Evaluation of Satellite Precipitation Products for Estimation of Floods in Data-Scarce Environment

Utilization of satellite precipitation products (SPPs) for reliable food modeling has become a necessity due to the scarcity of conventional gauging systems. Tree high-resolution SPPs, i.e., Integrated Multi-satellite Retrieval for GPM (IMERG), Global Satellite Mapping of Precipitation (GSMaP), and Climate Hazards Group InfraRed Precipitation with Station (CHIRPS), data were assessed statistically and hydrologically in the sparsely gauged Chenab River basin of Pakistan. Te consistency of rain gauge data was assessed by the double mass curve (DMC). Te statistical metrics applied were probability of detection (POD), critical success index (CSI), false alarm ratio (FAR), correlation coefcient (CC), root mean square error (RMSE), and bias (B). Te hydrologic evaluation was conducted with calibration and validation scenarios for the monsoon fooding season using the Integrated Flood Analysis System (IFAS) and fow duration curve (FDC). Sensitivity analysis was conducted using ± 20% calibrating parameters. Te rain gauge data have been found to be consistent with the higher coefcient of determination ( R 2 ). Te mean skill scores of GSMaP were superior to those of CHIRPS and IMERG. More bias was observed during the monsoon than during western disturbances. Te most sensitive parameter was the base fow coefcient (AGD), with a high mean absolute sensitivity index value. During model calibration, good values of performance indicators, i.e., R 2 , Nash − Sutclife efciency (NSE), and percentage bias (PBIAS), were found for the used SPPs. For validation, GSMaP performed better with comparatively higher values of R 2 and NSE and a lower value of PBIAS. Te FDC exhibited SPPs’ excellent performance during 20% to 40% exceedance time.


