial coverage (Tang et al. 2016).

Over the last two decades, several SPPs, such as Tropical Rainfall Measuring Mission (TRMM), Climate Prediction Center (CPC) Morphing (CMORPH), Precipitation Estimation from Remotely Sensed Information using Artificial Neural Networks (PERSIANN), Global Satellite Mapping of Precipitation (GSMaP), and other semi-global SPPs, were used broadly in hydrology, meteorology, and water resources management practices (Pakoksung & Takagi 2016; Liu et al. 2018; Soo et al. 2020). All satellite precipitation series currently available have their benefits and limitations in capturing the spatial and temporal characteristics of precipitation. In February 2014, a more precise SPP was obtained than the previous satellite precipitation series through a Global Precipitation Measurement (GPM) mission at higher spatial and temporal resolutions. The GPM is the latest generation and is the successor of the TRMM SPP. It was introduced with the collaboration of the National Aeronautics and Space Administration (NASA) and the Japan Aerospace Exploration Agency (JAXA) (Ning et al. 2016). The GPM mission provides the high-resolution global precipitation estimates through the international constellation of a core observatory satellite and approximately ten partner satellites. The core observatory satellite consists of a conical scanning multichannel GPM Microwave Imager (GMI, 10–183 GHz) and dual-frequency precipitation radar (DPR; Ku-band and Ka-band) (Ning et al. 2016).

The GPM products have a higher spatial and temporal resolution (0.1°×0.1° and 30 mins, respectively) than the TRMM precipitation products (0.25°×0.25° and 3 h, respectively) (Huffman et al. 2020). The GPM mission provides the Integrated Multi-satellitE Retrievals for GPM (IMERG), which consist of three SPPs: (i) Early (IMERG-E), (ii) Late (IMERG-L), and (iii) Final (IMERG-F) precipitation product. The first two products (IMERG-E and IMERG-L) are obtained by running the algorithms several times at each timestep and are, respectively, produced with 4 and 12 h of latency after sense. The first two products are near-real-time (NRT) products, which are useful for predicting peak events, while the third one is the post-real-time product, which is suitable for research purposes. The final product (IMERG-F) is obtained after 3.5 months of latency with ground gauge data evaluation or any other sources of precipitation measurements. NRT and post-real-time SPPs are beneficial in numerous applications like flood forecasting, early warning, and hydrometeorological modeling, which are available in the form of IMERG products. So, there is a need to evaluate the applicability of sub-daily and daily IMERG products over the Chenab River catchment and conduct a rigorous statistical analysis to authenticate their usage in this catchment.

Recently, numerous studies have compared the reliability of the GPM-based and other precipitation series on daily and monthly timescales (Lu et al. 2018; Su et al. 2019; Xu et al. 2019; Ahmed et al. 2020). For instance, Ahmed et al. (2020) compared the daily-based precipitation products (IMERG-F and TRMM (3B42V7)) over the Chenab River catchment and found that IMERG-F has better performance than 3B42V7 for the simulation of daily and monthly flows. Xu et al. (2019) evaluated the IMERG-F and 3B42V7 SPPs using daily precipitation data from 59 meteorological stations in the Huang-Huai-Hai Plain in China and found that 3B42V7 has a lower precipitation detection capability than IMERG-F at daily timescale with the probability of detection of 0.67 and 0.83, respectively. Su et al. (2019) evaluated IMERG products’ applicability over the Upper Huaihe River basin in China and noticed that IMERG-F demonstrates higher performance at daily hydrological simulations than IMERG-E and IMERG-L. They found that IMERG-F has a smaller false alarm ratio (FAR=0.31) than IMERG-E (FAR=0.37) and IMERG-L (FAR=0.32). Lu et al. (2018) compared the monthly IMERG (3IMERGM) and monthly TRMM (3B43V7) products over Xinjiang, China, and concluded that 3IMERGM has a smaller relative bias (RB=7.76%) than 3B43V7 (RB=10.24%), while 3IMERGM has a higher correlation coefficient (CC=0.68) than 3B43V7 (CC=0.62). Overall, their results showed that 3IMERGM outperformed the 3B43V7. Many studies have also been done in the South Asian monsoon region to detect the reliability of SPPs. For example, Yuan et al. (2017) conducted the performance evaluation of GPM IMERG-F and TRMM 3B42V7 precipitation products, and their applicability for streamflow simulation in the Chindwin River basin, Myanmar. Their results depicted that both IMERG and 3B42RT can potentially capture the historical precipitation events. Compared with 3B42V7, there was no significant improvement observed in IMERG-F. Moreover, 3B42V7 outperformed IMERG-F at daily and monthly scales and in heavy rain detections at four out of five gauges. Montes et al. (2021) evaluated the intense precipitation events during the monsoon period in Bangladesh. They focused on the analysis of four SPPs to monitor the intense rainfall events: the Climate Hazards Group Infrared Precipitation with Station Data (CHIRPS), the PERSIANN–Climate Data Record (PERSIANN-CDR), the IMERG, and the CMORPH. They concluded that IMERG is the best SPP followed by PERSIANN-CDR and CHIRPS. The IMERG exhibited the best performance for the events over a value of 20 mm/day. Reddy et al. (2019) evaluated the high-resolution SPPs over India for the southwest monsoon period. They evaluated the NRT and research versions of IMERG-V4 (NRT and Final (F)) and GSMaP-V6 (NRT and moving vector with Kalman filter (MVK)) against gridded gauge-based Indian Meteorological Department rainfall data on daily, monthly, and seasonal scales. In terms of biases, GSMaP (NRT and MVK) and IMERG (NRT and F) underestimated the rainfall (about 17, 23, 18, and 3%, respectively) for the whole Indian region. Overall, the IMERG-F precipitation product showed better agreement with the observed data compared to GSMaP.

