Available via license: CC BY 4.0
Content may be subject to copyright.
water
Article
Evaluation of Multi-Satellite Precipitation Products
for Streamflow Simulations: A Case Study for the
Han River Basin in the Korean Peninsula, East Asia
Thom Thi Vu, Li Li and Kyung Soo Jun *
Graduate School of Water Resources, Sungkyunkwan University, Suwon 16419, Korea;
vuthom.khtn@gmail.com (T.T.V.); lili0809@skku.edu (L.L.)
*Correspondence: ksjun@skku.edu; Tel.: +82-31-290-7515
Received: 27 March 2018; Accepted: 13 May 2018; Published: 16 May 2018
Abstract:
The accuracy and sufficiency of precipitation data play a key role in environmental
research and hydrological models. They have a significant effect on the simulation results of
hydrological models; therefore, reliable hydrological simulation in data-scarce areas is a challenging
task. Advanced techniques can be utilized to improve the accuracy of satellite-derived rainfall
data, which can be used to overcome the problem of data scarcity. Our study aims to (1) assess the
accuracy of different satellite precipitation products such as Tropical Rainfall Measuring Mission
(TRMM 3B42 V7), Precipitation Estimation from Remotely Sensed Information using Artificial
Neural Networks (PERSIANN), PERSIANN-Climate Data Record (PERSIANN-CDR), and China
Meteorological Assimilation Driving Datasets for the SWAT Model (CMADS) by comparing them
with gauged rainfall data; and (2) apply them for runoff simulations for the Han River Basin in South
Korea using the SWAT model. Based on the statistical measures, that is, the proportion correct (PC),
the probability of detection (POD), the frequency bias index (FBI), the index of agreement (IOA),
the root-mean-square-error (RMSE), the mean absolute error (MAE), the coefficient of determination
(R
2
), and the bias, the rainfall data of the TRMM and CMADS show a better accuracy than those of
PERSIANN and PERSIANN-CDR when compared to rain gauge measurements. The TRMM and
CMADS data capture the spatial rainfall patterns in mountainous areas as well. The streamflow
simulated by the SWAT model using ground-based rainfall data agrees well with the observed
streamflow with an average Nash-Sutcliffe efficiency (NSE) of 0.68. The four satellite rainfall products
were used as inputs in the SWAT model for streamflow simulation and the results were compared.
The average R
2
, NSE, and percent bias (PBIAS) show that hydrological models using TRMM (R
2
= 0.54,
NSE = 0.49, PBIAS = [
−
52.70–28.30%]) and CMADS (R
2
= 0.44, NSE = 0.42, PBIAS = [
−
29.30–41.80%])
data perform better than those utilizing PERSIANN (R
2
= 0.29, NSE = 0.13, PBIAS = [38.10–83.20%])
and PERSIANN-CDR (R
2
= 0.25, NSE = 0.16, PBIAS = [12.70–71.20%]) data. Overall, the results
of this study are satisfactory, given that rainfall data obtained from TRMM and CMADS can be
used to simulate the streamflow of the Han River Basin with acceptable accuracy. Based on these
results, TRMM and CMADS rainfall data play important roles in hydrological simulations and
water resource management in the Han River Basin and in other regions with similar climate and
topographical characteristics.
Keywords:
TRMM; PERSIANN; PERSIANN-CDR; CMADS; satellite-derived rainfall; streamflow
simulation; SWAT; Han River
1. Introduction
Precipitation is one of the most essential components of the hydrological cycle [
1
]. The quantity
and quality of the precipitation data used as the principal input to hydrological models affect the
Water 2018,10, 642; doi:10.3390/w10050642 www.mdpi.com/journal/water
Water 2018,10, 642 2 of 23
accuracy of the simulation results [
2
,
3
]. Rain gauges provide direct precipitation measures; however,
scarcity and irregularity problems with respect to the gauge network considerably influence the
data reliability [
4
,
5
]. In addition, an effective spatial coverage of precipitation over a large area
is difficult. Compared with rain gauges that provide rainfall data by accumulating rainfall over
a time interval, weather radar systems provide an instantaneous spatial measure of precipitation
and thus, produce rapid climate information [
6
]. However, Westrick et al. [
7
] investigated the
limitations of the radar network for quantitative precipitation measurement and showed that the
radar-derived precipitation estimates could not represent the regional precipitation since radar
coverage is limited to lowland areas. The drawbacks of radar-derived data, such as coverage area
limitations, costly infrastructure construction, and inaccuracy under complex atmospheric conditions,
result in the poor performance of hydrological models [
8
]. Currently, visible and thermal infrared
sensors onboard the geostationary Earth-orbiting satellites and passive microwave sensors onboard
the low-Earth-orbiting satellites provide more accurate rainfall estimates at a higher measurement
frequency. Based on the advancements of these techniques, several satellite-based precipitation
products with global high-resolution (up to 0.25
◦
) are now available such as those derived from
the Tropical Rainfall Measuring Mission (TRMM), Precipitation Estimation from Remotely Sensed
Information using Artificial Neural Networks (PERSIANN), and Climate Prediction Center morphing
technique (CMORPH) [
9
]. Those global and near-real-time rainfall estimates are extremely attractive
for hydrological and weather studies [
10
–
12
]. The rainfall estimates from PERSIANN, CMORPH,
and TRMM-based Multi-satellite Precipitation Analysis (TMPA), which combines satellite data from
different sensors and ground station data from the Global Precipitation Climatology Centre, have been
widely applied in numerous studies [
13
–
21
]. Recently, the China Meteorological Assimilation Driving
Datasets for the SWAT Model (CMADS), which consist of reanalyzed data based on assimilation
techniques, provided important basic data (that is, rainfall, maximum and minimum temperature,
solar radiation, relative humidity, and wind speed) that are extremely useful to analyze climate–water
cycles and “macro” energy balances in hydrological studies [
22
,
23
]. The CMADS was developed
by Dr. Xianyong Meng from the China Institute of Water Resources and Hydropower Research
(IWHR) and has received worldwide attention [
22
]. Based on the full coverage of East Asia and an
improved accuracy, CMADS promises to be one of the most useful satellite-derived weather datasets
for meteorological and hydrological research.
Distributed hydrological models have been widely applied in water resource management and
hydrological research [
24
]. They include the Hydrologic Simulation Program-Fortran (HSPF) [
25
],
MIKE SHE [
26
], the Hydrologic Modeling System (HEC-HMS) [
27
], and the Soil and Water Assessment
Tool (SWAT) [
28
]. Those models reduce the dependency on specific precipitation inputs and fully
use satellite-based hydrometeorological data [29]. The SWAT has been widely applied because many
studies showed that the SWAT model can simulate streamflow in regions with limited data well
[30–32]
.
Studies of SWAT applications in South Korea include those by Kang et al. [
33
], Kim et al. [
34
],
Bae et al. [35],
Kim et al. [
36
], Shope et al. [
37
], and Cho et al. [
38
]. Kim et al. [
34
] suggested the
integrated SWAT-MODFLOW model which can simulate the interaction between the river flow and the
saturated aquifer. Kim et al. [
36
] proposed a method for the evaluation of the flow regulation effects by
dams on river flow using the SWAT model for the Han River Basin. In those studies, the SWAT model
was successfully applied to mountainous areas and river basins with various sizes; however, a SWAT
model using satellite-based rainfall data has not been considered.
Satellite-derived rainfall estimates with high spatial resolution contribute to the water resources
management especially in areas where the ground-based climate data are limited. The hydrologic
performance of different satellite rainfall products varies regionally because of several factors such as
instrument characteristics and retrieval algorithms. Kim et al. [
39
] compared four satellite precipitation
products for the hydrological utility at a mountainous basin in South Korea and found that TMPAv6 and
TMPAv7 products were closer to ground-based rainfall than CMORPH and global satellite mapping of
Water 2018,10, 642 3 of 23
precipitation (GSMaP). In the streamflow simulation, TMPAv6 and TMPAv7 performed well while
CMORPH and GSMaP resulted in a large underestimation.
Although there are several flood control dams in the river, large floods have occurred in the
downstream area of the Han River Basin, causing severe damages. The use of advanced techniques to
predict precipitation that causes floods has attracted a large interest in recent years [
40
]. The application
of satellite precipitation data with high accuracy and resolution has been widely studied to be able to
respond quickly in real-world situations. Numerous studies successfully applied the SWAT model or
used satellite rainfall products for various regions of South Korea. However, combining satellite rainfall
products with the SWAT model was not considered. This study investigates the hydrologic application
of different satellite rainfall products in the Han River Basin of South Korea by applying the SWAT
model. The CMADS, a newly developed dataset with the full coverage of East Asia, was evaluated
in comparison with other satellite rainfall products. This is a case study for a specific period from
2008 to 2013 as the CMADS is only available from 2008. This study aimed to (1) compare four satellite
rainfall products (TRMM 3B42 V7, PERSIANN, PERSIANN-CDR, and CMADS) with gauge rainfall
data; and (2) evaluate the accuracy and suitability of the four satellite precipitation products as inputs
for streamflow simulations in the Han River Basin. The results of the study can provide information on
the performance of different satellite rainfall products in hydrologic modeling for the Han River Basin.
In addition, this study contributes to enriching the scientific database on hydrologic applications of
different satellite precipitation datasets, especially for data scarce regions.
2. Materials and Methods
2.1. Study Area
This study focused on the Han River Basin; the river originates from Mt. Taebaek and flows into
the Yellow Sea (Figure 1). The Han River Basin is the largest river basin (26,219 km
2
) in South Korea
and occupies approximately 27% of the country’s area [
41
]. The river has a total length of 5417 km and
comprises of two major branches, that is, the North Han River (NHR; 10,652 km
2
) and South Han River
(SHR; 12,514 km
2
). These branches converge at the immediate upstream of the Paldang Lake, which is the
major source of water supply to the Seoul metropolitan area and forms the main stream (Figure 1).
Water 2018, 10, x FOR PEER REVIEW 4 of 23
Figure 1. The Han River basin.
2.2. Precipitation Data from Rain Gauges
The ground-based precipitation data used in this study consist of daily records from 89 rain
gauges from 2007 to 2013 (Figure 1). The daily precipitation data were gathered from an extensive
ground-based data network consisting of 60 synoptic stations of the Water Management Information
System and 29 automatic weather stations (AWS) of the Korea Meteorological Administration. At the
AWS sites, rain is first detected by the sensor of the rain detector and then by the sensor of the tipping
bucket rain gauge, with an accuracy of 0.5 mm [45]. Homogenization and outlier processing were
used for the quality control of the gauged data [46].
In the Han River Basin, the precipitation data were collected from 89 gauge stations that could
be considered as a large number of stations. For a precise climatological/hydrological study, however,
89 stations may not be sufficient for a basin with the area of 26,219 km
2
. Observations with more
homogeneous distribution and fine spatial and temporal resolution could produce more accurate
results. The limitation of ground-based observations can be compensated by well-validated satellite-
derived estimates. Moreover, since various satellite-derived data are freely available, it could be a
cost-effective way to collect data, especially for data-scarce areas.
