Quantification of Gridded Precipitation Products for the Streamflow Simulation on the Mekong River Basin Using Rainfall Assessment Framework: A Case Study for the Srepok River Subbasin, Central Highland Vietnam

Abstract

Many fields have identified an increasing need to use global satellite precipitation products for hydrological applications, especially in ungauged basins. In this study, we conduct a comprehensive evaluation of three Satellite-based Precipitation Products (SPPs): Integrated Multi–satellitE Retrievals for GPM (IMERG) Final run V6, Soil Moisture to Rain (SM2RAIN)-Advanced SCATterometer (ASCAT) V1.5, and Multi-Source Weighted-Ensemble Precipitation (MSWEP) V2.2 for a subbasin of the Mekong River Basin (MRB). The study area of the Srepok River basin (SRB) represents the Central Highland sub-climatic zone in Vietnam under the impacts of newly built reservoirs during 2001–2018. In this study, our evaluation was performed using the Rainfall Assessment Framework (RAF) with two separated parts: (1) an intercomparison of rainfall characteristics between rain gauges and SPPs; and (2) a hydrological comparison of simulated streamflow driven by SPPs and rain gauges. Several key findings are: (1) IMERGF-V6 shows the highest performance compared to other SPP products, followed by SM2RAIN-ASCAT V1.5 and MSWEP V2.2 over assessments in the RAF framework; (2) MSWEP V2.2 shows discrepancies during the dry and wet seasons, exhibiting very low correlation compared to rain gauges when the precipitation intensity is greater than 15 mm/day; (3) SM2RAIN–ASCAT V1.5 is ranked as the second best SPP, after IMERGF-V6, and shows good streamflow simulation, but overestimates the wet seasonal rainfall and underestimates the dry seasonal rainfall, especially when the precipitation intensity is greater than 20 mm/day, suggesting the need for a recalibration and validation of its algorithm; (4) SM2RAIN-ASCAT had the lowest bias score during the dry season, indicating the product’s usefulness for trend analysis and drought detection; and (5) RAF shows good performance to evaluate the performance of SPPs under the impacts of reservoirs, indicating a good framework for use in other similar studies. The results of this study are the first to reveal the performance of MSWEP V2.2 and SM2RAIN-ASCAT V1.5. Additionally, this study proposes a new rainfall assessment framework for a Vietnam basin which could support future studies when selecting suitable products for input into hydrological model simulations in similar regions.

1. Introduction

Precipitation is one of the most critical factors when studying hydrological characteristics of river basins [1,2], but it remains a significant challenge for hydrologists to estimate [3,4]. The accuracy of precipitation estimations is crucial to improving hydrological modeling accuracy [5,6,7,8,9] and to improving the understanding of water resource management at both regional and global scales [10]. Traditionally, using conventional rain gauges provides an accurate, reliable, and simple approach to measuring rainfall data on ground-based points [11]. This does not require complicated pre- and postprocessing tasks before use [12,13]. However, it may be limited by the uneven and sparse distribution of rain gauge stations [14] or unavailable in some regions (i.e., coastal, high–land, and arid regions) [8]. Thus, it is more challenging to perform hydrological modeling in these regions. For transboundary rivers (i.e., Mekong River Basin—MRB), [15] confirmed the point mentioned by [16], in which the exchange of observation datasets (i.e., ground-based rain gauges) are becoming increasingly complicated due to the administrative, geographical, and technical reasons among different countries.

Many studies have indicated an alternative for filling this challenging constraint by using gridded precipitation products which combine information from both satellites and gauges. In recent years, they have been increasingly used in many hydrological models [2,6,17]. Many studies indicated the advantages of using them include finer spatiotemporal resolutions and wider spatial coverage compared to ground-based products [13,18,19,20,21], as well as the ability to provide real-time rainfall estimates which are useful for flash flood forecasting [2], or using readily available satellite observations to delineate flooded areas [22]. In this study, three Satellite-based Precipitation Products (SPPs) are used based on our review of the literature: the Integrated Multi-satellitE Retrievals for GPM (IMERG) [23], the Multi-Source Weighted-Ensemble Precipitation (MSWEP) [24], and the Soil Moisture to Rain (SM2RAIN)–Advanced SCATterometer (ASCAT) [25].

The Multi-Source Weighted-Ensemble Precipitation (MSWEP) is a fully global, historic SPP product. It has been used in numerous research works at local and regional scales [8,26,27,28]. This product has high accuracy with estimates of long-term precipitation means, primarily due to use of high-resolution gauge-based climatic datasets, optimal merging of multiple satellites, reanalysis of P datasets, daily gauge corrections, and accounting for gauge reporting times [1,28]. However, previous studies indicated that it generally overestimates observed rainfall (dry and wet seasons) for semi-arid regions [27], which may be due to its bias reduction algorithm and lack of uniformity in the spatial distribution of gauge data. This finding was confirmed by [26], in which the MSWEP underestimated rainfall patterns during the dry season in the Qinghai–Tibet Plateau (humid region). Additionally, the MSWEP dataset received limited attention for case studies in Southeast Asia regarding hydrological applications and has not been studied in Vietnam. Thus, we aim to conduct a comprehensive analysis and assessment of the application of this dataset as a first attempt to reveal its capability and robustness in the Central Highland region of Vietnam.

SM2RAIN is a new rainfall estimation dataset known as the Soil Moisture to Rain (SM2RAIN)–Advanced SCATterometer (ASCAT). Ref. [25] first released it in 2007 using the soil moisture (SM) dataset and the soil–water balance equation (bottom-up method) to directly measure rainfall. The state of SM and its temporal changes, employing the SM2RAIN algorithm, was applied to calculate rainfall variation [29]. The bottom-up technique has been indicated as a better approach than the top-down technique because of the limited number of satellite sensors and observations required for the accumulated precipitation estimates [30,31]. The SM2RAIN approach was applied to three surface soil moisture products collected from the Advanced SCATterometer (ASCAT), the Advanced Microwave Scanning Radiometer for Earth Observing System (AMSR–E), and the Soil Moisture and Ocean Salinity instruments (SMOS) [32]. In Ref [25] found that SM2RAIN–ASCAT outperforms the ground-based Global Precipitation Climatology Center (GPCC) product in central Asia. SM2RAIN limitations indicated in previous studies include its underestimation of peak rainfall during extreme events [25,31] and the presence of erroneous precipitation due to frequent SM changes, as initially found by [33,34]. However, this product’s performance has not been quantified in Vietnam, and has raised uncertainty about its utility for applications in this region. Thus, this newly released product was chosen for evaluation in this study. Our key finding regarding this dataset would be helpful to support disaster management, especially during drought-related events that have become more frequent in recent years [35,36] in the Srepok River Basin, Central Highland Vietnam, and other regions of Vietnam [37].