Introduction
A food is a frequently recurring destructive natural hazard that humanity encounters [1]. Te adverse fooding conditions have resulted in a signifcant deterioration of environmental sustainability [2] throughout the world, wreaking havoc with costly infrastructure, food production, and human lives [3]. Globally, climate change and anthropogenic activities have increased the frequency and intensity of foods and exacerbated riverine food threats in several world places [4]. Pakistan is located in a highly vulnerable region to climate change where foods occur almost every year since the last three decades [5,6]. Te devastating food in Pakistan's history has emphasized the critical need for an efcacious and steadfast food warning system [7]. Improved food modeling and remote sensing necessitate enhanced food analysis methodology to send timely warnings to populations and better reservoir operations [8]. Te early warning system in Pakistan has limited capabilities. Although substantial improvements in food forecasting have been made by utilizing the weather radar and telemetric systems in the warning system, there is still a need for many eforts to advance the food forecasting and warning system [9,10].
Precipitation is an important hydrological parameter used for watershed management, food forecasting, and climatological assessment [11]. It is also the most complex parameter because of its excessive spatiotemporal variations that traditional rain gauges and radar networks cannot record due to their sparsity [12]. Accurate and precise precipitation data with fne spatiotemporal resolution is important for watershed management and food analysis [13]. It necessitates the need for such techniques that supplements the rain gauge observations and provides exceptional precipitation data to support hydrological modeling issues [14][15][16].
Remote sensing satellites use refected light to detect, collect, measure, and record the electromagnetic energy from the earth's surface [17]. Advances in satellite remote sensing have made it an excellent data source, as it can provide metrological data to support hydrological modeling issues [18][19][20]. Moreover, precipitation data obtained from remote sensing have the potential to supplement the traditional rain gauge system [21].
Recent studies have shown that the precipitation data estimated through satellite-based observations contributed well to detecting rainfall distribution and severity in datascarce regions [22,23]. However, the satellite products may have errors due to indirect estimation, sampling uncertainty, and retrieval algorithms [24][25][26][27]. Te properties of these errors signifcantly vary in contrasting climates, storms, seasons, and altitudes [11,28]. Terefore, it is essential to validate the accuracy of satellite precipitation products and suitability for a broader range of environments. Satellite rainfall data can be validated using statistical analysis regarding ground-based on gauge data and a proper hydrological modeling framework [29]. Statistical analysis determines the accuracy and consistency of satellite precipitation data, while hydrological simulation elucidates the usefulness and application of the same datasets [30]. For reliable hydrological modeling, a proper calibration technique, parametric sensitivity, and model capability are of primary consideration. A suitable calibration technique is critical because errors in model calibration and input datasets contribute to incorrect outcomes [31]. Similarly, parametric sensitivity produces only important parameters, reduces the analysis time, and contributes to modeling calibration [32].
Furthermore, the ability of the hydrological model to simulate water fow can be examined using the FDC, "a key runof variability signature" [33]. Te FDC ofers additional details on the basins' hydrological modeling and underlying processes [34,35]. Generally, lumped, semidistributed, and distributed models are used for a watershed's hydrological modeling. Lumped models consider spatially uniform watershed characteristics. Te semidistributed model divides the watershed into subbasins with unique hydrological responses [36]. Conversely, distributed models consider the spatial and temporal variation of physical properties in the watershed. Tese models can interpolate the rainfall data and predict water fow at ungauged locations. However, these models require a considerable input data set for fow estimations [37].
Recently, studies reported that statistical assessment of precipitation data had not yielded reliable results that necessitated hydrological modeling. In Pakistan, limited studies have been conducted to assess satellite precipitation products' efectiveness, particularly utilizing the distributed IFAS model. In [38], the authors highlighted the scarcity of hydrological data and the signifcance of upstream fow boundary conditions when barrage operation standards are unknown in the Indus River. In [31], the authors used a lump and regional calibration approach to model the Jhelum river basin. In [39], the authors pointed out the difculty of food modeling at the confuence point of the Chenab and Jhelum basins. In [38], the authors explored that the performance of IFAS can be improved by utilizing local soil texture data in the Indus River. In [38], the authors evaluated the precipitation results from diferent sources for modeling the Indus River's middle reach. Aziz [40] demonstrated that IFAS could be used for hydrological modeling of the Kabul River with data scarcity. In [41], the authors investigated that integrating satellite and gauge rainfall data can enhance food forecasting in the Philippines-Cagayan River catchment. In [42], the authors recommended that improved satellite precipitation data be used to enhance food prediction in the Dungun River basin, Malaysia.
Assessment of SPPs with a fully distributed hydrological model under diferent calibration scenarios is yet to be evaluated in the study area. In addition, the representation of hydrological signatures with diferent rainfall data sets is yet to be explored. It mandates the investigation of satellite precipitation data sets using a distributed hydrological model for diferent applications. In this study, three SPPbased datasets, i.e., IMERG, GSMaP, and CHIRPS, have been evaluated statistically and hydrologically in a data-scar region, i.e., the Chenab River catchment of Pakistan. Te study utilized the IFAS model to generate streamfow for a sparsely gauged catchment by using satellite precipitation datasets and derived hydrological signatures.

Study Area and Data Description
2.1. Study Area. Te Chenab River starts in Himachal Pradesh, India, at the confuence of the Bhaga and Chandra streams and fows across Indian-controlled Kashmir to Pakistan [43]. Te catchment of the Chenab River covers an area of about 26,000 km 2 up to the Marala Barrage. It embraces 97% of this catchment area in India, while only 3% in Pakistan up to the Marala barrage [44]. In Pakistan, there are four streamfow gauge stations on the river, i.e., Marala barrage, Khanki barrage, Qadirabad barrage, and Trimmu barrage. For the present research, the study area ranged from the Marala barrage to the Trimmu barrage ( Figure 1).
Since the Chenab River and Jhelum River converged at the Trimmu barrage, hydrological modeling at the convergence point is not possible [39]. So, the study considered the assessment of outfow at Qadir Abad barrage with an assumption of free fow at Khanki barrage during the monsoon fooding season. Te selected catchment is situated between latitudes 72°-78°E and longitudes 32°-34°N, spanning over ∼16,000 km 2 , with a gradient of 0.4 m/km downstream of plain areas [43].
Te crosssectional characteristics of the basin comprise 85 km in length with an average width of 800 m. Te scarcity and sparsity of meteorological gauging stations are big concerns in this region. Tere are only four rain gauges organized by the Pakistan Meteorological Department (PMD), which do not meet the requirements of the World Meteorological Organization (WMO) and are inadequate for hydrological modeling for watershed management. Te central hydrology of the catchment is controlled by the summer monsoon and winter seasons, where the summer monsoon season dominates and has triggered signifcant fooding in this region.