2.3. Satellite-Derived Precipitation Products
2.3.1. TRMM 3B42 Precipitation Products
The TMPA generated precipitation data for a global coverage of 50° N–S and spatial resolution
of 0.25° × 0.25° based on various meteorological satellites [47]. The TRMM 3B42 V7 includes two types
of datasets, that is, the 3-hourly (corresponding to 3-hour intervals per day, that is, UTC 00:00, 03:00,
06:00, 09:00, 12:00, 15:00, 18:00, and 21:00) and the daily precipitation products. The datasets were
obtained by the TRMM Algorithm 3B42, which was calibrated by multiple independent precipitation
estimates from the optimal combination of 2B-31, 2A-12, Special Sensor Microwave Imager, Special
Sensor Microwave Imager/Sounder, Advanced Microwave Sounding Unit, Microwave Humidity
Figure 1. The Han River basin.
Water 2018,10, 642 4 of 23
The Han River Basin has four distinct seasons. The monthly average temperature varies from
−
6
◦
C to 3
◦
C during winter and from 23
◦
C to 25
◦
C during summer. The average precipitation
during the rainy season is 1272.5 mm, which accounts for 70% of the average annual precipitation.
A large amount of precipitation is concentrated in the mountainous area in the eastern part of the basin,
while less precipitation occurs in the lowlands of the western part due to the influence of western
North Pacific typhoons and summer monsoon precipitation [42,43].
Both precipitation and groundwater significantly affect the water resources of the Han River [
44
].
The surface runoff mainly originates from local precipitation during the summer monsoon season
(June to September) and is maintained by groundwater seepage during the remainder of the year [
44
].
Several hydraulic structures, such as flood control dams, significantly affect the river flow (Figure 1).
However, the flow of the Han River generally varies depending on the season, that is, it ranges from
150 m3/s in January to 4300 m3/s in August [44].
2.2. Precipitation Data from Rain Gauges
The ground-based precipitation data used in this study consist of daily records from 89 rain
gauges from 2007 to 2013 (Figure 1). The daily precipitation data were gathered from an extensive
ground-based data network consisting of 60 synoptic stations of the Water Management Information
System and 29 automatic weather stations (AWS) of the Korea Meteorological Administration. At the
AWS sites, rain is first detected by the sensor of the rain detector and then by the sensor of the tipping
bucket rain gauge, with an accuracy of 0.5 mm [
45
]. Homogenization and outlier processing were used
for the quality control of the gauged data [46].
In the Han River Basin, the precipitation data were collected from 89 gauge stations that
could be considered as a large number of stations. For a precise climatological/hydrological study,
however, 89 stations may not be sufficient for a basin with the area of 26,219 km
2
. Observations
with more homogeneous distribution and fine spatial and temporal resolution could produce more
accurate results. The limitation of ground-based observations can be compensated by well-validated
satellite-derived estimates. Moreover, since various satellite-derived data are freely available, it could
be a cost-effective way to collect data, especially for data-scarce areas.
2.3. Satellite-Derived Precipitation Products
2.3.1. TRMM 3B42 Precipitation Products
The TMPA generated precipitation data for a global coverage of 50
◦
N–S and spatial resolution of
0.25
◦×
0.25
◦
based on various meteorological satellites [
47
]. The TRMM 3B42 V7 includes two types of
datasets, that is, the 3-hourly (corresponding to 3-hour intervals per day, that is, UTC 00:00, 03:00, 06:00,
09:00, 12:00, 15:00, 18:00, and 21:00) and the daily precipitation products. The datasets were obtained
by the TRMM Algorithm 3B42, which was calibrated by multiple independent precipitation estimates
from the optimal combination of 2B-31, 2A-12, Special Sensor Microwave Imager, Special Sensor
Microwave Imager/Sounder, Advanced Microwave Sounding Unit, Microwave Humidity Sounder,
and microwave-adjusted merged geo-infrared (IR). In this study, TRMM 3B42 V7 satellite rainfall data
with a daily resolution were used for the period of 2008–2013. The data can be downloaded from the
Goddard Space Flight Center website (https://mirador.gsfc.nasa.gov).
2.3.2. PERSIANN and PERSIANN-CDR Products
The PERSIANN data were produced by using the artificial neural network algorithm to estimate
the rainfall rate based on longwave IR images from geostationary Earth-orbiting satellites. Rainfall data
with a spatial coverage of 60
◦
S–N and spatial resolution of 0.25
◦×
0.25
◦
are available for March 2000
to the present [
48
]. The PERSIANN-CDR rainfall data were generated by the PERSIANN algorithm
using Gridsat-B1 IR satellite data and the bias was adjusted using monthly products from the Global
Precipitation Climatology Project. The spatial resolution and coverage of this rainfall product are
Water 2018,10, 642 5 of 23
consistent with that of the PERSIANN dataset; data are available for January 1983 to April 2017. In this
study, both the PERSIANN and PERSIANN-CDR products on a daily time scale were obtained from
the Center for Hydrometeorology and Remote Sensing (http://chrsdata.eng.uci.edu/) from 2008
to 2013.
2.3.3. CMADS Precipitation Products
The CMADS is an atmospheric reanalysis dataset, which is obtained by using various assimilation
techniques and multi-source data. The Space and Time Multiscale Analysis System (STMAS)
assimilation technique combined with big data projection and processing methods was used to
produce the CMADS climate dataset. Multi-source data were obtained from the National Centers for
Environmental Prediction/National Center for the Atmospheric Research NCEP/NCAR-R1 reanalysis
dataset, the National Centers for Environmental Prediction-Department of Energy (NCEP-DOE)-(R2)
reanalysis dataset, the Climate Forecast System Reanalysis (CFSR) by NCEP, the European Centre
for Medium-Range Weather Forecasts 15-year Reanalysis (ECMWF ERA-15), the ECMWF Reanalysis
(ERA-40), the ECMWF Reanalysis-Interim (ERA-Interim), the Japanese 25-year Reanalysis project
(JRA-25), and the CMA Land Data Assimilation System (CLDAS) [
22
]. The CMADS rainfall data
provided by the China National Meteorological Information Center were produced using CMORPH’s
integrated precipitation products and validated with gauged rainfall data. In this study, CMADS
V1.1 data (2008–2013) with a spatial coverage of 60
◦
E–160
◦
E longitude and 0
◦
N–65
◦
N latitude,
daily resolution, and spatial resolution of 0.25
◦×
0.25
◦
were used. The detailed description of different
satellite rainfall datasets is shown in Table 1.
Table 1. The description of satellite-based rainfall datasets used in this study.
Dataset Version Spatial/Temporal Resolution Areal Coverage Time Coverage Sources
TRMM 3B42 V7 0.250/daily Near Global 1998–present Huffman et al. [47]
PERSIANN - 0.250/daily Near Global 2000–present Sorooshian et al. [48]
PERSIANN-CDR
CDR 0.250/daily Near Global 1983–2017 Ashouri et al. [49]
CMADS V1.1 0.250/daily East Asia 2008–2014 Meng [22]
2.4. SWAT Model
The SWAT is a conceptual and semi-distributed model that simulates, based on a daily time
step, various variables related to hydrology, weather, soil erosion, soil temperature, plant growth,
nutrients, pesticides, and land management [
28
]. The SWAT model uses the SCS curve number and
modified rational methods to calculate the surface runoff and estimate the peak discharge, respectively.
The Penman–Monteith and Muskingum methods are used to simulate the evapotranspiration and
channel routing, respectively [
50
]. Based on a delineation process, the basin is divided into sub-basins
and hydrologic response units utilizing three essential inputs including the digital elevation model,
land use map, and soil map.
Daily rainfall data from 89 stations and other climate data including air temperature, relative
humidity, wind speed, and solar radiation from 29 meteorology stations were used as inputs to
the model. The outflow of eight reservoirs (Figure 1) obtained from the Korea Water Resources
Management Information System were also used as model input. A detailed description of the input
data for the SWAT model is shown in Table 2. The SWAT-CUP sequential uncertainty fitting (SUFI-2)
program was used for the autocalibration and sensitivity analysis of the SWAT model [
51
]. In the
SUFI-2 program, the calibration objective was to maximize the NSE [
52
,
53
]. The SUFI-2 algorithm has
been widely used for the calibration of the SWAT model. Wu and Chen [
54
] and Khoi et al. [
55
] showed
that the SWAT model calibrated with the SUFI-2 algorithm makes better and reasonable predictions
than models calibrated with other auto-calibrated methods. To construct a homogeneous set of optimal
parameters for the model, a calibration was performed using the streamflow (2008–2010) measured at
16 gauging stations in the main branches. The calibrated parameters were then used for the model
validation (2011–2013). The data from 2007 were used for the model warm-up.
Water 2018,10, 642 6 of 23
Table 2. The description of the input data used in the Soil and Water Assessment Tool (SWAT).
Data Type Data Description Scale Data Source
Topography map Digital elevation map (DEM) 90 m USGS-HydroSHEDS
Land-use/Land cover map
Land use/Land cover classification 2010
1:1,250,000 Korea Ministry of Environment
Soil map Soil types (2007) 10 km Food and Agriculture Organization
Meteorology
Daily precipitation, Minimum and
maximum temperature, Solar radiation,
Relative humidity, Wind speed
1990–2013
Korea Meteorological Administration
and Water Resources Management
Information System
Hydrological data Discharge, Dam operation,
Reservoir characteristics 2008–2013 Water Resources Management
Information System
2.5. Statistical Measures for Precipitation and Runoff
The spatiotemporal variability of satellite precipitation products was compared to the
ground-based data by pixel-to-point comparison. To categorically evaluate and compare the daily
satellite-derived precipitation data with ground-based precipitation data, three statistical measures
were calculated, including the proportion correct (PC), probability of detection (POD), and frequency
bias index (FBI) [
56
,
57
]. Those statistical measures indicate the detection capability of satellite rainfall
data to estimate the possibility of precipitation events. A precipitation event represents a precipitation
day when the daily rainfall is greater than 1 mm/day. The POD, PC, and FBI are defined as follows:
PC = (a+d)/n(1)
POD = a/(a+c) (2)
FBI = (a+b)/(a+c), (3)
where a,b,c, and dare the numbers of precipitation events (a: satellite yes, observation yes; b: satellite
yes, observation no; c: satellite no, observation yes; and d: satellite no, observation no) and nis the
total number of satellite observation pairs.
The POD, which is also called the hit rate, determines the likelihood of detected rainfall data
and the PC represents the accuracy of detected rainfall data. The POD and PC values range from 0
to 1 and can be used to assess the level of agreement between satellite-based rainfall and the gauged
values. The perfect POD and PC score is 1. The FBI ranges from 0 to infinity, with a perfect score of 1.
These three statistical indicators were calculated using a 2 ×2 contingency table (Table 3).
Table 3. The contingency table for the satellite and gauged precipitations with a threshold of 1.0 mm.
Satellite Event
Observation Event Marginal Total
Yes (p≥1.0 mm) No (p< 1.0 mm)
Yes (p≥1.0 mm) a b a + b
No (p< 1.0 mm) c d c + d
Marginal total a+c b+d n=a+b+c+d
To evaluate the accuracy of satellite-derived rainfall data by comparing them with gauged
precipitation, five statistical indicators were adopted, including the index of agreement (IOA),
the root-mean-square-error (RMSE), the mean absolute error (MAE), the coefficient of determination
(R
2
), and the bias [
57
]. The IOA measures additive and proportional differences in the observed and
satellite-derived means and variances. The RMSE measures the average magnitude of the errors.
