agriculture

Article
Investigating Flood Impact on Crop Production under a
Comprehensive and Spatially Explicit Risk
Evaluation Framework
Xi Wang 1 , Zhanyan Liu 2 and Huili Chen 3, *

1 College of Urban and Environmental Sciences, Peking University, No.5 Yiheyuan Road, Haidian District,
Beijing 100871, China; [email protected]
2 Handan Hydrology and Water Resources Survey Bureau of Hebei Province, Handan 056001, China;
[email protected]
3 School of Architecture, Building and Civil Engineering, Loughborough University,
Loughborough LE11 3TU, UK
* Correspondence: [email protected]

Abstract: Due to the projected increased frequency of occurrence of extreme flood events, it is be-
coming increasingly important to pay attention to agricultural flood management. The middle and
lower reaches of the Yangtze River Basin (MLYRB), as one of the most important agricultural areas
in the world, frequently suffer from the ravages of long-duration extreme flood events. Compre-
hensive flood risk evaluation can provide important support for effective management strategies by
focusing on the combination of flood hazard and the consequences of flooding in areas exposed to
the inundation. Previous satellite-based flood disturbance detection methods intended for use in
single-cropping agricultural systems cannot be applied to the MLYRB with multi-cropping practices



and long-duration flood events. Additionally, comprehensive agricultural flood risk evaluations


traditionally neglect the characteristics of the impact of flooding with strong spatial and temporal
Citation: Wang, X.; Liu, Z.; Chen, H. variability. Thus, in this research, an integrated disturbance index (IDI) was developed to detect
Investigating Flood Impact on Crop
the impact of flood disturbance on crop growth, aiming to acquire a map of crop damage condition
Production under a Comprehensive
for a multi-cropping agricultural system with long-duration flood events that is spatially explicit
and Spatially Explicit Risk Evaluation
and has a sufficiently high spatial resolution. A coupled hydrological and 2D hydraulic model
Framework. Agriculture 2022, 12, 484.
parallelized using the GPU approach was employed to simulate flood flows, aiming at deriving
https://doi.org/10.3390/
agriculture12040484
sufficient meaningful detail at the local scale in terms of flood inundation patterns and processes over
the whole natural watershed. Additionally, a spatial map of the combined effects of flood hazard and
Academic Editor:
the consequences of flooding was used to investigate the relationship between flood characteristics
Tarendra Lakhankar
and associated loss extent with the random forest model. The comprehensive evaluation framework
Received: 25 January 2022 was applied for the 2010 flood event in the MLYRB. The evaluation results indicate that the detection
Accepted: 29 March 2022 results based on IDI are consistent with the governmental statistics, the most hard-hit areas in related
Published: 30 March 2022 reports, and the spatial characteristics of river floods. The coupled hydrological–hydraulic model
Publisher’s Note: MDPI stays neutral offers a clear picture of the flood characteristics over the whole basin, while simultaneously ensuring
with regard to jurisdictional claims in a sufficiently high spatial resolution. Our findings show that flood duration is the most important
published maps and institutional affil- predictor in predicting crop damage extent.
iations.
Keywords: flooding; crop production; disturbance detection index; two-dimensional hydraulic
model; spatial characteristics

Copyright: © 2022 by the authors.

Licensee MDPI, Basel, Switzerland.
This article is an open access article
1. Introduction
distributed under the terms and
conditions of the Creative Commons
Rainfall-induced flooding and accompanying waterlogging are agricultural disasters
Attribution (CC BY) license (https://
that occur frequently in many regions of the world and often cause remarkable damage to
creativecommons.org/licenses/by/ crop production [1]. Besides the direct rushing effect, the submergence creates complex
4.0/). abiotic stress in crops, including reduced light availability [2], oxygen depletion [3], and

Agriculture 2022, 12, 484. https://doi.org/10.3390/agriculture12040484 https://www.mdpi.com/journal/agriculture

Agriculture 2022, 12, 484 2 of 23

altered chemical characteristics of soil [4]. The combination of all these physical and
chemical alterations can substantially reduce crop stand and the growth and yield of
crops. Under current climate conditions, the reduction in crop yields owing to excess
rainfall events has been substantial, negatively influencing the grain supply and creating
food security [5]. The intensification of the hydrological cycle due to climate warming
is projected to alter the timing, magnitude, and frequency of extreme floods [6–8]. The
accompanying crop damage created by flooding is likely to be greater under future climate
conditions; consequently, the management of agricultural floods is expected to face more
challenges.
As one of the countries worst affected by flooding, China frequently suffers from the
ravages of extreme flood events. The Yangtze River is the longest river in China and the
third in the world in terms of its length and river flow. It is responsible for approximately
70% of China’s floods [9]. Historically, the Yangtze River Basin has been well known for
its frequent flood events, especially in its middle and lower reaches. Heavy rainfall in the
MLYRB due to strong and lengthy El Niño events brings about severe flooding and serious
loss of property and life. Among the recent largest floods are those of 1870, 1931, 1954, 1998,
and 2010. The 1954 flood event in the Yangtze River Basin inundated an area of 193,000 km2 ,
killed 33,169 people, and caused the displacement of 18,884,000 people [10]. The Yangtze
River Basin is affected by the joint influence of the tropical and subtropical East Asian
summer monsoons. Under the monsoonal climate, floods occur in the summer, especially
during June and July, when slow-moving cold fronts meet the humid and steady subtropical
air-mass and lead to excessive rainfall in the Yangtze River Basin [11]. These flood disasters
not only have serious consequences but are characterized by a high water level and long
duration (e.g., more than 70 days for the 1998 flood), especially in the MLYRB [12]. The
upper reaches of the Yangtze River are basically mountainous terrain, and the middle
and lower reaches are primarily a flattened floodplain. Large amounts of sediment (ca.
0.8 × 108 tons/a) are deposited in the middle reaches as a result of the decrease in slope
from the upstream mountain valley to the downstream flattened meandering river [13].
Along with the continuous deposition of sediment and embankment construction, the
current levee is 10–15 m higher than the outside areas along the networks, which makes
the flood evolution more complicated [14]. Additionally, the effects of the “suspending
river” lead to the surface of the floodplain being depressed and making it extremely
prone to becoming waterlogged. The high frequency and long duration of flood disasters,
in combination with poor water discharge, constitute a huge challenge to agricultural
production in the MLYRB. This is why flood disasters are the most serious form of natural
disaster in terms of agricultural production and account for 40% yield reductions in the
MLYRB [15].
Worse yet, obvious climate change and huge landscape changes occur in the Yangtze
River Basin. Some studies have investigated the statistical characteristics of historical
extreme precipitation in the Yangtze River Basin [11,16–19]. A significant increasing trend
in summer precipitation intensity over the MLYRB has been detected in the last 50 years.
Furthermore, due to global warming, the return periods of extreme precipitation events are
projected to become shorter in the MLYRB [20,21]. A significant increase in summer precip-
itation would lead to a higher possibility of the occurrence of flood hazards [22]. Besides
abnormally extreme rainfall events, the Yangtze River Basin has experienced extensive
landscape changes, including land use/cover changes (e.g., deforestation, urbanization,
and soil erosion) [23,24], the shrinkage of large lakes in the floodplain of the Yangtze
River Basin [25,26], and the construction of levees along riverbanks [12]. These landscape
changes caused by human activity may create extra pressure on the environment, increasing
the severity of major floods in the MLYRB [14,27,28]. Thus, the severity of this situation
prompts us to select the MLYRB as our study area, aiming to more deeply understand the
influence of flooding on crop production in order to cope with the projected more severe
and frequent floods.
Agriculture 2022, 12, 484 3 of 23

