*3.3. The TRMM, CMORPH, and CHIRPS Precipitation Datasets*

TRMM, CMORPH, and CHIRPS are three quasi-global gridded precipitation datasets. The TRMM TMPA products provide precipitation for the spatial coverage of 50◦ N–S from 1998 to the present with a spatial resolution of 0.25◦ × 0.25◦ [18]. In this study, the TRMM 3B42 product was used. The original temporal resolution of this dataset is 3 h, and the daily aggregated TRMM 3B42 product can be obtained from the NASA Goddard Earth Sciences (GES) Data and Information Services Center (DISC) (https://disc.gsfc.nasa.gov, accessed on 20 May 2020). The CMORPH products provide precipitation for the spatial coverage of 60◦ N–S from 1998 to the present [40]. The lasted version 1.0 includes three different products, including raw (satellite-only precipitation estimates), bias-corrected (CRT), and gauge-satellite blended (BLD). In this study, the CMORPH BLD product with a spatial resolution of 0.25◦ × 0.25◦ was used, which can be downloaded from https://ftp. cpc.ncep.noaa.gov/precip/CMORPH\_V1.0/BLD (accessed on 20 May 2020). The CHIRPS product provides precipitation for the spatial coverage of 50◦ N–S with a high spatial resolution (i.e., 0.05◦ × 0.05◦ ) from 1981 to the present [19]. The CHIRPS version 2.0 data were downloaded from https://data.chc.ucsb.edu/products/CHIRPS-2.0/global\_daily (accessed on 20 May 2020) and used in this study.

#### *3.4. SWAT Modelling Procedures*

Developed by the Agricultural Research Service of the United States Department of Agriculture, Agricultural Research Service [41], the Soil and Water Assessment Tool (SWAT) is a semi-distributed, process-based, and time-continuous watershed model. SWAT is capable of modeling hydrological processes, soil erosion, and water quality at basin scales [42]. In SWAT, the river basin is first divided into subbasins, then further to the hydrologic response units (HRUs) that represent a unique combination of soil type, land use, and slope. More details on the description of SWAT can be found in many sources [25,42,43], and thus this description was not repeated here for conciseness. As an easy-to-use toolbar in the QGIS interface, QSWAT (Version 2017.02\_1.4, Texas A&M University, College Station, TX, USA) was used to set up the SWAT model in this study.

To set up the SWAT model (Figure 2), the following data sources were used: (1) the elevation data was obtained from the Digital Elevation Model (DEM) product of ALOS

World 3D–30 m (AW3D30) which was released by the Japan Aerospace Exploration Agency (JAXA) with a horizontal spatial resolution of approximately 30 m. These data were downloaded from http://www.eorc.jaxa.jp/ALOS/en/aw3d30/ (accessed on 6 September 2019). (2) The land use map in 2010, with a scale of 1:100,000, was obtained from the National Earth System Science Data Sharing Platform of China (http://www.geodata.cn, accessed on 6 September 2019). (3) The soil map, with a scale of 1:500,000, was obtained from China Soil Database (CSDB, http://vdb3.soil.csdb.cn/, accessed on 6 September 2019). This study used eighteen scenarios that are generated through the combination of three meteorological (excluding precipitation) datasets (Gauge, CFSR, and CMADS) and six precipitation datasets (Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS), to study the impact of different input data on the streamflow simulation. were downloaded from http://www.eorc.jaxa.jp/ALOS/en/aw3d30/ (accessed on 6 September 2019). (2) The land use map in 2010, with a scale of 1:100,000, was obtained from the National Earth System Science Data Sharing Platform of China (http://www.geodata.cn, accessed on 6 September 2019). (3) The soil map, with a scale of 1:500,000, was obtained from China Soil Database (CSDB, http://vdb3.soil.csdb.cn/, accessed on 6 September 2019). This study used eighteen scenarios that are generated through the combination of three meteorological (excluding precipitation) datasets (Gauge, CFSR, and CMADS) and six precipitation datasets (Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS), to study the impact of different input data on the streamflow simulation.

World 3D–30 m (AW3D30) which was released by the Japan Aerospace Exploration Agency (JAXA) with a horizontal spatial resolution of approximately 30 m. These data

*Water* **2022**, *14*, x FOR PEER REVIEW 5 of 23

**Figure 2.** Flowchart of streamflow simulation driven by open-access gridded meteorological and remote sensing precipitation products. **Figure 2.** Flowchart of streamflow simulation driven by open-access gridded meteorological and remote sensing precipitation products.