Data Description.
Te data for the research was collected at a daily scale for the years 2015-2020 and consisted of a gauge rainfall dataset, observed streamfow, and a satellite precipitation dataset. In addition, the topographical data comprises a digital elevation model, land use, and soil type. Daily rainfall data for the selected rain gauge stations were acquired from the PMD. Streamfow data for stream gauging stations were collected from Pakistan's Flood Forecasting Division (FFD).
Satellite precipitation datasets at the daily time scale (Table 1) consisting of GSMaP, IMERGE, and CHIRPS were downloaded from the Japan Aerospace Exploration Agency (JAXA) (https://sharaku.eorc.jaxa.jp/GSMaP/), National Aeronautics and Space Administration (NASA) (https://pmm.nasa.gov/data-access/downloads/gpm), and the University of California, Santa Barbara's Climate Hazards Group (UC Santa Barbara) (https://chg.geog. ucsb.edu/data/chirps/), respectively. GSMaP consists of four types of products; two real-time (GSMaP-NRT, GSMaP-Gauge, and NRT) and two postreal-time (GSMaP-Gauge, GSMaP-MVK). In the present work, GSMaP-Gauge NRT (version 6) was used. In order to formulate the GSMaP-Gauge NRT precipitation predictions with a 4 h latency period, the error parameters estimated for the postreal-time product of GSMaP-Gauge are utilized. GSMaP-Gauge also employed a blending of passive microwave (PMW) and infrared (IR) data along with a unifed gauge-based analysis of the global daily precipitation dataset from the Climate Prediction Center (CPC) [22].

Advances in Meteorology
Similarly, Global Precipitation Measurement (GPM) is a multinational satellite project to integrate and enhance precipitation observations from diferent satellites. IMERG is a GPM-based level 3 multisatellite precipitation algorithm that incorporates all passive microwave and infrared-based observations in the constellation. Typically, three products of IMERG precipitation (early-run, late-run, and fnal-run) are mostly considered. Early-run uses only forward morphing with a 4 hour latency period; however, late-run and fnal-run use forward and backward morphing with latency periods of 14 hours and 3.5 months, respectively [45]. IMERG combines passive microwave, propagated pulse width modulation, and infrared radiation-based observations by the Kalman flter method to obtain precise estimation [46]. Tis research work utilized daily IMERGlate-run version 6 for statistical and hydrological assessments in the study area.
Moreover, CHIRPS is a quasi-global rainfall dataset that consists of three types of temporal data (daily, pentanal (5 days), and monthly) with two types of spatial resolution (0.05°, 0.25°). Te daily data are considered real-time data with a latency period of 2 days, while pentanal and monthly data are considered post-real-time datasets with a 21 days latency period. Its algorithm is based on cold cloud duration (CCD) and ground gauge observations to approximate the rainfall. Tis study employed the daily 0.05°grid CHIRPS version 2.0 dataset [47].
Furthermore, digital elevation model (DEM) was used to represent the catchment's topography and delineate the watershed. Te present study collected DEM and Global Map's land cover data (Version 2) from the International Steering Committee for Global Mapping (ISCGM). Finally, the soil type data based on the Digital Soil Map of the World (DSMW), provided by the Food and Agriculture Organization (FAO), was used.

Double Mass Curve
Analysis. DMC is employed to inspect the consistency of the hydrologic data and to adjust the inconsistent precipitation data. In this graphical approach, the cumulative data of a single station is compared with the pattern composed of cumulative data from other stations in the area. Likewise, for the used pattern, enough gauging stations must be included while checking the consistency of precipitation records so that inconsistency does not signifcantly infuence the average in one of the station' records. If there are less than 10 stations located in a specifc region, the consistency of each station must be examined. Terefore, at all four gauging stations located in the study area, the DMC technique was applied to examine the consistency of annual rainfall data.