The Srepok River Basin (SRB) is a major tributary of the MRB. The SRB is a central region of Vietnam, with agricultural land comprising most of the basin (Figure 1e). Variations in the water supply of the Srepok river can significantly affect the water resources management of the MRB middle and lower regions. In Refs [38,39] identified a need to investigate the operations of the 51 dams and reservoirs that were built in the 1990s in the lower MRB [36]. Since 2015, additional hydropower dams have been constructed in this region, bringing the total number of dams to 59 (Figure 1b). Their designed capacity exceeds 10,000 MW, which accounts for more than 35% of the hydropower capacity of the lower MRB (between 53,000 and 59,000 megawatts) [40]. The increase of newly built reservoirs with high capacities encouraged scientists to conduct further studies to better understand the hydrological flow regime changes in this region. However, as a transboundary river, it is difficult to estimate the hydrological characteristics of the MRB and the SRB due to the lack of important hydrological inputs (i.e., in situ precipitation data). Most of the available streamflow models for this region were performed using the outdated inputs, such as the 2012 Mekong Flows model (http://www.mekongflows.com) with the baseline period (1986–2006), or other models (1980s–2006) [41,42,43,44]. Thus, this study intends to fill previous gaps by conducting a comprehensive assessment of the chosen SPP products to improve the hydrological simulations in this region.

Understanding the lack of a reference framework to comprehensively evaluate the performance of SPPs under the impacts of reservoirs, this study proposes a Rainfall Assessment Framework (RAF) which is used to assess three gauge-corrected satellite-based rainfall products (GPM IMERGF-V6, MSWEP V2.2, and SM2RAIN–ASCAT V1.5) (

Table 1. Description of the SPPs chosen for this study.

Product	Temporal Coverage	Resolution	Spatial Coverage	Temporal Resolution	Latency	References
IMERGF–V6	2000–2021	0.1°	65°N–65°S	½ h	Months	Hou et al. (2014)
MSWEP V2.2	1979–present	0.1°	60°N–60°S	3 h	Days	Beck et al. (2019)
SM2RAIN–ASCAT V1.5	2007–present	0.1°	60°N–60°S (land)	Daily	N/A	Brocca et al. (2019)

) for the SRB, Central Highland Vietnam. This framework is intended to perform: (1) an intercomparison of rainfall characteristics between rain gauges and SPPs; and (2) a hydrological comparison of simulated streamflow driven by SPPs and rain gauges. The semi-distributed hydrological SWAT model has been chosen for this study because of its proven capability in previous case studies [13,45,46,47,48,49,50]. This study provides deeper insight into using the not-yet-evaluated MSWEP and SM2RAIN products and their performances in streamflow simulations under the operation of newly built reservoirs. Studies performing assimilation of satellite soil moisture into the SWAT model [51] and Land Information System (LIS) model [52] are still in their early stages.

2. Watershed

The Srepok River is an important Mekong River tributary. It originates from Dak Lak Province in the Vietnam Central Highlands, travels through Ratanakiri and Stung Treng areas, and joins the Mekong River. Its length varies from 406 to 450 km, with the first course within Vietnam ranging between 125 and 169 km and the last 281 km located in the Cambodian region (Figure 1a,b). The SRB covers an area in Vietnam of approximately 18,200 km² with more than 65% (12,000 km²) as the upper region of the SRB [36]. The upper SRB has been chosen in this study, which represents a mountainous region with a complex terrain profile and elevation ranging from 200 to 2240 m. The mean altitude is approximately 350 m in the northwest and 1000 m in the southeast [35,36] (Figure 1c,d). The average annual precipitation from 1998 to 2018 was 1920 mm at Giang Son station, 1937 mm at Duc Xuyen station, 1704 mm at Cau 14 station, and 1600 mm at Ban Don station (SRB outlet). 70% or more of the annual precipitation occurs during the wet season (May to October) and 41% of it becomes runoff [36,53]. Additionally, SBR has experienced a 10% increase in emigration and a 5% increase in local population growth. These have resulted in an uncontrolled expansion of cultivation [54]. Thus, water consumption for agriculture has increased significantly, accounting for an increasing proportion of the total annual water demand.

The major land use type (61.32% of the basin) is forest, followed by agricultural land (32.87%), urban area (4.15%), desert (0.59%), and water bodies (1.07%) (Figure 1e). The region’s agri-based product is rice. The majority of the SRB is forest which is located in the mountainous areas surrounding the SRB. The primary soil type is Ferric Acrisols (Af14–3C–1; 37.12%), followed by Dystric Cambisols (Bd30–2–3c–9; 27.68%), Orthic Acrisols (Ao39–2b–4; 14.64%), Eutric Cambisols (Be8–3c–24; 8.16%), Chromic Cambisols (Bc9–2b–8; 5.11%), Chromic Cambisols (Ao63–3b–6; 3.41%), Ferric Acrisols (Af32–2ab–3; 2.19%), and Dystric Cambisols (Bd31–2c–11; 1.69%). Land use for the lower Mekong River Basin has been created to aid in water management [55].

3. Data and Methods

In this study, the Rainfall Assessment Framework (RAF) has been constructed following two main steps: (1) an intercomparison of products’ characteristics between the chosen SPPs and rain gauges; and (2) a hydrological comparison using the simulated streamflow under the impacts of newly built reservoirs, focusing on the total runoff and flood peak over the chosen period (2001–2018), driven by SPPs and rain gauges. The details of RAF and chosen SPPs are given below.

3.1. The Rainfall Assessment Framework (RAF) for SPPs

In this study, the Rainfall Assessment Framework (RAF) was established and used to assess the performance of SPPs based on comparisons between SPPs and rain gauges in two separated parts: (1) an intercomparison between SPP and rain gauges; and (2) hydrological simulations using the SWAT model for streamflow comparisons. The order of the performing stages is shown in Figure 2. The first stage is the intercomparison where we compare the performance of SPPs and rain gauges in terms of (i) rainfall distribution (i.e., annual and seasonal rainfalls), (ii) detection metrics (i.e., Probability of Detection (POD), Critical Success Index (CSI), and False Alarm Ratio (FAR)), (iii) temporal dynamics metrics (i.e., Correlation Coefficient (CC), Relative Bias (RB), and Root Mean Square Error (RMSE)), and (iv) rainy days (i.e., annual and seasonal rainfalls). The second stage is the comparison of hydrological simulations using the SWAT model, in which we compare the performance of SPPs and rain gauges in terms of (i) statistical metrics (i.e., Nash–Sutcliffe Efficiency (NSE), the Percentage Bias (PBIAS), the Root Mean Square Error (RMSE), and the Coefficient of Determination (R²)), (ii) streamflow simulation, and (iii) flood peaks in extreme events.

