remote sensing

Article
The Impact of Satellite Soil Moisture Data Assimilation
on the Hydrological Modeling of SWAT in a Highly
Disturbed Catchment
Yongwei Liu 1, * , Wei Cui 2 , Zhe Ling 3 , Xingwang Fan 1 , Jianzhi Dong 4 , Chengmei Luan 5 , Rong Wang 1 ,
Wen Wang 6 and Yuanbo Liu 1

1 Key Laboratory of Watershed Geograpic Sciences, Nanjing Institute of Geography and Limnology,
Chinese Academy of Sciences, Nanjing 210008, China
2 Nanjing Hydraulic Research Institute, Nanjing 210029, China
3 Water Resources Department of Jiangsu Province, Nanjing 210029, China
4 School of Earth System Science, Tianjin University, Tianjin 300072, China
5 Hydrology and Water Resources Investigation Bureau of Jiangsu Province, Nanjing 210027, China;
[email protected]
6 State Key Laboratory of Hydrology-Water Resources and Hydraulic Engineering, Hohai University,
Nanjing 210098, China
* Correspondence: [email protected]

Abstract: The potential of satellite soil moisture (SM) in improving hydrological modeling has been
addressed in synthetic experiments, but it is less explored in real data cases. Here, we investigate
the added value of Soil Moisture and Passive (SMAP) and Advanced Scatterometer (ASCAT) SM
data to distributed hydrological modeling with the soil and water assessment tool (SWAT) in a highly
human disturbed catchment (126, 486 km2 ) featuring a network of SM and streamflow observations.
The investigation is based on the ensemble Kalman filter (EnKF) considering SM errors from satellite
data using the triple collocation. The assimilation of SMAP and ASCAT SM improved the surface
Citation: Liu, Y.; Cui, W.; Ling, Z.; (0–10 cm) and rootzone (10–30 cm) SM at >70% and > 50% stations of the basin, respectively. However,
Fan, X.; Dong, J.; Luan, C.; Wang, R.;
the assimilation effects on distributed streamflow simulation of the basin are un-significant and not
Wang, W.; Liu, Y. The Impact of
robust. SM assimilation improved the simulated streamflow at two upstream stations, while it
Satellite Soil Moisture Data
deteriorated the streamflow at the remaining stations. This can be largely attributed to the poor
Assimilation on the Hydrological
vertical soil water coupling of SWAT, suboptimal model parameters, satellite SM data quality, humid
Modeling of SWAT in a Highly
Disturbed Catchment. Remote Sens.
climate, and human disturbance to rainfall-runoff processes. This study offers strong evidence
2024, 16, 429. https://doi.org/ of integrating satellite SM into hydrological modeling in improving SM estimation and provides
10.3390/rs16020429 implications for achieving the added value of remotely sensed SM in streamflow improvement.

Academic Editors: Pierfranco

Keywords: soil moisture; data assimilation; SWAT; a disturbed catchment
Costabile, John Kalogiros, Venkatesh
Merwade and Jochen E. Schubert

Received: 20 October 2023

Revised: 9 January 2024 1. Introduction
Accepted: 12 January 2024
Soil moisture (SM) plays a crucial role in the energy and water cycle between land and
Published: 22 January 2024
atmosphere [1]. It not only exerts a strong control on the partitioning of incoming radiation
into latent and sensible heat fluxes, but also determines the partitioning of precipitation
into surface runoff and infiltration [2,3]. In hydrological modeling, rainfall-runoff processes
Copyright: © 2024 by the authors.
in a catchment are largely controlled by the initial state of SM [4]. Thus, an accurate
Licensee MDPI, Basel, Switzerland.
SM estimation is deemed to have great potential to improve hydrological simulation and
This article is an open access article prediction within a catchment [5].
distributed under the terms and The integration of SM observations in hydrological modeling via data assimilation
conditions of the Creative Commons (DA) techniques is considered a promising approach to improve SM estimation and rainfall-
Attribution (CC BY) license (https:// runoff modeling [6–9]. DA techniques provide an integrated framework to analyze all
creativecommons.org/licenses/by/ sources of uncertainties from modeling and observation, including the errors of model
4.0/). forcing inputs, model parameters, and model structures, as well as the observation errors

Remote Sens. 2024, 16, 429. https://doi.org/10.3390/rs16020429 https://www.mdpi.com/journal/remotesensing

Remote Sens. 2024, 16, 429 2 of 18

from in situ measurements and remote sensing [10]. The assimilation of high-quality in
situ SM data in hydrological modeling has been proven to benefit the SM and streamflow
estimation [11–13]. Nevertheless, the benefits are limited by sparse field measurement sites
with lacking capacity to fully account for the spatial heterogeneity of SM.
Remote sensing provides large-scale spatially and temporally continuous SM data.
The ever-improving observation and retrieval techniques and thereby data quality have
largely promoted remote sensing applications in DA in recent decades [14–20]. Currently,
the potential of remote sensing SM assimilation on the improvement in SM and runoff
estimation in hydrological modeling has been validated in synthetic experiments [21–26].
However, the performance of real DA remains largely unexplored. There is still no con-
sensus on the efficacy of remotely sensed SM for hydrological simulation and prediction.
Brocca et al. [27] assimilated the Advanced Scatterometer (ASCAT) SM and the Advanced
Microwave Scanning Radiometer for Earth observations (AMSR-E) SM into a distributed
rainfall-runoff model (MISDc) in six basins of Italy, Luxembourg, France, and the US. They
found a general positive impact on runoff prediction, but no improvement in mountainous
and snow-dominated regions. Alvarez-Garreton et al. [16] improved hydrological predic-
tion in a semiarid catchment by assimilating the AMSR-E SM into a probability distributed
model. Similar conclusions were obtained by Lievens et al. [17] who assimilated the Soil
Moisture Ocean Salinity (SMOS) SM into the Variable Infiltration Capacity (VIC) model over
the Murray Darling basin, Australia. Corato et al. [28] found that the DA performance of
remotely sensed SM was vegetation and time dependent for a catchment in the UK, whereas
Nayak et al. [29] underscored the significance of model structure in the DA performance.
Massari et al. [30] demonstrated that the remote sensing SM data quality and the accurate
model and observation error assumption and estimation have a significant impact on the
assimilation results. In view of the current unknown modelling and observation error,
the immature model structure, and the varying data quality of satellite SM, the efficacy of
remote sensing SM DA in hydrological modeling requires further investigation [31–33].
This study investigated the impact of the Soil Moisture Active and Passive (SMAP)
and the ASCAT SM DA on the distributed hydrological simulation with the soil and water
assessment tool (SWAT) model in a watershed highly disturbed by human activities in
China. The selection of the SWAT model considers its wide applicability in catchment
hydrological modeling and the unexplored performance of satellite SM retrievals [8,21,23].
The investigation is based on a robust data assimilation approach of the ensemble Kalman
filter (EnKF), considering the observation errors based on the triple collocation (TC) method.
Our focus is on the impact of the remotely sensed surface SM DA on profile SM and
streamflow estimation in SWAT for a highly disturbed catchment.

2. Methodology
2.1. SWAT Model
SWAT is a physically based basin scale distributed model, widely used to predict the
impacts of land management practices on water, sediment, and agricultural chemical yields
in large, complex catchments with varying land use, soil, and topography over long periods
of time [34]. For modeling purposes, a catchment is geographically partitioned into a
number of sub-catchments or subbasins; then, each subbasin is further divided into several
Hydrological Response Units (HRUs) with unique land cover, soil type, and slope steepness.
HRUs are the basic units for land phase of hydrological cycle modeling including surface
runoff generation, evapotranspiration, soil water routing, and groundwater generation. SM
plays a critical role in the rainfall-runoff process. SM modulates the redistribution of water
infiltrated or percolated to the soil profile, which can largely impact evapotranspiration,
surface runoff, and groundwater generation directly or indirectly, and finally river flow
through overland and channel flow concentration. The water balance equation for each
soil layer is as follows:

SW t,ly = SW t−1,ly + ∆w perc,ly − Qlat,ly − Ea,ly (1)

Remote Sens. 2024, 16, 429 3 of 18

where SWt− 1,ly and SWt,ly are the soil water content (mm) at the start and end of the day
for layer ly; ∆wperc,ly is the net water percolation received in layer ly; Qlat,ly is the lateral
flow generated from layer ly; and Ea,ly is the evapotranspiration drawn from layer ly.
The vertical movement of soil water in unsaturated soil layers is dominated by the
evapotranspiration process. Potential evapotranspiration (PET) is first calculated with
the Penman-Monteith equation [35]. The potential soil water evaporation is a function of
PET and leaf area index, and the potential plant uptake can also be estimated from the
Penman-Monteith equation. The total potential soil water evaporation and plant uptake
is allocated to each soil layer with the depth distribution function. The actual soil water
extraction is constrained by the available soil water of a given layer and not allowed to
be compensated from another layer. This deficiency is made up by a soil and a plant
compensation factor to adjust the depth distribution of the potential soil water evaporation
and ideal plant uptake.