The SWAT model was run at the daily time scales for the period 1 January 2008–31 July 2014. The first year (2008) was used as a warm-up period to alleviate the impact of initial hydrological simulation conditions. From 1 January 2009 to 31 December 2011, as a calibration period, sensitivity parameters were identified and further calibrated to make the simulated streamflow fit the observation as close as possible. The period 1 January 2012–31 July 2014 was used as the independent validation period to test the validity of calibrated parameters. In this study, the sequential uncertainty fitting algorithm version 2 The SWAT model was run at the daily time scales for the period 1 January 2008–31 July 2014. The first year (2008) was used as a warm-up period to alleviate the impact of initial hydrological simulation conditions. From 1 January 2009 to 31 December 2011, as a calibration period, sensitivity parameters were identified and further calibrated to make the simulated streamflow fit the observation as close as possible. The period 1 January 2012–31 July 2014 was used as the independent validation period to test the validity of calibrated parameters. In this study, the sequential uncertainty fitting algorithm version 2 (SUFI-2) in the SWAT-CUP tool [44,45] was adopted to perform sensitivity analysis and automatic calibration. Finally, 12 highly sensitive parameters for model calibration were

(SUFI-2) in the SWAT-CUP tool [44,45] was adopted to perform sensitivity analysis and automatic calibration. Finally, 12 highly sensitive parameters for model calibration were

have considerable impacts on runoff generation, and more runoff will be generated under wetter conditions [46]. Even for forest ecosystems, finite amounts of precipitation can be retained during extreme rainfall events [47–49]. The SCS curve number (CN) in Table 1 is for average soil moisture conditions (CN2), the SWAT model updates CN values on each day according to the current soil moisture levels, and whether the ground is frozen. The

selected, as presented in Table 1. It should be noted that antecedent moisture conditions have considerable impacts on runoff generation, and more runoff will be generated under wetter conditions [46]. Even for forest ecosystems, finite amounts of precipitation can be retained during extreme rainfall events [47–49]. The SCS curve number (CN) in Table 1 is for average soil moisture conditions (CN2), the SWAT model updates CN values on each day according to the current soil moisture levels, and whether the ground is frozen. The CN value is the highest (referred to as CN3) when the soil is at field capacity, and the lowest (referred to as CN1) when the soil is at wilting point. Both CN1 and CN3 are functions of CN2, and the details can be found in the theoretical documentation of SWAT [42].

**Table 1.** List of parameters used for calibration and their default values and ranges for calibration ("a\_\_", "v\_\_", and "r\_\_" mean an absolute increase, a replacement, and a relative change from the initial parameter values, respectively).


Two iterations were carried out in the process of calibration, with 1000 simulations in each iteration (a total of 2000 simulations were carried out in the calibration period) by using the objective function of the Nash–Sutcliffe efficiency coefficient (NSE) [50]. After each iteration, the parameter ranges were updated based on the new parameter ranges recommended by SWAT-CUP and the physical boundaries of the parameters. The best model result among the 2000 simulations was selected as the final model output to evaluate and compare different models' performance. Three indicators, including NSE, the coefficient of determination (R<sup>2</sup> ), and the percent bias (PBIAS), were used to evaluate and compare the model outputs. This study referred to the widely used guideline [51] to classify the model performance in terms of NSE; models would be classified as unsatisfactory (NSE ≤ 0.50), satisfactory (0.50 < NSE = 0.65), good (0.65 < NSE = 0.75), and very good (NSE > 0.75). PBIAS measures the average tendency of the simulated data to be larger or smaller than their observed counterparts. PBIAS is ideally zero, with positive values indicating model overestimation and negative values indicate model underestimation. The low-magnitude PBIAS values indicate accurate model simulations [52].

#### **4. Results**

#### *4.1. Comparison of Different Meteorological Inputs*

Figure 3 illustrates the cumulative fraction of daily precipitation at the watershed scale from six sources (i.e., Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS) during the period 2009–2014. The probability of dry day for Gauge was 54%, and the values were much lower (i.e., 32%) for CMADS and much higher for CHIRPS (i.e., 72%). The other three datasets had similar values to that of Gauge (i.e., 45% for CFSR, 51% for TRMM, and 48% for CMORPH). For precipitation intensity within 1–5 mm/day, CFSR had obviously lower frequencies than Gauge, while CHIRPS had obviously higher frequencies. CMADS had higher frequencies of precipitation within 5–20 mm/day. The average annual precipitation

for Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS were 1178, 1230, 966, 1260, 1208, and 1238 mm/year during the entire period 2009–2014.

Figure 4 shows the monthly precipitation of six datasets averaged over the watershed from 2009 to 2014. All six datasets displayed similar rainy seasons centered from June to September. CMADS had generally less monthly precipitation than Gauge, and there were considerably higher peaks in several months in the CFSR precipitation, for instance, August in 2009 and 2012. The three satellite-based precipitation datasets (i.e., TRMM, CMORPH, and CHIRPS) were generally close to gauge measurements, except for some lower peaks, such as in July and September of 2010. *Water* **2022**, *14*, x FOR PEER REVIEW 7 of 23