In the early era of SPPs, most of the studies were performed on a broader temporal scale. For instance, Hughes (2006) used the Pitman monthly model to perform rainfall–runoff modeling by using satellite rainfall. At that time, they found that there were some corrections required for use of satellite data for rainfall–runoff modeling. Artan et al. (2007) demonstrated the usefulness of remotely sensed precipitation data for hydrologic modeling. They concluded that SPP is considered more reliable when the hydrologic model is calibrated with the use of satellite rainfall data. Collischonn et al. (2008) merged three-hourly TRMM rainfall estimates to produce daily hydrographs in the river Tapajos, Amazon. They found that TRMM-based hydrographs are comparable with those obtained using rain gauge data. For the comparative evaluation of the SPPs on a finer/sub-daily temporal scale, only a limited number of studies have been performed (Yuan et al. 2019; Tang et al. 2020; Zhu et al. 2021). For instance, Yuan et al. (2019) compared the statistical and hydrological performance of sub-daily-based SPPs, i.e., GPM and TRMM, in a poorly gauged watershed in Myanmar. They concluded that GPM IMERG-F reveals higher accuracy than its predecessor, TRMM 3B42V7, and found that GPM-era is a perfect replacement for TRMM-era. Zhu et al. (2021) used five sub-daily SPPs, i.e., three IMERG products and two TRMM products, to simulate the flood over a humid region of China. They found that IMERG-F produced the best hourly flood simulation with a Nash–Sutcliffe Efficiency (NSE) value of 0.71. Additionally, they concluded that IMERG-E is a perfect replacement for 3B42RT for sub-daily flood simulation. Tang et al. (2020) evaluated three IMERG-Hourly products (IMERG-HHE, IMERG-HHL, and IMERG-HHF) on a grid scale over the Sichuan basin of China. They found that IMERG-HHF and IMERG-HHL outperformed IMERG-HHE in terms of CC and root-mean-square error (RMSE), whereas IMERG-HHL has less RB than IMERG-HHF by 21.1%.

Several hydrological models have been developed to simulate the stream flows, including HEC-HMS, MIKE SHE, SWMM, IFAS, and HSPF. The soil and water assessment tool (SWAT) is a time-continuous and semi-distributed basin-scale model, which is good at simulating the discharge for a longer time (Arnold & Fohrer 2005). It has been extensively used in simulating the runoff at yearly, monthly, and daily timescales (Shrivastava et al. 2004; Ghoraba 2015; Saddique et al. 2019; Rahman et al. 2020; Chen & Chang 2021). However, limited studies have been carried out on the sub-daily timescale to simulate the flows in large river basins (Yang et al. 2016; Yu et al. 2018). Hydrological processes are efficiently captured and validated by using sub-daily SWAT models. Thus, many researchers studied the sub-daily SWAT model for the simulation of discharge in small watersheds. For instance, Maharjan et al. (2013) set up a sub-daily SWAT model on a small agricultural watershed (0.8 ha) to predict runoff and found that the sub-daily SWAT model proved to be efficient, while the daily SWAT model was out of the acceptable limits. Jeong et al. (2010) made some improvements in the SWAT model to simulate the floods with hourly precipitation input in a small watershed (1.94 km²) present in Austin, TX, USA. They concluded that the SWAT model performance improved with its modification to the sub-daily model compared to the daily model. Yang et al. (2016) compared the SWAT model with hourly and daily precipitation inputs for daily discharge simulation in the upper Huai River Basin of China and revealed that the model with hourly precipitation input performed better than the model with daily precipitation input, primarily due to its better ability to simulate peak flows during the flood season.

Flooding has become a common phenomenon in Pakistan nowadays during the monsoon season. Floods in Jhelum and Indus rivers are mostly controlled by the Mangla and Tarbela reservoirs, respectively (Figure 1). There is no big reservoir present in the transboundary Chenab River on the Pakistan side, which results in flooding problems almost every monsoon season (Tariq & van de Giesen 2012). From 2010 onwards, the Chenab River is facing high floods at the Marala Barrage almost every year (Federal Flood Commission Islamabad 2014). The authors selected the Chenab River catchment, as a study area, to simulate the runoff with the help of freely available remote sensing information because Pakistan does not have access to the point rainfall data from the Indian side of the catchment of Chenab River (Riaz et al. 2017). Therefore, rainfall–runoff modeling is a tiresome job in this catchment due to the unavailability of ground data, which arouses the need for flow simulation with the help of SPPs to better understand the usefulness of freely available remote sensing data.