3.2. In Situ Data

The daily 2001–2018 runoff data and other hydrological inputs (daily precipitation and temperature) required for the SWAT model were collected at Giang Son station, and hydrometeorological data for Giang Son, Buon Me Thuot, Buon Ho, M’Drak, Dak Lak, Krong Buk, Duc Xuyen, Dak Nong, Cau 14, Ban Don, and Ea So (Figure 1a) were collected from the Vietnam National Center for Hydrometeorological Forecasting (https://kttv.gov.vn/, (accessed on 23 January 2022)). These collected data were preprocessed at the regional stations before use in this study.

3.3. DEM, Land-Use, and Soil

The Digital Elevation Model (DEM) was extracted from the HydroSHEDS v1.0 database (https://www.hydrosheds.org/; accessed on 3 January 2022) with a resolution of 90 × 90 m (Figure 1c). This product has been used widely in previous studies [56,57,58,59,60], including custom modifications [61].

Land-use data were obtained from the Mekong River Council field survey in the lower Mekong Basin data collection project (https://portal.mrcmekong.org/land–cover; accessed on 8 January 2022) in 2010 and can be found in the MRC land-use catalog of the lower MRB. In this study, we reclassified the original land-use types into five major types: forest, agricultural land, urban area, barren land, and water bodies (Figure 1e).

The soil data was obtained from the FAO Harmonized World Soil Database at a resolution of 30 × 30 m (HWSD v1.2; https://www.fao.org/; accessed on 8 January 2022) (Figure 1f). The documentation for the Harmonized World Soil Database is by [62,63,64] and the documentation for the Global Agro–Ecological Zone V4 Model is by [65]. Figure 1f shows the eight major soil types available in the SWAT soil library and that can be used in the hydrological SWAT model.

3.4. Satellite-Based Precipitation Products

3.4.1. GPM IMERG Final Run V06

The Integrated Multi-satellite Retrievals for GPM (IMERG) is a level 3 rainfall dataset that employs the Global Precipitation Measurement (GPM) estimation algorithm and aims to intercalibrate, integrate, and interpolate various satellite microwave precipitation estimations [7,23]. This mission combined precipitation gauge analyses, microwave-calibrated infrared satellite estimates, and other precipitation estimators with a spatial resolution of 0.1°. The IMERG dataset has a wider coverage (65°S–65°N) and a shorter time interval (30 min) than TMPA products [23]. The IMERG Final run V6.0 (henceforth IMERGF-V6) was used instead of the IMERG Early or Late version. This half-hour 0.1° gridded product can be downloaded from the NASA Goddard Space Flight Center website (https://pmm.nasa.gov/data–access/downloads/gpm, (accessed on 7 January 2022)).

3.4.2. MSWEP V2.2

The Multi-Source Weighted-Ensemble Precipitation (MSWEP) is a newly released, global historic, 3-hourly temporal, and 0.1° spatial resolution precipitation dataset [24,66,67]. The MSWEP product has generated accurate global precipitation estimates by combining the capabilities of gauge, satellite, and reanalysis-based datasets. The long-term mean of MSWEP (V1.0) was derived from the CHPclim dataset but replaced with more accurate regional datasets where available [24,66]. Since MSWEP (V2.0), a remarkable change has been made which includes an improvement in spatial resolution from 0.25° to 0.1°. The precipitation frequency corrections and cumulative distribution methods were applied, and satellite thermal infrared imaging precipitation estimates were used. In this study, the daily 0.1° gridded MSWEP V2.2 dataset was obtained via the GloH2O website (http://www.gloh2o.org/mswep/, (accessed on 5 January 2022)).

3.4.3. SM2RAIN–ASCAT V1.5

The Soil Moisture to Rain (SM2RAIN) from the Advanced SCATterometer (ASCAT)is a new global-scale rainfall, satellite soil moisture-based product collected by the European Meteorological Satellite (EUMETSAT) Organization operational satellite MetOp. It uses the SM2RAIN algorithm as the bottom-up approach to calculate rainfall variation based on SM data [25,29,31]. It was derived from soil moisture measurements (ASCAT) from Metop-A and Metop-B satellites, transformed into SM using the soil WAter Retrieval Package (WARP v.5) algorithm [68]. In this study, the daily 0.1° gridded SM2RAIN–ASCAT V1.5 dataset (2007–2018) was downloaded via the website (https://zenodo.org/record/6136294, (accessed on 3 January 2022)).

3.5. SWAT Model

In this study, the semi-distributed hydrological (SWAT) model was used. It was developed and supported by the Agriculture Research Service (ARS) and the U.S. Department of Agriculture (USDA) [69]. In recent years, this well-known watershed-scale model has been used more frequently in the United States, Europe, and other regions [70] due to its high capability to solve hydrological-related issues. Many studies have used SWAT to reveal the impacts of land-use and land cover (LULC), climate change [71], robustness of SPPs [13], and pesticide and agricultural chemical pollutions on streamflow or sediment loads [5,13,48,70,72,73,74,75,76]. The first version of this model was developed for ArcGIS, as explained by [69,76]. A new version of this model (QSWAT) has been released in recent years due to the wide application of QGIS (an open–source GIS interface) (https://www.qgis.org/, (accessed on 23 July 2021)) in many fields [72,77,78,79,80,81,82].

In this study, the model was set up with reservoirs (see Appendix A), on a daily timescale with two years of warm-up period (2001–2002) over the chosen simulation period of 18 years (2001–2018). For IMERGF-V6 and MSWEP, nine years were chosen (2003–2011) for calibration. The validation period was chosen between 2012 and 2018. In addition, due to the difference in the dataset availability, SM2RAIN–ASCAT V1.5 (2007–2018) has been established with different periods: warm-up period (2007–2008), calibration period (2009–2013), and validation period (2014–2018).

In this study, twenty-three parameters with their names, methods, descriptions, ranges, and fitted values were chosen for calibration and validation (

Table 2. SWAT and reservoir parameters for calibration in this study.