It is important to remember that the MLYRB is one of the most important agricultural
areas in the world and is famous for being the “Land of Rice and Fish”. The warm and
humid climate, abundant light and heat, and equal periods of rain and heat are beneficial
to agricultural production. The cropping systems used in the MLYRB are double-cropping
or triple-cropping in one year, with rape–rice–rice cropping primarily used in the southern
region and wheat–rice cropping primarily used in the northern region [29]. The farming
land is highly productive, and the crops planted are very diverse, with the major ones being
rice, wheat, cotton, rape, and silkworm. Thus, the spatial distribution of crop types and crop
rotations have obvious heterogeneity due to the soil natural regime and multiple cropping
system applied. In addition, flooding often has obvious temporal and spatial variations, as
it is influenced by local topographic and weather conditions; thus, the influence of flooding
variations is always localized [30]. Thus, given the complex agricultural conditions in the
MLYRB and the localized impact of flooding, it is difficult to derive the localized influence
of flooding on crop production over such a large basin.
Satellite remote sensing seems to be a feasible tool in terms of obtaining spatially
explicit maps of the localized impacts of (or damage caused by) flooding over large-scale
areas with a sufficiently high spatial resolution. Some studies have attempted to use the
difference in crop growth condition between pre-flood and post-flood periods [31–33] or
between the flood years and normal years of post-flood periods [34] from remote sensing
imagery for this purpose. Our previous study [35] uses the difference of vegetation index
between the flood year and harvest year to obtain the crop yield losses caused by flood
disaster. However, these extreme flood events in the Yangtze River are characterized by
a long duration (often more than a month). Over such long periods, crop systems may
naturally grow or die. The intra-annual and inter-annual variations in crop systems’ vege-
tation greenness are relatively large even without external disturbances [36,37]. Obviously,
the differences in crop growth conditions between pre-flood and post-flood periods or
between flood years and normal years will be affected by both flood disturbances and
natural variations, meaning that this method cannot be applied to the MLYRB. In another
previous study [38], the comparison of maximum crop greenness between the years with
and without flood disturbances was used in order to quantity the influence of flooding
on crop growth, avoiding including the effect of crop natural growth during the growing
period. Yet, the comparison of maximum greenness only applies to single-cropping sys-
tems and flood events occurring before crops reach maximum greenness. Thereby, a new
satellite-based index is developed in this study to identify the spatially explicit influence
of flooding on crop production for long-duration floods and multi-cropping agricultural
systems.
Besides the consequences of flooding, the importance of the comprehensive evaluation
of flood risk has been increasingly taken into account by combining data on hazard and
vulnerability of flood-affected areas [39,40]. At present, comprehensive evaluations of flood
risk are largely implemented through laboratory testing or model evaluation. Waterlogging
testing under controlled depth and duration offers basic information on the biophysical
response of crops to submersion [41–44]. However, such experimental findings have limited
value in extrapolation to the real world due to the extent of differences between places. To
avoid the limitations associated with laboratory testing, a variety of mathematical models
have been developed to examine the relationships between crop yield losses and flood
events. Mathematical models of this type mainly fall into two categories—i.e., statisti-
cal [30,45] and mechanistic [46]. Statistical models relate crop yield fluctuation to various
flood or meteorological variables by regression or other more sophisticated statistical meth-
ods. This type of statistical estimation can be used to quantify the effects of flooding on
agricultural production from a broad-scale perspective—e.g., the national or province/state
scale. However, these statistical methods are incapable of considering the influence of
localized flooding variations. Mechanistic models utilize the biophysical principles of
crop, soil, weather, and management interactions to predict the crop yield [47]. Due to the
high data demand for the determination of parameters associated with crops, soil, and
Agriculture 2022, 12, 484 4 of 23

management in mathematical terms, the data are often sparse; consequently, these mech-
anistic models cannot be applied to large-scale area. Thus, in order to comprehensively
evaluate agricultural flood risk, spatially explicit, temporally dynamic, and sufficiently
high-spatial-resolution maps of flood characteristics are needed.
Recently developed 2D hydraulic models offer the possibility of obtaining spatial
and temporal field information of flood evolution. Water flows are described in terms of
continuity and momentum principles and discretized over a grid of cells, which allows
the models to depict 2D dynamic flow fields based on the topography of the riverbed and
floodplain [48]. Thus, 2D hydraulic models can provide spatially explicit and temporally
dynamic pictures of flood characteristics for the whole computational domain. While
some existing 2D hydraulic models are capable of capturing inundation fields, often they
only show a very small basin [49–53] or a single river reach [48,54], or have a coarse
resolution [55] due to the high computational expense involved. One practicable approach
proposed to improve the computation speed of 2D hydraulic models is model parallel
computation with a graphics processing unit (GPU), which can provide catchment-scale
2D flood simulations at high resolutions involving millions of computational cells. Some
studies have made exploratory attempts and achieved feasible results [56–60]. Thus, the
accessibility of high-performance computing could make it possible to derive a spatially
explicit and temporally dynamic map of flood characteristics over a large-scale watershed,
ensuring the consistency of spatial resolution between flood damage evaluations based on
remote sensing imagery and 2D flood simulation. Additionally, it is possible to investigate
the highly localized impacts of flooding on crops—i.e., the flooding of individual fields—
and implement a comprehensive evaluation of the agricultural flood risk.
Therefore, this study aimed to comprehensively evaluate the agricultural flood risk,
focusing on the combination of flood hazard and the consequences of flooding on crops,
considering the characteristics of flooding impacts with a strong spatial and temporal
variability. Specifically, three problems were solved: (a) what the spatial–temporal charac-
teristics are of the consequences of flooding on crop growth; (b) what the features are of
flood inundation patterns and evolution processes; (c) what the relationship is between
flood characteristics and crop damage from flooding. The investigation framework con-
sisted of the following three parts: (a) an integrated disturbance index (IDI) based on
remote sensing imagery was developed in this study to identify the flood impact on crops
from the backdrop of natural fluctuations in the crop systems for long-duration floods and
multi-cropping agricultural systems, aiming to acquire spatially explicit and sufficiently
high spatial resolution maps of the crop damage condition over large-scale areas; (b) a 2D
hydraulic model parallelized using the GPU approach was employed to simulate the flood
dynamics, aiming to derive enough meaningful details at the local scale in terms of flood
inundation patterns and processes over the whole natural watershed; (c) the random forest
model was used to relate the crop yield loss and flood characteristics, aiming to investigate
the relationship between flood hazard and consequent damage. The proposed evaluation
framework will be tested in the MLYRB, where the severe flood event in 2010 caused great
agricultural losses during the crop growing season.