2.2. The Ensemble Kalman Ffilter (EnKF)

The ensemble Kalman filter (EnKF) is proposed based on the theory of the linear
Kalman filter by incorporating the Monte Carlo method to generate a state ensemble
representing the probability distribution of the state, for example SM [36,37]. The predicted
f
state ensemble Xt at time t is given by the following:
 
f
Xt = F Xta−1, It + µt µt ∼ N 0, δ2


(2)

where F indicates the SWAT model; Xta−1 represents the updated state ensemble at time
t − 1. In this study, Xta−1 includes the three-layer profile SM for each subbasin (Table 1).
It indicates the model forcing inputs (e.g., precipitation and temperature) at time t. µt is the
stochastic perturbation of the forecast state and represents the modeling error, assumed to
f
have a normal distribution with zero mean and a variance of δ2 . The state update of Xt is
obtained by the following:
h  i
f f
Xta = Xt + Kt Yt − H Xt (3)

where Yt is the observation ensemble at time t, which is generated by adding a stochastic

perturbation with a normal distribution of zero mean and a variance of σ2 on the rescaled
remote sensing SM retrievals. Yt covers the SM observations of all remote sensing grids that
overlapped the subbasin being assimilated. H represents the observation operator to map
model states to observations. In this study, H is simplified as a matrix of 3 × m constructed
by 1 (first column) and 0, where m is the number of remote sensing grids overlapped the
subbasin being assimilated. Kt is the Kalman gain matrix at time t, which indicates the
weight of modeling and observation. Kt is estimated based on the forecast and observation
error covariance:
Kt = Mcs,t ( Ms,t + Rs,t )−1 (4)
f
where Mcs,t is the cross error covariance of the predicted state Xt and the measurement
f
prediction H(Xt ) at time t, Ms,t and Rs,t are the error covariance of the measurement
prediction and the observations at time t, respectively.

3. Study Area and Data

3.1. Study Area
The study catchment is located in the upper Huai River basin of China (Figure 1).
It covers an area of 126,486 km2 with an elevation ranging from −11 to 2122 m. The
catchment is in the transitional zone between the northern warm temperate semi-humid
and the southern subtropical humid monsoon climate. The mean annual precipitation is
884 mm, of which over 60% falls in June to September. The average air temperature is 15 °C.
The dominant land covers are agriculture (69.9%), forest (11.5%), and residential medium
 Soil Profile
CLAY SILT SAND ROCK
Soil Name Layers Depth Stratification Area (%)
(%) (%) (%) (%)
(cm) (cm)
HEAVYCLAY 3 100 67 21 12 7 0–10–30–100 0.01
Remote Sens. 2024, 16, 429 CLAY 3 100 49.54 29.23 21.23 4.77 0–10–30–100 4 of 18 18.19
CLAYLOAM 3 100 30 31.25 38.75 7 0–10–30–100 7.29
SILTLOAM 3 100 21 50 29 10 0–10–30–100 14.49
LOAM
density land (12.5%),3 with significant
100 20.51
anthropogenic41.16 influences
38.34 (e.g., 9.44irrigation,
0–10–30–100
reservoir) 48.13
SANDYCLAYLOAM
on the basin’s hydrology.3 100 23 24 53 5 0–10–30–100 0.75
SANDYLOAM 3 100 10.25 13.375 76.38 9.125 0–10–30–100 2.13
LOAMYSAND 3 100 8.1 11.7 80.2
Table 1. Soil classification and its area proportion in upper Huai River basin. 10 0–10–30–100 7.64
SAND 3 100 4.2 6 89.8 10 0–10–30–100 0.25
SHUITI
Soil Depth CLAY 2 SILT100 0
SAND 0ROCK 0 0
Profile 0–10–30 1.13
Soil Name Layers Area (%)
(cm) (%) (%) (%) (%) Stratification (cm)
3. Study Area and Data
HEAVYCLAY 3 100 67 21 12 7 0–10–30–100 0.01
CLAY 3 100 3.1. Study Area
49.54 29.23 21.23 4.77 0–10–30–100 18.19
CLAYLOAM 3 100 30 The study 31.25catchment is located in 7the upper Huai
38.75 River basin of China
0–10–30–100 7.29 (Figure 1). It
SILTLOAM 3 100 21
covers an area50of 126,486 km292 with an elevation
10 0–10–30–100
ranging from −11 to 2122 m. 14.49
The catchment
LOAM 3 100 20.51 41.16
is in the transitional 38.34
zone between the 9.44northern warm
0–10–30–100 48.13
temperate semi-humid and the
SANDYCLAYLOAM 3 100 23
southern 24
subtropical humid 53 monsoon climate.
5 The 0–10–30–100
mean annual precipitation 0.75 is 884 mm,
SANDYLOAM 3 100 10.25 13.375
of which over 60% falls 76.38 9.125
in June to September. The 0–10–30–100
average air temperature 2.13is 15 ℃. The
LOAMYSAND 3 100 8.1
dominant land 11.7covers are80.2 10
agriculture (69.9%), forest0–10–30–100
(11.5%), and residential 7.64medium den-
SAND 3 100 4.2 6 with significant
sity land (12.5%), 89.8 10
anthropogenic 0–10–30–100
influences 0.25reservoir) on
(e.g. irrigation,
SHUITI 2 100 0 0
the basin’s hydrology. 0 0 0–10–30 1.13

Figure 1. Basic information of the upper Huai River basin including location, elevation, reaches,
Figure 1. Basic information of the upper Huai River basin including location, elevation, reaches,
location of the meteorological,
location of theprecipitation, and
meteorological, hydrological
precipitation, andstations, andstations,
hydrological the subbasin delineation
and the subbasin delinea-
in SWAT modeling. tion in SWAT modeling.

3.2. Data Preparation for SWAT Model Building

The SWAT model building requires meteorological forcing and underlying land sur-
face data. The meteorological forcing data include precipitation, maximum and minimum
air temperature, solar radiation, wind speed, and relative humidity. The precipitation
data were collected from 63 precipitation and meteorological stations within the catch-
ment (Figure 1). The other meteorological variables were recorded from 12 meteorological
stations. The land surface data include terrain, land use, and soil category. The terrain
data were obtained from the digital elevation model (DEM) of the basin, download from
the 90 m Shuttle Radar Topography Mission (SRTM) (https://search.earthdata.nasa.gov/,
accessed on 1 October 2022) (Figure 1). The land use land cover data were extracted from
the1km remote sensing based Chinese national land use map (Figure 2a). The catchment
is predominantly covered by agricultural land, mixed forest, and residential land. The
soil category data were extracted from the 1km Harmonized World Soil Database (HWSD)
 (Figure 1). The other meteorological variables were recorded from 12 meteorological sta-
tions. The land surface data include terrain, land use, and soil category. The terrain data
were obtained from the digital elevation model (DEM) of the basin, download from the
90m Shuttle Radar Topography Mission (SRTM) (https://search.earthdata.nasa.gov/, ac-
Remote Sens. 2024, 16, 429 cessed on 1 October 2022) (Figure 1). The land use land cover data were extracted from
5 of 18
the1km remote sensing based Chinese national land use map (Figure 2a). The catchment
is predominantly covered by agricultural land, mixed forest, and residential land. The soil
category data were extracted from the 1km Harmonized World Soil Database (HWSD)
database (Figuredatabase
2b). The soil texture,
(Figure 2b). The profile stratification,
soil texture, and area proportion
profile stratification, of this of this
and area proportion
basin were listedbasin
in Table
were 1. Based
listed on these
in Table aforementioned
1. Based meteorological
on these aforementioned and underly-
meteorological and underly-
ing surface data,ing
the upperdata,
surface HuaitheRiver
upperbasin
Huai was
Riverdivided
basin wasinto 37 subbasins
divided and further
into 37 subbasins and further
delineated
delineated into 107 HRUs.into 107 HRUs.

Figure 2. (a) LandFigure 2. (a)

use and Land
land use and
cover, land
and (b)cover, and (b) soil
soil category in category in the
the upper upper
Huai Huai
River RiverAGRL
basin. basin. AGRL
(69.9%), FRST (11.5%), PAST (3.2%), URMD (12.5%), URML (0.006%), and WATR (2.9%) represent
(69.9%), FRST (11.5%), PAST (3.2%), URMD (12.5%), URML (0.006%), and WATR (2.9%) represent the
the general agricultural land, mixed forest, pasture, residential medium density land, residential
general agricultural land, mixed
medium and low forest, pasture,
density residential
land, and medium density land, residential medium
water, respectively.
and low density land, and water, respectively.
3.3. Remote Sensing and In Situ SM Data
3.3. Remote Sensing and In Situ SM Data
The SMAP L3 and ASCAT L2 SM data were used in this study. The SMAP SM (V8)
The SMAP is L3anand ASCAT
estimate L2 surface
of the SM dataSMwere
withinused
top in this
5 cm ofstudy. The SMAP
soil column based onSM the(V8) is dual
baseline
an estimate of the surface
channel SM within
algorithm [38]. top 5 cm
Here, the of soil column
passive based on the
L-band radiometer baseline
36 km dualproduct
resolution
channel algorithm was[38].
used. The ASCAT
Here, SM was
the passive retrieved
L-band from the ASCAT
radiometer backscatterproduct
36 km resolution measurements
was using
used. The ASCAT a time-series-based
SM was retrieved change detection
from approach
the ASCAT [39,40]. ASCAT
backscatter is a real-aperture
measurements usingradar in-
a time-series-based strument
changemeasuring radar
detection backscatter
approach at C-band
[39,40]. in VV is
ASCAT polarization and scans
a real-aperture the Earthʹs
radar
instrument measuring radar backscatter at C-band in VV polarization and scans the Earth’s with a
surface in descending and ascending overpasses [41]. ASCAT provides SM data
spatial resolution of 25 km and a revisit time of 1–3 days. The SMAP SM was first
surface
Remote Sens. 2024, 16, x FOR in descending
PEER REVIEW and ascending overpasses [41]. ASCAT provides SM data 6 ofwith
20 a
resampled to 25 km using the nearest neighboring method to make it consistent with the
spatial resolution of 25 km and a revisit time of 1–3 days. The SMAP SM was first resampled
ASCAT SM. The spatial distribution of the resampled SMAP and ASCAT grids is shown
to 25 km using the nearest
in Figure
neighboring method to make it consistent with the ASCAT SM.
3a. were validated against the in situ 10 cm SM values, considering the fact
satellite retrievals
The spatial distribution of the resampled
SM observations SMAP and
were collected from ASCAT grids
51 in situ is shown
stations in Figure
with varying 3a. lengths
record
that SM in the top 5 cm and 10 cm layers have marginal differences [42,43].
between 2016 and 2018 (hydrographic office of Henan and Shandong province in China)
(Figure 3a,b). The ground-based data were used to validate the satellite SM DA efficacy
on SM estimation. The validation was between the grid-based SM estimation and the in
situ SM observation falling in a given grid pixel. In situ SM data were observed at the
depths of 10 cm, 20 cm and 40 cm. According to the stratification scheme of soil profile,
we obtained 30 cm SM values based on linear interpolation. Model simulations and