ID	Parameter	Method	Description	Range
1	CN2.mgt	Relative	SCS runoff curve number f	−0.25–0.25
2	ALPHA_BF.gw	Replace	Baseflow alpha factor for bank storage	0–1
3	GW_DELAY.gw	Replace	Groundwater delay (days)	0–500
4	GWQMN.gw	Replace	Threshold depth of water in the shallow aquifer required for return flow to occur (mm)	0–5000
5	GWHT.gw	Replace	Initial groundwater height (m)	0–1
6	RCHRG_DP.gw	Replace	Deep aquifer percolation fraction	0–1
7	GW_REVAP.gw	Replace	Groundwater “revap” coefficient	0.02–0.2
8	REVAPMN.gw	Replace	Percolation to the deep aquifer to occur (mm. H2O)	0–1000
9	ESCO.hru	Replace	Soil evaporation compensation factor	0–1
10	EPCO.hru	Replace	Plant uptake compensation factor	0–1
11	HRU_SLP.hru	Relative	Average slope steepness	−0.25–0.25
12	SLSUBBSN.hru	Relative	Average slope length	−0.25–0.25
13	CANMX.hru	Replace	Maximum canopy storage	0–25
14	OV_N.hru	Replace	Manning’s “n” value for overland flow	0.01–1
15	LAT_TTIME.hru	Replace	Lateral flow travel time	0–180
16	CH_N2.rte	Replace	Manning’s “n” value for the main channel	0–0.3
17	CH_K2.rte	Replace	Effective hydraulic conductivity in main channel alluvium	0–500
18	ALPHA_BNK.rte	Replace	Baseflow alpha factor for bank storage	0–1
19	SOL_AWC(.).sol	Relative	Available water capacity of the soil layer	−0.25–0.25
20	SOL_K(.).sol	Relative	Saturated hydraulic conductivity	−0.25–0.25
21	SOL_BD(.).sol	Relative	Moist bulk density	−0.25–0.25
22	SOL_Z(.).sol	Relative	Depth from soil surface to bottom of layer	−0.5–0.5
23	SURLAG.bsn	Replace	Surface runoff lag time	0–25
1	NDTARGR	Replace	Number of days the reservoir would be filled (days)	1–365
2	RES_K	Replace	Hydraulic conductivity of the reservoir bottom (m/s)	10−2–11

) based on findings of [69,75,83]. Giang Son station was chosen for the calibration and validation, and the observed streamflow (2001–2018) was extracted and used throughout this study. This data has undergone the preprocessing stage to ensure the dataset gap is below 3.0%. This study used the Sequential Uncertainty Fitting method version 2 (SUFI–2) in the SWAT-CUP (V5.2.1) program for calibration and validation [84,85]. The calibrating objective function is NSE (Nash–Sutcliffe Efficiency), used in SWAT-CUP program during the newly built reservoir operation period (2009–2018).

3.6. Performance Metrics

Table 3. Performance metrics for evaluating precipitation and hydrological models.

	Metric	Equation	Optimal Value	Evaluation Range
Product Performance Metrics	POD	$\frac{N_{11}}{N_{11}+N_{01}}$	1
	CSI	$\frac{N_{11}}{N_{11}+N_{01}+N_{10}}$	1
	FAR	$\frac{N_{10}}{N_{11}+N_{10}}$	0
	CC	$\frac{{{\displaystyle \sum }}_{i=1}^N\left(R{G}_i-\overline{RG}\right)(SP{E}_i-\overline{SPP})}{\sqrt{{{\displaystyle \sum }}_{i=1}^N{\left(R{G}_i-\overline{RG}\right)}^2{{\displaystyle \sum }}_{i=1}^N{\left(SP{P}_i-\overline{SPP}\right)}^2}}$	1
	RB	$\frac{mean\left(SPP\right)}{mean\left(RG\right)}-1$	0
	RMSE	$\sqrt{\frac{1}{n}{\displaystyle {\displaystyle \sum }_{i=1}^N}{\left(SP{P}_i-R{G}_i\right)}^2}$	0
Streamflow Performance Metrics	NSE	$1-\frac{{{\displaystyle \sum }}_{i=1}^N\left(ob{s}_i-si{m}_i\right)}{{{\displaystyle \sum }}_{i=1}^N{\left(ob{s}_i-\overline{obs}\right)}^2}$	1	$\mathrm{VG}: 1 \ge \mathrm{NSE} \ge 0.8, \mathrm{G}: 0.80 \ge \mathrm{NSE} \ge 0.70, \mathrm{S}: 0.70 \ge \mathrm{NSE} \ge 0.50, \mathrm{NS}: \mathrm{NSE}<0.50$
	PBIAS	$1-\frac{{{\displaystyle \sum }}_{i=1}^N\left(ob{s}_i-si{m}_i\right)}{{{\displaystyle \sum }}_{i=1}^{N}ob{s}_i}$	0	$\mathrm{VG}: \mathrm{PBIAS} \ge \pm 5, \mathrm{G}: 5\le \mathrm{PBIAS} \le \pm 10, \pm 10 \le \mathrm{PBIAS} \le \pm 15, \mathrm{NS}: \mathrm{PBIAS}>\pm 15$
	R2	$1-\frac{{{\displaystyle \sum }}_{i=1}^N{\left(ob{s}_i-si{m}_i\right)}^2}{{{\displaystyle \sum }}_{i=1}^N{\left(ob{s}_i-\overline{obs}\right)}^2}$	1	$\mathrm{VG}$$$$0.8, \mathrm{G}: 0.8 \ge$$$$0.7, \mathrm{S}: 0.70 \ge$$$$$ 0.5

Note:$N_{11}$ is the precipitation measured simultaneously by the rain gauge and the satellite. $N_{10}$ is the precipitation observed by the satellite but not by the rain gauge. $N_{01}$ is contrary to $N_{10}$. ${\mathrm{sim}}_{\mathrm{i}}$ represents the simulated while ${\mathrm{obs}}_{\mathrm{i}}$ represents the observed streamflow (m³/s). $\overline{\mathrm{sim}}$ is the average simulated and $\overline{\mathrm{obs}}$ is the average observed streamflow (m³/s). ${\sigma }_{OBS}$ is the standard deviation in the observations and ${\sigma }_{SIM}$ is the standard deviation in the simulations. Very Good (VG), Good (G), Satisfactory (S), and Not Satisfactory (NS).

represents the formulas and the perfect scores for performance metrics used throughout this study. In the first stage of the RAF framework, the three following decision metrics were used: (1) POD, (2) CSI, and (3) FAR. For the temporal dynamics evaluation of the SPP products, the three following metrics were used: (1) CC, (2) RB, and (3) RMSE.

For the comparison of streamflow performance in the second stage of the RAF framework, (1) NSE, (2) PBIAS, (4) RMSE, and R² [86,87] have been used.

POD is the number of actual precipitation events divided by the total number of precipitation events found by SPP products. FAR shows how many false precipitation events SPP products detected out of all the precipitation events they found. The most accurate and well-balanced detection metric is CSI, which is based on POD and FAR. In this study, 0.6 mm/day⁻¹ was chosen as the rainy-day threshold. The RB and RMSE show the bias and error of the satellite estimates, respectively. CC is a measure of the similarity using a linear relationship but may not capture it if the relationship is nonlinear. NSE shows how close the simulation and observation are. PBIAS looks at how often the simulated streamflow is more or less than the observed streamflow on average.