Since the errors are squared before averaged, RMSE gives relatively high weights to large errors.
The MAE is a linear score which means that all the individual differences are weighted equally in the
Water 2018,10, 642 7 of 23
average. The R
2
is a measure of the proportion of variance, and bias measures the difference between
gauged and satellite-derived data. These indicators are defined as follows:
IOA =1−∑n
i=1(Mi−Oi)2
∑n
i=1Mi−O+Oi−O2(4)
RMSE =s1
n
n
∑
i=1
(Mi−Oi)2(5)
MAE =1
n
n
∑
i=1
|Mi−Oi|(6)
R2=
n
∑
i=1
(Oi−O)×(Mi−M)
sn
∑
i=1
(Oi−O)2×(Mi−M)2
(7)
Bias =1
n
n
∑
i=1
(Mi−Oi), (8)
where M
i
is the estimated grid-scale precipitation from satellite products (that is, TRMM, PERSIANN,
PERSIANN-CDR, and CMADS); O
i
are ground-based measurement data;
O
and
M
are the average
values of ground-based measurements and satellite precipitation data, respectively; and nis the total
number of data.
The suitability of satellite-derived rainfall data for the streamflow simulation was evaluated using
the Nash–Sutcliffe efficiency (NSE), R
2
, and percent bias (PBIAS), which are widely used performance
measures in hydrologic studies [
58
]. NSE is a normalized statistic that determines the relative
magnitude of the residual variance compared to the measured data variance, and PBIAS measures the
average tendency of the estimated data to be larger or smaller than their observed counterparts.
NSE =1−
n
∑
i=1
(Oi−Pi)2
n
∑
i=1Oi−O2
(9)
R2=
n
∑
i=1
(O−O)×(P−P)
sn
∑
i=1
(O−O)2×(P−P)2
(10)
PBIAS =100.
n
∑
i=1
(Oi−Pi)
n
∑
i=1
Oi
(11)
where O
i
is the ith observation; P
i
is the ith predicted value;
O
and
P
are the mean observed and
predicted values, respectively; and nis the total number of observations.
3. Results
3.1. Evaluation of Different Satellite-Derived Precipitation Data
In this study, the applicability of four satellite-derived precipitation datasets (2008–2013),
that is, from TRMM 3B42 V7, PERSIANN, PERSIANN-CDR, and CMADS, were evaluated.
Water 2018,10, 642 8 of 23
First, these satellite-based rainfall data were estimated and compared with gauged rainfall data.
The contingency measures of the POD, PC, and FBI were used for categorical data analysis.
Unlike gauged rainfall data that can be measured in a quantitative manner, it is difficult to determine
the threshold for the precipitation and non-precipitation events of satellite-based rainfall data. Based on
the studies by Dai [
59
] and Dinku et al. [
60
], a minimum threshold of 1.0 mm per day was used to
discriminate between rain (
≥
1.0 mm/day) and no rain (<1.0 mm/day). Figure 2shows the statistical
measures calculated for the four satellite-based precipitation datasets. With respect to the POD values,
TRMM has the largest POD with an average of 0.73 (ranging from 0.21 to 0.88), which is closest
to the perfect score of 1. The CMADS has a slightly higher POD than PERSIANN, with a mean
of 0.62 (ranging from 0.47 to 0.76), while the PERSIANN mean is 0.58 (ranging from 0.44 to 0.75).
The PERSIANN-CDR has the smallest POD with a mean of 0.52 (ranging from 0.21 to 0.75; Figure 2a).
With respect to the PC values, TRMM shows the best performance, with an average of 0.78 (ranging
from 0.71 to 0.89) and CMADS has the second-best performance, with an average of 0.70 (ranging
from 0.56 to 0.82). The mean PERSIANN and PERSIANN-CDR values are 0.60 (ranging from 0.40
to 0.77) and 0.50 (ranging from 0.24 to 0.69), respectively (Figure 2b). With respect to the FBI values,
TRMM has an average value of 2.0 (ranging from 1.50 to 3.0), which is the smallest value and closest to
the perfect score of 1. The average FBI of CMADS is 2.75, ranging from 1.25 to 4.25. The PERSIANN
and PERSIANN-CDR averages are 2.75 (ranging from 1.50 to 5.50) and 4.50 (ranging from 1.85 to 5.54),
respectively (Figure 2c). These results indicate that the TRMM data show the best agreement with the
gauged data, while the PERSIANN-CDR data display the biggest difference from the gauged data.
The TRMM has an average IOA value of 0.30 (ranging from 0.25 to 0.37). The CMADS has a
slightly bigger average value of 0.32 (ranging from 0.21 to 0.34). The PERSIANN and PERSIANN-CDR
have relatively lower mean values of 0.24 (ranging from 0.20 to 0.26) and 0.23 (ranging from 0.19
to 0.26), respectively (Figure 2d). Therefore, the average IOA of CMADS is the closest to the
perfect value of 1. The average RMSEs of the TRMM, CMADS, PERSIANN, and PERSIANN-CDR
are 11.22 mm/day (ranging from 2.15 mm/day to 21.29 mm/day), 10.72 mm/day (ranging from
2.05 mm/day to 17.47 mm/day), 11.61 mm/day (ranging from 1.81 mm/day to 22.53 mm/day),
and 11.35 mm/day (ranging from 1.04 mm/day to 22.46 mm/day), respectively (Figure 2e). This shows
that the four satellite-based rainfall datasets have similar RMSE averages and ranges. The average MAE
of the CMADS is 3.53 mm/day (ranging from 1.52 mm/day to 4.80 mm/day), which is smaller than
that of the TRMM, PERSIANN, and PERSIANN-CDR, with average values of 3.96 mm/day (ranging
from 2.75 mm/day to 5.0 mm/day), 3.80 mm/day (ranging from 2.01 mm/day to 5.25 mm/day),
and 3.84 mm/day (ranging from 2.0 mm/day to 5.30 mm/day), respectively (Figure 2f). In Figure 2,
it is difficult to see whether the RMSEs (Figure 2e) and MAE (Figure 2f) of each satellite rainfall data are
significantly different. In order to determine if there are significant differences among the mean RMSEs
and the mean MAEs of the four satellite rainfall data, one-way Analysis of Variance (ANOVA) was
carried out. The results of ANOVA showed that the mean RMSEs of the four satellites’ rainfall data
are not significantly different (p-value = 0.692) at a confidence level of 0.05. However, the p-value for
MAE was 0.013, which is smaller than 0.05, and the multiple comparisons showed that the mean MAE
of CMADS is significantly greater than that of TRMM. Based on the average IOA, RMSE, and MAE
values, the CMADS data show the best accuracy when compared with the gauged data.
The spatial correlation and bias patterns of the four satellite-derived rainfall datasets for the
quantitative verification of precipitation are illustrated in Figures 3and 4. Overall, the TRMM shows a
better correlation (average of 0.58, ranging from 0.16 to 0.80) with the gauged data and a smaller bias
pattern (ranging from
−
5.1 mm/day to 4.0 mm/day) for most gauges, which indicates that the TRMM
data are similar to the gauge observations. The average R
2
of the CMADS data is 0.52, which is slightly
smaller than that of the TRMM data. The bias of the CMADS data varies between
−
5.3 mm/day and
6.0 mm/day. The average R2of the PERSIANN and PERSIANN-CDR data is 0.41 (ranging from 0.20
to 0.58) and 0.46 (ranging from 0.22 to 0.61), respectively, and the bias varies between
−
5.3 mm/da
and 7.1 mm/day, and between −5.5 mm/day and 7.9 mm/day, respectively.
Water 2018,10, 642 9 of 23
Water 2018, 10, x FOR PEER REVIEW 9 of 23
mm/day and 6.0 mm/day. The average R
2
of the PERSIANN and PERSIANN-CDR data is 0.41
(ranging from 0.20 to 0.58) and 0.46 (ranging from 0.22 to 0.61), respectively, and the bias varies
between −5.3 mm/da and 7.1 mm/day, and between −5.5 mm/day and 7.9 mm/day, respectively.
Figure 2. The box plots for the statistical measures of (a) probability of detection (POD), (b) proportion
correct (PC), (c) frequency bias index (FBI), (d) index of agreement (IOA), (e) root mean square error
(RMSE), and (f) mean absolute error (MAE). The square symbol represents the mean value. The
median is presented by the middle line in the box. Each box ranges from the lower (25th) to upper
quartile (75th).
Figure 3. The spatial correlation pattern for ground-based and satellite-derived rainfall during 2008–2013.
The circles represent the gauge stations. (a) TRMM 3B42 V7, (b) CMADS, (c) PERSIANN, (d)
PERSIANN-CDR.
Figure 2.
The box plots for the statistical measures of (
a
) probability of detection (POD), (
b
) proportion
correct (PC), (
c
) frequency bias index (FBI), (
d
) index of agreement (IOA), (
e
) root mean square error
(RMSE), and (
f
) mean absolute error (MAE). The square symbol represents the mean value. The median
is presented by the middle line in the box. Each box ranges from the lower (25th) to upper quartile (75th).
Water 2018, 10, x FOR PEER REVIEW 9 of 23
mm/day and 6.0 mm/day. The average R
2
of the PERSIANN and PERSIANN-CDR data is 0.41
(ranging from 0.20 to 0.58) and 0.46 (ranging from 0.22 to 0.61), respectively, and the bias varies
between −5.3 mm/da and 7.1 mm/day, and between −5.5 mm/day and 7.9 mm/day, respectively.
Figure 2. The box plots for the statistical measures of (a) probability of detection (POD), (b) proportion
correct (PC), (c) frequency bias index (FBI), (d) index of agreement (IOA), (e) root mean square error
(RMSE), and (f) mean absolute error (MAE). The square symbol represents the mean value. The
median is presented by the middle line in the box. Each box ranges from the lower (25th) to upper
quartile (75th).
Figure 3. The spatial correlation pattern for ground-based and satellite-derived rainfall during 2008–2013.
The circles represent the gauge stations. (a) TRMM 3B42 V7, (b) CMADS, (c) PERSIANN, (d)
PERSIANN-CDR.
Figure 3.
The spatial correlation pattern for ground-based and satellite-derived rainfall during
2008–2013. The circles represent the gauge stations. (
a
) TRMM 3B42 V7, (
b
) CMADS, (
c
) PERSIANN,
(d) PERSIANN-CDR.
Water 2018,10, 642 10 of 23
Water 2018, 10, x FOR PEER REVIEW 10 of 23
Figure 4. The spatial bias pattern for ground-based and satellite-derived rainfall during 2008–2013.
The circles represent the gauge stations. (a) TRMM 3B42 V7, (b) CMADS, (c) PERSIANN, (d)
PERSIANN-CDR.