Figure 3. (a) Location of the 51 soil moisture observation stations and the spatial distribution of the
Figure 3. (a) Location of the 51 soil moisture observation stations and the spatial distribution of the
remote sensing raster data of SMAP and ASCAT, and (b) the probability distribution of the data
remote sensingrecord
raster data The
lengths. of SMAP andpixels
colored grid ASCAT, anda single
represent (b) thedayprobability distribution
of SMAP retrievals of2016).
(2 February the data
record lengths. The colored grid pixels represent a single day of SMAP retrievals (2 February 2016).
4. Data Assimilation (DA) Setup
4.1. Triple Collocation (TC) Based Error Analysis
TC analysis is a method to estimate the total random error variances of three collo-
cated measurements of the same geophysical variable [44]. It allows the intercomparison
of errors obtained for three independent datasets after scaling into the same space. Cur-
rently, TC is one of the most widely used error evaluation methods for SM products in the
absence of in situ SM measurements [45–48]. In this study, the three independent SM da-
tasets are SMAP, ASCAT, and SWAT modelled SM, respectively. To ensure unbiased er-
Remote Sens. 2024, 16, 429 6 of 18

SM observations were collected from 51 in situ stations with varying record lengths
between 2016 and 2018 (hydrographic office of Henan and Shandong province in China)
(Figure 3a,b). The ground-based data were used to validate the satellite SM DA efficacy on
SM estimation. The validation was between the grid-based SM estimation and the in situ
SM observation falling in a given grid pixel. In situ SM data were observed at the depths of
10 cm, 20 cm and 40 cm. According to the stratification scheme of soil profile, we obtained
30 cm SM values based on linear interpolation. Model simulations and satellite retrievals
were validated against the in situ 10 cm SM values, considering the fact that SM in the top
5 cm and 10 cm layers have marginal differences [42,43].

4. Data Assimilation (DA) Setup

4.1. Triple Collocation (TC) Based Error Analysis
TC analysis is a method to estimate the total random error variances of three collocated
measurements of the same geophysical variable [44]. It allows the intercomparison of errors
obtained for three independent datasets after scaling into the same space. Currently, TC is
one of the most widely used error evaluation methods for SM products in the absence of in
situ SM measurements [45–48]. In this study, the three independent SM datasets are SMAP,
ASCAT, and SWAT modelled SM, respectively. To ensure unbiased errors from the three
independent sources, SMAP and ASCAT SM data were rescaled to the reference model
estimates of SWAT by matching the cumulative distribution function (CDF) of satellite SM
retrievals to that of the modelled SM [49,50]. The CDF matching was achieved by ranking
datasets of satellite derived and modelled SM and fitting a five-order polynomial to the
differences. Under the assumption that the three independent datasets are linearly related
to the true value with additive random error in TC, the error variance (σ2 ) of each dataset
can be estimated from the temporal variance and covariance between datasets, respectively,
as follows [45]:
Q x,y Q x,z
σx2 = ( Q x,x − ) (5)
Qy,z
Q x,y Qy,z
σy2 = ( Qy,y − ) (6)
Q x,z
Qz,y Q x,z
σz2 = ( Qz,z − ) (7)
Q x,y
where x, y and z represent the rescaled SMAP, ASCAT and SWAT SM, respectively, and
Q denotes the temporal variance and covariance between the three datasets.

4.2. EnKF Implementation in SWAT Model

DA aims to merge forecasted model states (i.e., the background) and observations
considering their respective uncertainties to minimize the error variance in the updated
(analyzed) model states. The EnKF requires an appropriate ensemble of model states to
represent the background error. In this study, the ensemble of SM model states was gener-
ated by an integrative perturbation of model forcing inputs of precipitation, temperature,
and model simulated SM. Precipitation at each station was perturbed by a multiplicative
lognormal error with a mean of 1.0 and a standard deviation of 0.3 combing the error setting
in previous research (e.g., [25,51]) and the robustness of EnKF in error processing. The max-
imum and minimum air temperature were perturbed by an additive normally distributed
error with a standard deviation of 1.0 ◦ C [18,21]. The modelled SM states were perturbed
by an additive normally distributed error with a standard deviation of 0.02 m3 /m3 for the
surface layer (0–10 cm) and rootzone (10–30 cm), and a standard deviation of 0.01 m3 /m3
for the deep layer (30–100 cm).
In this study, the SM DA was based on the calibrated SWAT model considering ex-
clusively the state update. The implementation of the calibrated SWAT model with error
perturbations on precipitation, temperature, and simulated SM was regarded as reference
(Openloop). Preprocessing was implemented with satellite SM retrievals before DA. Sys-
 perturbations on precipitation, temperature, and simulated SM was regarded as reference
(Openloop). Preprocessing was implemented with satellite SM retrievals before DA. Sys-
tematic biases between satellite and modeled SM were corrected using the CDF rescaling
method after the spatial mapping of the grid pixels and subbasins. The rescaled satellite
Remote Sens. 2024, 16, 429 SM was assimilated into SWAT modeling at each time step with available EnKF-based SM
7 of 18
state updates and TC-based error characterization. Figure 4 provides the schematic over-
view of the Openloop, DA preprocessing, and EnKF scenario.
The DA period was 2016–2018 after the model calibration of 2010–2015. The model
tematic biases between
calibrationsatellite and
is based on themodeled
streamflow SM were corrected
observations using
of all six the CDF
hydrological rescaling
stations availa-
blespatial
method after the at the mainstream,
mapping i.e. Xixian,
of the gridHuaibin,
pixels Wangjiaba, Lutaizi,The
and subbasins. Bengbu and Xiaoliuxiang
rescaled satellite
(Figureinto
SM was assimilated 1). The distributed
SWAT parameter
modeling optimization
at each time stepfollows
withthe principleEnKF-based
available from upstream
to downstream controlling stations. The calibration processes combine the auto-calibra-
SM state updates and TC-based error characterization. Figure 4 provides the schematic
tion of the Sequential Uncertainty Fitting [52] and the manual fine-tuning method. The
overview of thesensitive
Openloop, DA preprocessing, and EnKF scenario.
parameters that need to be adjusted are listed in Table 2.

Figure 4. Schematic overview of the model Openloop, EnKF simulation, and the DA preprocessing
Figure 4. Schematic overview of the model Openloop, EnKF simulation, and the DA
scheme.
preprocessing scheme.

The DA period was 2016–2018 after the model calibration of 2010–2015. The model
calibration is based on the streamflow observations of all six hydrological stations available
at the mainstream, i.e., Xixian, Huaibin, Wangjiaba, Lutaizi, Bengbu and Xiaoliuxiang
(Figure 1). The distributed parameter optimization follows the principle from upstream to
downstream controlling stations. The calibration processes combine the auto-calibration of
the Sequential Uncertainty Fitting [52] and the manual fine-tuning method. The sensitive
parameters that need to be adjusted are listed in Table 2.

Table 2. SWAT model parameters in calibration.

Hydrological Modules Paramters Descriptions

CN2 moisture condition II curve number
surface runoff
surlag surface runoff lag coefficient
esco soil evaporation compensation coefficient
evapotranspiration epco plant uptake compensation factor
canmx maximum canopy storage
AWCly available water capacity of soil layer
soil water dynamics
Ksat saturated hydraulic conductivity of soil
ground water aqshthr,q threshold water level in shallow aquifer for base flow
αgw baseflow recession constant
βrev revap coefficient
aqshthr,rvp threshold water level in shallow aquifer for revap
Kch effective hydraulic conductivity of channel
channel routing
n Manning’s n value for main channels

4.3. Evaluation Metrics

The DA performance was evaluated by comparing the EnKF and Openloop results
against the in situ SM and streamflow measurements. The bias (Bias), percent bias (PBias),
Pearson’s correlation coefficient (r2 ), root mean square error (RMSE), unbiased RMSE
(ubRMSE), and Nash–Sutcliffe coefficient of efficiency (NSE) were used as metrics. Mean-
while, the effectiveness criterion (EFF) [53] and the normalized error reduction index (NER)
were used to directly assess the streamflow improvement in the EnKF over the Openloop.
EFF and NER can be obtained as follows:
Remote Sens. 2024, 16, 429 8 of 18

 2
∑nt=1 QtEnKF − Qobs
t
EFF(%) = 100·(1 −
) (8)
obs 2
∑nt=1 Qol
t − Qt
RMSE EnKF
NER(%) = 100·(1 − ) (9)
RMSEol
where n is the total time step of DA; QtEnKF , Qol obs are the model predicted
t , and Qt
streamflow in the EnKF and the Openloop scenario; and the observed discharge at time
t, RMSEEnKF and RMSEol are the RMSE of the modelled streamflow in the EnKF and
Openloop scenario, respectively.