4. Results

4.1. Comparison of Characteristics between Rain Gauges and SPP Products

In the first stage of the RAF framework, we evaluate the SPPs’ characteristics, in which the three SPP products (IMERGF-V6, MSWEP, and SM2RAIN–ASCAT) were compared with precipitation data from rain gauges over the SRB (Figure 1). The precipitation values generated from SPP grids with the locations of rain gauges were compared directly. If there were many rain gauges in the same grid, we averaged their values before comparing them.

4.1.1. Evaluation of Annual and Seasonal Rainfall Distributions

Figure 3 shows the spatial distributions of rainfall estimates by SPPs on annual and seasonal scales. The wet season of the SRB has been defined from May to October, while the dry season is from November to April [36,53]. In general, while IMERGF showed a lower rainfall estimate over the SRB, especially in the high-land region and the SRB’s outlet (Figure 1c,d), SM2RAIN–ASCAT and MSWEP showed higher rainfall estimates that were detected in these regions. For the comparison of seasonal rainfall, SM2RAIN–ASCAT and MSWEP showed a large difference compared to IMERGF in the SRB central region, in which IMERGF detected less rainfall. MSWEP captured a higher amount of rainfall (annual and wet season) in the low-land region (Southwest SRB) (Figure 3) and was similar to IMERGF, whereas SM2RAIN–ASCAT showed that the SRB was hotter in this region but had more rainfall in the high-land region (Southeast SRB). During the dry season, these three SPPs showed a similar spatial distribution of rainfall estimates in which a higher amount of rainfall was detected in the high-land region (Southeast SRB), and it became drier in the SRB outlet (Figure 3). The higher difference of spatial distribution of rainfall shown by SM2RAIN–ASCAT showed its sensitivity to detect rainfall events, indicating a high probability of having higher total amounts of rainfall in comparison with other SPPs, mentioned in the following assessments.

Figure 4 shows the annual, dry, and wet rainfall seasons of the SRB. In general, while IMERGF-V6 overestimated the rainfall compared to rain gauges, MSWEP and SM2RAIN–ASCAT underestimated the wet season rainfall but overestimated rainfall during the dry season. IMERGF-V6 showed the smallest differences in the mean rainfall estimates (5.5% for annual and 0.2% for wet season) compared to rain gauges over the SRB, indicating a good capability to estimate the annual and wet season rainfall. MSWEP showed good rainfall detection with the smallest difference (5.5% difference) compared to rain gauges during the dry season. However, IMERGF-V6 showed a better performance in capturing the annual rainfall (i.e., 2004, 2006, 2007, 2009, 2013, etc.) (Figure 4a) and the seasonal rainfalls (i.e., 2004, 2007, 2011, 2016, etc.) compared to rain gauges and other SPPs (Figure 4b,c). SM2RAIN–ASCAT could be recognized as having underestimated overall rainfall during the wet season, with the highest difference of nearly 500 mm (i.e., 2011) compared to rain gauges (Figure 4b). Remarkable underestimations of rainfall during the wet season in the high-land region (Southeast SRB) (Figure 3) were found in MSWEP and SM2RAIN–ASCAT, reflecting their algorithmic challenges in estimating rainfall in mountainous regions. It implies that these products show a significant underestimation of terrain precipitation, similar to previous findings by [26,33,88].

Figure 5a–c show the daily precipitation scatter density plots and each statistical metric of the chosen SPPs with rain gauges. The color bar in each scatter plot represents the frequency of occurrence, which was calculated based on the corresponding grids between rain gauges and SPPs in the multi-year average precipitation map. The seasonal period was divided into wet and dry seasons [53]. The horizontal axis represents the paired value of gauge-estimated and SPP-estimated precipitation while the vertical axis shows the rainfall estimates from SPPs.

IMERGF-V6 showed the best correlation compared to rain gauges, followed by SM2RAIN–ASCAT and MSWEP. The density-colored scatter plots of IMERGF-V6 showed long tails compared to other SPPs (Figure 5). However, some points deviate significantly from the 1:1 line when the precipitation intensity is greater than 20 mm/day, indicating that the precipitation is underestimated over different seasons. The complex terrain of the SRB may be one of the important factors contributing to this phenomenon.

To reveal the relationship between rain gauges and SPPs, the RMSE, MAE, CC, and RB were calculated and shown in Figure 5. The estimated precipitation of SPPs and the precipitation measured by rain gauges vary for each product. IMERGF-V6 showed the highest correlation with the CC values of about 0.65 for the annual rainfall, while it is lowest during the wet and dry seasons (median CC of 0.53). SM2RAIN–ASCAT showed similar results, in which it underestimated the rainfall when the precipitation intensity is greater than 20 mm/day over seasons, with a lower correlation found during the wet season that could be explained due to its problem estimating rainfall under the saturated soil conditions [29] (Figure 5f). In summary, IMERGF-V6 performs the best, with the highest CC score (median CC of 0.55), lowest MAE score (median MAE of 3.03 mm/day), followed by SM2RAIN–ASCAT (median CC of 0.40 and median MAE of 3.11 mm/day). SM2RAIN–ASCAT performed better during the dry season and worsened in the wet season. It implies that SM2RAIN–ASCAT remarkably overestimated the wet season rainfall (RB value of 234.4%) and underestimated the dry season rainfall (RB value of –11.2%), especially when the rainfall intensity is greater than 20 mm/day (Figure 5c,f). Additionally, SM2RAIN–ASCAT was found with the lowest RMSE score (median RMSE of 5.7.00 mm/day) compared to IMERGF-V6 (median RMSE of 6.27 mm/day), which could be explained due to the point-by-point calibration of SM2RAIN parameter values with the objective function root mean square error (RMSE) [25]. On the other hand, among the three SPPs, MSWEP performs the worst, with the highest RMSE score (median RMSE of 8.89 mm/day), lowest CC score (median CC of 0.21), and highest MAE score (median MAE of 4.01 mm/day). Discrepancies were found with the observation of MSWEP during the dry and wet seasons, in which MSWEP exhibited very low correlation compared to rain gauges when the precipitation intensity is greater than 15 mm/day (Figure 5b,e). This finding is similar to the previous work of [88] but contrasts with [89] for MSWEP’s performance in mountainous regions with (p > 10 mm/day). We also found that MSWEP overestimated the wet season rainfall (RB of 195.3%) and underestimated the dry season rainfall (RB of –103.8%).

All SPPs could generally reflect rainfall’s seasonal distribution over the SRB. IMERGF-V6 performed well in estimating the seasonal and annual rainfalls, while SM2RAIN–ASCAT performed better during the dry season but worse in the wet season. MSWEP showed the worst performance compared to other SPPs, especially during the wet season.