The objectives of this study were twofold. The first objective was to simulate the discharge in a transboundary Chenab River catchment with three satellite-based precipitation products and compare their results for better understanding and applicability in future hydrological simulations. The second objective was to analyze the SWAT model's performance in continuous flow simulation at sub-daily and daily timescales in a relatively larger catchment. This study is valuable because fewer studies have been done previously on the transboundary large river catchments using a sub-daily routine of Arc-SWAT and SWAT-CUP. This study is also useful, in a combination of the SWAT model and comparative evaluation of the latest daily and sub-daily SPPs, to understand the flow behavior of the Chenab River catchment.

The Chenab River is an eastern tributary of the Indus River and originates from the Kulu and Kangra districts of Himachal Pradesh in India (Shahid et al. 2017) (Figure 1). It flows through Indian-administrated Jammu and Kashmir and finally enters the plains of the province of Punjab in Pakistan. The Marala Barrage (32°40′ N, 74°27′ E) is the first gauging site of the Chenab River after entering Pakistan. The total catchment area of the Chenab River, upstream of the Marala Barrage gauging site, is approximately 28,000 km² and its extent lies between 73 and 78° E (longitude), and 32 and 35° N (latitude). The elevation of the Chenab River catchment at the Marala Barrage ranges from 224 to 7,085 m in the upper snowy catchment (Figure 2). The Chenab River slope varies from 0.4 m/km (in plains) to 25 m/km (in most steep parts) (Awan 2003).

The Chenab catchment lies in an active monsoon belt, which occurs between 15 June and 15 October. River flow is also contributed by snowmelt from upstream of the catchment. Most of the annual river flows (about 84%) occur during the pre-monsoon and monsoon periods (April–September), particularly from June to September due to synchronized flow from snowmelt and precipitation (Singh et al. 1997). About 97% of the catchment, upstream of the Marala Barrage, lies on the Indian side. Therefore, there is a need to evaluate the latest SPPs for better hydrological simulations in the Chenab River.

The Digital Elevation Model (DEM) is a representation of topographies and an important aspect of hydrological applications. The DEM is used to delineate and obtain the physical characteristics of a catchment, i.e., reaches, subbasins, outlets, and flow paths. In this study, the Advanced Spaceborne Thermal Emission and Reflection Radiometer's (ASTER) Global Digital Elevation Model (GDEM), with a 30 m resolution, was downloaded from the Jet Propulsion Laboratory of the National Aeronautics and Space Administration (NASA) website, https://asterweb.jpl.nasa.gov/data.asp, and extracted in the ArcGIS environment. The delineated Chenab River catchment, with ASTER GDEM, is shown in Figure 2.

Soil types and land use represent the hydrological processes and their governing systems very effectively. In the present study, soil data was downloaded and extracted from the website of the United Nations Food and Agriculture Organization (FAO) harmonized world soil database (HWSD), http://www.fao.org/soils-portal/data-hub/soil-maps-and-databases/harmonized-world-soil-database-v12/en/, as displayed in Figure 3. Eight different soil types were found in the study catchment, as shown in Table 1, in which 38.08% of the soil is Lithic, 23.26% is Orthic Luvisol, and 16.84% is glacier cover. Land-use information on the Chenab River catchment was extracted from the SWAT website's global database, https://swat.tamu.edu/data/. The catchment represents the 18 different land-use classes (Figure 4), and their description is tabulated in Table 2. About 35.91% of the land use consists of grasslands and shrublands, 16.11% comprises irrigated croplands, and 11.23% includes snow or ice covers.

Precipitation datasets of the GPM (version 6), IMERG-E, IMERG-L, and IMERG-F, were downloaded from NASA's earth data platform of GES DISC, https://disc.gsfc.nasa.gov/. In this study, precipitation data with two different temporal resolutions, i.e., sub-daily (half-hourly) and daily (24-hourly), were downloaded for the duration of 10 years (2003–2012). The nomenclature of the downloaded GPM products and their descriptions are tabulated in Table 3. Sub-daily/half-hourly data is represented by ‘HH’, whereas ‘D’ represents daily data. IMERG-HHE, IMERG-HHL, and IMERG-HHF are the sub-daily early, late, and final precipitation products, respectively, while IMERG-DE, IMERG-DL, and IMERG-DF are the daily early, late, and final precipitation products, respectively. Twenty-seven sub-catchments were extracted by delineating the study area catchment in Arc-SWAT (Figure 1). Each precipitation dataset was extracted from the closest pixels of the centroids of sub-catchments by writing and running a script in R.

The Climate Forecast System Reanalysis (CFSR) dataset is developed by the National Center for Environmental Prediction (NCEP) of the National Oceanic and Atmospheric Administration (NOAA) and is derived from the Global Forecast System (Fuka et al. 2014). The CFSR finds broad application in hydrological simulations due to its reliability, high spatial resolution, and long-time series. In the present study, maximum and minimum temperatures, sunshine hours, relative humidity, and wind speed data were collected from the CFSR dataset. The CFSR dataset can be retrieved from the SWAT official website, https://globalweather.tamu.edu/.