5. Results
5.1. Catchment Applicability of SWAT Model
Figure 5 compares the SWAT model simulated against the observed daily streamflow
at the catchment outlet (Xiaoliuxiang station) in model calibration and validation phases.
The hydrographs of both streamflow data generally match well during the calibration
and validation phases. The evaluation statistics at the catchment outlet show good consis-
tency of the modeling and observational results: the PBias values within ±10%, r2 > 0.8,
NSE > 0.65, and RMSE < 600 m3 /s (station Xiaoliuxiang in Table 3). The evaluation statis-
tics at the other five interior hydrological stations (Table 3) indicate the relatively good
performance of the SWAT model in the distributed rainfall-runoff simulation of the upper
Huai River basin. For both calibration and validation phases, almost all stations have PBias
values within ± 15%, r2 > 0.8, and NSE > 0.6 except for the station Xixian at the most
Remote Sens. 2024, 16, x FORThe
upstream. PEERcomparatively
REVIEW high consistent hydrograph and the good statistics reflect the 9 of 20

good applicability of the SWAT model in the study catchment.

Figure 5.(Observation)
Figure 5. The observed The observed (Observation) and simulated
and simulated (Simulation)
(Simulation) daily streamflow
daily streamflow at catchment
at catch-
ment outlet of theoutlet
upperof the upper Huai River basin in the model calibration (2010–2016) and validation (2016–
Huai River basin in the model calibration (2010–2016) and validation
2018) phases.
(2016–2018) phases.
Table 3. Statistics for the model simulated streamflow at the six hydrological stations in the model
calibration
Table 3. Statistics for the modelandsimulated
validation phases.
streamflow at the six hydrological stations in the model
calibration and validation phases. Calibration (2010–2015) Validation (2016–2018)
Hydrological Stations
NSE PBias (%) r2 RMSE (m3/s) NSE PBias (%) r2 RMSE (m3/s)
Calibration (2010–2015) Xixian 0.41 5.81 0.65Validation
123.60 (2016–2018)
−0.15 21.75 0.48 162.11
Hydrological Huaibin 0.63 21.89 0.80 126.92 0.66 −4.78 0.81 180.49
RMSE −23.07 RMSE
Stations NSE PBias (%) r2 Wangjiaba 0.68 NSE
6.73 0.83PBias137.44
(%) r2
0.58 0.78 265.47
(m3 /s)
Lutaizi 0.64 −5.11 0.80 358.82 0.68
3 /s)
0.12 (m0.83 503.70
Bengbu 0.66 11.61 0.82 462.01 0.74 4.49 0.86 550.57
Xixian 0.41 5.81 0.65 123.60 −0.15 21.75 0.48 162.11
Xiaoliuxiang 0.66 9.50 0.82 464.63 0.73 6.07 0.86 578.43
Huaibin 0.63 21.89 0.80 126.92 0.66 −4.78 0.81 180.49
Wangjiaba 0.68 6.73 0.83 137.44 0.58 −23.07 0.78 265.47
Lutaizi 0.64 −5.11 5.2.
0.80Modeling358.82
and Observation 0.68
Errors 0.12 0.83 503.70
Bengbu 0.66 11.61 0.82 Figure 6 presents
462.01 the rescaled
0.74 SMAP, ASCAT
4.49 and SWAT-modeled
0.86 SM errors at each
550.57
Xiaoliuxiang 0.66 9.50 grid
0.82 pixel and the error probability
464.63 0.73 distribution
6.07 of all grid
0.86pixels within
578.43the catchment
based on the TC method (Figure 3). The error standard deviation of SMAP SM and ASCAT
SM vary between 0.01 and 0.03 m3/m3, and the error standard deviation of SWAT-modelled
SM is less than 0.02 m3/m3. The random error of remote sensing SM is substantially larger
than that of the modelled SM. Large error differences are present in the grid pixels in space
with varying topography, land cover and soil (Figure 2), in particular for the ASCAT SM
and SWAT modelled SM. Meanwhile, large error differences are present in the three SM
datasets for a given grid pixel, demonstrating the potential benefit of SM data fusion and
 NSE PBias (%) r RMSE (m /s) NSE PBias (%) r RMSE (m /s)
Xixian 0.41 5.81 0.65 123.60 −0.15 21.75 0.48 162.11
Huaibin 0.63 21.89 0.80 126.92 0.66 −4.78 0.81 180.49
Wangjiaba 0.68 6.73 0.83 137.44 0.58 −23.07 0.78 265.47
Lutaizi 0.64 −5.11 0.80 358.82 0.68 0.12 0.83 503.70
Remote Sens. 2024, 16, 429 Bengbu 0.66 11.61 0.82 462.01 0.74 4.49 0.86 550.57
9 of 18
Xiaoliuxiang 0.66 9.50 0.82 464.63 0.73 6.07 0.86 578.43

5.2. Modeling and Observation Errors

5.2. Modeling and Observation Errors
Figure 6 presents the rescaled SMAP, ASCAT and SWAT-modeled SM errors at each
Figure 6 presents the rescaled SMAP, ASCAT and SWAT-modeled SM errors at each
grid pixel and the error probability distribution of all grid pixels within the catchment
grid pixel and the error probability distribution of all grid pixels within the catchment
based on
based on the
the TC
TC method
method (Figure 3). The
(Figure 3). The error
error standard
standard deviation
deviation of of SMAP
SMAP SM SM and
and ASCAT
ASCAT
SM vary between 0.01 and 0.03 m 3/m3, and the error standard deviation of SWAT-modelled
3 3
SM vary between 0.01 and 0.03 m /m , and the error standard deviation of SWAT-modelled
SM is
SM is less
less than
than 0.02
0.02 m
m33/m
/m.3 .The
3
Therandom
randomerror
errorofofremote
remotesensing
sensing SM SM is
is substantially
substantially larger
larger
than that of the modelled SM. Large error differences are present
than that of the modelled SM. Large error differences are present in the grid in the grid pixels
pixels in
in space
space
with varying
with varying topography,
topography, land
land cover
cover and
and soil
soil (Figure
(Figure 2),
2), in
in particular
particular forfor the
the ASCAT
ASCAT SM SM
and SWAT
and SWAT modelled
modelled SM.SM. Meanwhile,
Meanwhile, large
large error
error differences
differences areare present
present in the three
in the three SM
SM
datasets for a given grid pixel, demonstrating the potential benefit of SM
datasets for a given grid pixel, demonstrating the potential benefit of SM data fusion anddata fusion and
assimilation. The
assimilation. The very
very small
small values
values of
of the
the SWAT-SM
SWAT-SM error at aa few
error at few grids
grids are
are likely
likely caused
caused
by the unchanged modelling SM, for example the
by the unchanged modelling SM, for example the water body. water body.

Figure 6. Standard deviation of the error (Std of error) for the rescaled remote sensing (SMAP and
Figure 6. Standard deviation of the error (Std of error) for the rescaled remote sensing (SMAP and
ASCAT) and SWAT modelled soil moisture based on the TC method (i.e. the σx , σy and σz) at each
ASCAT) and SWAT modelled soil moisture based on the TC method (i.e., the σx , σy and σz ) at each
grid pixel (a) and the error probability density (b).
grid pixel (a) and the error probability density (b).
5.3. Data
5.3. Data Assimilation
Assimilation Effects
Effects on
on Soil
Soil Moisture
Moisture
Figure 7 compares the data accuracy of the simulated SM in EnKF and Openloop sce-
narios based on the 51 in situ SM stations available to illustrate the SMAP SM DA effects on
profile SM modeling. For the surface layer (0–10 cm), most stations have bias values within
−0.1 and 0.1 m3 /m3 , r values within 0.2–0.6, RMSE values within −0.05–0.15 m3 /m3 , and
ubRMSE values within 0.04–0.06 m3 /m3 for the Openloop, EnKF and rescaled SMAP SM
(Figure 7a–d). The corresponding statistics of the rootzone (0–30 cm) SM are comparable to
those of the surface layer (Figure 7e–h). SMAP SM DA marginally reduces the biases of
SM modeling (Figure 7a), yet significantly increases the correlations between the model
simulated and in situ SM (Figure 7b). Meanwhile, the RMSE and ubRMSE values are largely
reduced through SMAP SM assimilation (Figure 7c,d). Similar to the surface layer, the
improvement in the rootzone is mainly manifested in the improved r, RMSE, and ubRMSE
values (Figure 7f–h). Generally, SMAP SM is superior to SWAT simulated SM in Openloop
at ~47% stations (Table 4). SMAP SM has a relatively larger (0.7%) bias, but a higher (7%)
r, and a slightly lower (−0.8%) RMSE and (−2%) ubRMSE than Openloop. Integrating
SMAP SM largely improves SWAT simulated SM for surface and rootzone layers at over
70% of stations. The overall improvements are −10.5%/−9.5% for Bias, 11.7%/8.8% for r,
−9.5%/−10.1% for RMSE, and −7.9%/−10.9% for ubRMSE in the surface/rootzone.
 Remote Sens. 2024, 16, x FOR PEER REVIEW 11
Remote Sens. 2024, 16, 429 10 of 18

Figure 7. Bias Figure

(a,e), r 7. BiasRMSE
(b,f), (a,e), (c,g),
r (b,f),ubRMSE
RMSE (c,g),
(d,h)ubRMSE (d,h) ofSMAP
of the rescaled the rescaled
SM andSMAP SM and the estim
the estimated
soil
soil moisture in moistureand
Openloop in Openloop
EnKF based andonEnKF based SM
the SMAP on the SMAP SMfor
assimilation assimilation
the surfacefor the cm)
(0–10 surface (0–10
and rootzone
and rootzone (10–30 cm). (10–30 cm).