4.1.2. Detection Metrics Assessment

In the second stage of the RAF framework, the chosen SPPs were evaluated using detection metrics, as shown in Figure 6 and Figure 7, and

Table 4. The median performance metrics of three SPPs were based on daily rain gauge measurements (2001–2018). Bold values indicated the best score. Larger numbers show better performance except for RMSE and FAR.

Metric		IMERGF	MSWEP	SM2RAIN–ASCAT
POD	Dry	0.738	0.745	0.899
	Wet	0.939	0.927	0.999
	All	0.887	0.879	0.972
FAR	Dry	0.322	0.362	0.538
	Wet	0.042	0.059	0.087
	All	0.120	0.148	0.264
CSI	Dry	0.546	0.524	0.439
	Wet	0.902	0.876	0.912
	All	0.791	0.763	0.721
CC	Dry	0.794	0.566	0.624
	Wet	0.705	0.363	0.608
	All	0.745	0.458	0.697
RB	Dry	0.472	0.651	0.130
	Wet	−0.063	−0.229	−0.265
	All	0.026	−0.083	−0.195
RMSE	Dry	3.964	7.347	5.321
	Wet	6.557	10.029	6.943
	All	5.665	8.801	5.976

.

IMERGF-V6 achieved the highest performance and outperformed other SPPs for seasonal and annual comparisons (POD of 0.887–rank 2; CSI of 0.791–rank 1; FAR of 0.120–rank 1) (Figure 6 and Figure 7, and Table 4), consistent with the key findings of [13,90], reflecting the quality of the IMERG new version (V6) retrieval algorithms for research products [91,92]. It implies IMERGF-V6 showed an overall good ability to capture rainfall events, with a high POD score and the lowest FAR score that would be useful to detect rainfall events.

MSWEP was ranked as the second best (POD of 0.879–rank 3; CSI of 0.763–rank 2; FAR of 0.148–rank 2), followed by SM2RAIN–ASCAT with the highest POD score (POD of 0.972), lowest FAR score (FAR of 0.264), and lowest CSI score (CSI of 0.721), indicating a discrepancy in the precipitation retrieval algorithm, showing a higher sensitivity with the detection of rainfall events compared to IMERGF-V6 and MSWEP. We found that for heavy precipitation (p > 10 mm/day), IMERGF-V6 exhibited a good performance over the year and during the wet and dry seasons with good skill scores (POD, FAR, and CSI). When the rainfall increases, MSWEP rapidly showed a considerable decrease in performance and was ranked as the worst after SM2RAIN–ASCAT, which contrasts with the findings of [89]. Apart from this, all SPP products showed a higher ability to detect rainfall events during the wet season compared to the dry season, consistent with the key findings of [13,19,20,90,93]. In summary, the IMERGF-V6 product outperformed the other SPP datasets in these comparisons, followed by SM2RAIN–ASCAT, and the poorest performance was MSWEP.

4.1.3. Temporal Dynamic Metrics Assessment

In this section, we evaluate the chosen SPPs using the temporal dynamics metrics, mentioned as one of the assessments within the performance of the RAF framework (Figure 2). Based on the results of Figure 4 and Figure 5, we found that SPPs remarkably underestimated the terrain precipitation over the SRB, especially SM2RAIN–ASCAT and MSWEP. IMERGF-V6 was ranked as the best product and outperformed other SPPs with the highest correlation (CC of 0.745), followed by SM2RAIN–ASCAT (CC of 0.697), and MSWEP (CC of 0.458), similar to the findings of [13,66]. Additionally, IMERGF-V6 remained the best in seasonal assessment with the highest correlation during the wet and dry seasons, while MSWEP showed the poorest performance, which greatly underestimated the rainfall (RB of 0.229) during the wet season (Table 4). Similarly, SM2RAIN–ASCAT showed a higher underestimation of rainfall estimates (RB of 0.265) compared to MSWEP during the wet season, similar to the finding of [89]. This could be explained due to the SM2RAIN rainfall retrieval algorithm that uses the SM dataset to detect rainfall events [29]. The SRB’s Central Highlands climate, in which winter is highly dry while a high amount of rainfall characterizes summer, could be a problem affecting the rainfall estimate. It implies this product algorithm would have a problem measuring or detecting rainfall under the impacts of very dry or saturated soil conditions, as indicated previously by [25]. Another reason is the limitation of the bottom-up approach of this product. The performance of this approach was described as highly dependent on the land characteristics, and it could only estimate terrestrial rainfall [31]. Thus, the SRB, with a high variation of rainfall patterns over the year [53], dense vegetation coverage (i.e., agricultural region), and tropical forests with complex topography [36], could remarkably affect the rainfall estimates of this product. However, SM2RAIN–ASCAT, with the lowest RB score (RB of 0.13), good RMSE score (RMSE of 5.321 mm/day), and high CC score (CC of 0.624), showed a good ability to detect drought events (Table 4), similar to the CHIRPS product [13].

For MSWEP, the underestimation of rainfall could be explained due to the low density of rain gauges in the SRB region, resulting in low gauge reporting times [66]. This means that for the same grid cell, the daily precipitation accumulations are computed by considering the inferred gauge reporting times which resulted in low-quality gauge-corrected MSWEP due to temporal mismatches between the uncorrected MSWEP and gauge data. This suggests a recalibration or evaluation of the algorithm to solve this problem for this region, instead of solely using the gauge data to calibrate the uncorrected MSWEP. In addition, there was no remarkable difference between the dry and wet seasons for the evaluation of streamflow using temporal dynamic metrics. During the dry season, the average RMSE for SPPs was 6.814 mm d⁻¹, which was lower than the rainy season (7.843 mm d⁻¹), indicating that rainfall variability is lower during the dry season. While IMERGF-V6 performed the best in this assessment, SM2RAIN–ASCAT was ranked second, followed by MSWEP.

4.1.4. Rainy-Days Detection

The detection of the number of rainy days is a crucial criterion to evaluate the performance of SPP products as well as to support drought-related studies (Figure 8). In this section, the assessment of SPPs’ performances in terms of rainy-day detection was performed in the fourth stage of RAF framework (Figure 2). The threshold of 0.6 mm/day for rainy days was chosen based on the suggestion of the Vietnam National Centre for Hydro–Meteorological Forecasting (https://www.nchmf.gov.vn/kttv/, (accessed on 5 January 2022)). The daily rainfall rate of the basin was calculated so it would be counted as a rainy day if the rainfall rate is higher than 0.6 mm/day.