3.2. SWAT Calibration and Validation
The SWAT model was calibrated and validated using gauged precipitation as model input. The
simulated streamflow was then compared to the data observed at 16 stream gauges that are
distributed near homogeneously and have no missing data. A sensitivity analysis was used to
identify the key parameters required for model calibration [61]. Six parameters, that is, CN2,
ALPHA_BF, CH_K2, CH_N2, CANMX, and CH_N1, were selected to calibrate the model. The initial
and calibrated values of those six parameters are shown in Table 4. An automated baseflow
separation technique was used [31] to obtain a more reasonable ALPHA BF (ratio of surface runoff
to baseflow) value for the Han River Basin. Based on the obtained values of ALPHA_BF at different
streamflow stations, the initial range of ALPHA_BF was set for the SWAT model calibration. The
calibration (2008–2010) and validation (2011–2013) were performed using the SUFI-2 algorithm. The
objective function of the SWAT model calibration is to maximize the NSE. The distribution of
behavioral model parameters (1000 runs) for the model calibration is shown in Figure 5. Figure 5
shows that most of the identified parameters are distributed with the NSE ranging from 0.4 to 0.6.
The statistical measures for model performances computed using daily streamflow observations are
listed in Table 5. The results show an overall good agreement between the observed and simulated
streamflow; the NSE, R2, and PBIAS values vary in the ranges of 0.50 to 0.94, 0.51 to 0.95, and −14.10%
to 24.30%, respectively, during the calibration and from 0.50 to 0.90, 0.51 to 0.90, and −38.80% to
21.47%, respectively, during the validation. The NSE, R2, and PBIAS values at the outlet of the basin
(Haengjudaegyo Station) are 0.58%, 0.59%, and −10.00%, respectively, during the model calibration
and 0.77%, 0.81%, and −38.80%, respectively, during the model validation.
Figure 4.
The spatial bias pattern for ground-based and satellite-derived rainfall during 2008–2013.
The circles represent the gauge stations. (
a
) TRMM 3B42 V7, (
b
) CMADS, (
c
) PERSIANN,
(d) PERSIANN-CDR.
3.2. SWAT Calibration and Validation
The SWAT model was calibrated and validated using gauged precipitation as model input.
The simulated streamflow was then compared to the data observed at 16 stream gauges that are
distributed near homogeneously and have no missing data. A sensitivity analysis was used to identify
the key parameters required for model calibration [
61
]. Six parameters, that is, CN2, ALPHA_BF,
CH_K2, CH_N2, CANMX, and CH_N1, were selected to calibrate the model. The initial and calibrated
values of those six parameters are shown in Table 4. An automated baseflow separation technique
was used [
31
] to obtain a more reasonable ALPHA BF (ratio of surface runoff to baseflow) value for
the Han River Basin. Based on the obtained values of ALPHA_BF at different streamflow stations,
the initial range of ALPHA_BF was set for the SWAT model calibration. The calibration (2008–2010) and
validation (2011–2013) were performed using the SUFI-2 algorithm. The objective function of the SWAT
model calibration is to maximize the NSE. The distribution of behavioral model parameters (1000 runs)
for the model calibration is shown in Figure 5. Figure 5shows that most of the identified parameters
are distributed with the NSE ranging from 0.4 to 0.6. The statistical measures for model performances
computed using daily streamflow observations are listed in Table 5. The results show an overall good
agreement between the observed and simulated streamflow; the NSE, R
2
, and PBIAS values vary in
the ranges of 0.50 to 0.94, 0.51 to 0.95, and
−
14.10% to 24.30%, respectively, during the calibration and
from 0.50 to 0.90, 0.51 to 0.90, and
−
38.80% to 21.47%, respectively, during the validation. The NSE, R
2
,
and PBIAS values at the outlet of the basin (Haengjudaegyo Station) are 0.58%, 0.59%, and
−
10.00%,
respectively, during the model calibration and 0.77%, 0.81%, and
−
38.80%, respectively, during the
model validation.
Water 2018,10, 642 11 of 23
Table 4. The initial and calibrated parameters selected for the SWAT model.
Parameters Parameter Description Initial Range Calibrated Range Best Value
rCN2 Initial SCS CN II value −0.20 to 0.20 −0.07 to 0.79 0.18
vALPHA_BF Baseflow alpha factor 0.0035 to 0.80 0.28 to 0.80 0.72
vCH_K2
Effective hydraulic conductivity of the main channel
−0.01 to 500 −0.01 to 268 4.65
vCH_N2 Manning’s value for main channels −0.01 to 0.30 −0.10 to 0.17 0.03
vCANMX Maximum canopy storage 0 to 100 0 to 55.40 55.19
vCH_N1 Manning’s value for tributary channels 0.01 to 30 10 to 30 28.46
rThe parameter was multiplied by one plus a given value; vThe parameter was replaced by a given value.
Water 2018, 10, x FOR PEER REVIEW 11 of 23
Table 4. The initial and calibrated parameters selected for the SWAT model.
Parameters Parameter Description Initial Range Calibrated Range Best Value
r
CN2 Initial SCS CN II value −0.20 to 0.20 −0.07 to 0.79 0.18
v
ALPHA_BF Baseflow alpha fa ctor 0.0035 to 0.80 0.28 to 0.80 0.72
v
CH_K2 Effective hydraulic conductivity of the main channel −0.01 to 500 −0.01 to 268 4.65
v
CH_N2 Manning’s value for main channels −0.01 to 0.30 −0.10 to 0.17 0.03
v
CANMX Maximum canopy storage 0 to 100 0 to 55.40 55.19
v
CH_N1 Manning’s value for tributary channels 0.01 to 30 10 to 30 28.46
r
The parameter was multiplied by one plus a given value;
v
The parameter was replaced by a given
value.
Figure 5. The distribution of behavioral model parameters.
Figure 5. The distribution of behavioral model parameters.
Water 2018,10, 642 12 of 23
Table 5. The statistical measurements of the model performance for the streamflow simulation.
Code Station Calibration (2008–2010) Validation (2011–2013)
NSE R2PBIAS (%) NSE R2PBIAS (%)
SG6 PanUn 0.53 0.54 −7.50 - - -
SG8 YeongWeol1 0.57 0.59 14.70 0.64 0.72 −33.41
SG9 YeongChun 0.59 0.59 3.50 0.61 0.71 −26.99
SG10 DalCheon 0.63 0.67 24.30 0.75 0.78 −17.19
SG11 Mokgyegyo 0.94 0.95 14.70 0.61 0.73 −10.10
SG15 Yeojudaegyo 0.82 0.82 −0.20 - - -
SG17 Heukcheongyo 0.51 0.53 20.00 0.61 0.61 −7.73
SG19 WeonTong 0.53 0.67 11.30 0.50 0.51 21.47
SG20 NaeLinCheon 0.69 0.70 17.10 0.58 0.66 −12.44
SG23 Jueumchigyo 0.50 0.51 16.90 0.56 0.56 18.90
SG25 Bangokgyo 0.67 0.71 24.20 0.58 0.63 −6.78
SG26 Daeseongri 0.71 0.73 0.60 0.61 0.73 −14.71
SG28 Sumthlgyo 0.63 0.67 20.50 0.75 0.78 20.77
SG31 Gwangjingyo 0.56 0.63 −14.10 0.56 0.77 −24.96
SG33 Jungranggyo 0.56 0.58 18.40 0.90 0.90 −11.80
SG37 (outlet) Haengjudaegyo 0.58 0.59 −13.00 0.77 0.81 −38.80
3.3. Streamflow Simulation Using Four Satellite-Derived Rainfall Datasets
The SWAT model calibrated using ground-based rainfall data was used for the evaluation of the
hydrologic performance of four satellite-derived rainfall products. Propagation of the uncertainties
in the parameters leads to uncertainties in the model output, that is, the discharge of the streamflow,
which are expressed as the 95% probability distributions. These are calculated at the 2.5% and 97.5%
levels of the cumulative distribution of the discharge. This is referred to as the 95% prediction
uncertainty, or 95PPU. Figures 6–10 illustrate the prediction uncertainty at the outlet (Haengjudaegyo
station) of the Han River Basin from May to October 2008. These figures show that the TRMM rainfall
data have a better performance than other satellite rainfall data in streamflow simulations.
Water 2018, 10, x FOR PEER REVIEW 12 of 23
Table 5. The statistical measurements of the model performance for the streamflow simulation.
Code Station
Calibration (2008–2010) Validation (2011–2013)
NSE R
2
PBIAS (%) NSE R
2
PBIAS (%)
SG6 PanUn 0.53 0.54 −7.50 - - -
SG8 YeongWeol1 0.57 0.59 14.70 0.64 0.72 −33.41
SG9 YeongChun 0.59 0.59 3.50 0.61 0.71 −26.99
SG10 DalCheon 0.63 0.67 24.30 0.75 0.78 −17.19
SG11 Mokgyegyo 0.94 0.95 14.70 0.61 0.73 −10.10
SG15 Yeojudaegyo 0.82 0.82 −0.20 - - -
SG17 Heukcheongyo 0.51 0.53 20.00 0.61 0.61 −7.73
SG19 WeonTong 0.53 0.67 11.30 0.50 0.51 21.47
SG20 NaeLinCheon 0.69 0.70 17.10 0.58 0.66 −12.44
SG23 Jueumchigyo 0.50 0.51 16.90 0.56 0.56 18.90
SG25 Bangokgyo 0.67 0.71 24.20 0.58 0.63 −6.78
SG26 Daeseongri 0.71 0.73 0.60 0.61 0.73 −14.71
SG28 Sumthlgyo 0.63 0.67 20.50 0.75 0.78 20.77
SG31 Gwangjingyo 0.56 0.63 −14.10 0.56 0.77 −24.96
SG33 Jungranggyo 0.56 0.58 18.40 0.90 0.90 −11.80
SG37 (outlet) Haengjudaegyo 0.58 0.59 −13.00 0.77 0.81 −38.80
3.3. Streamflow Simulation Using Four Satellite-Derived Rainfall Datasets
The SWAT model calibrated using ground-based rainfall data was used for the evaluation of the
hydrologic performance of four satellite-derived rainfall products. Propagation of the uncertainties
in the parameters leads to uncertainties in the model output, that is, the discharge of the streamflow,
which are expressed as the 95% probability distributions. These are calculated at the 2.5% and 97.5%
levels of the cumulative distribution of the discharge. This is referred to as the 95% prediction
uncertainty, or 95PPU. Figures 6–10 illustrate the prediction uncertainty at the outlet (Haengjudaegyo
station) of the Han River Basin from May to October 2008. These figures show that the TRMM rainfall
data have a better performance than other satellite rainfall data in streamflow simulations.
Figure 6. The 95% prediction uncertainty (95PPU) for streamflow simulations using gauged rainfall
from May to October (2008).
Figure 6.
The 95% prediction uncertainty (95PPU) for streamflow simulations using gauged rainfall
from May to October (2008).
Water 2018,10, 642 13 of 23
Water 2018, 10, x FOR PEER REVIEW 13 of 23
Figure 7. The 95PPU for streamflow simulations using Tropical Rainfall Measuring Mission (TRMM
3B42 V7) rainfall estimates from May to October (2008).
Figure 8. The 95PPU for streamflow simulations using China Meteorological Assimilation Driving
Datasets for the SWAT Model (CMADS) rainfall estimates from May to October (2008).