Evaluation Statistics.
Te efcacy of selected satellite precipitation products was assessed against four-gauge station records with categorical and continuous metrics on a daily, ten daily, monthly, and seasonal scales from 2015 to 2020. Diferent approaches have been used by comparing point precipitation data observed by rain gauges with pixel precipitation data recorded by remote sensing satellites. Usually, such procedures are based upon the upscale interpolation of point values to grid scale data and the downscaling of grid data towards point values. Te reanalyzed gridded data always vary from the station observations and vice versa in several aspects. To avoid inaccuracies caused by such upscale interpolation methods and downscaling, a more direct approach has been proposed and used. In this approach, precipitation observed at stations falling within a grid cell will be averaged to obtain an estimate for the observed precipitation at the center of that grid cell and then compared to the gridded value [21]. Tis approach has been used in this study for comparison between rain gauge and satellite data sets. For this purpose, satellite precipitation data at the daily time scale was downloaded and then converted into ten daily, monthly, and seasonal scales. In selecting the tile of satellite precipitation data, PMD gauge value recording time (8:00 am daily) was kept in focus. Event detection capability was evaluated with categorical metrics that include POD, FAR, and CSI. POD refects the ratio of accurately identifed rainfall events by the satellite concerning gauge rainfall data. FAR demonstrates the fraction of rainfall events in which the satellite predicts precipitation while the rain gauge does not observe it. CSI represents typically the fraction of rainfall occurrence accurately recognized by the satellite. Continuous metrics measure the quantitative diference between observed and predicted precipitation. Tese metrics include bias, CC, and RMSE. Bias is the mean discrepancy between satellite estimation and rain gauge data. Depending on the quality of the rainfall data, its value could be positive or negative, indicating Advances in Meteorology overestimation and underestimation, respectively. CC estimates the degree of agreement between the satellite and rain gauge precipitation data. RMSE depicts the mean dispersion of predicted precipitation around the known value of gauge observations. It is used to evaluate the precision of the rainfall dataset [48]. From historical data, two rainy seasons, monsoon (June to September) and westerly disturbance (November to February), have been established in the study area: these were considered for seasonal evaluation at the daily scale. Tese statistical metrics are given as follows: where H, M, and F exhibit the number of hit, miss, and false alarm events, while Gi and Si denote the gauge and satellite precipitation, Gm and Sm represent the mean of gauge and satellite precipitation data, and N indicates the total number of events used for evaluation.

Explication of Hydrological (IFAS)
Model. IFAS is a succinct runof analysis toolkit designed for food prediction in basins with insufcient hydrological and geophysical information in developing countries. It is categorized as a physically distributed framework that can integrate gauge rainfall data, satellite-based precipitation data, evaporation data, snowmelt data, and geophysical data to simulate river course fow. It integrates grid-based datasets of topography, geology, and land cover to estimate the parameters of the physical conditions of a basin [49]. Te model can generate the channel network using topographical data to defne the basin, sub-basin boundaries, fow direction, and drainage patterns. It employs the Public Works Research Institute Distributed Hydrological Model (PWRI-DHM), consisting of a two-or three-tank structure and a routing model for runof simulation. Te three-tank structure comprises a surface, sub surface, and aquifer, while the routing model comprises a kinematic hydraulic river course routing tank. PWRI-DHM uses a nonlinear relationship to calculate each cell's outfow based on the tank model philosophy, considering Manning's equation, Darcy's law, and hyperbolic approximations. It uses a kinematic wave equation to calculate the discharge in the river course tank [38,50,51].

Model Formulation.
For the development of the IFAS model, the extent of the target study area was defned by determining the latitude and longitude of the selected catchment. Te IFAS model with a two-layered and threetank structure was created by customizing the digital-based land cover, elevation, and soil type data to the appropriate grid size. Te shapefle of the study area was imported into the basin manager function of IFAS to defne basin and sub basin boundaries. Te surface tank parameters were estimated utilizing the land cover data, while the aquifer tank parameters were tuned according to soil type data. Te essential aspect of hydrological modeling in a basin is accurately estimating runof and water level initial conditions in the river course that afect the parameter optimization of the model [31]. Te model was run six months before the calibration of the food event to generate proper initial conditions until hydrological equilibrium was achieved. Te principle of equifnality dictates that many combinations of parameters are possible that give good agreement with the observed streamfow data. Boundary conditions are essential, especially when hydrological data is scarce and standards for barrage operations are unknown. IFAS has an integrated water resources management (IWRM) interface that contains various techniques to incorporate barrage operating tasks. Te discharge fle technique was applied using the IWRM function to give the boundary condition in this study condition. Tis technique employed daily discharge data in the CSV fle to represent barrage operations. Marala barrage outfows were considered boundary conditions in this research due to data scarcity and unknown barrage operations upstream of the catchment.