2. Materials and Methods

2.1. Conceptual Model for Monitoring Flood Impacts on Crop Growth
Excessive rainfall and associated waterlogging can seriously impede plant growth and
lead to significant yield losses in many crop species. Flash floods may wash away or ruin
entire swaths of agricultural land and completely destroy crops. Besides the direct damage
caused by flooding, waterlogging resulting from heavy rainfall with a long duration also
adversely affects crop production, mainly through restricting gaseous exchange in the
soil [61]. Consequently, crop yield subjected to soil submergence can decrease signifi-
cantly [62,63]. At the same time, previous experiments have suggested that crops exposed
to waterlogging have the capacity to recover after stress is released [64–66]. Thus, the final
flooding damage caused to crops depends not only on the stress conditions but also on
Agriculture 2022, 11, x FOR PEER REVIEW 5 of 23

Agriculture 2022, 12, 484 nificantly [62,63]. At the same time, previous experiments have suggested that crops ex-
5 of 23
posed to waterlogging have the capacity to recover after stress is released [64–66]. Thus,
the final flooding damage caused to crops depends not only on the stress conditions but
also on the subsequent
the subsequent recoveryrecovery status
status after after
stress is stress is released.
released. The Moderate The Moderate
ResolutionResolution
Imaging
Imaging Spectroradiometer (MODIS) with its high-frequency
Spectroradiometer (MODIS) with its high-frequency sampling (twice daily) provides sampling (twice daily) pro-
a valid
vides a valid tool for tracking the temporal trajectories of crop growth
tool for tracking the temporal trajectories of crop growth dynamics. The vegetation indices dynamics. The veg-
etation
(VI) from indices
MODIS (VI)arefrom MODISused
commonly are tocommonly used
statistically to statistically
correlate correlate
to crop growth to crop
conditions
growth
and yield across the world [67,68]. The trajectory of vegetation indices can help us toindices
conditions and yield across the world [67,68]. The trajectory of vegetation detect
can help us
flooding to detect flooding
disturbance disturbance
and characterize and characterize
the subsequent recoverythe of
subsequent
crops, thus recovery
makingofit
crops, thus
ideal for making itthe
identifying ideal forimpact
final identifying the final
of flooding onimpact of flooding on crop production.
crop production.
Previous
Previous efforts have shown the potential of VI temporal trajectory
efforts have shown the potential of VI temporal trajectoryanalysis
analysisiningen-
gen-
erating
erating parameters for crop yield estimation, such as the integrated greenness and peak
parameters for crop yield estimation, such as the integrated greenness and peak
greenness.
greenness.The Theintegrated
integrated greenness—i.e.,
greenness—i.e., summing
summing thetheVI values at regular
VI values intervals
at regular dur-
intervals
ing the crop
during growing
the crop growingseason—can
season—can be used to roughly
be used represent
to roughly the accumulation
represent of green
the accumulation of
leaf
greenarea.
leafItarea.
has been
It hasshown to be an
been shown to excellent measure
be an excellent of the of
measure produced biomass
the produced and crop
biomass and
yield [38,67,69–71].
crop yield GivenGiven
[38,67,69–71]. that thethatcrop growth
the crop dynamics
growth dynamicsduring the full
during thetime period
full time can
period
be
can be considered in the integrated VI, it is more ideal than individual VI observations for
considered in the integrated VI, it is more ideal than individual VI observations for
long-duration
long-duration flood flood events.
events. Additionally,
Additionally, the the use
use of
of integrated
integrated VI VI helps
helps usus to
to distinguish
distinguish
the
the effects of the
the growing
growingseason
seasonofofflood-affected
flood-affectedcrops crops inin a multi-crop
a multi-crop agricultural
agricultural sys-
system.
tem.
BetterBetter yet, integrated
yet, integrated VI canVI can combine
combine theofeffects
the effects both the of both the disturbance
flooding flooding disturbance
(indicated
(indicated
by the redby thein
area red area in
Figure 1) Figure
and the 1) possible
and the possible
recoveryrecovery
(indicated (indicated by thearea
by the green green in
Figure
area 1). Thus,
in Figure 1).the integrated
Thus, VI is adopted
the integrated in this study.
VI is adopted in this study.

Figure 1.
Figure The conceptual
1. The conceptual model
model illustrating
illustrating the
the effect
effect of
of flood
flood disturbance
disturbance on
on crop
crop production.
production. The
The
red
red area represents
represents the
theeffect
effectofofthe
thedisturbance
disturbanceofof flood
flood onon crop
crop growth,
growth, andand
thethe green
green areaarea repre-
represents
sents the subsequent
the subsequent recovery
recovery of theof the crop.
crop.

Recently, different
Recently, differenttypes
typesofof VIs
VIs have
have beenbeen proposed.
proposed. The Normalized
The Normalized Difference
Difference Veg-
Vegetation Index (NDVI) and the Enhanced Vegetation Index (EVI) are the
etation Index (NDVI) and the Enhanced Vegetation Index (EVI) are the most extensively most extensively
used indices
used indices forfor the
the statistical
statistical analysis
analysis of
of crop
crop growth
growth conditions
conditions [72].
[72]. The
The newly
newly released
released
EVI has
EVI has been
been shown
shown to tohave
havegreater
greaterpotential
potentialin inmonitoring
monitoringcrop
cropyield
yieldthan
thanNDVI
NDVI[73,74],
[73,74],
owing to
owing to fact
factthat
thatEVI
EVIremains
remainssensitive in dense
sensitive vegetation
in dense whilewhile
vegetation NDVINDVIis saturated [75,76].
is saturated
Therefore,
[75,76]. EVI andEVI
Therefore, integrated EVI (IEVI)
and integrated EVI are(IEVI)
adopted
are in this study
adopted andstudy
in this may beand calculated
may be
by the following
calculated by the formula:
following formula:
EVI==2.5
EVI 2.5
××(R(R
Nir −−RR )/(R
Red ) / +66×
(RNir + ×RRRed −
− 7.5
7.5×× RRBlue +
+ 11)
) (1)
(1)
where R , R , and R refer to the reflectance of the near-infrared, red, and blue
where RNir , RRed , and RBlue refer to the reflectance of the near-infrared, red, and blue bands
bands of remote sensing images, respectively.
of remote sensing images, respectively.
IEVI = ∑ EVI (2)
IEVI = ∑ EVIt (2)
where t represents the time period of the crop growing season and EVI is the EVI value
in timetperiod
where t. the time period of the crop growing season and EVIt is the EVI value in
represents
Flood impact
time period t. on crop production can be seen as one type of terrestrial disturbance
that occurs outsideonthe
Flood impact range
crop of natural
production can variations [77].type
be seen as one In the absence of
of terrestrial disturbances,
disturbance that
occurs outside the range of natural variations [77]. In the absence of disturbances, crops
are expected to reach their maximum greenness for a specific environment, and natural
fluctuations in hydrothermal balance manifest physiologically as interannual variability
Agriculture 2022, 12, 484 6 of 23