Table 4. Comparison of bias, r, RMSE and ubRMSE for total 51 SM stations based on SMAP and
ASCAT DA.

Scenarios Statistics Bias r RMSE ubRMSE

% of stations with SMAP
37 53 47 49
SMAP DA superiors to Openloop
(10 cm) % of stations with EnKF
27 63 88 76
improves upon the Openloop
SMAP DA % of stations with EnKF
29 71 82 84
(30 cm) improves upon the Openloop
% of stations with ASCAT
33 27 25 25
ASCAT DA superiors to Openloop
(10 cm) % of stations with EnKF
33 31 69 57
improves upon the Openloop
ASCAT DA % of stations with EnKF
24 41 76 57
(30 cm) improves upon the Openloop

Figure 8 shows the bias, r, RMSE, and ubRMSE values for the EnKF and Openloop
SM based on ASCAT SM assimilation. ASCAT SM is inferior to SWAT simulated SM in
Remote Sens. 2024, 16, 429 11 of 18

Openloop at most stations, particularly for r, RMSE and ubRMSE (Figure 8a–d). Only ~28%
stations show better ASCAT SM than Openloop SM (Table 3). It seems that the ASCAT SM
DA effects on SM modeling are not robust for all stations. The ASCAT SM retrievals with
high data quality tend to improve SM simulation (Figure 8a–h). Statistically, the ASCAT SM
Remote Sens. 2024, assimilation improved the surface and rootzone SM at over 50% of stations, in particular
16, x FOR PEER REVIEW 12 of 20
for the RMSE and ubRMSE values (Table 3). The RMSE and ubRMSE are improved by
−4.9%/−9.3% and −2.9%/−5.3% in the surface/rootzone.

Figure
Figure 8. Bias (a,e), 8. BiasRMSE
r (b,f), (a,e), r (c,g),
(b,f), RMSE (c,g),(d,h)
ubRMSE ubRMSE (d,h)
of the of the ASCAT
rescaled rescaled ASCAT
SM andSM
theand the estimated
estimated
soil moisture in Openloop and EnKF based on the ASCAT SM assimilation for the surface (0–10 cm)
soil moisture in Openloop and EnKF based on the ASCAT SM assimilation for the surface (0–10 cm)
and rootzone (10–30 cm).
and rootzone (10–30 cm).
Table 4. Comparison of bias, r, RMSE and ubRMSE for total 51 SM stations based on SMAP and
Overall, SMAP
ASCAT SM
DA. DA largely improved the surface (0–10 cm) and rootzone (10–30 cm)
SM estimation in the upper Huai River basin. The effects of ASCAT SM assimilation are
Scenarios Statistics Bias r RMSE ubRMSE
moderate, caused by its poor data quality in this
% of stations basin.superiors to Openloop
with SMAP 37 53 47 49
SMAP DA (10 cm)
% of stations with EnKF improves upon the Openloop
27 63 88 76
5.4. Data Assimilation
SMAP DA Impacts
(30 cm) on Streamflow
% of stations with EnKF improves upon the Openloop
29 71 82 84
% of stations with ASCAT superiors to Openloop 33 27 25
Table 5 shows
ASCATthe
DA effectiveness
(10 cm) criterion (EFF) and the normalized error reduction25
% of stations with EnKF improves upon the Openloop 33 31 69 57
index (NER) values
ASCATof
DAthe
(30simulated
cm) % of streamflow based
stations with EnKF on SMAP
improves upon theand ASCAT24SM
Openloop 41 assimila-
76 57
tion at five interior hydrological stations (Xixian, Huaibin, Wangjiaba, Lutaizi, and Bengbu)
and the station atData
5.4. the catchment
Assimilation outlet
Impacts(Xiaoliuxiang).
on Streamflow SMAP and ASCAT SM assimilation
improves the modelled streamflow at two
Table 5 shows the effectiveness upstream interior
criterion (EFF)stations, i.e., the Huaibin
and the normalized and
error reduction
Wangjiaba satations. However,
index (NER) thethe
values of DA does not
simulated improve based
streamflow the modelled
on SMAPstreamflow
and ASCAT atSMtheassimi-
other four stations, in five
lation at particular
interiorfor the downstream
hydrological stationsstation
(Xixian,(Bengbu)
Huaibin, and the catchment
Wangjiaba, Lutaizi, and
Bengbu) and the station at the catchment outlet (Xiaoliuxiang). SMAP and ASCAT SM
assimilation improves the modelled streamflow at two upstream interior stations, i.e. the
Huaibin and Wangjiaba satations. However, the DA does not improve the modelled
streamflow at the other four stations, in particular for the downstream station (Bengbu)
Remote Sens. 2024, 16, 429 12 of 18

outlet. It means that both SMAP SM and ASCAT SM assimilation produce un-robust
impacts on the distributed streamflow modeling in the Huai River basin.

Table 5. Statistics of the SMAP and ASCAT soil moisture assimilation impacts on streamflow
simulation at the six hydrological stations in upper Huai River basin.

Hydrological EnKF-SMAP EnKF-ASCAT

Stations EFF (%) NER (%) EFF (%) NER (%)
Xixian −2.8 −5.8 −4.5 −9.1
Huaibin 2.4 4.7 1.9 3.8
Wangjiaba 5.3 10.3 4.1 8.1
Lutaizi 0.3 0.6 −2.9 −5.9
Bengbu −5.8 −12.0 −5.3 −10.9
Xiaoliuxiang −8.7 −18.1 −5.6 −11.4

Figure 9 compares the SMAP SM assimilation-based and the Openloop streamflow at

the Wangjiaba and Xiaoliuxiang stations. The DA has a marginal impact on streamflow
modeling in the study catchment. The EnKF streamflow does not show much difference
from the Openloop streamflow (the red and blue lines in Figure 9a,b). The marginal effects
in streamflow can be partly impacted by the semihumid and humid climate conditions
of the catchment. SM content is relatively high throughout the year, with catchment
averages in most days exceeding 0.3 m3 /m3 , in particular for the rootzone and deep layers
(Figure 10). Thereby, it leaves a limited room for improvement of streamflow simulation
by enhancing SM status. Moreover, both positive and negative effects are present (see the
light green and pink bands in Figure 10), suggesting the un-robust DA performance, in
particular for the basin outlet. The un-robust effects might be related to the sub-optimal
model parameters and the unsatisfied model structure, which cannot effectively account for
Remote Sens. 2024, 16, x the
FOR disturbances
PEER REVIEW of the complex and dense human activities on the rainfall-runoff14processes
of 20

of the basin, for example the reservoir influences (see the gray rectangle in Figure 10b).

Figure 9. The daily

Figure streamflow
9. The at theatstation
daily streamflow of of
the station (a)(a)Wangjiaba and(b)(b)
Wangjiaba and Xiaoliuxiangs
Xiaoliuxiangs in EnKF
in EnKF and and
Openloop Openloop
scenariosscenarios
of SMAP of SMAP SMassimilation
SM data data assimilation over
over 2016–2018.
2016–2018.
 RemoteSens.
Remote Sens.2024,
2024,16,
16,429
x FOR PEER REVIEW 15 of 18
13 of 20

Figure 10. The three-layer soil moisture of the basin between 2016 and 2018.
Figure 10. The three-layer soil moisture of the basin between 2016 and 2018.

6.6. Discussion
Discussion
Wedemonstrated
We demonstrated thethe benefits
benefits of high-quality
of high-quality satellitesatellite SM to SWAT
SM to improve improve SWAT
modeling
modeling
of of the
SM at both SM surface
at bothlayer
the surface layer and
and rootzone. Therootzone.
improvedThe improved SM
SM estimation estimation
confirmed the
confirmed the
significance of significance of integrating
integrating remotely sensedremotely sensed SM in
SM in distributed distributed modeling
hydrological hydrological of
modeling of SWAT. Nevertheless, the improvements in profile SM
SWAT. Nevertheless, the improvements in profile SM estimation are not robust, particularly estimation are not
robust,
for thoseparticularly
based on thefor those based
low-quality on the
ASCAT SMlow-quality
data. Analyzed ASCAT SM data.isAnalyzed
SM accuracy deterioratedSM
accuracy
at is deteriorated
several stations, at several
even when SMAPstations, even when(Figures
SM is assimilated SMAP SM 7 and is assimilated (Figure 7
8, Table 3). Different
and Figure
model 8, Table 3). Different
error parameterizations model
still cannot errorimprove
largely parameterizations
the DA effects still
on cannot largely
SM estimation.
We found error
improve the DA parameterizations
effects on SM had no significant
estimation. We foundimpacterror
on SM estimation as differences
parameterizations had no
in ubRMSE impact
significant values onareSMmarginal (see vertical
estimation legendininubRMSE
as differences Figure 11a,b).
values Itare seems that low
marginal (see
ubRMSE values can be obtained in the cases of high SM states and large
vertical legend in Figure 11a,b). It seems that low ubRMSE values can be obtained in the precipitation errors.
However,
cases of highlarger
SM SM perturbations
states at each time
and large precipitation step likely
errors. However,cause biases
larger SMtoperturbations
streamflow
modeling
at each timeduestep
to the truncation
likely errorstoofstreamflow
cause biases the updatedmodeling
SM by the duesaturation water capacity,
to the truncation errors
particularly
of the updated for deep
SM bysoil
thelayers. Thus,
saturation the error
water standard
capacity, deviation
particularly forwasdeepsetsoil 0.02 m3Thus,
as layers. /m3
3 3
and 0.01 m /m , respectively, for surface/rootzone and deep-layer SM.
 Remote Sens. 2024, 16, x FOR PEER REVIEW 16 of 20

Remote Sens. 2024, 16, 429 14 of 18

was set as 0.02 m3/m3 and 0.01 m3/m3, respectively, for surface/rootzone and deep-layer
SM.