Calibration Scenarios and Model Performance Indicators.
Since the outputs of hydrological models are rarely capable of accurately refecting nature in its completeness, their performance must be evaluated before they can be employed in any decision-making process. Te IFAS is designed for food analysis; therefore, it was calibrated and validated utilizing precipitation data collected during the monsoon seasons (July to October) of 2015 and 2017, considering medium and high fooding years, respectively (PMD/FFD). Te model was calibrated individually using CHIRPS, GSMaP, and IMERG precipitation datasets. All datasets were validated against each calibration scenario. Te calibration of the model was achieved through a trial-and-error process. Te model performance was evaluated using model performance indicators (MPI), R 2 , NSE, and PBIAS. Te R 2 is a statistical indicator representing the fraction of the dependent variable's variance predicted by the independent variable. Te ideal value is 1, while a lower value than 1 reveals the variation of model output. Te model performance with R 2 > 0.5 is acceptable. Likewise, the NSE is the most often used method for determining correlation to test the efcacy of hydrological models. Te literature reveals a variety of acceptable, very good, and excellent value categories for NSE. Te calibration tolerance criteria are very subjective. Calibration with NSE is generally perceived as good if it is higher than 0.6 and excellent if it is more Advances in Meteorology signifcant than 0.8 in the literature. Te model validation criteria are less restrictive than the calibration levels. A value of NSE greater than 0.5 is acceptable for validation, while NSE greater than 0.7 is considered highly excellent. Te PBIAS examines the average tendency of the simulated fows to be greater or smaller than their observed fows. Te perfect value of PBIAS is 0, and lower values represent accurate model reproduction [52].
Tese MPIS are given as follows: where Oi and Pi represent the observed and simulated fows, Om and Pm are denoted by the mean values of the observed and simulated fows, and n is the total number of events.

Implications of Sensitivity Analysis Technique.
Te sensitivity analysis helps to examine the nonlinear variation of highly uncertain parameters in the complex models. Te IFAS model was investigated to determine the most sensitive calibrating parameters severely afecting the calibrating hydrograph. In this regard, the values of all calibrated parameters were frst increased and then decreased by 20%, one by one of their calibrated values. C +20% is the 20 percent change in the calibrated parameter value and is determined using equation (3). So are the simulation results of a calibrated hydrograph, and S +20% is the change that occurred in the simulation results when one of the calibrating parameters' values changed to +20%. Te mean percentage change in the simulation results by increasing or decreasing the calibrating parameter value to 20% is called the sensitivity index (I) and can be determined using equation (4). Te mean absolute sensitivity index (MASI) can be determined using equation (5).
3.7. Flow Assessment Using Hydrological Signature. Te FDC is an infuential streamfow variability signature that describes hydrological behavior. It is the graphical representation of fows and the percentage of time that the fows equal or surpass each other. Satellite precipitation datasets were evaluated through the FDC, in which the observed streamfow was taken as the baseline and the variation in simulated fow was evaluated. Furthermore, the ability of each precipitation dataset to generate high and medium fows was examined through dependable fow exceedance, where the extreme food events were represented in the range of Q5-Q25 dependable fows, the medium fow required for irrigation was designated by Q50 dependable fows, and Q70 dependable fows correspond to the water availability for domestic supply.

Consistency of Gauge Rainfall Data.
Double mass analysis has been used to check the consistency of rainfall data records at four stations, i.e., Gujrat, Sialkot, Sargodha, and Jhang. Te cumulative data of a single station was compared with the cumulative data of other stations in the respective area. Te straight line shows data consistency, whereas any change in the straight line manifests a change in the data collection method that afects the relationship. For any station, the rise of the curve from the trend line shows that there was more annual rainfall than in other stations. For all gauging stations, the R 2 values were 0.98 to 0.99, and the annual rainfall data were consistent with the DMC technique ( Figure 2). where CAR � is the cumulative annual rainfall.

Statistical Evaluation of SPPs at Diferent Temporal Scales.
Tis study identifed and quantifed the errors associated with satellite datasets. Te efcacy of selected satellite products (CHIRPS, IMERG, and GSMaP) was assessed statistically at daily, 10-daily, monthly, annual, and seasonal scales using precipitation data recorded at PMD stations.