**Figure 3.** The cumulative fraction of daily precipitation from six datasets (Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS) at the watershed scale during 2009–2014. The left subfigure shows the cumulative fraction for daily precipitation ranging from 0 to 100 mm/day; in order to show more details, the two subfigures in the right show the cumulative fraction for daily precipitation ranging from 0 to 5 mm/day and 5 to 40 mm/day, respectively. **Figure 3.** The cumulative fraction of daily precipitation from six datasets (Gauge, CFSR, CMADS, TRMM, CMORPH, and CHIRPS) at the watershed scale during 2009–2014. The left subfigure shows the cumulative fraction for daily precipitation ranging from 0 to 100 mm/day; in order to show more details, the two subfigures in the right show the cumulative fraction for daily precipitation ranging from 0 to 5 mm/day and 5 to 40 mm/day, respectively. *Water* **2022**, *14*, x FOR PEER REVIEW 8 of 23

**Figure 4.** Comparison of monthly precipitation totals from six datasets at the watershed scale during **Figure 4.** Comparison of monthly precipitation totals from six datasets at the watershed scale during 2009–2014, including (**a**) Gauge CFSR, and CMADS, (**b**) Gauge, TRMM, CMORPH, and CHIRPS.

Figure 5 shows the cumulative fraction of daily maximum and minimum air temper-

2009–2014, including (**a**) Gauge CFSR, and CMADS, (**b**) Gauge, TRMM, CMORPH, and CHIRPS.

(PET) over the basin from three datasets between 2009 and 2014. Figure 6 displays the monthly mean of daily maximum and minimum air temperature, solar radiation, wind speed, relative humidity, and the sum of PET over the entire period. Taking an overall look at Figures 5 and 6, despite some fluctuations, three datasets had a fairly good agreement in the maximum temperature, minimum temperature, and relative humidity. However, large discrepancies among the three datasets can be clearly seen for the solar radiation and wind speed. For solar radiation, the two gridded products had larger fluctuations than Gauge. For wind speed, CMADS was considerably lower than the Gauge and CFSR datasets which display very good agreement with each other. To compute the PET in SWAT, the Penman–Monteith method was used, which requires air temperature, solar radiation, relative humidity, and wind speed as inputs. Despite these large discrepancies in daily solar radiation and wind speed, overall monthly PET totals from the three datasets were in good agreement, as shown in Figures 5 and 6. This suggests the discrepancies in individual weather variables (wind speed and solar radiation in particular in this case) cancelled each other to a certain degree when integrated into a further output, the PET in this case. Since PET would affect the computation of water balance in SWAT, considering good agreement in monthly PET totals from all the three datasets, it was expected that solar radiation and wind speed inputs from three datasets had little influence on the

SWAT modelling results in this studied basin.

Figure 5 shows the cumulative fraction of daily maximum and minimum air temperature, solar radiation, wind speed, relative humidity, and potential evapotranspiration (PET) over the basin from three datasets between 2009 and 2014. Figure 6 displays the monthly mean of daily maximum and minimum air temperature, solar radiation, wind speed, relative humidity, and the sum of PET over the entire period. Taking an overall look at Figures 5 and 6, despite some fluctuations, three datasets had a fairly good agreement in the maximum temperature, minimum temperature, and relative humidity. However, large discrepancies among the three datasets can be clearly seen for the solar radiation and wind speed. For solar radiation, the two gridded products had larger fluctuations than Gauge. For wind speed, CMADS was considerably lower than the Gauge and CFSR datasets which display very good agreement with each other. To compute the PET in SWAT, the Penman–Monteith method was used, which requires air temperature, solar radiation, relative humidity, and wind speed as inputs. Despite these large discrepancies in daily solar radiation and wind speed, overall monthly PET totals from the three datasets were in good agreement, as shown in Figures 5 and 6. This suggests the discrepancies in individual weather variables (wind speed and solar radiation in particular in this case) cancelled each other to a certain degree when integrated into a further output, the PET in this case. Since PET would affect the computation of water balance in SWAT, considering good agreement in monthly PET totals from all the three datasets, it was expected that solar radiation and wind speed inputs from three datasets had little influence on the SWAT modelling results in this studied basin. *Water* **2022**, *14*, x FOR PEER REVIEW 9 of 23

**Figure 5.** The cumulative fraction of daily maximum and minimum air temperature, solar radiation, wind speed, relative humidity, and potential ET from three different datasets (Gauge, CFSR, and CMADS) at the watershed scale during 2009–2014. **Figure 5.** The cumulative fraction of daily maximum and minimum air temperature, solar radiation, wind speed, relative humidity, and potential ET from three different datasets (Gauge, CFSR, and CMADS) at the watershed scale during 2009–2014.

**Figure 6.** Comparison of the monthly mean of daily maximum (TMX) and minimum (TMN) air temperature, solar radiation (SOL), wind speed (WIND), and relative humidity (HMD) from the Gauge, CFSR, and CMADS datasets, and their resulting potential evapotranspiration (PET) estimates at the watershed scale during 2009–2014. **Figure 6.** Comparison of the monthly mean of daily maximum (TMX) and minimum (TMN) air temperature, solar radiation (SOL), wind speed (WIND), and relative humidity (HMD) from the Gauge, CFSR, and CMADS datasets, and their resulting potential evapotranspiration (PET) estimates at the watershed scale during 2009–2014.