Figure11.
Figure 11.The
TheubRMSE
ubRMSEof ofthe
theSMAP
SMAPdata
dataassimilation-based
assimilation-based SM
SMat
atthe
the(a)
(a)surface
surface(0–10
(0–10cm)
cm)and
and
(b) rootzone (10–30 cm).
(b) rootzone (10–30 cm).

Thelimited
The limitedsuccess
successof ofactual
actualsurface
surfaceSM SMassimilation
assimilationon onprofile
profilesoil
soilwater
waterestimation
estimation
in SWAT has also been found in previous studies (e.g. [21,51]),
in SWAT has also been found in previous studies (e.g., [21,51]), largely attributed to largely attributed tothe
the
inadequate vertical soil water coupling strength of SWAT. The correlation
inadequate vertical soil water coupling strength of SWAT. The correlation between surface between surface
androotzone
and rootzoneSM SMisissubjected
subjectedto tounderestimation
underestimationand andoverestimation,
overestimation,varying
varyingwith withsites
sites
(Figure12),
(Figure 12),despite
despitethe the consideration
consideration of of evaporation
evaporation andand evapotranspiration
evapotranspiration compensa-
compensation
tion factors
factors in SWAT in SWAT
to accountto account
for varyingfor varying vertical coupling
vertical coupling in real
in real cases. cases. Decoupling
Decoupling (underes-
(underestimation)
timation) limits the
limits the EnKF’s EnKF’s
ability ability
to update SMtostates
update SM states
of deeper of deeper
layers, layers, while
while over-coupling
over-coupling (overestimation)
(overestimation) leads to overcorrectionleads to overcorrection
of deeper SM of deeper
states [54].SM states [54].
Attempts Attempts
to improve
the SWAT inherent
to improve the SWAT vertical
inherentcoupling
vertical ofcoupling
soil layersof were madewere
soil layers using the ensemble
made of soil
using the ensem-
Remote Sens. 2024, 16, x FOR PEER REVIEW 17 of 20
storages andstorages
ble of soil the coupledand thesoil coupled
moisturesoil analytical
moisture relationship
analyticalbased on the EnKF
relationship based[55,56].
on the
These
EnKFattempts marginally
[55,56]. These attemptsbenefit the covariance-based
marginally state updates in EnKF.
benefit the covariance-based However,
state updates in
itEnKF.
should be acknowledged
However, it should be that SWAT still lacks
acknowledged that aSWAT
physical
stillmechanism
lacks a physicalin itsmechanism
soil water
also
in itsasoil
dynamic promising
module, approach to couple
so integrating
water dynamic module, aso big data with
physically
integrating based hydrological
module, based
a physically formodels
example by separately
module, theforRichards
example ac-
counting
equation, for
shouldmultiple
be a sources
candidate of uncertainty
solution to under
this
the Richards equation, should be a candidate solution to this issue. the
issue. DA framework [62–64].
In addition to the vertical coupling issue of the model, suboptimal model calibration
is also responsible for the un-robust improvements in SM, particularly for streamflow. It
is a challenging task to seek optimal parameters for distributed hydrological models, due
to the large number of parameters and the well-known parameter equifinality issue
[57,58]. In the current study, we can hardly fine-tune the best-for-all parameters for the
rainfall-runoff module based on data from six discharge stations. Extreme streamflow
(low and high) is inclined to be more biased with suboptimal parameters [59]. Thus, the
pre-parameter calibration is highly needed to improve streamflow modeling in the state-
update DA scheme. The combined state-parameter update likely provides a viable solu-
tion (e.g. [60,61,8]).
Human manipulation of river discharges partly explains the poor DA performance
on streamflow simulation. We obtained improved streamflow in the upstream regions
with less intensive human activities, whereas neutral impacts or even deterioration of ac-
curacy were found in the densely populated downstream regions (Figure 1, Table 3). Hu-
man alterations of hydrological processes increase the uncertain effectiveness of DA. It
requires a more sophisticated model structure to account for these human activities. The
humid and subhumid climate of the catchment are also responsible for the marginal DA
effects12.
Figure onRelationship
streamflow.ofThe relatively
the soil
soil moisture high SM throughout
correlation in the year tends to leave limited
Figure 12. Relationship of the moisture correlation in SWAT
SWAT modeling
modeling and and in
in situ
situ observation
observation
room forthe
between streamflow
surface layer improvement
and rootzone.by enhancing SM status. Additionally, data quality of
between the surface layer and rootzone.
satellite SM is another hinderance to improve streamflow simulations based on DA. Fu-
7. Conclusion
ture enhancement
In addition to the onvertical
satellitecoupling
SM observation
issue of the and retrieval
model, is essential
suboptimal model forcalibration
its effective is
application
also responsiblein hydrological
for the modeling
un-robust and
improvements prediction.
in SM, Assimilating
particularly
The DA impacts of satellite SM on distributed hydrological modeling of SWAT are multisource
for streamflow. data
It is ais
challenging task to seek optimal parameters for distributed hydrological
investigated based on the EnKF considering the observation error using the TC method. models, due to the
large number of parameters
The investigation focused onand the DAthe well-known
effects on theparameter
surface and equifinality
rootzone SM, issueas [57,58].
well as theIn
the current study,
streamflow we canofhardly
estimation a highlyfine-tune
human thedisturbed
best-for-all parameters
catchment in for the rainfall-runoff
China, i.e. the upper
Huai River basin with dense in situ SM and streamflow observations. Our results show
the following:
(1) SWAT has good applicability in the daily rainfall-runoff simulation of the upper Huai
River basin. The PBias values are generally within ±15%, r2 > 0.8, and NSE > 0.60 at
Remote Sens. 2024, 16, 429 15 of 18

module based on data from six discharge stations. Extreme streamflow (low and high) is
inclined to be more biased with suboptimal parameters [59]. Thus, the pre-parameter cali-
bration is highly needed to improve streamflow modeling in the state-update DA scheme.
The combined state-parameter update likely provides a viable solution (e.g., [8,60,61]).
Human manipulation of river discharges partly explains the poor DA performance
on streamflow simulation. We obtained improved streamflow in the upstream regions
with less intensive human activities, whereas neutral impacts or even deterioration of
accuracy were found in the densely populated downstream regions (Figure 1, Table 3).
Human alterations of hydrological processes increase the uncertain effectiveness of DA. It
requires a more sophisticated model structure to account for these human activities. The
humid and subhumid climate of the catchment are also responsible for the marginal DA
effects on streamflow. The relatively high SM throughout the year tends to leave limited
room for streamflow improvement by enhancing SM status. Additionally, data quality
of satellite SM is another hinderance to improve streamflow simulations based on DA.
Future enhancement on satellite SM observation and retrieval is essential for its effective
application in hydrological modeling and prediction. Assimilating multisource data is
also a promising approach to couple big data with hydrological models by separately
accounting for multiple sources of uncertainty under the DA framework [62–64].

7. Conclusions
The DA impacts of satellite SM on distributed hydrological modeling of SWAT are
investigated based on the EnKF considering the observation error using the TC method.
The investigation focused on the DA effects on the surface and rootzone SM, as well as
the streamflow estimation of a highly human disturbed catchment in China, i.e., the upper
Huai River basin with dense in situ SM and streamflow observations. Our results show
the following:
(1) SWAT has good applicability in the daily rainfall-runoff simulation of the upper Huai
River basin. The PBias values are generally within ±15%, r2 > 0.8, and NSE > 0.60 at
the validation stations.
(2) The random error standard deviation of SMAP and ASCAT SM varies between
0.01 m3 /m3 and 0.03 m3 /m3 for all grid pixels within the basin. Large error differ-
ences are present in the SM datasets and in different grid pixels.
(3) SMAP SM DA largely improves the surface and rootzone SM estimation. Nevertheless,
ASCAT SM DA gains mixed impacts on performance in SM estimation, primarily due
to poor data quality.
(4) The satellite SM DA does not improve streamflow simulation as effectively as SM
itself. The effects of SMAP and ASCAT SM assimilation on distributed streamflow
simulation are un-significant and not robust.
Overall, this study presents strong evidence of integrating satellite SM in hydrological
modeling in improving SM estimation and provides implications for achieving the added
value of remotely sensed SM in streamflow improvement. Satellite SM retrievals do have
great capacity to improve the distributed hydrological simulation of SWAT, in particular for
SM estimation. However, current benefits from DA effects are still limited by the vertical
soil water coupling of SWAT, suboptimal model parameters, satellite SM data quality,
humid climate, and the human disturbances on catchment rainfall-runoff process. In future,
more efforts should be paid to the improvement in model structure, satellite SM retrieval,
and multi-source data assimilation.