Daily and 10-Daily Scale.
Statistical evaluations of selected SPPs at daily and 10-daily levels are presented in Table 2. In the case of categorical metrics, the mean POD of GSMaP was better than CHIRPS and IMERG on both temporal scales. Te mean POD for CHIRPS and IMERG was lower by 50.79% and 22.22% for the daily scale with reference to GSMaP. In terms of mean FAR values, IMERG and GSMaP showed good agreement, and CHIRPS underperformed. CSI gives more stable results due to the characteristics of the blending of POD and FAR. Te performance of GSMaP, IMERG, and CHIRPS was improved by 54%, 58%, and 55%, respectively, for CSI at the 10-daily scale. Remarkably better values of categorical metrics were given by all the used satellite products at the 10 daily time scale as compared to the daily time scale.
In the case of continuous metrics, the mean BIAS of GSMaP was better than the other two products at daily scale. A slight diference was observed between the mean RMSE values of the selected products on both temporal scales. However, it was noted that the used SPPs showed less errors (BIAS and RMSE) on a daily scale, compared to 10 daily. Probably this was due to a reduction in sample size at a larger time scale as compared to a smaller time scale. All SPPs did not show good agreement with rain gauge data at the daily scale. Te correlation coefcient was low at 0.21-0.30 at the daily scale, while it was high at 0.71-0.74 at the 10-daily scale, showing better performance of SPPs at a larger time scale, identical with categorical metrics. Overall, the statistical performance of SPPs was lower on a daily scale and higher on a 10-daily scale. Figure 3 shows a monthly comparison of IMERG, GSMaP, and CHIRPS precipitation observations with the reference data for the entire study period (January 2015 to December 2020). Te GSMaP precipitation product represented the best monthly precipitation temporal pattern. However, both IMERG and CHIRPS were also capable of representing the temporal variability of observed precipitation over the study area, albeit with notable overestimation. In July and August of 2016, 2017, and 2018, all precipitation data sources (gauges, IMERG, GSMaP, and CHIRPS) revealed increased precipitation magnitude. Almost all data sources exhibited a similar temporal pattern in monthly estimates from January to December 2015. All SPPs signifcantly overestimated precipitation for July through September 2020.

Monthly and Annual Scale.
On an annual time scale, a comparison was made between the rain gauge values and the used SPP values shown in Figure 4. Te annual average precipitation in the study area, as estimated from observations from 2015 to 2020 from four gauging stations, was 691 mm/year. It has been observed that the selected satellite-based precipitation products overestimated the annual precipitation amounts. IMERG and GSMaP showed overestimations of 23.47% and 7.17%, respectively, while CHIRPS showed an overestimation of 1.08% with reference to rain gauge values.
Te result showed some diference in estimating precipitation magnitudes by the IMERG products over the Chenab River basin of Pakistan, but the performance of CHIRPS and GSMaP encourages the utilization of SPPs in the study area at an annual time scale. Several researchers have also reported identical fndings in diferent regions of the world.   Figure 6 shows the performance of SPPs towards estimation of precipitation during winter due to the western disturbance season (westerly waves) for the entire study period based on daily precipitation data. Te event detection capability revealed that the POD of GSMaP was higher and better than the other two products. Te CHIRPS underperformed in terms of POD values. In the case of FAR, GSMaP outerperformed than IMERG and CHIRPS. In the case of CSI, all selected SPPs revealed better performance during western disturbances. In the case of bias value, IMERG overestimated the precipitation, while CHIRPS showed excellent performance. While considering the results of RMSE, an agreement was observed between the median values of CHIRPS and GSMaP. Te box plot results showed the RMSE values ranged from 2.5 to 7.5 mm/day for the selected satellite products, and higher values were produced by the IMERG. A strong agreement between CHIRPS and IMERG was observed for the CC results. Intercomparison revealed that SPPs showed comparatively better statistical performance during western disturbance than monsoon season. Conclusively, the statistical performance of GSMaP is better than other SPPs, as also reported in [22,53,54], in other regions of the world. Figure 7 shows the sensitivity analysis of the IFAS model for surface, aquifer, and river course tank parameters based on the mean absolute sensitivity index (MASI). In the case of surface tank parameters, the surface tank height (HFMND) and fnal  Meanwhile, the analysis of the river course tank parameters indicated that the parameters related to the coefcients of a crosssection of a river, i.e., RLCOF and RBS, were sensitive to the simulated hydrograph. Te aquifer tank parameters are more sensitive than any other tank parameters also reported by [55], and river tank parameters played signifcantly less in calibration. Terefore, it is suggested to introduce the option of actual groundwater conditions in the IFAS model for the target area.