in crops’ EVI value. The magnitude of the interannual fluctuations is usually large in
agricultural ecosystems. Therefore, a fine-tuning of change thresholds is necessary in order
to separate true disturbances from natural fluctuations. The integrated disturbance index
(IDI) is defined as follows:
IDI = 1 − IEVI/IEVImax (3)
where IEVI is the integrated EVI value of the current year and IEVImax is the multi-year
mean of IEVI without external disturbances. In this study, the analysis is based on the
number of years (17 years) of available data, and the top 50% of the IEVI is used in the
calculation to derive the crop’s IEVI without external disturbances. As for the range of
natural fluctuations, we use one standard deviation (SD) from the mean IEVI without
external disturbances for each individual pixel as the disturbance threshold. Pixels that fall
within ±1 SD of the multi-year mean are considered to be within the natural fluctuations
for that individual pixel. Pixels that depart significantly (beyond 1 standard deviation)
from the long-term mean IEVI are identified as areas of potential disturbance. When a flash
flood occurs, the IDI in the damaged area will decrease to a certain level that is obviously
less than the multi-year mean. The negative impact from this flood event (as shown in
Figure 1 left) will be detected when it shifts outside the range of natural variations. As
the damaged area recovers after stress is released, the IDI should fall inside the range of
natural fluctuations (as shown in Figure 1 right). Although simple, the IDI is thus ideal for
identifying the final impact of flooding on crop production.

2.2. Coupled Hydrological and 2D Hydraulic Model for Flooding Simulation

The coupled model uses the SCS-CN hydrological model as a rainfall-runoff generator
and implements the routing scheme with the 2D hydraulic model in order to simulate
localized flood depths, durations, and velocities. The Soil Conservation Service-Curve
Number (SCS-CN) model [78] is an event-based model that estimates hydrologic losses and
transforms rainfall to runoff with a rainfall-runoff curve. The curve varies according to a
parameter called the curve number (CN). The dimensionless CN represents the antecedent
potential water retention of a watershed [79]. The SCS-CN model has been widely used all
around the world over various hydro-meteorological conditions because the CN value can
be tabulated considering hydrologic conditions, land use types, and soil types. Detailed
information on it can be found in the study of Mishra and Singh [80]. In this study, the
surface runoff is computed with the SCS-CN for every cell in every time step in order to be
consistent with the spatio-temporal resolution of the 2D hydraulic model. The calculation
procedure used can be found in Caviedes-Voullième et al. [49].
For the hydrodynamic component, surface flow is simulated by means of the depth-
averaged 2D shallow water equations. Floodplain flow modeling using shallow water
equations has been extensively explored [81–86]. Shallow water equations can be thought
of as a simplification of the Navier–Stokes equations, assuming the pressure distribution to
be hydrostatic—i.e., neglecting the vertical acceleration term [87]. In shallow water models,
the horizontal wavelength is much larger than the vertical dimension and the streamlines
remain approximately parallel to each other [88]. The conservative matrix form of the 2D
shallow water models may be expressed as follows:

∂q ∂f ∂g
+ + =s (4)
∂t ∂x ∂y

where t is the time; x and y are the Cartesian coordinates; q is the flow-variable vector; f
and g denote the flux vectors in the x and y direction, respectively; and s is the source term
vector.
Agriculture 2022, 12, 484 7 of 23

   
h qx
q =  qx  f =  uq x + 21 gh2 
qy uqy
  
R
 (5)
qy √ ∂z 
g= vq x s =  −C f u u2 + v2 − gh ∂xb 
 
1 2
√
vqy + 2 gh −C f v u2 + v2 − gh ∂z b
∂x

where h denotes the water depth (m); q x and qy denote the unit-width discharges in the
y and y directions (m2 /s), respectively; u and v are the depth-averaged velocities in the
y and y directions (m/s), respectively; q x = uh and qy = vh; R is the generated runoff rate
from the SCS-CN model (m/s); zb is the bed elevation (m); and C f is the bed roughness
coefficient (s m−1/3 ).
The shallow water equations are solved by means of a first-order Godunov-type finite
volume scheme. Finite volume schemes allow simulations of flood propagation to be
carried out on real irregular topography, and its predictions are robust and accurate [57,89].
However, it is widely known that finite volume schemes heavily rely on mesh size and
have a high computational cost as the number of cells within the domain increases [49]. The
increased computational time can become a crucial issue, since the complex topography of
large-scale catchments leads to a very large number of cells. To overcome this issue, this
study carries out flood simulations on Graphics Processing Units (GPUs) using NVIDIA’s
parallel computing architecture CUDA (compute unified device architecture) [90], thus
taking advantage of the parallel numerical schemes of modern graphics cards.
The topography of channels and floodplain is another of the most important factors de-
termining the performance of 2D hydraulic models. However, the currently available global
topographic data have limitations regarding their application in flood modeling, mainly
due to their random errors and insufficient spatial resolution [91]. To ensure the modeling
accuracy of the 2D hydraulic model, we apply a newly released DEM (i.e., Multi-Error
Removed DEM) which corrects major error components in the Shuttle Radar Topographic
Mission DEM (SRTM DEM) [92]. Additionally, a hydraulic correction method [93] is em-
ployed for the MERIT DEM to improve the flow connectivity of river networks, which can
reduce the limitation of insufficient spatial resolution to some extent. Then, the hydrauli-
cally corrected DEM is used to enhance the flood inundation modeling.