Author Contributions: Y.L. (Yongwei Liu): Conceptualization, data curation, formal analysis, investi-
gation, methodology, validation, visualization, writing—original draft, writing—review editing. W.C.
and X.F.: methodology, writing—review and editing. J.D., Y.L. (Yuanbo Liu) and W.W.: methodology,
conceptualization and supervision. Z.L., C.L. and R.W.: Data collection and processing. All authors
have read and agreed to the published version of the manuscript.
Remote Sens. 2024, 16, 429 16 of 18

Funding: This work was supported by the National Key R&D Program of China (2018YFE0105900),
the National Natural Science Foundation of China (42271040, 41901049, 41961134003), the Key
Deployment Projects of Sino-Africa Joint Research Center, Chinese Academy of Sciences (SAJC202106),
and the International Collaboration Program of Chinese Academy of Science (151542KYSB20200015).
Data Availability Statement: The data that support the findings is available upon reasonable request.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Wanders, N.; Bierkens, M.F.P.; Jong, S.M.; Roo, A.; Karssenberg, D. The benefits of using remotely sensed soil moisture in
parameter identification of large-scale hydrological models. Water Resour. Res. 2014, 50, 6874–6891. [CrossRef]
2. Seneviratne, S.I.; Corti, T.; Davin, E.L.; Hirschi, M.; Jaeger, E.B.; Lehner, I.; Orlowsky, B.; Teuling, A.J. Investigating soil
moisture–climate interactions in a changing climate: A review. Earth Sci. Rev. 2010, 99, 125–161. [CrossRef]
3. Liu, Y.W.; Liu, Y.B.; Wang, W.; Zhou, H.; Tian, L. Historical droughts manifest an abrupt shift to a wetter Tibetan Plateau. Hydrol.
Earth Syst. Sci. 2022, 26, 3825–3845. [CrossRef]
4. Brocca, L.; Melone, F.; Moramarco, T. On the estimation of antecedent wetness conditions in rainfall–runoff modelling. Hydrol.
Process. 2008, 22, 629–642. [CrossRef]
5. Massari, C.; Camici, S.; Ciabatta, L.; Brocca, L. Exploiting satellite-based surface soil moisture for flood forecasting in the
Mediterranean area: State update versus rainfall correction. Remote Sens. 2018, 10, 292. [CrossRef]
6. Crow, W.; Bindlish, R.; Jackson, T.J. The added value of spaceborn passive microwave soil moisture retrievals for forecasting
rainfall-runoff partitioning. Geophys. Res. Lett. 2005, 32. [CrossRef]
7. Leroux, D.J.; Pellarin, T.; Vischel, T.; Cohard, J.-M.; Gascon, T.; Gibon, F.; Mialon, A.; Galle, S.; Peugeot, C.; Seguis, L. Assimilation
of SMOS soil moisture into a distributed hydrological model and impacts on the water cycle variables over the Ouémé catchment
in Benin. Hydrol. Earth Syst. Sci. 2016, 20, 2827–2840. [CrossRef]
8. Liu, Y.W.; Wang, W.; Hu, Y. Investigating the impact of surface soil moisture assimilation on state and parameter estimation in
SWAT model based on the ensemble Kalman filter in upper Huai River basin. J. Hydrol. Hydromech. 2017, 65, 123–133. [CrossRef]
9. Yang, H.; Xiong, L.; Liu, D.; Cheng, L.; Chen, J. High spatial resolution simulation of profile soil moisture by assimilating
multi-source remote-sensed information into a distributed hydrological model. J. Hydrol. 2021, 597, 126311. [CrossRef]
10. Wang, D.; Chen, Y.; Cai, X. State and parameter estimation of hydrologic models using the constrained ensemble Kalman filter.
Water Resour. Res. 2009, 45, w11416. [CrossRef]
11. Aubert, D.; Loumagne, C.; Oudin, L. Sequential assimilation of soil moisture and streamflow data in a conceptual rainfall–runoff
model. J. Hydrol. 2003, 280, 145–161. [CrossRef]
12. Fu, X.; Zhang, Y.; Zhong, Q.; Lü, H.; Ding, Y.; Li, Z.; Yu, Z.; Jiang, X. Soil moisture estimation by assimilating in-situ and SMAP
surface soil moisture using unscented weighted ensemble Kalman filter. Water Resour. Res. 2023, 59, e2023WR034506. [CrossRef]
13. Yu, Z.; Fu, X.; Luo, L.; Lü, H.; Ju, Q.; Liu, D.; Kalin, A.D.; Huang, D.; Yang, C.; Zhao, L. One-dimensional soil temperature
simulation with Common Land Model by assimilating in situ observations and MODIS LST with the temperature simulation
with Common Land Model by assimilating in situ observations and MODIS LST with the ensemble particle filter. Water Resour.
Res. 2014, 50, 6950–6965. [CrossRef]
14. Sahoo, A.K.; Lannoy, G.J.M.D.; Reichle, R.H.; Houser, P.R. Assimilation and downscaling of satellite observed soil moisture
over the Little River Experimental Watershed in Georgia, USA. Adv. Water Resour. 2013, 52, 19–33. Available online: https:
//linkinghub.elsevier.com/retrieve/pii/S0309170812002357 (accessed on 9 January 2024). [CrossRef]
15. Ines, A.V.M.; Das, N.N.; Hansen, J.W.; Njoku, E.G. Assimilation of remotely sensed soil moisture and vegetation with a crop
simulation model for maize yield prediction. Remote Sens. Environ. 2013, 138, 149–164. [CrossRef]
16. Alvarez-Garreton, C.; Ryu, D.; Western, A.W.; Su, C.H.; Crow, W.T.; Robertson, D.E.; Leahy, C. Improving operational flood
ensemble prediction by the assimilation of satellite soil moisture: Comparison between lumped and semi-distributed schemes.
Hydrol. Earth Syst. Sci. 2015, 19, 1659–1676. [CrossRef]
17. Lievens, H.; Tomer, S.K.; Al Bitar, A.; De Lannoy, G.J.M.; Drusch, M.; Dumedah, G.; Hendricks Franssen, H.J.; Kerr, Y.H.; Martens,
B.; Pan, M.; et al. SMOS soil moisture assimilation for improved hydrologic simulation in the Murray Darling Basin. Australia.
Remote Sens. Environ. 2015, 168, 146–162. [CrossRef]
18. Liu, Y.W.; Wang, W.; Liu, Y. ESA CCI Soil Moisture Assimilation in SWAT for Improved Hydrological Simulation in Upper Huai
River Basin. Adv. Meteorol. 2018, 2018, 7301314. [CrossRef]
19. Khaki, M.; Hendricks Franssen, H.J.; Han, S.C. Multi-mission satellite remote sensing data for improving land hydrological
models via data assimilation. Sci. Rep. 2020, 10, 18791. [CrossRef]
20. Santis De, D.; Biondi, D.; Crow, W.T.; Camici, S.; Modanesi, S.; Brocca, L.; Massari, C. Assimilation of satellite soil moisture
products for river flow prediction: An extensive experiment in over 700 catchments throughout Europe. Water Resour. Res. 2021,
57, e2021WR029643. [CrossRef]
21. Chen, F.; Crow, W.T.; Starks, P.J.; Moriasi, D.N. Improving hydrologic predictions of a catchment model via assimilation of surface
soil moisture. Adv. Water Resour. 2011, 34, 526–536. [CrossRef]
Remote Sens. 2024, 16, 429 17 of 18