Figure 9. The 95PPU for streamflow simulations using Precipitation Estimation from Remotely
Sensed Information using Artificial Neural Networks (PERSIANN) rainfall estimates from May to
October (2008).
Figure 7.
The 95PPU for streamflow simulations using Tropical Rainfall Measuring Mission
(TRMM 3B42 V7) rainfall estimates from May to October (2008).
Water 2018, 10, x FOR PEER REVIEW 13 of 23
Figure 7. The 95PPU for streamflow simulations using Tropical Rainfall Measuring Mission (TRMM
3B42 V7) rainfall estimates from May to October (2008).
Figure 8. The 95PPU for streamflow simulations using China Meteorological Assimilation Driving
Datasets for the SWAT Model (CMADS) rainfall estimates from May to October (2008).
Figure 9. The 95PPU for streamflow simulations using Precipitation Estimation from Remotely
Sensed Information using Artificial Neural Networks (PERSIANN) rainfall estimates from May to
October (2008).
Figure 8.
The 95PPU for streamflow simulations using China Meteorological Assimilation Driving
Datasets for the SWAT Model (CMADS) rainfall estimates from May to October (2008).
Water 2018, 10, x FOR PEER REVIEW 13 of 23
Figure 7. The 95PPU for streamflow simulations using Tropical Rainfall Measuring Mission (TRMM
3B42 V7) rainfall estimates from May to October (2008).
Figure 8. The 95PPU for streamflow simulations using China Meteorological Assimilation Driving
Datasets for the SWAT Model (CMADS) rainfall estimates from May to October (2008).
Figure 9. The 95PPU for streamflow simulations using Precipitation Estimation from Remotely
Sensed Information using Artificial Neural Networks (PERSIANN) rainfall estimates from May to
October (2008).
Figure 9.
The 95PPU for streamflow simulations using Precipitation Estimation from Remotely
Sensed Information using Artificial Neural Networks (PERSIANN) rainfall estimates from May to
October (2008).
Water 2018,10, 642 14 of 23
Water 2018, 10, x FOR PEER REVIEW 14 of 23
Figure 10. The 95PPU for streamflow simulations using PERSIANN-Climate Data Record
(PERSIANN-CDR) rainfall estimates from May to October (2008).
To quantify the fit between the simulation results, expressed as 95PPU, and the observations,
the P-factor and R-factor were calculated. The P-factor is the percentage of observed data enveloped
by the out modeling result, the 95PPU. The R-factor is the thickness of the 95PPU envelope. A P-factor
of 1 and an R-factor of zero is a simulation that exactly corresponds to the measured data. The two
statistics for simulations using gauged rainfall and satellite rainfall estimates are shown in Table 6.
Table 6. The P-factor and R-factor due to different rainfall datasets
.
Statistics Gauged Rainfall
(Calibration)
Gauged Rainfall
(Validation) TRMM CMADS PERSIANN PERSIANN-CDR
P-factor 0.54 0.47 0.51 0.40 0.39 0.33
R-factor 0.47 0.56 0.42 0.42 0.37 0.43
Table 7 shows that the daily streamflow simulation using TRMM data, with an average R
2
of
0.54 (ranging from 0.29 to 0.81), average NSE of 0.49 (ranging from 0.27 to 0.79), and the PBIAS
ranging from −52.70% to 28.30%, performs better than the simulations using the other three satellite-
derived rainfall datasets. The streamflow simulation using the CMADS data shows the second-best
performance, with an average R
2
of 0.44 (ranging from 0.22 to 0.70), average NSE of 0.42 (ranging
from 0.3 to 0.62), and the PBIAS ranging from −29.3% to 41.8%. The average R
2
, NSE, and PBIAS show
that models using the PERSIANN and PERSIANN-CDR perform relatively poor.
Table 7. The statistical indicators for the streamflow simulations using different rainfall datasets.
Code Station Product R
2
NSE PBIAS (%)
SG6 PanUn
Rain gauge 0.54 0.53 −7.50
PERSIANN 0.44 0.23 55.70
PERSIANN-CDR 0.21 0.19 35.00
TRMM 3B42 V7 0.48 0.42 −52.70
CMADS 0.39 0.37 11.20
SG8 YeongWeol1
Rain gauge 0.66 0.61 −9.36
PERSIANN 0.55 0.21 67.20
PERSIANN-CDR 0.55 0.21 67.20
TRMM 3B42 V7 0.46 0.45 −18.40
CMADS 0.49 0.43 23.20
SG9 YeongChun Rain gauge 0.65 0.60 −11.75
Figure 10.
The 95PPU for streamflow simulations using PERSIANN-Climate Data Record
(PERSIANN-CDR) rainfall estimates from May to October (2008).
To quantify the fit between the simulation results, expressed as 95PPU, and the observations,
the P-factor and R-factor were calculated. The P-factor is the percentage of observed data enveloped by
the out modeling result, the 95PPU. The R-factor is the thickness of the 95PPU envelope. A P-factor of 1
and an R-factor of zero is a simulation that exactly corresponds to the measured data. The two statistics
for simulations using gauged rainfall and satellite rainfall estimates are shown in Table 6.
Table 6. The P-factor and R-factor due to different rainfall datasets.
Statistics Gauged Rainfall
(Calibration)
Gauged Rainfall
(Validation) TRMM CMADS PERSIANN PERSIANN-CDR
P-factor 0.54 0.47 0.51 0.40 0.39 0.33
R-factor 0.47 0.56 0.42 0.42 0.37 0.43
Table 7shows that the daily streamflow simulation using TRMM data, with an average R
2
of 0.54
(ranging from 0.29 to 0.81), average NSE of 0.49 (ranging from 0.27 to 0.79), and the PBIAS ranging from
−
52.70% to 28.30%, performs better than the simulations using the other three satellite-derived rainfall
datasets. The streamflow simulation using the CMADS data shows the second-best performance,
with an average R
2
of 0.44 (ranging from 0.22 to 0.70), average NSE of 0.42 (ranging from 0.3 to 0.62),
and the PBIAS ranging from
−
29.3% to 41.8%. The average R
2
, NSE, and PBIAS show that models
using the PERSIANN and PERSIANN-CDR perform relatively poor.
Table 7. The statistical indicators for the streamflow simulations using different rainfall datasets.
Code Station Product R2NSE PBIAS (%)
SG6 PanUn
Rain gauge 0.54 0.53 −7.50
PERSIANN 0.44 0.23 55.70
PERSIANN-CDR
0.21 0.19 35.00
TRMM 3B42 V7
0.48 0.42 −52.70
CMADS 0.39 0.37 11.20
SG8 YeongWeol1
Rain gauge 0.66 0.61 −9.36
PERSIANN 0.55 0.21 67.20
PERSIANN-CDR
0.55 0.21 67.20
TRMM 3B42 V7
0.46 0.45 −18.40
CMADS 0.49 0.43 23.20
Water 2018,10, 642 15 of 23
Table 7. Cont.
Code Station Product R2NSE PBIAS (%)
SG9 YeongChun
Rain gauge 0.65 0.60 −11.75
PERSIANN 0.49 0.25 59.50
PERSIANN-CDR
0.28 0.25 31.30
TRMM 3B42 V7
0.59 0.54 −33.00
CMADS 0.44 0.42 20.90
SG10 DalCheon
Rain gauge 0.73 0.69 3.56
PERSIANN 0.41 0.15 68.90
PERSIANN-CDR
0.29 0.23 46.30
TRMM 3B42 V7
0.33 0.33 −14.80
CMADS 0.36 0.33 10.10
SG11 Mokgyegyo
Rain gauge 0.84 0.78 2.30
PERSIANN 0.61 0.18 71.10
PERSIANN-CDR
0.60 0.42 50.10
TRMM 3B42 V7
0.81 0.79 −7.30
CMADS 0.70 0.62 35.50
SG15 Yeojudaegyo
Rain gauge 0.82 0.82 −0.20
PERSIANN 0.06 0.04 40.50
PERSIANN-CDR
0.05 0.03 31.20
TRMM 3B42 V7
0.60 0.43 −13.57
CMADS 0.31 0.31 −29.30
SG17 Heukcheongyo
Rain gauge 0.57 0.56 6.14
PERSIANN 0.08 0.03 69.90
PERSIANN-CDR
0.06 0.03 58.80
TRMM 3B42 V7
0.48 0.42 12.70
CMADS 0.30 0.38 41.80
SG19 WeonTong
Rain gauge 0.59 0.52 16.39
PERSIANN 0.40 0.04 83.20
PERSIANN-CDR
0.44 0.17 71.20
TRMM 3B42 V7
0.59 0.58 28.30
CMADS 0.57 0.49 27.10
SG20 NaeLinCheon
Rain gauge 0.68 0.64 2.33
PERSIANN 0.32 0.12 72.60
PERSIANN-CDR
0.23 0.17 54.10
TRMM 3B42 V7
0.59 0.58 −4.50
CMADS 0.31 0.45 19.50
SG23 Jueumchigyo
Rain gauge 0.52 0.51 17.90
PERSIANN 0.04 0.02 73.80
PERSIANN-CDR
0.04 0.02 59.90
TRMM 3B42 V7
0.58 0.42 11.30
CMADS 0.50 0.43 22.90
SG25 Bangokgyo
Rain gauge 0.67 0.63 8.71
PERSIANN 0.14 0.02 73.70
PERSIANN-CDR
0.08 0.02 62.00
TRMM 3B42 V7
0.56 0.55 16.00
CMADS 0.47 0.41 20.70
SG26 Daeseongri
Rain gauge 0.73 0.66 −7.06
PERSIANN 0.45 0.35 48.90
PERSIANN-CDR
0.50 0.44 36.00
TRMM 3B42 V7
0.55 0.51 −8.90
CMADS 0.53 0.51 24.30
Water 2018,10, 642 16 of 23
Table 7. Cont.
Code Station Product R2NSE PBIAS (%)
SG28 Sumthlgyo
Rain gauge 0.73 0.69 20.64
PERSIANN 0.13 0.20 75.30
PERSIANN-CDR
0.07 0.10 67.40
TRMM 3B42 V7
0.62 0.57 25.10
CMADS 0.59 0.55 23.90
SG31 Gwangjingyo
Rain gauge 0.70 0.56 −14.53
PERSIANN 0.08 0.05 38.10
PERSIANN-CDR
0.06 0.01 12.70
TRMM 3B42 V7
0.57 0.44 −51.20
CMADS 0.49 0.42 −10.70
SG33 Jungranggyo
Rain gauge 0.74 0.73 3.30
PERSIANN 0.19 0.04 75.80
PERSIANN-CDR
0.30 0.15 62.50
TRMM 3B42 V7
0.29 0.27 13.40
CMADS 0.39 0.30 21.80
SG37 (Outlet)
Haengjudaegyo
Rain gauge 0.70 0.68 −25.90
PERSIANN 0.23 0.16 52.60
PERSIANN-CDR
0.18 0.16 32.90
TRMM 3B42 V7
0.46 0.47 −32.50
CMADS 0.22 0.32 14.60
Average
Rain gauge 0.68 0.64 0.31
PERSIANN 0.29 0.13 64.18
PERSIANN-CDR
0.25 0.16 48.66
TRMM 3B42 V7
0.54 0.49 −8.13
CMADS 0.44 0.42 17.34
Figure 11 shows the daily streamflow simulated by using the gauged rainfall and different
satellite-derived rainfall datasets as inputs. Based on Figure 11, the models using the four satellite-based
rainfall datasets capture the behavior of the streamflow in most cases by characterizing the rising and
falling of flow. At SG11, SG26, SG31, and SG37, the hydrological models driven by the TRMM and
CMADS data simulate the streamflow very well. However, the models using the four satellite-derived
rainfall datasets tend to underestimate the peak flow of most flood events.