For calibration and validation purposes, observed daily and monthly discharge (2003–2012) at the Marala Barrage gauging station were obtained from the flood forecasting division, Pakistan Meteorological Department (PMD).

In this study, the Arc-SWAT model was intended to set up for rainfall–runoff modeling in the transboundary catchment of the Chenab River to ensure the possibility of using GPM IMERG SPPs. The traditional DEM-based watershed delineation method, in Arc-SWAT, was used to delineate the Chenab River catchment into 27 sub-catchments (Figure 1). These sub-catchments were further divided into 2,079 different HRUs by incorporating the slope, soil, and land-use information in the model.

Considering the availability of IMERG and observed data, the ‘temporal split-sample approach’ was adopted to select the calibration and validation time periods. It is an arbitrary division of the time-series data into two subsets, i.e., earliest 50% for calibration and later 50% for validation. This method is thoroughly explained by Arnold et al. (2012) and Daggupati et al. (2015). The initial two years, i.e., 2003 and 2004, were selected as a warmup period for model stabilization. Subsequently, the model was calibrated from 2005 to 2008 and validated from 2009 to 2012.

Sub-daily and daily SWAT models were developed for the transboundary Chenab River catchment using the latest version of SWAT 2012. The SWAT model uses the Penman–Monteith method to estimate potential evapotranspiration (Arnold et al. 2011). This study calculated the surface runoff using the Soil Conservation Service-Curve Number (SCS-CN) method for daily simulations and the Green & Ampt Infiltration method for sub-daily simulations. The Green and Ampt equation continuously simulates the infiltration process. It assumes the homogeneous soil profile and antecedent moisture distributed in the soil profile. The Green and Ampt equation uses a direct relationship between rainfall and infiltration rate based on physical parameters, which allows continuous surface runoff simulation. Since SCS-CN is used to simulate the direct runoff using total runoff and watershed properties (Garen & Moore 2005), this method is considered inappropriate for sub-daily runoff simulation (Jeong et al. 2010). For channel routing, the SWAT model comprises variable flow routing (Williams 1969) and Muskingum routing (Overton 1966). Variable flow routing was selected in this study. The lateral flow was estimated by using the kinematic method. Further details about the model can be obtained from the SWAT users’ manual (Arnold et al. 2011).

Sub-daily and daily SWAT models were supplied by HH and daily IMERG precipitation datasets on the centroid of each sub-catchment, respectively (Figure 1). For this purpose, time-series data of SPPs were downloaded in the form of NETCDF files, and then precipitation data of the closest pixel from the centroid was extracted using R. Subsequently, it was supplied to the Arc-SWAT model for rainfall–runoff simulation. Initially, six models were set up in Arc-SWAT for sub-daily and daily (IMERG-E, IMERG-L, and IMERG-F). Each model was calibrated and validated against daily and monthly observed discharge. In the end, we obtained 24 calibrated and validated model results. Figure 5 shows the devised framework of the methodology for hydrological evaluation of sub-daily and daily GPM IMERG SPPs.

Various parameters are present in the SWAT model, which governs the hydrological process. Some of them are not sensitive enough to be considered during model calibration. Selecting the most sensitive parameters and reducing the calibration time of the model require a sensitivity analysis. In this study, a global sensitivity analysis (GSA) was performed to sort out the sensitive and non-sensitive parameters in SWAT-Calibration and Uncertainty Programs (CUP) (Abbaspour 2015). The SWAT-CUP is a calibration helper tool, which many scholars have applied recently to ease their calibration processes (Yang et al. 2008; Capra et al. 2009; Sao et al. 2020). SWAT-CUP introduced various uncertainty and calibration analysis algorithms, including Sequential Uncertainty Fitting (SUFI-2), Generalized Likelihood Uncertainty Estimation (GLUE), Parameter Solution (ParaSol), and Markov Chain Monte Carlo (MCMC). This paper applied the SUFI-2 algorithm (Abbaspour et al. 2007). The SWAT-CUP uses p-value and t-statistic to determine the sensitivity of the parameters. If the absolute value of the p-value is small and the t-statistic is large, then the parameter is classified as sensitive. The GSA (1,000 simulations) was performed on sub-daily and daily timescales in SWAT-CUP. Table 4 lists the description of the parameters selected for the sensitivity analysis of sub-daily and daily model runs.

Table 5 represents the sensitivity ranks of these selected parameters. It can be noted from Table 5 that, in sub-daily simulation, GW_REVAP, SUB_SFTMP, ESCO, SUB_SMFMN, and SUB_SMTMP have a strong effect on the runoff generation. Whereas, in daily simulation, the ALPHA_BF, CN2, CH_K2, ESCO, and GW_REVAP have a vibrant effect on the runoff estimation, as they have a smaller value of p and a larger value of t-stat. Therefore, the top nine parameters, having minimum values of p (i.e., p<0.20), were selected from Table 5 and the model calibration was executed, i.e., 500 simulations with several iterations were performed. Both sub-daily and daily models of IMERG-E, IMERG-L, and IMERG-F were calibrated for daily and monthly runoff simulations. In total, 12 models were calibrated from 2003 to 2008 with 2 years of warmup periods. A single iteration from 2009 to 2012 with the same number of simulations (500), as in the last iteration of calibration, was carried out to perform validation for all 12 models. The NSE was selected as an objective function for the whole calibration process. The best parameter set produces the maximum value of the objective function and it is also evaluated by using additional performance evaluation indices, discussed in the next section.