Parametric Sensitivity Assessment.
Te IFAS model was calibrated and validated for the river Chenab at Qadir Abad barrage outlet for the monsoon periods of 2015 and 2017, respectively, utilizing the selected SPPs. In the calibration process, initially the default parameters for surface and aquifer tanks were used to run the model. Te surface parameters were based on digital land cover data, while the aquifer parameters were based on soil type data for the selected basin. Te parameters were tuned and optimized with the trial-and-error technique to bring them into sound agreement with the observed fow data. Te critical parameters considered for a successful calibration of the model are the coefcient of base fow regulation (AGD) for the aquifer tank, the surface tank height (HFMND), the fnal infltration capacity of the soil (SKF), and the initial height of infltration (HFOD) for the surface tank. Since HFOD is a surface parameter, it signifcantly infuences the adjustment of the peak of the hydrograph. Due to the surface tank's fve distinct feature classes, successful peak calibration requires fne-tuning of the land cover parameter. Land cover classes from the IFAS graphical module were used to calibrate the model, which was then fne-tuned using a trial-and-error method. Another surface tank parameter (FALFX) was tuned from 0 to 1 to control the subsurface fow to calibrate the model. Te values of FALFX parameters were subsequently decreased to adjust the hydrograph in the calibration process.
Tree diferent calibration scenarios were established to investigate the capacity of selected SPPs to calibrate the IFAS model and to examine their efectiveness for diferent applications. In the frst scenario, the model was calibrated utilizing the CHIRPS satellite precipitation data, and then the validation process was completed using the GSMaP and IMERG. For the second scenario, the IFAS model was calibrated utilizing the GSMaP, and the model was validated by using CHIRPS and IMERG for evaluation. In the third scenario, the IMERG precipitation dataset was utilized to calibrate the IFAS model and then validated against GSMaP and CHIRPS. Calibration and validation of the IFAS model were evaluated using the model performance indicators, i.e., NSE, R2, and PBIAS. Te model's performance on each scenario and comparison among the performance of the three scenarios are presented in Table 3.
For the frst calibration scenario, the statistical performance indicators R 2 , NSE, and PBIAS were 0.89, 0.86, and −0.16, respectively. Te intercomparison results of model validation for this scenario revealed the better performance of the GSMaP dataset with R 2 , NSE, and PBIAS values of 0.85, 0.83, and 0.16, respectively. Te IMERG and CHIRPS datasets showed slightly lower performance during the model validation process. From the graphical presentation of scenario 1 in Figure 8, some variations in simulating low and high fows were observed by the SPPs.
For the second calibration scenario, the statistical performance indicators (R 2 , NSE, and PBIAS) were 0.97, 0.96, and −0.03, respectively. According to the calibration criteria, this scenario displayed excellent performance, demonstrating that GSMaP precipitation data resulted in a robust and trustworthy testing model with utility and accuracy that could be used to check and compare the results produced from the IMERG and CHIRPS precipitation models. Comparison of model validation results revealed that the GSMaP dataset outperformed the other datasets, with R 2 , NSE, and PBIAS values of 0.9, 0.89, and 0.14, respectively. A strong agreement was observed between IMERG and CHIRPS-based simulated fows. Te ability of the GSMaP     Advances in Meteorology dataset to optimize the parameters and calibrate the model was better when compared with the CHIRPS precipitation model. A graphical presentation of scenario 2 is shown in Figure 9. It depicted a trend identical to scenario 1, but a bit improved simulation was observed in predicting low and high fows.
For the third calibration scenario, the statistical performance indicators R 2 , NSE, and PBIAS for model calibration were observed at 0.92, 0.91, and −0.11, respectively, which exhibited excellent performance of this model according to the calibration rating described by [52]. Te R 2 of the GSMaP, IMERG, and CHIRPS datasets were 0.87,