In general, IMERGF-V6 showed the most similar result compared to rain gauges, followed by MSWEP and SM2RAIN–ASCAT (Figure 8). IMERGF-V6 showed an overestimation (2.9% higher) of the number of rainy days during the dry season and an underestimation (0.8% lower) during the wet season, consistent with results of [13], indicating that frequent temporal rainfall sampling (every half hour) of the IMERG product [23] could help to detect short-period rainfall and drought events. The underestimation of the number of rainy days during the wet season contributes to the findings for IMERGF-V6 using detection and temporal dynamic metrics (see Section 4.1.1, Section 4.1.2 and Section 4.1.3), in which there is a correlation between the lack of rainy days detected and the underestimation of rainfall, especially during the wet season (RB of −0.063; Table 4).

MSWEP showed similar results compared to IMERGF-V6 during the wet season but larger estimates than rain gauges (2.3% higher) for the entire period, which could be explained by the low density of rain gauges in the SRB region. Thus, it has a low number of gauge reporting times [66], resulting in fewer rainy days detected. This result suggested a recalibration or evaluation of the algorithm for this region in terms of rainy day detection, even if it has been measured more frequently (every 3 h) [24,66] compared to other SPPs (Table 1). The SM2RAIN–ASCAT was ranked as the worst product in this assessment, as it overestimated the number of rainy days in both wet (4.4% higher) and dry seasons (14.6% higher) compared to rain gauges (Figure 8). This reflects the impact of the SM2RAIN algorithm in detecting rainfall using SM data [29] and suggests a recheck for its algorithm in terms of seasonal rainfall detection is needed.

4.2. Streamflow Simulation Driven by SPPs and Rain Gauges

For the second part of the RAF framework, we compare the streamflow simulations driven by SPPs and rain gauges using statistical metrics (i.e., NSE, PBIAS, RMSE, and R²) (Figure 2; Table 3). There were many scenarios with unsatisfactory levels (|PBIAS| greater than 15%) shown in the validation period that could be explained due to the operation of newly built reservoirs since 2009 [36].

IMERGF-V6 showed a Good streamflow simulation after the rain-gauge-driven simulation, followed by SM2RAIN–ASCAT with good performances for daily and monthly simulated streamflow based on the chosen statistical metrics (Table 3).

Figure 9 represents the daily streamflow simulations performed by rain gauges and SPPs at the SRB (2001–2018). In general, the rain-gauge-driven simulation showed good daily NSE scores (NSE of 0.67 and 0.63) for the calibration and validation, respectively (Figure 9a). These figures from IMERGF-V6 were nearly the same during the model calibration, with an NSE score of 0.60, but worse during the validation period (NSE of 0.36) (Figure 9b), which could be explained by the operation of newly built reservoirs, generating a remarkable decrease in the total runoff and the change of the flow regime at the SRB outlet in the low-land region of the SRB (Figure 1a). Similarly, SM2RAIN–ASCAT showed correlated results with good performance of streamflow simulation during the calibration (NSE of 0.68) but worse during the validation period (NSE of 0.38) (Figure 9d;

Table 5. Summary of SPPs’ performances using statistical metrics.

Product	Calibration				Validation
	NSE	PBIAS	RMSE	R2	NSE	PBIAS	RMSE	R2
	Daily
Rain gauge	0.67	−12.43	62.99	0.68	0.63	−13.48	44.85	0.80
IMERGF	0.60	−13.57	69.39	0.62	0.36	−37.19	30.11	0.50
MSWEP	0.36	5.08	87.18	0.40	0.15	−22.39	34.74	0.20
SM2RAIN	0.68	−6.94	50.83	0.68	0.38	−36.50	63.86	0.49
	Monthly
Rain gauge	0.77	−12.60	43.76	0.79	0.72	−13.59	31.05	0.88
IMERGF	0.71	−13.83	48.71	0.75	0.52	−37.44	40.75	0.69
MSWEP	0.49	4.69	64.93	0.52	0.15	−22.79	53.99	0.22
SM2RAIN	0.76	−7.11	37.91	0.80	0.49	−36.88	44.27	0.64

). In this assessment, streamflow simulation driven by MSWEP was found with the lowest correlation (NSE of 0.49 and 0.15) compared to the rain gauge during the calibration and validation periods, respectively (Figure 9 and Figure 10). In general, while IMERGF-V6 was ranked the best to use after rain gauges, MSWEP was exhibited as the worst under the impacts of operating reservoirs.

For monthly simulation, MSWEP remained the worst performance, with both calibration and validation periods categorized as unsatisfactory levels using the chosen statistical metrics (Table 3). The rain-gauge-driven simulations showed the highest performance compared to other SPPs while IMERGF-V6 was ranked the second best, followed by SM2RAIN–ASCAT, and MSWEP (Figure 9, Figure 10 and Figure 11; Table 5).

In general, the streamflow simulations driven by rain gauges indicated a good capability for the ground-based dataset under the impacts of reservoirs, while there are significant discrepancies for streamflow simulations between the calibration and validation periods driven by SPPs. It implies that the performance of SPPs on streamflow simulation was affected considerably by the impacts of reservoirs. However, due to the difference in the coverage period of the chosen datasets (Table 1), IMERGF-V6 and SM2RAIN–ASCAT were further evaluated to identify the best-performing product in the following sections.

4.3. Other Hydrological Outputs at the SRB Outlet

In this section, the RAF framework was performed to reveal the hydrological outputs, in which we assessed the monthly, yearly, flood peak, and seasonal (dry and wet) runoff at the SRB outlet (Figure 12).

IMERGF-V6 showed the best performance compared with the gauge–based product (Figure 12). For monthly streamflow, IMERGF-V6 achieved the least difference compared to the rain gauge, followed by SM2RAIN–ASCAT and MSWEP. IMERGF-V6 and SM2RAIN–ASCAT showed good correlations compared to the monthly streamflow driven by rain gauges, while MSWEP showed the highest difference compared to the rain gauge. It implies that MSWEP showed the least runoff between May and September, and remarkably higher in November, compared to rain gauges (Figure 12a). For seasonal streamflow, IMERGF-V6 showed the least difference compared to gauge-based products, with 18.7% lower (wet season) and 30.6% higher (dry season), whereas other SPPs showed higher differences: SM2RAIN–ASCAT with 21.4% and 40.5%, followed by MSWEP (30.9% and 50.2%), for wet and dry seasons, respectively (Figure 12d). For annual streamflow, IMERGF-V6 showed good performance in simulating the SRB annual streamflow compared to the rain gauge, followed by SM2RAIN–ASCAT. The worst product was MSWEP, with a high difference in the annual runoff (i.e., 2010 with 42.36%) (Figure 12b). At this point, we could see that MSWEP likely gives incorrect flood peak estimates, especially during the wet season. It also overestimated the rainfall estimates over the year.