2.3. Random Forest Model for Identifying Flood Impact on Crop Production
In the MODIS-based crop yield losses and 2D hydraulic flooding simulation results,
it is essential to identify the relationship between crop response and predictors and then
recognize the most important influence variables of floods that impact crop production. As
a machine learning method, random forest has a good tolerance to outliers and noise, and
strong robustness, which leads to a wide range of application prospects [94]. Random forest
provides two kinds of importance measures: Mean Decrease Impurity (MDI) based on
the Gini index and Mean Decrease Accuracy (MDA) based on Out-of-Bag (OOB) data [95].
In this study, the random forest model is used to relate the crop yield losses to flood
characteristics. Random forest models create a “forest” out of an ensemble of decision trees.
Each tree is created using a sample of two-thirds of the data and a randomly permutated
subset of the explanatory variables. One-third of the data are left out and used to test
the performance of that tree. For each explanatory variable, the importance metric can
be determined by the difference in performance between decision trees with and without
a given variable permuted. Using a random forest model, the functional forms relating
response and predictors can be displayed using partial dependence plots [96]. In this study,
a random forest model was created using the “Random Forest” package [97] in R [98].
 Agriculture 2022, 11, x FOR PEER REVIEW 8 of 23
Agriculture 2022, 12, 484 8 of 23

2.4.The
2.4. TheStudy
StudyArea
Areaand andEvents
Events
Themiddle
The middleand andlowerlower reaches
reaches ofof
thethe Yangtze
Yangtze River
River Basin
Basin (MLYRB)
(MLYRB) are are located
located at 24atto24
35to◦ N
35° N 109
and andto109
122to◦ E,122°
mainlyE, mainly
includingincluding the provinces
the provinces of Hubei,of Hubei,
Hunan, Hunan,
Jiangsu, Jiangsu,
Zhejiang,
Zhejiang, Anhui, and Jiangxi and Shanghai municipality (Figure
Anhui, and Jiangxi and Shanghai municipality (Figure 2). The annual average temperature 2). The annual average
ranges from 14 to 18 C, and the annual mean rainfall is approximately 1000–1400 mm1000–
temperature ranges ◦ from 14 to 18 °C, and the annual mean rainfall is approximately [99].
1400
The mmis[99].
basin The basinby
characterized is acharacterized
cold-dry winter by aandcold-dry
warm-wet winter and warm-wet
summer, and 70–80% summer,
of its
and 70–80%
rainfall occursofinits rainfall
the summer occurs
[14].inAppropriate
the summerclimate
[14]. Appropriate
conditions makeclimate conditions
this region one make
of
this
the region
most one of the
important most important
agricultural areas inagricultural
the world.areas in the world.
It accounts It accounts
for 24.5% of China’sfor crop-
24.5%
of China’sarea
harvesting crop-harvesting
and accountsarea and accounts
for nearly 28.6% offorthenearly
grain 28.6% of thein
production grain
China.production
However,in
China. However, due to the fact that the Yangtze River is almost parallel
due to the fact that the Yangtze River is almost parallel to the subtropical monsoon rain belt to the subtropical
monsoon
(i.e., Meiyurain belt
front), the(i.e., Meiyu front),
agricultural the agricultural
production is vulnerableproduction
to floodsisand
vulnerable to floods
always afflicted
and always afflicted by
by crop failure from waterlogging. crop failure from waterlogging.

Figure2.2.The
Figure Thestudy
studyarea:
area:(a)
(a)China;
China;(b)
(b)the
themiddle
middleand
andlower
lowerreaches
reachesofofYangtze
YangtzeRiver;
River;(c)
(c)the
theHan
Han
River.
River.

From
FromJune
June to
to August
August inin 2010,
2010, an
anextreme
extremeflood
floodevent
eventwith
with a record
a record highhigh water
water dis-
discharge
charge hit the Yangtze River, causing immense economic losses and casualties(see
hit the Yangtze River, causing immense economic losses and casualties (see
http://en.wikipedia.org/wiki/2010_China_floods,
http://en.wikipedia.org/wiki/2010_China_floods, accessed accessedonon2424January
January2022).
2022).According
Accord-
ing
to to GRACE
GRACE data
data analysis,the
analysis, thedischarge
dischargeand
andprecipitation
precipitation of
of this
this rainstorm
rainstormdisaster
disasterwere
were
38%
38% and 19% higher than the average for previous years [100]. Additionally,this
and 19% higher than the average for previous years [100]. Additionally, thisflood
flood
was
wascharacterized
characterizedby bya along
longflood
floodduration
durationandandwide
wideinfluence
influencescope;
scope;thus,
thus,itithas
hasbeen
been
identified
identified as the most serious flood disaster since 2000. According to annual reporton
as the most serious flood disaster since 2000. According to annual report on
natural
naturaldisasters
disastersininChina,
China,approximately
approximately100 100million
millionpeople
peoplewere
wereaffected;
affected;2027
2027people
people
were killed, 620 thousand houses were damaged, 52 thousand square kilometers of land
were killed, 620 thousand houses were damaged, 52 thousand square kilometers of land
were affected, and the economic losses may have been as high as 125 billion yuan [101].
were affected, and the economic losses may have been as high as 125 billion yuan [101].
A flood inundation simulation utilizing the coupled hydrological–hydraulic model
was implemented in the Han River, the largest tributary of the MLYRB, and then used to
Agriculture 2022, 12, 484 9 of 23

identify how flood characteristics affect crop failure (Figure 2). The Han River is located at
30◦ 100 –34◦ 200 N and 106◦ 150 –114◦ 200 E, with a 159,000 km2 basin area. The basin is situated
in the humid zone with a sub-tropic monsoon climatic environment. The annual average
temperature ranges from 12 to 16 ◦ C. The annual average precipitation is 700–1800 mm,
with high intra-annual and inter-annual variations [102], and 80% of the precipitation is
concentrated during the period from May to October [103]. These extreme rainstorms
occurring in the early summer and the long periods of rainfall often lead to flooding and
waterlogging disasters occurring in the Han River Basin.

2.5. Data Acquisition

The MODIS EVI dataset for the period from 2000 to 2016 were provided by the
NASA Earth Observing System (EOS). Although MODIS sensors provide nearly complete
Earth coverage every day, the daily measurements are frequently contaminated by clouds,
aerosols, and cloud shadows [104]. The cloud contamination is more serious during
flooding events [105]. To help reduce the effects of cloud interference, the MOD13Q1
version 6 product was retrieved, which provides the 16-day EVI at a 250 m resolution. The
16-day dataset is a maximum value composite (MVC), in which the pixel observation with
the highest EVI value is adopted to represent the compositing period (16 days). MVC selects
the least contaminated observations and tends to adopt the closest near-nadir view [106].
Moreover, all unreliable pixels are removed according to the MOD13Q1 reliability layer
(value) before the EVI data are adopted.
The HJ-1A/B satellites are China’s two small environment satellites; they were
launched on 6 September 2008. The CCD images of the HJ-1A/B satellites with a 30 m
resolution, which were acquired during the 2010 flood event, were derived from the China
Centre for Resources Satellite Data and Application and used to map the water surface area
and validate the flood simulation result. The hourly precipitation dataset was provided by
the National Climate Center (NCC) of the China Meteorological Administration (CMA).
This dataset was constructed based on a 0.1◦ attitude-longitude grid covering the period
from 2008 to now by merging two kinds of information sources with different characteris-
tics, including more than 30 thousand automatic meteorological stations and the CMORPH
satellite-based global rainfall product. Land use data were obtained from the GlobCover
2009 dataset, which was created based on 300 m MEdium-spectral Resolution Imaging Spec-
trometer (MERIS) data by the European Space Agency (ESA) and Université Catholique
de Louvain (UCL) [107]. Soil type information was provided by the harmonized world
soil database [108]. Additionally, this study adopted the “Multi-Error-Removed Improved-
Terrain DEM” (MERIT DEM) with a spatial resolution of 90 m [92]. River networks with a
15 arc-second resolution were derived from the HydroSHEDS datasets [109]. In order to
maintain consistency with the spatial resolution of the MODIS EVI product, the precipi-
tation data, land use data, and DEM were all resampled to a pixel size of 250 m × 250 m
using the nearest neighbor algorithm. In addition, the flood-affected crop areas for the
2010 flood event were collected from information provided by the Ministry of Agriculture.
When more than 10% of the expected crop yield of the cultivated land was reduced by
flooding, the affected areas were counted in the agricultural disaster statistics [110].