Water 2018, 10, x FOR PEER REVIEW 16 of 23
TRMM 3B42 V7 0.62 0.57 25.10
CMADS 0.59 0.55 23.90
SG31 Gwangjingyo
Rain gauge 0.70 0.56 −14.53
PERSIANN 0.08 0.05 38.10
PERSIANN-CDR 0.06 0.01 12.70
TRMM 3B42 V7 0.57 0.44 −51.20
CMADS 0.49 0.42 −10.70
SG33 Jungranggyo
Rain gauge 0.74 0.73 3.30
PERSIANN 0.19 0.04 75.80
PERSIANN-CDR 0.30 0.15 62.50
TRMM 3B42 V7 0.29 0.27 13.40
CMADS 0.39 0.30 21.80
SG37 (Outlet) Haengjudaegyo
Rain gauge 0.70 0.68 −25.90
PERSIANN 0.23 0.16 52.60
PERSIANN-CDR 0.18 0.16 32.90
TRMM 3B42 V7 0.46 0.47 −32.50
CMADS 0.22 0.32 14.60
Average
Rain gauge 0.68 0.64 0.31
PERSIANN 0.29 0.13 64.18
PERSIANN-CDR 0.25 0.16 48.66
TRMM 3B42 V7 0.54 0.49 −8.13
CMADS 0.44 0.42 17.34
Figure 11 shows the daily streamflow simulated by using the gauged rainfall and different
satellite-derived rainfall datasets as inputs. Based on Figure 11, the models using the four satellite-
based rainfall datasets capture the behavior of the streamflow in most cases by characterizing the
rising and falling of flow. At SG11, SG26, SG31, and SG37, the hydrological models driven by the
TRMM and CMADS data simulate the streamflow very well. However, the models using the four
satellite-derived rainfall datasets tend to underestimate the peak flow of most flood events.
Figure 11. Cont.
Water 2018,10, 642 17 of 23
Water 2018, 10, x FOR PEER REVIEW 17 of 23
Figure 11. The comparison of the observed and computed flows obtained by using ground-based
rainfall and the four satellite-derived rainfall datasets.
Figure 11.
The comparison of the observed and computed flows obtained by using ground-based
rainfall and the four satellite-derived rainfall datasets.
4. Discussion
Our study indicates that the TRMM and CMADS rainfall data show higher accuracy than the
PERSIANN and PERSIANN-CDR data. Additionally, the models using the TRMM and CMADS
data show a better performance than those using the PERSIANN and PERSIANN-CDR data for
Water 2018,10, 642 18 of 23
streamflow simulation in the Han River Basin. The results show that the PERSIANN-CDR data,
which are bias-adjusted products, and the PERSIANN data have a similar accuracy. The hydrologic
models using these two rainfall datasets as inputs show similar performances. Contrary to the results
of this study, Ashouri et al. [
49
] evaluated an extreme weather event (Hurricane Katrina) in the United
States and showed that PERSIANN-CDR data have a higher correlation with gauged rainfall than the
TMPA data.
While the satellite rainfall data either overestimate or underestimate the gauged rainfall data at
different stations due to the spatiotemporal uncertainty of satellite rainfall products, the streamflow
simulation results show that the SWAT model using different satellite rainfall datasets mostly
underestimates the peak flow. Figures 12–15 show the comparison of spatially averaged gauged rainfall
and satellite-derived rainfall estimates, and Figure 16 shows the comparison of spatially averaged
annual maximum daily rainfall. These figures show that except TRMM, the other three satellites’
data tend to underestimate the spatially averaged gauged data. Therefore, the underestimation
of streamflow could be attributed to the overall underestimation of satellite rainfall data. Such a
tendency may or may not be only for the Han River Basin and/or for the specific period from
2008 to 2013. Hromadka and McCuen [
62
], Maskey et al. [
63
], and Jones et al. [
64
] indicated that
one of the main sources of uncertainty in streamflow simulations using rainfall–runoff models
is the spatiotemporal uncertainty of the catchment rainfall. This means that the satellite-based
rainfall estimates have a significant effect on the streamflow simulation [
65
]. The uncertainty in
satellite-based rainfall estimates can have several causes. Gebregiorgis and Hossain [
66
] characterized
the errors of satellite-derived rainfall data based on climate type and topography (elevation).
They found that the uncertainty of satellite rainfall data depends more on the topography of the
region than the regional climate.
Xu et al. [67]
found that a large amount of uncertainty in a satellite
precipitation dataset can be explained by the normalized difference vegetation index, digital elevation
model, and land surface temperature. Bitew and Gebremichael [
68
] evaluated the performance of
satellite rainfall products in streamflow simulations (that is, CMORPH, TMPA 3B42RT, TMPA 3B42,
and PERSIANN) for two different watersheds of the Ethiopian highlands and showed that the
application of different satellite rainfall products is related to the watershed area. All these factors can
explain the errors in streamflow simulations using satellite-derived rainfall data. An error propagation
from satellite-derived rainfall to streamflow simulation of hydrological models is inevitable. Thus,
the improvement of the accuracy of satellite-derived rainfall data is very important and a systematic
error correction of satellite data for hydrological applications should be considered. Several studies
showed that the use of satellite data for the calibration of hydrological models results in a better
streamflow prediction [
18
,
69
]. However, such a recalibration may fail due to errors in the satellite data,
which are more difficult to control than those of ground-based data.
Water 2018, 10, x FOR PEER REVIEW 18 of 23
4. Discussion
Our study indicates that the TRMM and CMADS rainfall data show higher accuracy than the
PERSIANN and PERSIANN-CDR data. Additionally, the models using the TRMM and CMADS data
show a better performance than those using the PERSIANN and PERSIANN-CDR data for
streamflow simulation in the Han River Basin. The results show that the PERSIANN-CDR data,
which are bias-adjusted products, and the PERSIANN data have a similar accuracy. The hydrologic
models using these two rainfall datasets as inputs show similar performances. Contrary to the results
of this study, Ashouri et al. [49] evaluated an extreme weather event (Hurricane Katrina) in the
United States and showed that PERSIANN-CDR data have a higher correlation with gauged rainfall
than the TMPA data.
While the satellite rainfall data either overestimate or underestimate the gauged rainfall data at
different stations due to the spatiotemporal uncertainty of satellite rainfall products, the streamflow
simulation results show that the SWAT model using different satellite rainfall datasets mostly
underestimates the peak flow. Figures 12–15 show the comparison of spatially averaged gauged
rainfall and satellite-derived rainfall estimates, and Figure 16 shows the comparison of spatially
averaged annual maximum daily rainfall. These figures show that except TRMM, the other three
satellites’ data tend to underestimate the spatially averaged gauged data. Therefore, the
underestimation of streamflow could be attributed to the overall underestimation of satellite rainfall
data. Such a tendency may or may not be only for the Han River Basin and/or for the specific period
from 2008 to 2013. Hromadka and McCuen [62], Maskey et al. [63], and Jones et al. [64] indicated that
one of the main sources of uncertainty in streamflow simulations using rainfall–runoff models is the
spatiotemporal uncertainty of the catchment rainfall. This means that the satellite-based rainfall
estimates have a significant effect on the streamflow simulation [65]. The uncertainty in satellite-
based rainfall estimates can have several causes. Gebregiorgis and Hossain [66] characterized the
errors of satellite-derived rainfall data based on climate type and topography (elevation). They found
that the uncertainty of satellite rainfall data depends more on the topography of the region than the
regional climate. Xu et al. [67] found that a large amount of uncertainty in a satellite precipitation
dataset can be explained by the normalized difference vegetation index, digital elevation model, and
land surface temperature. Bitew and Gebremichael [68] evaluated the performance of satellite rainfall
products in streamflow simulations (that is, CMORPH, TMPA 3B42RT, TMPA 3B42, and PERSIANN)
for two different watersheds of the Ethiopian highlands and showed that the application of different
satellite rainfall products is related to the watershed area. All these factors can explain the errors in
streamflow simulations using satellite-derived rainfall data. An error propagation from satellite-
derived rainfall to streamflow simulation of hydrological models is inevitable. Thus, the
improvement of the accuracy of satellite-derived rainfall data is very important and a systematic error
correction of satellite data for hydrological applications should be considered. Several studies
showed that the use of satellite data for the calibration of hydrological models results in a better
streamflow prediction [18,69]. However, such a recalibration may fail due to errors in the satellite
data, which are more difficult to control than those of ground-based data.
Figure 12. The comparison of spatially averaged gauged rainfall and CMADS.
Figure 12. The comparison of spatially averaged gauged rainfall and CMADS.
Water 2018,10, 642 19 of 23
Water 2018, 10, x FOR PEER REVIEW 19 of 23
Figure 13. The comparison of spatially averaged gauged rainfall and TRMM.
Figure 14. The comparison of spatially averaged gauged rainfall and PERSIANN.
Figure 15. The comparison of spatially averaged gauged rainfall and PERSIANN-CDR.
Figure 13. The comparison of spatially averaged gauged rainfall and TRMM.
Water 2018, 10, x FOR PEER REVIEW 19 of 23
Figure 13. The comparison of spatially averaged gauged rainfall and TRMM.
Figure 14. The comparison of spatially averaged gauged rainfall and PERSIANN.
Figure 15. The comparison of spatially averaged gauged rainfall and PERSIANN-CDR.
Figure 14. The comparison of spatially averaged gauged rainfall and PERSIANN.
Water 2018, 10, x FOR PEER REVIEW 19 of 23
Figure 13. The comparison of spatially averaged gauged rainfall and TRMM.
Figure 14. The comparison of spatially averaged gauged rainfall and PERSIANN.
Figure 15. The comparison of spatially averaged gauged rainfall and PERSIANN-CDR.
Figure 15. The comparison of spatially averaged gauged rainfall and PERSIANN-CDR.
Water 2018,10, 642 20 of 23
Water 2018, 10, x FOR PEER REVIEW 20 of 23
Figure 16. The comparison of annual maximum daily rainfall of different datasets.
5. Conclusions
The main purpose of this study was to evaluate the utility of satellite-derived precipitation data
in a hydrological model for streamflow simulation in the Han River Basin. Four different satellite-
derived rainfall datasets were compared with ground-based rainfall data at a daily time step. The
TRMM and CMADS rainfall data have relatively higher accuracy. The runoff simulations indicate
that the use of ground-based rainfall data in the SWAT model leads to an overall good agreement
between observed and computed streamflow. Based on the comparison of the use of the four satellite
rainfall products for streamflow simulation in the Han River Basin, hydrological models using the
TRMM and CMADS rainfall data perform better than those using PERSIANN and PERSIANN-CDR
data. The results of this study indicate that the TRMM and CMADS rainfall data can play significant
roles in water resource management and flood control in the Han River Basin. This study also
indicates that the CMADS will provide important basic data with acceptable accuracy for
hydrological research in East Asia, especially in areas with scarce data.