In the SUFI-2 algorithm of SWAT-CUP, two factors, P-factor and R-factor, were introduced to examine the calibration's acceptability. The percentage of observed data falling into the 95% of the prediction boundary is termed as 95 percent prediction uncertainty (95PPU), known as P-factor. The P-factor value ranges from 0 to 1, in which a value close to 1 represents that 100% of the observed data is bracketed by the model prediction uncertainty. The R-factor represents the average thickness of the PPU band, which ranges from 0 to ∞. R-factor values close to 0 shows the best model performance. Both ideal values of the P-factor and R-factor cannot be achieved simultaneously, requiring a compromise. Generally, a value close to 0.7 for P-factor and a value less than 1.5 for R-factor is suggested for calibration of discharge (Abbaspour et al. 2015). The Coefficient of determination (R²), the NSE, and the PBIAS indicate the goodness of fit for the best model simulation. R² is the linear correlation between observed and simulated discharge (Moriasi et al. 2007) and ranges from 0 to 1. The NSE depicts how precisely observed and simulated flows match the 1:1 line, and its value lies between −∞ and 1 (Santhi et al. 2001). The zero value of the PBIAS is the model's optimal condition; however, a negative value shows overestimation, and a positive value shows underestimation (Gupta et al. 1999). The model consists of satisfactory results with PBIAS values between ±20 and ±40% (Van Liew et al. 2005). Besides R², NSE, and PBIAS, the Taylor diagram (Taylor 2001) was also introduced to evaluate the performance of flood simulations. The Taylor diagram is constructed between RMSE, Pearson's correlation (R), and standard deviation (SD) and is beneficial for comparing the performances of multiple model simulations in a single diagram.

As per Table 5, out of 15 potentially relevant parameters, the nine most sensitive sub-daily SWAT model simulation parameters (sensitivities in descending order) were GW_REVAP, SUB_SFTMP, ESCO, SUB_SMFMN, SUB_SMTMP, CH_N2, GW_DELAY, SOL_AWC, and CN2. In contrast, the most sensitive parameters (sensitivities in descending order) in the daily model simulation were ALPHA_BF, CN2, CH_K2, ESCO, GW_REVAP, SOL_AWC, SUB_SMTMP, SUB_SMFMN, and GW_DELAY. CH-K2 and ALPHA_BF were observed to be more sensitive in daily simulation instead of sub-daily simulation, while SUB_SFTMP and CH_N2 were observed to be more sensitive in sub-daily simulation instead of daily simulation. In both sub-daily and daily simulations, SUB_SMFMN, SUB_SMTMP, ESCO, CN2, GW_REVAP, GW_DELAY, and SOL_AWC were significantly sensitive, which means snow, soil, curve number, and groundwater parameters are the most important for flow simulation in this catchment. The main reason for the difference between the sensitivity of some parameters is the selection of surface runoff calculation methods. As in this study, the SCS-CN method was used for daily simulation, while the Green & Ampt Infiltration method was selected for sub-daily simulation. ESCO is a sensitive parameter in many studies because it controls evapotranspiration and hence water balance (Feyereisen et al. 2007; Kannan et al. 2007; Cibin et al. 2010). CN2 is used to calculate the effective hydraulic conductivity (Nearing et al. 1996), which is an important variable in determining infiltration rate and then surface runoff (Bauwe et al. 2017). SOL_AWC is the soil available water capacity, which is influential in both tests and does not have any nominal correlation with model timestep. Table 5 suggests that the groundwater parameters (ALPHA_BF, GW_REVAP, and GW_DELAY) are more sensitive in daily simulation rather than sub-daily simulation, which confirms the result of Jeong et al. (2010), who noticed the increasing trend of groundwater parameters with additional simulation timesteps. In daily flow simulation, CH_K2 is effective hydraulic conductivity in the main channel alluvium. Since precipitation occasionally occurs as snow in the winter season, snow-related parameters (SUB_SMTMP, SUB_SFTMP, and SUB_SMFMN) appeared to be more sensitive in sub-daily flow simulation as compared to daily simulation.