22. Crow, W.T.; Ryu, D. A new data assimilation approach for improving runoff prediction using remotely-sensed soil moisture
retrievals. Hydrol. Earth Sys. Sci. 2009, 13, 1–16. [CrossRef]
23. Han, E.; Merwade, V.; Heathman, G.C. Implementation of surface soil moisture data assimilation with watershed scale distributed
hydrological model. J. Hydrol. 2012, 416–417, 98–117. [CrossRef]
24. Reichle, R.H.; Crow, W.T.; Keppenne, C.L. An adaptive ensemble Kalman filter for soil moisture data assimilation. Water Resour.
Res. 2008, 44, 423. [CrossRef]
25. Xie, X.; Zhang, D. Data assimilation for distributed hydrological catchment modeling via ensemble Kalman filter. Adv. Water
Resour. 2010, 33, 678–690. [CrossRef]
26. Lei, F.; Huang, C.; Shen, H.; Li, X. Improving the estimation of hydrological states in the SWAT model via the ensemble Kalman
smoother: Synthetic experiments for the Heihe River Basin in northwest China. Adv. Water Resour. 2014, 67, 32–45. [CrossRef]
27. Brocca, L.; Moramarco, T.; Dorigo, W.; Wagner, W. Assimilation of satellite soil moisture data into rainfall-runoff modelling for
several catchments worldwide. In Proceedings of the 2013 IEEE International Geoscience and Remote Sensing Symposium-
IGARSS, Melbourne, VIC, Australia, 21–26 July 2013; pp. 2281–2284. [CrossRef]
28. Corato, G.; Matgen, P.; Fenicia, F.; Schlaffer, S.; Chini, M. Assimilating satellite derived soil moisture products into a distributed
hydrological model. In Proceedings of the 2014 IEEE Geoscience and Remote Sensing Symposium, Quebec City, QC, Canada,
13–18 July 2014; pp. 3315–3318. [CrossRef]
29. Nayak, A.K.; Biswal, B.; Sudheer, K.P. Role of hydrological model structure in the assimilation of soil moisture for streamflow
prediction. J. Hydrol. 2021, 598, 126465. [CrossRef]
30. Massari, C.; Brocca, L.; Tarpanelli, A.; Moramarco, T. Data assimilation of satellite soil moisture into rainfall-runoff modelling: A
complex recipe? Remote Sens. 2015, 7, 11403–11433. [CrossRef]
31. Fu, X.; Jiang, X.; Yu, Z.; Ding, Y.; Lü, H.; Zheng, D. Understanding the key factors that influence soil moisture estimation using
the unscented weighted ensemble Kalman filter. Agric. For. Meteorol. 2022, 313, 108745. [CrossRef]
32. Brocca, L.; Moramarco, T.; Melone, F.; Wagner, W.; Hasenauer, S.; Hahn, S. Assimilation of Surface-and Root-Zone ASCAT Soil
Moisture Products Into Rainfall–Runoff Modeling. IEEE Trans. Geosci. Remote Sens. 2015, 50, 2542–2555. [CrossRef]
33. Dorigo, W.A.; Gruber, A.; De Jeu, R.A.M.; Wagner, W.; Stacke, T.; Loew, A.; Kidd, R. Evaluation of the ESA CCI soil moisture
product using ground-based observations. Remote Sens. Environ. 2015, 162, 380–395. [CrossRef]
34. Neitsch, S.L.; Arnold, J.G.; Kiniry, J.R.; Williams, J.R. Soil and Water Assessment Tool Theoretical Documentation Version 2009; TR-406,
Texas Water Resources Institute Technical Report No.406; Texax A&M University: College Station, TX, USA, 2011; Available
online: http://swat.tamu.edu/media/99192/swat2009theory.pdf (accessed on 9 January 2024).
35. Monteith, J.L. Evaporation and the environment. In 19th Symposia of the Society for Experimental Biology: The State and Movement of
Water in Living Organisms; Cambridge University Press: London, UK, 1965; pp. 205–234.
36. Evensen, G. Sequential data assimilation with a nonlinear quasi-geostrophic model using Monte Carlo methods to forecast error
statistics. Journal of Geophysical methods to forecast error statistics. J. Geophys. Res.-Ocean 1994, 99, 10143–10162. [CrossRef]
37. Evensen, G. The Ensemble Kalman Filter: Theoretical formulation and practical implementation. Ocean Dynam. 2003, 53, 343–367.
[CrossRef]
38. O’Neill, P.E.; Chan, S.; Njoku, E.G.; Jackson, T.; Bindlish, R.; Chaubell, J. L3 Radiometer Global Daily 36 km EASE-Grid Soil Moisture,
Version 8; NASA National Snow and Ice Data Center Distributed Active Archive Center: Boulder, CO, USA, 2021. [CrossRef]
39. Wagner, W.; Lemoine, G.; Rott, H. A method for estimating soil moisture from ERS scatterometer and soil data. Remote Sens.
Environ. 1999, 70, 191–207. [CrossRef]
40. Naeimi, V.; Scipal, K.; Bartalis, Z.; Hasenauer, S.; Wagner, W. An improved soil moisture retrieval algorithm for ERS and METOP
scatterometer observations. IEEE Trans. Geosci. Remote Sens. 2009, 47, 1999–2013. [CrossRef]
41. Zeng, J.; Li, Z.; Chen, Q.; Bi, H.; Qiu, J.; Zou, P. Evaluation of remotely sensed and reanalysis soil moisture products over the
Tibetan Plateau using in-situ observations. Remote Sens. Environ. 2015, 163, 91–110. [CrossRef]
42. Shellito, P.J.; Small, E.E.; Livneh, B. Controls on surface soil drying rates observed by smap and simulated by the noah land
surface model. Hydrol. Earth Syst. Sci. 2018, 22, 1649–1663. [CrossRef]
43. Dong, J.Z.; Crow, W.T.; Tobin, K.J.; Cosh, M.H.; Bosch, D.D.; Starks, P.J.; Seyfried, M.; Collins, C.H. Comparison of microwave
remote sensing and land surface modeling for surface soil moisture climatology estimation. Remote Sens. Environ. 2020,
242, 111756. [CrossRef]
44. Stoffelen, A. Toward the true near-surface wind speed: Error modeling and calibration using triple collocation. J. Geophys. Res.
1998, 103, 7755–7766. [CrossRef]
45. McColl, K.A.; Vogelzang, J.; Konings, A.G.; Entekhabi, D.; Piles, M.; Stoffelen, A. Extended triple collocation: Estimating errors
and correlation coefficients with respect to an unknown target. Geophys. Res. Lett. 2014, 41, 6229–6236. [CrossRef]
46. Dorigo, W.; Wagner, W.; Albergel, C.; Albrecht, F.; Balsamo, G.; Brocca, L.; Chung, D.; Ertl, M.; Forkel, M.; Gruber, A. ESA CCI
Soil Moisture for improved Earth system understanding: State-of-the art and future directions. Remote Sens. Environ. 2017,
203, 185–215. [CrossRef]
47. Crow, W.; Van den Berg, M. An improved approach for estimating observation and model error parameters in soil moisture data
assimilation. Water Resour. Res. 2010, 46, W12519. [CrossRef]
48. Su, C.H.; Ryu, D.; Crow, W.T.; Western, A.W. Beyond triple collocation: Applications to soil moisture monitoring. J. Geophys.
Res.-Atmos. 2014, 119, 6419–6439. [CrossRef]
Remote Sens. 2024, 16, 429 18 of 18

49. Reichle, R.H.; Koster, R.D. Bias reduction in short records of satellite soil moisture. Geophys. Res. Lett. 2004, 31, L19501. [CrossRef]
50. Scipal, K.; Drusch, M.; Wagner, W. Assimilation of a ERS scatterometer derived soil moisture index in the ECMWF numerical
weather prediction system. Adv. Water Resour. 2008, 31, 1101–1112. [CrossRef]
51. Clark, M.P.; Rupp, D.E.; Woods, R.A.; Zheng, X.; Ibbitt, R.P.; Slater, A.G.; Schmidt, J.; Uddstrom, M.J. Hydrological data
assimilation with the ensemble Kalman filter: Use of streamflow observations to update states in a distributed hydrological
model. Adv. Water Resour. 2008, 31, 1309–1324. [CrossRef]
52. Abbaspour, K.; Johnson, C.; Van Genuchten, M.T. Estimating uncertain flow and transport parameters using a sequential
uncertainty fitting procedure. Vadose Zone J. 2004, 3, 1340–1352. [CrossRef]
53. Chen, H.; Yang, D.; Hong, Y.; Gourley, J.J.; Zhang, Y. Hydrological data assimilation with the ensemble square-root-filter: Use
of streamflow observations to update model states for real-time flash flood forecasting. Adv. Water Resour. 2013, 59, 209–220.
[CrossRef]
54. Kumar, S.V.; Reichle, R.H.; Koster, R.D.; Crow, W.T.; Peters-Lidard, C.D. Role of Subsurface Physics in the Assimilation of Surface
Soil Moisture Observations. J. Hydrometeor. 2009, 10, 1534–1547. [CrossRef]
55. Patil, A.; Ramsankaran, R. Improving streamflow simulations and forecasting performance of SWAT model by assimilating
remotely sensed soil moisture observations. J. Hydrol. 2017, 555, 683–696. [CrossRef]
56. Patil, A.; Ramsankaran, R. Improved streamflow simulations by coupling soil moisture analytical relationship in EnKF based
hydrological data assimilation framework. Adv. Water Resour. 2018, 121, 173–188. [CrossRef]
57. Beven, K.; Freer, J. Equifinality, data assimilation, and uncertainty estimation in mechanistic modelling of complex environmental
systems using the GLUE methodology. J. Hydrol. 2001, 249, 11–29. [CrossRef]
58. Abbaspour, K.C.; Rouholahnejad, E.; Vaghefi, S.; Srinivasan, R.; Yang, H.; Kløve, B. A continental-scale hydrology and water
quality model for Europe: Calibration and uncertainty of a high-resolution large-scale SWAT model. J. Hydrol. 2015, 524, 733–752.
[CrossRef]
59. Dong, J.Z.; Lei, F.N.; Crow, W. Land transpiration-evaporation partitioning errors responsible for modeled summertime warm
bias in the central United States. Nat. Commun. 2022, 13, 336. [CrossRef] [PubMed]
60. Xie, X.; Zhang, D. A partitioned update scheme for state-parameter estimation of distributed hydrologic models based on the
ensemble Kalman filter. Water Resour. Res. 2013, 49, 7350–7365. [CrossRef]
61. Sun, L.; Seidou, O.; Nistor, I. Data assimilation for streamflow forecasting: State-parameter assimilation versus output assimilation.
J. Hydrol. Eng. 2017, 22, 04016060. [CrossRef]
62. Sun, L.; Seidou, O.; Nistor, I.; Goïta, K.; Magagi, R. Simultaneous assimilation of in situ soil moisture and streamflow in the SWAT
model using the Extended Kalman Filter. J. Hydrol. 2016, 543, 671–685. [CrossRef]
63. Avellaneda, P.M.; Ficklin, D.L.; Lowry, C.S.; Knouft, J.H.; Hall, D.M. Improving hydrological models with the assimilation of
crowdsourced data. Water Resour. Res. 2020, 56, e2019WR026325. [CrossRef]
64. Azimi, S.; Dariane, A.B.; Modanesi, S.; Bauer-Marschallinger, B.; Bindlish, R.; Wagner, W.; Massari, C. Assimilation of Sentinel 1
and SMAP–based satellite soil moisture retrievals into SWAT hydrological model: The impact of satellite revisit time and product
spatial resolution on flood simulations in small basins. J. Hydrol. 2020, 581, 124367. [CrossRef]

Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual
author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to
people or property resulting from any ideas, methods, instructions or products referred to in the content.