2.6. Study Framework

As shown in Figure 3, this study comprises three main steps:
(1) The calculation of integrated disturbance index (IDI). The integrated disturbance
index (IDI) was applied to the MODIS EVI dataset for 17 years.
(2) Flood Simulation. The spatial surface characteristics and precipitation were input into
the coupled SCS-CN and 2D hydraulic model, and the flood process was simulated
for MLYRB with GPU parallel computing.
(3) The integrated impacts of flood on crop yield loss. Random forest was adopted to
predict the crop damage extent with different flood characteristics.
 (2) Flood Simulation. The spatial surface characteristics and precipitation were input
into the coupled SCS-CN and 2D hydraulic model, and the flood process was simu-
lated for MLYRB with GPU parallel computing.
Agriculture 2022, 12, 484 10 of 23
(3) The integrated impacts of flood on crop yield loss. Random forest was adopted to
predict the crop damage extent with different flood characteristics.

3. The study framework.

Figure 3.

3.
3. Results
Results
3.1. The Characteristics of Flood Disturbances on Crop Yields
3.1. The Characteristics of Flood Disturbances on Crop Yields
The integrated disturbance index (IDI) proposed in this study was applied to the
The integrated disturbance index (IDI) proposed in this study was applied to the
MODIS MOD13Q1 data for the years 2000–2016 across the MLYRB. In the study area, two
MODIS MOD13Q1 data for the years 2000–2016 across the MLYRB. In the study area, two
or three harvests a year are grown instead of one. In order to investigate the influence of
or three harvests a year are grown instead of one. In order to investigate the influence of
flooding on the current crop, the IDI value was based on the EVI of crops after flooding.
flooding
The 2010 onfloodtheevent
current crop,took
mainly the IDI
place value
in thewas basedfrom
MLYRB on the EVI
late of to
June crops
July.after flooding.
We analyzed
The 2010 flood event mainly took place in the MLYRB from late
the MODIS EVI from late June to early September, including day 209, day 225, and day June to July. We analyzed
the MODIS
241 EVI from
of the 16-day late JuneFigure
composites. to early 4aSeptember,
displays the including
IDI values dayacross
209, day the 225, and day
croplands of
241 of the 16-day composites. Figure 4a displays the IDI values
the MLYRB. Given that the Ministry of Agriculture defines a 10% loss of crop yield as across the croplands of the
the
MLYRB. Given
threshold recorded thatinthe
theMinistry
national of Agriculture
disaster dataset,defines
an IDIaof10% moreloss
thanof crop
10% and yieldbeyond
as the
threshold recorded in the national disaster dataset, an IDI of more
one standard deviation was determined as the criterion for flagging a pixel as disturbed. than 10% and beyond
one standard
Based on this deviation
criterion, was85,887determined as the criterion
km2 of croplands across forthe flagging
MLYRB were a pixeldetermined
as disturbed. to
Based on this criterion, 85,887 km 2 of croplands across the MLYRB were determined to be
be impacted by external disturbance. The statistics released by the government showed a
impacted by external
flood-affected disturbance.
area of 81,404 km2 inThe statistics The
the MLYRB. released by the government
satellite-based showed
detection result wasa
flood-affected
very close to that area of 81,404
shown by thekm 2 in the MLYRB.
statistical data, andThethesatellite-based
error between detection
them was result was
only 5.5%.
very close to
According to that shown by the statistical
the governmental reports and data, and the
related newserror between
reports, the them
mainwas onlyof
streams 5.5%.
the
According
Yangtze to the
River governmental
(including reports and
Jianli, Luoshan, related Jiujiang,
Hankou, news reports, the mainDongting
and Datong), streams of the
Lake
Yangtze River (including Jianli, Luoshan, Hankou, Jiujiang,
Basin, Poyang Lake Basin, and Han River Basin were the areas that were most heavilyand Datong), Dongting Lake
Basin, Poyang
damaged during Lake Basin,
the 2010 andevent.
flood Han River Basin were
Additionally, the FutheRiver,
areasXin
thatRiver,
wereand mostGan heavily
River
damaged
within the during
Poyangthe Lake2010 flood
Basin were event. Additionally,
the most the Fu
hard-hit areas, withRiver, Xin River,reaching
the discharge and Gan a
River within
50-year returnthe Poyang
period. TheLake
mainBasin
river were
in thethe most hard-hit
Dongting areas,the
Lake Basin, with
XiangtheRiver,
discharge
was
reaching
also a 50-year
seriously damaged,returnwith
period.
the The
water main river
level in the Dongting
reaching Lake Basin,
the third largest level the Xiang
recorded.
River,tributaries
Some was also seriously
of the upperdamaged,
reacheswith of thetheHanwater
River level
werereaching the third
also among largest
the most level
hard-hit
recorded.
areas. From Some
Figure tributaries
4a, it canofbethe
seen upper reaches
that most of the
of the Han River
cropland pixelswere
that also
wereamong
obviouslythe
disturbed by flooding (shown in red) were basically distributed around Dongting Lake and
Poyang Lake and along the main stream of the Yangtze River. The satellite-based detection
results were, on the whole, consistent with the most hard-hit areas recorded (shown in blue
in Figure 4a).
 most hard-hit areas. From Figure 4a, it can be seen that most of the cropland pixels
were obviously disturbed by flooding (shown in red) were basically distributed ar
Agriculture 2022, 12, 484 Dongting Lake and Poyang Lake and along the main stream of the 11 Yangtze
of 23 River
satellite-based detection results were, on the whole, consistent with the most hard-h
eas recorded (shown in blue in Figure 4a).

Figure 4. FloodFigure 4. Flood

disturbance disturbance
detection detection
results results
obtained from obtained
the IDI infrom the middle
(a) the IDI in (a)
andthe middle and
lower
reaches of the Yangtze River Basin (MLYRB) and (b) the Han River Basin.
reaches of the Yangtze River Basin (MLYRB) and (b) the Han River Basin.