12
Advances in Meteorology 0.85, and 0.84, respectively, which depicts better GSMaP product performance for the third scenario also. A graphical presentation of this scenario, Figure 10, shows identical performance in forecasting low and high fows. It was noted that GSMaP outerperformed in terms of parameter optimization and fne-tuning of the IFAS model during the calibration procedure with R 2 , NSE, and PBIAS     the hydrological model for the high monsoon food of 2017, GSMaP outperformed in each scenario. Overall, the hydrological performance of GSMaP was more satisfactory than that of IMERG and CHIRPS, IMERG was ranked second, while the CHIRPS exhibited a lower performance.

Hydrological
Signature-Based Assessment. Te hydrological performance of all selected satellite precipitation datasets was evaluated through FDC, in which daily observed stream fows were taken as the baseline and the variation in simulated fow was determined. Te FDC results displayed that the selected satellite dataset has relatively inferior performance in capturing extreme fooding conditions. While considering the medium fow, all datasets showed excellent performance in the range of 20% to 40% exceedance time, as displayed in Figure 11.
Similarly, these precipitation datasets do not yield satisfactory results for the simulation of low fows. For all precipitation datasets at the catchment outlet stage, an exceedance fow analysis was used to estimate the dependable fow exceedance of Q5, Q10, Q25, Q50, and Q70. Q5 denotes a fow that exceeds 5% of the analysis time, and so forth. Extreme food events are revealed by 5% and 10% stream fows, while 50% dependability designates the median fow, 70% dependable fow resembles the water availability for agriculture, and higher dependable fows correspond to the water availability for domestic supplies. Te performances of SPPs to generate high, medium, and low fows was analyzed through these dependable fows. It was found that all SPPs data sets' performance was lower, corresponding to Q5 and Q70. Te SPPs datasets can generate medium fow in the range of Q25-Q50 Figure 12.

Conclusion
Te present study evaluated three high-resolution multisatellite precipitation estimation products statistically and hydrologically in the Chenab River catchment. Te consistency of rain gauge data observed by PMD was examined by double mass analysis. Numerous statistical indicators were applied at daily, monthly, and seasonal scales to detect and quantify errors associated with these products. Tree diferent calibration scenarios were established for the hydrological assessment to analyze the satellite precipitation datasets. A sensitivity analysis was performed to study the most sensitive parameters of the distributed IFAS model. Te hydrological signature was used to assess the potential of satellite products to generate high, medium, and low fows. Te existence of about 62 percent of the catchment area in Indian-held Kashmir and the occurrence of only four gauging stations in the rest of the catchment area are the major limitations of the study towards hydrological and statistical assessment of the satellite products in the study area, respectively. From the fndings of this study, it was observed as follows: (1) PMD rain gauge-based precipitation data are consistent and can be used for the assessment of satellitebased precipitation datasets.
(2) Statistical evaluation revealed that the efcacy of GSMaP has been better, while CHIRPS showed more biases. Te performance of SPPs improved at 10daily and monthly time scales than at the daily time scale. Higher values of uncertainties (bias and RMSE) were observed during the monsoon season than during the western disturbances. Missed and false alarms were the main errors associated with SPPs due to spatial mobility and the sudden bursting of clouds, specifcally during the monsoon season. (3) Te stativity analysis revealed that the aquifer tank parameters were found to be the most sensitive. Te base fow coefcient (AGD) was found to be the most sensitive parameter in calibrating the IFAS model to simulate fows using SPPs. (4) Te model calibration and validation scenarios indicated that the GSMaP precipitation dataset has better capability to calibrate and validate the model compared to IMERG and CHIRPS, with the highest R 2 , NSE and lower PBIAS values. It was also observed that the SPPs have relatively poor performance in capturing extreme fooding events. While considering the medium fows, in the range of 20%-40% exceedance time, all datasets showed excellent performance.
Findings of this study suggested that direct utilizations of satellite-based precipitation products were not promising at daily scales and bias correction is recommended. For food modeling, the hydrological IFAS model should be calibrated based on peak fow, considering the combination of statistical and error indicators. Further studies may be carried out to assess the efectiveness of the available sensitivity analysis techniques in this study area.