In the study, a pixel-to-point comparison was made between satellite rainfall data and gauged
data. However, a point-to-pixel comparison using some statistical interpolation methods such as
multiple linear regression, optimal interpolation or Kriging should be further discussed in the future
study.
One of the main purposes of the study is to evaluate the newly developed CMADS in
comparison with gauged data and other satellite rainfall products. In this study, data for a limited
period (2008–2013) were used since the CMADS is only available from 2008. However, from a
climatological point of view, using data for only 6 years may not lead to robust results since it is too
short to include all the different characteristics of the rainfall regime of the area. Using data from
different periods could show different results in the rainfall data comparison as well as in the SWAT
model simulation. In addition, 3-year data (2008–2010) was used for the SWAT model calibration;
however, data from different periods may also produce different results. Therefore, the findings and
the conclusions of this study may not be generalized for different time periods. Further studies using
a longer period of data would produce more robust results.
Author Contributions: T.T.V. designed the framework and analyzed the data of this study; T.T.V. and L.L.
collected the data and wrote the paper; K.S.J. provided significant suggestions on the methodology and structure
of the manuscript. All authors read and approved the final manuscript.
Funding: This research was funded by Ministry of Land, Infrastructure and Transport of Advanced Water
Management Research Program (13AWMP-B066744-01).
Conflicts of Interest: The authors declare no conflict of interest.
Figure 16. The comparison of annual maximum daily rainfall of different datasets.
5. Conclusions
The main purpose of this study was to evaluate the utility of satellite-derived precipitation data in
a hydrological model for streamflow simulation in the Han River Basin. Four different satellite-derived
rainfall datasets were compared with ground-based rainfall data at a daily time step. The TRMM
and CMADS rainfall data have relatively higher accuracy. The runoff simulations indicate that the
use of ground-based rainfall data in the SWAT model leads to an overall good agreement between
observed and computed streamflow. Based on the comparison of the use of the four satellite rainfall
products for streamflow simulation in the Han River Basin, hydrological models using the TRMM
and CMADS rainfall data perform better than those using PERSIANN and PERSIANN-CDR data.
The results of this study indicate that the TRMM and CMADS rainfall data can play significant roles in
water resource management and flood control in the Han River Basin. This study also indicates that
the CMADS will provide important basic data with acceptable accuracy for hydrological research in
East Asia, especially in areas with scarce data.
In the study, a pixel-to-point comparison was made between satellite rainfall data and gauged data.
However, a point-to-pixel comparison using some statistical interpolation methods such as multiple
linear regression, optimal interpolation or Kriging should be further discussed in the future study.
One of the main purposes of the study is to evaluate the newly developed CMADS in comparison
with gauged data and other satellite rainfall products. In this study, data for a limited period
(2008–2013) were used since the CMADS is only available from 2008. However, from a climatological
point of view, using data for only 6 years may not lead to robust results since it is too short to include
all the different characteristics of the rainfall regime of the area. Using data from different periods
could show different results in the rainfall data comparison as well as in the SWAT model simulation.
In addition, 3-year data (2008–2010) was used for the SWAT model calibration; however, data from
different periods may also produce different results. Therefore, the findings and the conclusions of this
study may not be generalized for different time periods. Further studies using a longer period of data
would produce more robust results.
Author Contributions:
T.T.V. designed the framework and analyzed the data of this study; T.T.V. and L.L. collected
the data and wrote the paper; K.S.J. provided significant suggestions on the methodology and structure of the
manuscript. All authors read and approved the final manuscript.
Funding:
This research was funded by Ministry of Land, Infrastructure and Transport of Advanced Water
Management Research Program (13AWMP-B066744-01).
Conflicts of Interest: The authors declare no conflict of interest.
Water 2018,10, 642 21 of 23
References
1.
González-Rouco, J.F.; Jiménez, J.L.; Quesada, V.; Valero, F. Quality control and homogeneity of precipitation
data in the southwest of Europe. J. Clim. 2001,14, 964–978. [CrossRef]
2.
Price, K.; Purucker, S.T.; Kraemer, S.R.; Babendreier, J.E.; Knightes, C.D. Comparison of radar and gauge
precipitation data in watershed models across varying spatial and temporal scales. Hydrol. Process.
2014
,28,
3505–3520. [CrossRef]
3.
Sikorska, A.; Seibert, J. Importance of precipitation data quality for streamflow predictions. Geophys. Res. Abstr.
2015,17, 13369.
4.
Buarque, D.C.; De Paiva, R.C.D.; Clarke, R.T.; Mendes, C.A.B. A comparison of Amazon rainfall
characteristics derived from TRMM, CMORPH and the Brazilian national rain gauge network. J. Geophys.
Res. Atmos. 2011,116. [CrossRef]
5.
Kidd, C.; Bauer, P.; Turk, J.; Huffman, G.J.; Joyce, R.; Hsu, K.L.; Braithwaite, D. Intercomparison of
high-resolution precipitation products over northwest Europe. J. Hydrometeorol.
2012
,13, 67–83. [CrossRef]
6.
Thorndahl, S.; Einfalt, T.; Willems, P.; Nielsen, J.E.; Veldhuis, M.C.; Arnbjerg-Nielsen, K.; Rasmussen, M.R.;
Molnar, P. Weather radar rainfall data in urban hydrology. Hydrol. Earth Syst. Sci.
2017
,21, 1359–1380. [CrossRef]
7.
Westrick, K.J.; Mass, C.F.; Colle, B.A. The limitation of the WSR-88D radar network for quantitative precipitation
measurement over the coastal western United States. Bull. Am. Meteorol. Soc. 1999,80, 2289–2298. [CrossRef]
8.
Rico-Ramirez, M.A.; Liguori, S.; Schellart, A.N.A. Quantifying radar-rainfall uncertainties in urban drainage
flow modeling. J. Hydrol. 2015,528, 17–28. [CrossRef]
9.
Joyce, R.J.; Janowiak, J.E.; Arkin, P.A.; Xie, P. CMORPH: A method that produces global precipitation
estimates from passive microwave and infrared data at high spatial and temporal resolution. J. Hydrometeorol.
2004,5, 487–503. [CrossRef]
10. Kidd, C.; Huffman, G. Global precipitation measurement. Meteorol. Appl. 2011,18, 334–353. [CrossRef]
11.
Kidd, C.; Levizzani, V. Status of satellite precipitation retrievals. Hydrol. Earth Syst. Sci.
2011
,15, 1109–1116.
[CrossRef]
12.
Hou, A.Y.; Kakar, R.K.; Neeck, S.; Azarbarzin, A.A.; Kummerow, C.D.; Kojima, M.; Oki, R.; Nakamura, K.;
Iguchi, T. The global precipitation measurement mission. Bull. Am. Meteorol. Soc. 2014,95, 701–722. [CrossRef]
13.
Su, F.; Hong, Y.; Lettenmaier, D.P. Evaluation of TRMM multisatellite precipitation analysis (TMPA) and its
utility in hydrologic prediction in the La Plata basin. J. Hydrometeorol. 2008,9, 622–639. [CrossRef]
14.
Collischonn, B.; Collischonn, W.; Tucci, C.E.M. Daily hydrological modeling in the Amazon basin using
TRMM rainfall estimates. J. Hydrol. 2008,360, 207–216. [CrossRef]
15.
Scheel, M.L.M.; Rohrer, M.; Huggel, C.; Villar, D.S.; Huffman, G.J. Evaluation of TRMM multi-satellite
precipitation analysis (TMPA) performance in the Central Andes region and its dependency on spatial and
temporal resolution. Hydrol. Earth Syst. Sci. 2011,15, 2649–2663. [CrossRef]
16.
Ouma, Y.O.; Owiti, T.; Kibiiy, J.; Ouma, Y.O.; Tateishi, R.; Kipkorir, E. Multitemporal comparative analysis of
TRMM-3B42 satellite-estimated rainfall with surface gauge data at basin scales: Daily, decadal and monthly
evaluations. Int. J. Remote Sens. 2012,33, 7662–7684. [CrossRef]
17.
Xue, X.; Hong, Y.; Limaye, A.S.; Gourley, J.J.; Huffman, G.J.; Khan, S.I.; Dorji, C.; Chen, S. Statistical and
hydrological evaluation of TRMM-based multi-satellite precipitation analysis over the Wangchu Basin of
Bhutan: Are the latest satellite precipitation products ready for use in ungauged basins? J. Hydrol.
2013
,499,
91–99. [CrossRef]
18.
Stisen, S.; Sandholt, I. Evaluation of remote-sensing-based rainfall products through predictive capability in
hydrological runoff modelling. Hydrol. Process. 2010,24, 879–891. [CrossRef]
19.
Behrangi, A.; Khakbaz, B.; Jaw, T.C.; AghaKouchak, A.; Hsu, K.; Sorooshian, S. Hydrologic evaluation of
satellite precipitation products over a mid-size basin. J. Hydrol. 2011,397, 225–237. [CrossRef]
20.
Shen, Y.; Xiong, A.; Wang, Y.; Ha, P. Performance of high-resolution satellite precipitation products over
China. J. Geophys. Res. Atmos. 2010,115. [CrossRef]
21.
Hirpa, F.A.; Gebremichael, M.; Hopson, T. Evaluation of high-resolution satellite precipitation products over
very complex terrain in Ethiopia. J. Appl. Meteorol. Climatol. 2010,49, 1044–1051. [CrossRef]
22.
Meng, X.; Wang, H. Significance of the China meteorological assimilation driving datasets for the SWAT
model (CMADS) of East Asia. Water (Switzerland) 2017,9, 765. [CrossRef]
Water 2018,10, 642 22 of 23
23.
Meng, X.Y.; Wang, H.; Lei, X.H.; Cai, S.Y.; Wu, H.J. Hydrological modeling in the Manas River Basin using
Soil and Water Assessment Tool driven by CMADS. Teh. Vjesn. 2017,24, 525–534.
24.
Setegn, S.G.; Donoso, M.C. Sustanability of Intergrated Water Resources Management: Water Governace, Climate
and Ecohydrology; Springer: Basel, Switzerland, 2015.
25.
Bicknell, B.R.; Imhoff, J.C.; Kittle, J.L.; Donigian, A.S.; Johanson, R.C. Hydrologic Simulation Program—Fortran:
User’s Manual for Release 11; Environmental Research Laboratory, Office of Research and Development, U.S.
Environmental Protection Agency: Athens, GA, USA, 1996.
26.
Refsgaard, J.C.; Storm, B. MIKE SHE. In Computer Models of Watershed Hydrology; Sing, V.P., Ed.;
Water Resource Publications: Highland Ranch, CO, USA, 1995; pp. 806–846.
27.
USACE. HEC-5 Simulation of Flood Control and Conservation System; US Army Corps of Engineers, Hydrologic
Engineering Center: Davis, CA, USA, 1998.