In addition to validating the detection results obtained throughout the whole study,
their accuracy was also verified with spatially compatible field data. Given the difficulty
of deriving field investigation data over large areas through postsurveys, the spatial char-
acteristics of the obviously disturbed croplands were investigated based on the physical
characteristics of the river floods. In addition to areas where surface water concentrates, the
Agriculture 2022, 12, 484 12 of 23

interaction between rivers and floodplains is crucial in the evolution of flood events. Often,
flood inundation emanates from overflow river systems and expands into the surrounding
areas. Tehrany et al. [111] showed that flooding mainly occurs near to the riverbank and
that a distance from the drainage network of within 1700 m is positively correlated with
flood occurrence in flat areas. Kazakis et al. [112] also indicated that areas near a drainage
network are highly vulnerable to flood hazards, and that flood hazards decrease at dis-
tances from the drainage network >2000 m by analyzing records of historical floods that
occurred in the Rhodope-Evros region, Greece. Other similar results can be found in Samela
et al. [113] and Tang et al. [114]. Distance from streams usually has a major influence on the
velocity and extent of flooding in flat areas. From Figure 4b, in the Han River Basin, it can be
seen that most of the pixels that are obviously disturbed by flooding (in red) are distributed
near to steams. In order to quantitatively and objectively evaluate our results, the distance
between disturbed pixels and the nearest rivers was calculated in ArcGIS, and the analysis
domain does not include these concentrated disturbed areas (around Doting Lake, Poyang
Lake, and the main stream of the Yangtze River). The results are shown in Table 1 as 47%
and 74% of the disturbed pixels are, respectively, within 1000 m and 2000 m from river
streams. This indicates that most of these disturbed pixels are concentrated in the riparian
zones. The spatial characteristics of flood-affected crop areas were in agreement with the
potential flood-prone areas, as expected. This indirectly verifies the spatial characteristics of
the detected flood-disturbed areas as well as the validity of the satellite-based IDI proposed
in this study for flood impact detection.

Table 1. Percentage of the flood-disturbed areas within a certain distance to the nearest stream.

Distance (km) <0.1 [0.1, 0.2) [0.2, 0.5) [0.5, 1.0) [1.0, 1.5) [1.5, 2.0) [1.5, 2.0)
Percentage 7 13 28 47 62 74 83

Traditionally, flood impact detection is implemented by combining the identified

agricultural lands affected by inundations and “loss functions”. Firstly, inundation maps are
derived by different approaches, including DEM analysis [115], the interpretation of remote
sensing imagery [116], and hydrodynamic simulation [117–121]. Then, the agricultural
lands affected by flooding are identified by overlying the inundation maps with land use
maps. Lastly, “loss functions”, which describe the relationship between flood intensity
(e.g., water depth) and associated damage, are used to estimate the crop damage caused to
affected areas. However, it is hard to derive the loss functions suitable for specific study
areas, which are regarded as the key factors in traditional methods, because historical flood
damage data are generally scarce and unlikely to be available [120]. In this study, we make
use of the information of MODIS imagery with high frequent sampling and a relatively
long data archive, which can be used to track the temporal trajectory of the vegetation index
during the crop growing season. An effective and scalable index was designed to detect
the flood disturbance signal from the natural variation signal for the complicated multi-
cropping agricultural systems. Relative to previous methods, this evaluation index had the
advantages of a higher efficiency and lower computation burden. It had the capacity to
quantify the actual flooding damage, taking into consideration the direct influence of the
flood and the subsequent crop recovery status. Moreover, this index is relatively simple
and mainly based on globally accessible data; it does not necessarily require the division of
crop growing seasons and is especially useful for complex agricultural systems that follow
multi-cropping practices.

3.2. Flood Simulation Results

The coupled hydrological and hydraulic model developed in this study was imple-
mented to predict time-varying and grid-based flood depths, velocities, and durations
within the whole Han River Basin for the July 2010 event. The spatial resolution of the
Agriculture 2022, 12, 484 13 of 23

2D hydraulic model is 250 m. The calculating duration is the same as the precipitation
duration, and the length of dynamic calculating step is about 1 s.
Figure 5 highlights the spatial distribution of the maximum water depth, flow velocity,
and flood duration for each pixel. The strength of the coupled model is exemplified by this
series of maps, as they offer a clear picture in terms of the flood process and pattern while
Agriculturesimultaneously ensuring a sufficiently high spatial resolution, which makes it possible to
2022, 11, x FOR PEER REVIEW 14 of 23
investigate the highly localized impacts of flooding on individual crop fields.

Figure 5. Flood simulation results of (a) maximum water depth, (b) flow velocity and (c) flood
Figure 5. Flood simulation results
duration for ofRiver
the Han (a) maximum
Basin. water depth, (b) flow velocity and (c) flood
duration for the Han River Basin.
Agriculture 2022, 12, 484 14 of 23

According to the simulation results, the total precipitation was 117.5 mm for the entire
storm, which resulted in a mean maximum water depth of 80.3 mm (as shown in Figure 5a).
The predicted maximum water depth spatially varied from 0 mm to 30.3 m during this
flood. Therefore, the percentage of water with a depth of more than 0.01 m and 0.1 m
was 20.1% and 9.6%, respectively. Besides flooding depth, the coupled model provided a
refined description of flow velocity, highly influenced by the topography. The velocity of
flow in the floodplain was found to vary between 0 and 1.5 m/s (as shown in Figure 5b).
Moreover, the proposed model could also be used to predict the flood duration (as shown
in Figure 5c).
The evaluation of the flood inundation simulations was implemented using HJ-1A/B
CCD images of the event acquired from 21 July to 3 August. The surface water area was
extracted from these images using a supervised classification model. The surface water
area, including rivers, lakes, and pond areas, can be most easily identified when image
pixels are not impacted by cloud cover. Based on previous research on this topic [122–124],
extracting water body information from HJ-1 A/B CCD datasets is an effective method,
and the overall accuracy can reach as high as 93.82% [122].
The simulated maximum water depth of 2D hydraulic model and the surface water
areas extracted from HJ-1 A/B CCD images were overlapped, and the intersecting part
was analyzed across the Han River Basin. In Figure 6, the intersection of the water body
area of HJ-1 A/B CCD (as the observed results) and the simulated results is shown in
different colors for different depths. As the extent of the surface water areas is significantly
lower than that of the actual flood footprint because overland runoff can flow past certain
areas without forming open water (i.e., boundaries of inundated areas) [125], we defined
those areas where the simulated maximum water depth was above 20 mm as the predicted
correct areas. As Figure 6 shows, the predicted correct area accounted for 87.6% of the
intersection area, which means that the accuracy of the validation was 87.6%. Therefore,
the percentage of predicted water depth of above 50 mm and 100 mm in the surface water
area was 76.3% and 50.7%, respectively. This result shows a favorable agreement between
the numerical modeling and remote sensing data for the coupled hydrological and 2D
hydraulic model.
This study may contain biases for neglecting the control facilities. For example, in the
case of continuous levees, the lowest point of the levee height overflows or breaks the levee.
The artificial levee would raise the flood risk by decreasing the water storage capacity of
river, which leads to flood water either overtopping the bank or breaking the levee and
creating the flood during heavy rain [126]. The flood inundation area, water depth and
flood duration could be reduced by storing the flood water in the reservoir behind the dam
during flooding time. A case study in Bago River Basin, Myanmar, shows that the dam
operation for flood control would reduce the flood inundation area by approximately 10%
and crop field damage by approximately 10% [127]. However, based on the overlapped
images of simulated water depth of hydraulic model and water body area from HJ-1A/B
CCD images, the validation of the flood simulation, in the hypothesis that the area where
the simulation water depth was less than 20 mm would be excluded, are reliable.
 those areas where the simulated maximum water depth was above 20 mm as the predicted
correct areas. As Figure 6 shows, the predicted correct area accounted for 87.6% of the
intersection area, which means that the accuracy of the validation was 87.6%. Therefore,
the percentage of predicted water depth of above 50 mm and 100 mm in the surface water
Agriculture 2022, 12, 484
area was 76.3% and 50.7%, respectively. This result shows a favorable agreement between
15 of 23
the numerical modeling and remote sensing data for the coupled hydrological and 2D
hydraulic model.