Figure 12c shows the flood peak simulation by SPPs and rain gauges. In contrast to the primary findings of [89,94], IMERGF-V6 performed the best, while MSWEP performed the worst, with overestimations of flood peaks (i.e., 2009, 2010, and 2012) and underestimations (i.e., 2013, 2016, and 2017). This suggests a consideration when using IMERGF-V6 and SM2RAIN–ASCAT to observe the flood peak in flood-forecasting studies.

In this comparison, MSWEP has the lowest PBIAS scores (Table 5), indicating its usefulness for trend analysis and drought detection. Based on the findings presented, we could conclude that MSWEP performed poorly in predicting flood peaks, particularly when considering the effects of newly built reservoirs. SM2RAIN–ASCAT outperformed MSWEP, but typically underestimated flood peaks (Figure 12c), which is likely due to its algorithms for rainfall retrieval (see Section 4.1.3), and it is comparable to its product family (SM2RAIN–CCI) for mountainous regions [30,94]. It suggests that this product, when combined with the rainfall retrieval algorithm that uses SM to estimate rainfall, is considered to have a disadvantage for simulating the flood peaks during extreme floods. In general, for streamflow simulation under the impacts of reservoirs or dams, IMERGF-V6 exhibited the best performances over daily and monthly scales, while SM2RAIN–ASCAT and MSWEP need to be improved, especially during extreme rainfall events.

4.4. Limitations and the Need for Further Studies

Our study’s limitation is that it evaluated the chosen SPPs for a basin in Central Highland Vietnam which is limited by basin samples. An increase in basin samples would give higher accuracy for our key findings regarding the MSWEP and SM2RAIN–ASCAT products. This gap could be a potential idea for future flooding and drought-related studies, in which we would conduct a comprehensive assessment of more SPPs over a higher basin sample representing different sub–climate zones of Vietnam. In addition, the chosen period (2001–2018) should be extended until the latest product release. However, our study was limited because of the challenges of obtaining the after-2018 observed streamflow and rain gauge dataset, and GPM IMERGF-V6 is presently available only until mid-2021.

For future works, we would also include the validation of hydrological variables (i.e., soil moisture) using the recent global 400 m and 1 km downscaled Soil Moisture Active Passive (SMAP) [95,96] with the method established in [97,98,99]. This dataset has been used to quantify hydrology in the lower Mekong River Basin [100] and droughts in Australia [101]. Validation of the hydrological model with soil moisture observations will help constrain models and obtain accurate streamflow forecasts and other fluxes.

5. Conclusions

This study established the Rainfall Assessment Framework (RAF) to assess the performance of three SPPs (IMERGF-V6, MSWEP, and SM2RAIN–ASCAT) for the SRB under the impacts of newly built reservoirs in the Central Highland climate of Vietnam [53]. This work consists of two main parts: (1) an intercomparison between SPP and rain gauges, and (2) hydrological simulations using the SWAT model for streamflow comparisons. The major findings are summarized:

• For the intercomparison, IMERGF-V6 showed the highest correlation with the CC values of about 0.65 for the annual rainfall, while it is lowest during the wet and dry seasons (median CC of 0.53). For the hydrological application, IMERGF-V6 achieved the highest accuracy for streamflow simulation (i.e., daily and monthly) among the chosen SPPs (IMERGF-V6, MSWEP, and SM2RAIN–ASCAT) compared to rain-gauge-driven simulations. IMERGF-V6 showed a good performance in estimating the annual and seasonal streamflow after rain-gauge-driven simulations, followed by SM2RAIN–ASCAT and MSWEP. Regarding flood peak prediction, IMERGF-V6 showed high similarities compared to rain gauges. Our study made the second attempt after [13] to evaluate the performance of GPM IMERG in the Central Highlands of Vietnam, suggesting that this product is well-suited for hydrological applications, especially for high-land regions of the SRB and is recommended for continued usage in the context of reservoirs and dams.
• Our study made the first attempt to assess the MSWEP product for the Vietnam basin, in which MSWEP showed the worst performance compared to other SPPs. Key findings indicated that this product over- and underestimated the flood peaks in many extreme rainfall events, especially under the impacts of reservoirs. MSWEP showed discrepancies during the dry and wet seasons, exhibiting very low correlation compared to rain gauges when the precipitation intensity is greater than 15 mm/day, which could be explained by the low density of rain gauges in the SRB region [1,28], resulting in low gauge reporting times that affect the quality of MSWEP in the SRB.
• This study is the first attempt to evaluate the SM2RAIN–ASCAT product for the Vietnam Basin. Results indicated that SM2RAIN–ASCAT was ranked as the second best product after the rain gauge and IMERGF-V6, exhibiting a potential development in the future for Vietnam high-land regions. However, it fails to estimate rainfall under highly saturated soil conditions, in which the rainfall retrieval algorithm can be affected by SM [29]. It implies SM2RAIN–ASCAT remarkably overestimated the wet season rainfall (RB value of 234.4%) and underestimated the dry season rainfall (RB value of −11.2%), especially when the rainfall intensity was greater than 20 mm/day. However, SM2RAIN–ASCAT had the lowest PBIAS score during the dry season, indicating the product’s usefulness for trend analysis and drought detection, similar to CHIRPS in previous works by [13,83,102]. This study suggests using these findings as the reference for further improvements of this product’s algorithm, which would also benefit the SM2RAIN–CCI product because of similar rainfall retrieval problems [30,94] for mountainous regions.
• The SPP products performed slightly better regarding rainfall detection metrics during the wet season than during the dry season, using detection and temporal dynamic metrics. In Central Highland Vietnam, the temporal dynamics of SPPs were not significantly different between the two seasons in this study.
• The Rainfall Assessment Framework (RAF) showed good performance throughout this study in evaluating the performance of SPPs comprehensively using different terms of assessment (i.e., the spatial distribution of rainfall, decision and temporal dynamic metrics, rainy days detection, flood peak prediction, etc.). Thus, this framework could be used as a good reference framework for other studies and could be further developed for different case studies based on their characteristics (i.e., climate, land-use, land cover, and topography).

This study continues previous works to evaluate the performance of chosen SPP products for rainfall estimates using the newly established framework (RAF). Our key findings would support this region in finding alternative rainfall data for hydrological simulations in the SRB and the MRB. This would also contribute to improved regional hydrological understanding, serve as a scientific basis for future plans, and support decision-makers in managing water resources for this region. Our findings re-evaluate the quality of the IMERGF-V6 product, and are the first attempt to reveal the performance of MSWEP and the new rainfall product derived from SM, SM2RAIN–ASCAT, in a subbasin of the MRB. This work would be useful for choosing appropriate SPPs for hydrological simulations, and the methodology could be used for other similar characteristic regions, especially in Southeast Asia.