28.
Arnold, J.G.; Srinivasan, R.; Muttiah, R.S.; Williams, J.R. Large area hydrologic modeling and assessment
part I: Model development. J. Am. Water Resour. Assoc. 1998,34, 73–89. [CrossRef]
29.
Sivapalan, M.; Takeuchi, K.; Franks, S.W.; Gupta, V.K.; Karambiri, H.; Lakashmi, V.; Liang, X.; McDonnell, J.J.;
Mendiondo, E.M.; O’connell, P.E.; et al. IAHS Decade on Predictions in Ungauged Basins (PUB), 2003–2012:
Shaping an exciting future for the hydrological sciences. Hydrol. Sci. J. 2003,48, 857–880. [CrossRef]
30.
Eckhardt, K.; Arnold, J.G. Automatic calibration of a distributed catchment model. J. Hydrol.
2001
,251,
103–109. [CrossRef]
31.
Arnold, J.G.; Allen, P.M. Automated methods for estimating baseflow and ground water recharge from
streamflow records. J. Am. Water Resour. Assoc. 1999,35, 411–424. [CrossRef]
32.
White, K.L.; Chaubey, I. Sensitivity analysis, calibration, and validations for a multisite and multivariable
SWAT model. J. Am. Water Resour. Assoc. 2005,41, 1077–1089. [CrossRef]
33.
Kang, M.S.; Park, S.W.; Lee, J.J.; Yoo, K.H. Applying SWAT for TMDL programs to a small watershed
containing rice paddy fields. Agric. Water Manag. 2006,79, 72–92. [CrossRef]
34.
Kim, N.W.; Chung, I.M.; Won, Y.S.; Arnold, J.G. Development and application of the integrated
SWAT-MODFLOW model. J. Hydrol. 2008,356, 1–16. [CrossRef]
35.
Bae, D.H.; Jung, I.W.; Lettenmaier, D.P. Hydrologic uncertainties in climate change from IPCC AR4 GCM
simulations of the Chungju Basin, Korea. J. Hydrol. 2011,401, 90–105. [CrossRef]
36.
Kim, N.W.; Lee, J.E.; Kim, J.T. Assessment of flow regulation effects by dams in the Han River, Korea, on the
downstream flow regimes using SWAT. J. Water Resour. Plan. Manag. 2012,131, 24–35. [CrossRef]
37.
Shope, C.L.; Maharjan, G.R.; Tenhunen, J.; Seo, B.; Kim, K.; Riley, J.; Arnhold, S.; Koellner, T.; Ok, Y.S.;
Peiffer, S.; et al. Using the SWAT model to improve process descriptions and define hydrologic partitioning
in South Korea. Hydrol. Earth Syst. Sci. 2014,18, 539–557. [CrossRef]
38.
Cho, K.H.; Pachepsky, Y.A.; Kim, M.; Pyo, J.; Park, M.H.; Kim, Y.M.; Kim, J.W.; Kim, J.H. Modeling seasonal
variability of fecal coliform in natural surface waters using the modified SWAT. J. Hydrol.
2016
,535, 377–385.
[CrossRef]
39.
Kim, J.P.; Jung, I.; Park, K.W.; Yoon, S.K.; Lee, D. Hydrological utility and uncertainty of multi-satellite
precipitation products in the mountainous region of South Korea. Remote Sens. 2016,8, 608. [CrossRef]
40.
Smith, P.J.; Panziera, L.; Beven, K.J. Forecasting flash floods using data-based mechanistic models and NORA
radar rainfall forecasts. Hydrol. Sci. J. 2014,59, 1343–1357. [CrossRef]
41.
KOWACO. The Pre-Investigation Report for Groundwater Resources; Korea Water Resources Corporation:
Daejeon, Korea, 1993; p. 340.
42.
Korea Meteorological Administration. Annual Report 2016; Korea Meteorological Administration: Seoul,
Korea, 2016; p. 49.
43.
Kim, J.S.; Jain, S.; Yoon, S.K. Warm season streamflow variability in the Korean Han River Basin: Links with
atmospheric teleconnections. Int. J. Climatol. 2012,32, 635–640. [CrossRef]
44.
Lee, K.S.; Bong, Y.S.; Lee, D.; Kim, Y.; Kim, K. Tracing the sources of nitrate in the Han River watershed in
Korea, using δ15N-NO3and δ18O-NO3values. Sci. Total Environ. 2008,395, 117–124. [CrossRef] [PubMed]
45.
Heo, B.-H.; Kim, K.-E.; Kang, S.-G. Removals of noises from automatic weather station data and radial velocity
data of doppler weather radar using modified median filter. J. Korean Meteorol. Soc. 1999,35, 127–135.
46.
Li, D.; Christakos, G.; Ding, X.; Wu, J. Adequacy of TRMM satellite rainfall data in driving the SWAT
modeling of Tiaoxi catchment (Taihu lake basin, China). J. Hydrol. 2018,556, 1139–1152. [CrossRef]
Water 2018,10, 642 23 of 23
47.
Huffman, G.J.; Adler, R.F.; Bolvin, D.T.; Gu, G.; Nelkin, E.J.; Bowman, K.P.; Hong, Y.; Stocker, E.F.; Wolff, D.B.
The TRMM multi-satellite precipitation analysis (TMPA): Quasi-global, multi-year, combined-sensor
precipitation estimates at fine scale. J. Hydrometeorol. 2007,8, 38–55. [CrossRef]
48. Sorooshian, S.; Hsu, K.L.; Gao, X.; Gupta, H.V.; Imam, B.; Braithwaite, D. Evaluation of PERSIANN system
satellite-based estimates of tropical rainfall. Bull. Am. Meteorol. Soc. 2000,81, 2035–2046. [CrossRef]
49.
Ashouri, H.; Hsu, K.L.; Sorooshian, S.; Braithwaite, D.K.; Knapp, K.R.; Cecil, L.D.; Nelson, B.R.; Prat, O.P.
PERSIANN-CDR: Daily precipitation climate data record from multisatellite observations for hydrological
and climate studies. Bull. Am. Meteorol. Soc. 2015,96, 69–83. [CrossRef]
50.
Neitsch, S.L.; Arnold, J.G.; Kiniry, J.R.; Williams, J.R. Soil and Water Assessment Tool Theoretical Documentation
Version 2009; Texas Water Resources Institute: College Station, TX, USA, 2011.
51.
Abbaspour, K.C.; Johnson, C.A.; van Genuchten, M.T. Estimating uncertain flow and transport parameters
using a sequential uncertainty fitting procedure. Vadose Zone J. 2004,3, 1340–1352. [CrossRef]
52.
Yang, J.; Reichert, P.; Abbaspour, K.C.; Xia, J.; Yang, H. Comparing uncertainty analysis techniques for a
SWAT application to the Chaohe Basin in China. J. Hydrol. 2008,358, 1–23. [CrossRef]
53.
Narsimlu, B.; Gosain, A.K.; Chahar, B.R.; Singh, S.K.; Srivastava, P.K. SWAT model calibration and uncertainty
analysis for streamflow prediction in the Kunwari River Basin, India, using sequential uncertainty fitting.
Environ. Process. 2015,2, 79–95. [CrossRef]
54.
Wu, H.; Chen, B. Evaluating uncertainty estimates in distributed hydrological modeling for the Wenjing
River watershed in China by GLUE, SUFI-2, and ParaSol methods. Ecol. Eng. 2015,76, 110–121. [CrossRef]
55.
Khoi, D.N.; Thom, V.T. Parameter uncertainty analysis for simulating streamflow in a river catchment of
Vietnam. Glob. Ecol. Conserv. 2015,4, 538–548. [CrossRef]
56.
Willmott, C.J.; Matsuura, K. Advantages of the mean absolute error (MAE) over the root mean square error
(RMSE) in assessing average model performance. Clim. Res. 2005,30, 79–82. [CrossRef]
57.
Baik, J.; Choi, M. Spatio-temporal variability of remotetely sensed precipitation data from COMS and TRMM:
Case study of Korean peninsula in East Asia. Adv. Space Res. 2015,56, 125–1138. [CrossRef]
58.
Moriasi, D.N.; Arnold, J.G.; Van Liew, M.W.; Bingner, R.L.; Harmel, R.D.; Veith, T.L. Model evaluation guidelines
for systematic quantification of accuracy in watershed simulations. Trans. ASABE 2007,50, 885–900. [CrossRef]
59. Dai, A. Precipitation characteristics in eighteen coupled climate models. J. Clim. 2006,19, 4605–4630. [CrossRef]
60.
Dinku, T.; Chidzambwa, S.; Ceccato, P.; Connor, S.J.; Ropelewski, C.F. Validation of high-resolution satellite
rainfall products over complex terrain. Int. J. Remote Sens. 2008,29, 4097–4110. [CrossRef]
61.
Ma, L.; Ascough Ii, J.C.; Ahuja, L.R.; Shaffer, M.J.; Hanson, J.D.; Rojas, K.W. Root Zone Water Quality Model
sensitivity analysis using Monte Carlo simulation. Trans. ASAE 2000,43, 883–895. [CrossRef]
62.
Hromadka, T.V.; McCuen, R.H. Uncertainty estimates for surface runoff models. Adv. Water Resour.
1998
,11,
2–14. [CrossRef]
63.
Maskey, S.; Guinot, V.; Price, R.K. Treatment of precipitation in rainfall-runoff modelling: A fuzzy set
approach. Adv. Water Resour. 2004,27, 889–898. [CrossRef]
64.
Jones, P.D.; Lister, D.H.; Wilby, R.L.; Kostopoulou, E. Extended riverflow reconstructions for England and
Wales, 1865–2002. Int. J. Climatol. 2006,25, 219–231. [CrossRef]
65.
Andreassian, V.; Perrin, C.; Michel, C.; Usart-Sanchez, I.; Lavabre, J. Impact of imperfect rainfall knowledge
on the efficency and the parameters of watershed models. J. Hydrol. 2001,250, 206–223. [CrossRef]
66.
Gebregiorgis, A.S.; Hossain, F. Understanding the dependence of satellite rainfall uncertainty on topography
and climate for hydrologic model simulation. IEEE Trans. Geosci. Remote Sens.
2013
,51, 704–718. [CrossRef]
67.
Xu, S.G.; Niu, Z.; Shen, Y. Understanding the dependence of the uncertainty in a satellite precipitation data
set on the underlying surface and a correction method based on geogrphically weighted reggression. Int. J.
Remote Sens. 2014,35, 6508–6521. [CrossRef]
68.
Bitew, M.M.; Gebremichael, M. Assessment of satellite rainfall products for streamflow simulation in medium
watersheds of the Ethiopian highlands. Hydrol. Earth Syst. Sci. 2011,15, 1147–1155. [CrossRef]
69.
Artan, G.; Gadain, H.; Smith, J.L.; Asante, K.; Bandaragoda, C.J.; Verdin, J.P. Adequacy of satellite derived
rainfall data for streamflow modeling. Nat. Hazards 2007,43, 167–185. [CrossRef]
©
2018 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access
article distributed under the terms and conditions of the Creative Commons Attribution
(CC BY) license (http://creativecommons.org/licenses/by/4.0/).