Several iterations were run for models’ calibration (with 500 simulations each) in SWAT-CUP until no apparent changes were observable in the outcomes. Calibrated parameter values of sub-daily and daily models of IMERG-E, IMERG-L, and IMERG-F are populated in Tables 6 and 7. Both sub-daily and daily models were calibrated for daily and monthly simulations. Hence, six models (three with sub-daily precipitation inputs and three with daily precipitation inputs) were calibrated for daily simulations, and six models (three with sub-daily precipitation inputs and three with daily precipitation inputs) were calibrated for monthly simulations. Every model was then validated with a similar number of simulations, i.e., 500 (as they were in the last iteration of calibration). Models were calibrated from 2003 to 2008 with 2 years of warmup periods. Then, calibrated models were validated from 2009 to 2012. The complete picture of calibrated and validated, sub-daily and daily, models for the simulation of daily and monthly discharge is shown in Figure 6. Figure 6(a) and 6(c) represents the daily flow simulations with sub-daily and daily IMERG products, respectively. Figure 6(b) and 6(d) displays the monthly flow simulations with sub-daily and daily IMERG products, respectively. Figure 6 suggests that IMERG-HHF and IMERG-DF have better performance than the rest of IMERG sub-daily or daily products for daily and monthly simulations. It is also revealed from Figure 6 that IMERG-E has better performance than IMERG-L after IMERG-F for both sub-daily and daily products. This can be observed in monthly simulations, i.e., Figure 6(b) and 6(d), in which IMERG-L shows underestimation from IMERG-F and IMERG-E.

To verify the validity of satellite precipitation data for the Chenab River catchment, it is required to compare the daily and monthly simulated runoff with the observed runoff at the Marala Barrage gauging station. Performance evaluation of three sub-daily IMERG-based SPPs and three daily IMERG-based SPPs was done by five metrics (P-factor, R-factor, R², NSE, and PBIAS). The calculated statistical metrics of calibrated and validated discharge values, at daily and monthly timescales, are shown in Table 8. Different comparisons between flow simulations with different IMERG precipitation inputs are presented in the subsequent sub-section.

For comparison of daily and monthly flow simulations with sub-daily and daily IMERG products, monthly calibration and validation show more favorable results than daily calibration and validation (Table 8). Monthly simulated flows are fairly in line with the observed flow, as the model performance in capturing the peaks at the monthly timescale is better than the daily timescale. For example, PBIAS values in the validation of monthly simulations with IMERG-DE, IMERG-DL, and IMERG-DF present 18.8, 26.4, and 22.4, respectively, which are less than daily simulations, i.e., 24.1, 31, and 27.9, respectively. On the other hand, PBIAS values in the validation of monthly simulations with IMERG-HHE, IMERG-HHL, and IMERG-HHF present 5.6, 15.1, and 2.4, respectively, which are less than daily simulations, i.e., 5.9, 15.6, and 15.4, respectively. A similar trend for the NSE is noted in Table 8, where the validation of monthly simulations with IMERG-DE, IMERG-DL, and IMERG-DF exhibit large values, i.e., 0.75, 0.67, and 0.81, respectively, whereas daily simulations show small values, i.e., 0.58, 0.50, and 0.57, respectively. For the validation of monthly simulation with IMERG-HHE, IMERG-HHL, and IMERG-HHF, the NSE values are 0.79, 0.71, and 0.82, respectively, which are less than the daily validation results, i.e., 0.61, 0.55, and 0.63, respectively. The values of R² for monthly simulations exhibit a similar trend of better values than daily simulations (Table 8). The values of R² for monthly simulations with IMERG-DE, IMERG-DL, and IMERG-DF display 0.82, 0.81, and 0.83, respectively, whereas for daily simulations, the values are 0.65, 0.63, and 0.66, respectively. In the case of validation of monthly simulation with IMERG-HHE, IMERG-HHL, and IMERG-HHF, the values of R² are 0.83, 0.82, and 0.84, respectively, while for daily simulations, the values are 0.61, 0.58, and 0.66, respectively. The above discussion indicates that the monthly flow simulations perform better than the daily flow simulations.

According to Table 8, the values of P-factor and R-factor are well balanced for flow simulation with the sub-daily model than the daily model. For both daily and monthly flow simulations, the results of IMERG-HH are more reliable than those of IMERG-D for each early, late, and final SPP. For both daily and monthly flow simulations, sub-daily IMERG outperformed each evaluation indices. However, the only R² value of IMERG-HHE is less than that of IMERG-DE for both daily and monthly simulations, which means the IMERG-DE model has better goodness of fit than IMERG-HHE. Contrary, IMERG-HHE has better NSE than IMERG-DE for both daily and monthly simulations. According to the values of the PBIAS, the IMERG-HHE-based model has less underestimation than the IMERG-DE-based model for both simulations. Thus, IMERG-HHE has better performance than IMERG-DE. IMERG-HHL and IMERG-HHF have also outperformed the IMERD-DL and IMERG-DF models, respectively, for both daily and monthly simulations. Same as IMERG-HHE and IMERG-DE, both IMERG-HHL and IMERG-HHF have fewer underestimations than IMERG-DL and IMERG-DF, respectively. Conclusively, all sub-daily models exhibit better performance than daily models for both daily and monthly simulations.

The selection of the best IMERG product for flow simulation in the Chenab River catchment depends on the evaluation indices. Table 8 demonstrates that the value of R² for daily validation with IMERG-DE, IMERG-DL, and IMERG-DF display 0.65, 0.63, and 0.66, respectively, while for monthly validation, they are 0.82, 0.81, and 0.83, respectively. The values of the NSE for daily validation with IMERG-DE, IMERG-DL, and IMERG-DF display 0.58, 0.50, and 0.57, respectively, while for monthly validation, they represent 0.75, 0.67, and 0.81, respectively. The PBIAS for daily validation with IMERG-DE, IMERG-DL, and IMERG-DF depict 24.1, 31, and 27.9, respectively, whereas, for monthly validation, they exhibit 18.8, 26.4, and 22.4, respectively. In accordance with the above discussion, these results show that IMERG-DF performs better, followed by IMERG-DE and IMERG-DL. Table 8 also shows the values of R² for daily validation with IMERG-HHE, IMERG-HHL, and IMERG-HHF, i.e., 0.61, 0.58, and 0.66, respectively, while for monthly validation, they are 0.83, 0.82, and 0.84, respectively. The NSE values for daily validation with IMERG-HHE, IMERG-HHL, and IMERG-HHF display 0.61, 0.55, and 0.63, respectively, while for monthly validation, they show 0.79, 0.71, and 0.82, respectively. PBIAS values for daily validation with IMERG-HHE, IMERG-HHL, and IMERG-HHF display 5.9, 15.6, and 15.4, respectively, whereas, for monthly validation, they present 5.6, 15.1, and 2.4, respectively. This discussion verifies that IMERG-HHF has better performance, followed by IMERG-HHE and IMERG-HHL.

For a complete understanding of models’ performance, Taylor diagrams were constructed to statistically summarize the agreement between the simulated and observed discharge values at the Marala Barrage gauging station. In the Taylor diagram (Figure 7), the RMSE in simulated discharge is associated with the distance from the point marked as ‘observed’ on the x-axis, the Pearson correlation (R) is related to the azimuthal angle, and the SD of simulated discharge is linked to the radial distance from the origin. Near the point marked as ‘observed’ on the x-axis, the model simulation values will be considered the best fit with the observed values. For model representation in Figure 7, filled shapes depict sub-daily models and hollow shapes depict daily models. Figure 7(a) and 7(b) represents the daily discharge calibrations and validations, with red, green, and blue colors showing the IMERG-E, IMERG-L, and IMERG-F models, respectively. Figure 7(c) and 7(d) depicts the monthly flow calibrations and validations, with pink, purple, and orange colors showing the IMERG-E, IMERG-L, and IMERG-F models, respectively.

It is evident from Figure 7(a) and 7(b) that among calibration and validation of six models, the IMERG-HHF-based model's performance appears to be more favorable than all the other IMERG-based models for daily flow simulation. This means that the discharge simulated by IMERG-HHF provides the best approximation to the observed daily discharge. The IMERG-HHL-based model shows poor performance in daily flow calibration, while the IMERG-DL-based model performs poorly in daily flow validation. It can be seen in Figure 7(a) that IMERG-HHF has a higher value of R, lower RMSE, and higher SD. Contrary, IMERG-HHL has a lower value of R and SD.

Figure 7(c) displays the calibration of monthly flow simulation, in which IMERG-HHF shows a strong correlation with the observed flow and less RMSE, while IMERG-DF has more SD than IMERG-HHF. In Figure 7(d), IMERG-DF displays better results in the validation. However, the overall performance of IMERG-HHF is outstanding for daily and monthly simulations. Like daily simulation, a similar trend exists for monthly flow simulation driven by the IMERG-HHL-based model, which shows poor performance in both calibration and validation phases (Figure 7). IMERG-DL shows poor performance, but it is better than that of IMERG-HHL. Nevertheless, IMERG-E performs better than IMERG-L for both daily and monthly simulations in this catchment. This observation provides a positive outcome for this research because IMERG-E is the NRT IMERG product, which can be used for NRT flow simulation in future studies. Furthermore, according to Figure 7, trends from daily and monthly flow simulations depict that the results from monthly simulations dominate results from daily simulations. The monthly flow simulations represented a higher correlation coefficient, less RMSE, and higher SD close to the point marked as ‘Observed’. Table 9 also shows the Taylor diagram statistics in a tabular form.

The SWAT model has been widely used in continuous simulations for daily, monthly, and yearly simulations with the different resolutions of spatial precipitation inputs. However, few studies have been carried out to understand the impact of temporal resolution of precipitation on the SWAT model performance in large catchments. This study focused on developing and comparing the sub-daily and daily SWAT models for a large-sized transboundary Chenab River catchment in Pakistan and India.

Three GPM IMERG precipitation products, i.e., IMERG-E, IMERG-L, and IMERG-F, were downloaded at HH/sub-daily and daily. The precipitation information of the closest pixel from the centroid of each sub-catchment was extracted and supplied to Arc-SWAT for the simulation of discharge. Six sub-daily and six daily precipitation input models were calibrated and validated from 2003 to 2012 (with 2 years of warmup periods – 2003 and 2004) for daily and monthly flow simulations. The GSA and calibration were performed by using the SWAT-CUP model. The SWAT model simulations were compared with observed hydrographs at the Marala Barrage gauging site in Pakistan. In this study, the performance indices and the Taylor diagram depicted that IMERG-F showed better results than IMERG-E and IMERG-L. This study also endorses the previous studies (i.e., Wang et al. 2017; Jiang et al. 2018; Gan et al. 2020), in which IMERG-F outperformed IMERG-E and IMERG-L.

After a comparison of the model simulations with sub-daily and daily precipitation inputs, this study concluded that the temporal resolution of precipitation could affect the runoff output for both daily and monthly flow simulations. The sub-daily SWAT model showed an improvement in simulating the flows and showed its capacity to simulate the peak flows more precisely than the daily SWAT model. This study was endorsed by Yang et al. (2016), who demonstrated that the SWAT model with sub-daily precipitation inputs performed better than the model with daily precipitation inputs for daily streamflow simulation. Wetterhall et al. (2011) also compared the daily and sub-daily precipitation inputs for flood forecasting and found that the sub-daily precipitation input can improve the flood forecasting results. Moreover, Ochoa-Rodriguez et al. (2015) also evaluated the effects of temporal resolution on the flow simulation, and they concluded that temporal resolution of precipitation affects the hydrodynamic modeling results more strongly than spatial resolution.

This study focused on the suitability of the latest, NRT, or post-real-time IMERG products for reliable simulation in a transboundary Chenab River catchment. The Chenab River at Marala Barrage shares 97% of its catchment with India, where there are no rain gauges’ data accessibility from Pakistan, as there is less practice of data sharing between Pakistan and India due to the Jammu and Kashmir conflict. In recent years, Pakistan had to face many damages from floods in the Chenab River because of weak data sharing practices of precipitation and discharge data from the transboundary Indian side catchment. Overall, this research validates the usefulness of SPPs in the transboundary Chenab River catchment. Some researchers have also selected the SPPs, in the recent past, for flood forecasting purposes in the Chenab River catchment. For instance, Shahid et al. (2017) used TRMM 3B42 data for predicting the peak flows at Marala Barrage, Pakistan through event-based hydrological modeling. They found that the TRMM 3B42 precipitation data is well capable of capturing the peak flows. Asghar et al. (2019) used GPM GSMaP-Gauge for flood and inundation forecasting in the transboundary Chenab River catchment. They applied the bias correction on the GSMaP-Gauge product with the help of station data from a nearby catchment to overcome the underestimation. They concluded that their technique can become more reliable by further studies in the Chenab River catchment. However, there is still no study done using sub-daily IMERG data in this catchment. This study could be a breakthrough for selecting the IMERG data for flow simulation in this catchment. The results from this study will encourage water managers in Pakistan to choose IMERG precipitation products for the Chenab River catchment or nearby similar catchments for better hydrologic modeling.

This study compared the IMERG-based ‘Early’, ‘Late’, and ‘Final’ precipitation products for flow simulation in a transboundary catchment of Chenab River at Marala Barrage, Pakistan. Twelve models were set up in the SWAT model and calibrated and validated for the study catchment. The results showed that IMERG precipitation products have a good potential to simulate the runoff. IMERG-HHF outperformed the IMERG-HHE followed by IMERG-HHL, and similarly, IMERG-DF outperformed the IMERG-DE followed by IMERG-DL for both daily and monthly simulations. IMERG-HHF performed better than IMERG-DF for daily flow simulations. In comparison to daily flow simulations, the performance of IMERG-E, IMERG-L, and IMERG-F depicted that the flow simulations at the monthly timescale have high values of R², NSE, and R, but fewer values of the PBIAS and RMSE. Conclusively, IMERG-HHE is perfect for NRT flow simulation, while IMERG-HHF is considered best for post-real-time flow simulation in the Chenab River catchment.

This research demonstrated the applicability of GPM-era IMERG SPPs in a transboundary Chenab River catchment using the SWAT model. While GPM-era SPPs are also available in the form of GSMaP at higher temporal and spatial resolutions (1 h and 0.1°), which has to have consideration in future studies for continuous flow simulations in this catchment. Moreover, this study was performed from 2003 to 2012. This means that the use of the latest IMERG SPPs for hydrologic simulation in this catchment is worth studying. The model was set up with the provision of satellite precipitation data to the nearest cells of the catchments’ centroids. Further studies are required in a full-scale distributed model to support the sub-daily simulation over daily simulation.

The authors are thankful to the GPM research communities for making the satellite precipitation data available for this work. The authors would like to extend their gratitude to the Flood Forecasting Division, PMD, for the provision of valuable data for this research. The authors also give appreciation to the reviewers for their thoughtful comments. E.A. is grateful to the Higher Education Commission of Pakistan (HEC) and the German Academic Exchange Service (DAAD) for providing him scholarship opportunity to pursue PhD degree. E.A. would like to thank the facilities provided by the Institute of Urban and Industrial Water Management. E.A. is also thankful to IWA Publishing for Subscribe to Open (S₂O) platform.

E.A. conceptualized the study, formulated the study methodology, and involved in the software setup and execution; E.A. and W.Y. did the validation; A.A. performed the formal analysis; N.S. did the data curation; E.A. wrote the original draft preparation; F.A.J. wrote, reviewed, and edited the manuscript; F.A.J. was involved in visualization; and P.K. supervised the study.

The authors did not receive support from any organization for the submitted work.

There are no relevant financial or non-financial competing interests to report.

All relevant data used in this research are referred/cited in the paper.