Figure 6. Comparing the modeled water surface area with HJ imagery.

Figure 6. Comparing the modeled water surface area with HJ imagery.
ThisRelationship
3.3. The study maybetween
containFlood
biases for neglecting
Characteristic andthe control
Crop Damagefacilities.
ExtentFor example, in the
case After
of continuous levees, the lowest point of the levee
accomplishing the detection of flood disturbance of crops height overflows
based onor thebreaks the
IDI index
levee. The artificial levee would raise the flood risk by decreasing
from the MODIS EVI dataset and flood inundation simulation via coupled hydrological– the water storage ca-
pacity
hydraulicof river,
model, which leadsdamage
the crop to floodextent
waterand
either
floodovertopping the bank
characteristics couldorbebreaking
derived the
for
levee and creating the flood during heavy rain [126]. The flood inundation
each pixel over the whole basin. Subsequently, the average values of each flood variable area, water
depth and the
(including flood durationwater
maximum coulddepth,
be reduced by storing
mean water depth,the
flowflood waterand
velocity, in the reservoir
duration) for
behind the dam during flooding time. A case study in Bago River Basin,
pixels with the same crop damage extent, and thereby the distribution of flood variables Myanmar, shows
that theevery
against dam operation
1‰ crop damagefor flood control
extent, would
could reduce the
be acquired. Weflood
then inundation
adopted randomarea by ap-
forest
proximately 10% and
analyses to predict thecrop
cropfield damage
damage by with
extent approximately
these flood10% [127]. However,
characteristic basedThe
predictors. on
the overlapped
variable images
importance of simulated
calculation water
ranked thedepth of hydraulic
importance of flowmodel and(h),
duration water
flowbody area
velocity
from HJ-1A/B
(m/s), maximum CCD images,
water depththe validation
(m), and meanofwater
the flood
depthsimulation, in the hypothesis
(m) for the integrated that
disturbance
the
indexarea where
(IDI). Thethe simulation
ntree was set towater
500,depth
and thewas less than
width 20 mmofwould
and height be were
the tree excluded,
set toare
14
reliable.
and 8, respectively. Additionally, relationship between crop yield losses and each variable
is plotted in Figure 7.
The results from the random forest analysis suggest that the flood duration and flow
velocity are positively correlated with the crop damage extent, as expected. However, the
water depths, including maximum and mean depths, did not show obvious correlation
with the crop damage extent. The flood duration was the most important predictor in our
analysis. Although in previous evaluations of flood damage to agricultural production,
water depth is most commonly used and is considered to be the most influential flood
parameters [128]. Compared to previous research by Vempi and Komori in 2021 [129], the
most important flood parameter of the maximum water depth, maximum flood duration,
and maximum flow velocity to crop yield loss is the maximum water depth. The rela-
tionship between flood characteristics and crop yield losses varies for different physical
condition of the study area. Yet, the Han River Basin shows different features mainly due
to its own watershed characteristics. The Han River is characterized by a high level of
precipitation and siltation, but unfavorable water discharge. Due to siltation, the riverbed
has kept rising over the last nearly 400 years; consequently, there is a difference of several
 3.3. The Relationship between Flood Characteristic and Crop Damage Extent
After accomplishing the detection of flood disturbance of crops based on the IDI in-
dex from the MODIS EVI dataset and flood inundation simulation via coupled hydrolog-
ical–hydraulic model, the crop damage extent and flood characteristics could be derived
for each pixel over the whole basin. Subsequently, the average values of each flood varia-
Agriculture 2022, 12, 484 16 of 23
ble (including the maximum water depth, mean water depth, flow velocity, and duration)
for pixels with the same crop damage extent, and thereby the distribution of flood varia-
bles against every 1‰ crop damage extent, could be acquired. We then adopted random
meters between
forest analysesthe
to riverbed andcrop
predict the the damage
floodplain [130].
extent withMost floodplain
these areas are more
flood characteristic predic-
low-lying than
tors. The the riverbank
variable of the
importance Lower Han
calculation rankedRiver.
theOnce long-lasting
importance of flowrains arrive(h),
duration (justflow
as in the 2010
velocity flood
(m/s), event), the
maximum flood
water water
depth (m),is and
hardmean
to discharge and (m)
water depth manyforparts of the
the integrated
floodplain can become
disturbance water-impregnated,
index (IDI). The ntree was set leading
to 500,toand
poorthe
productivity.
width and Meanwhile,
height of theintree
previous experimental research, crop yield has been identified to be
were set to 14 and 8, respectively. Additionally, relationship between cropreduced linearly
yieldwith
losses
theand
duration of waterlogging [41,42,131].
each variable is plotted in Figure 7. Thus, the flow duration can be considered as the
most influential flood parameter in predicting damage to crop production in the Han River
Basin.

Figure 7. The relationship between (a) flow duration, (b) flow velocity, (c) maximum water depth,
Figure 7. The relationship between (a) flow duration, (b) flow velocity, (c) maximum water depth,
(d) mean water
(d) mean depth
water andand
depth cropcrop
damage extent
damage detected
extent by the
detected random
by the forest
random model.
forest TheThe
model. predictors
predictors
are are
sorted by their importance.
sorted by their importance.
4. Discussion
The results from the random forest analysis suggest that the flood duration and flow
The flood
velocity areparameters can be adopted
positively correlated to construct
with the “lossextent,
crop damage functions” for agriculture.
as expected. However,Thethe
parameters include water depth, water duration, flow velocity, water salinity, deposits, and
contamination by pollution [132]. The flood parameter most frequently adopted in “loss
functions” is water depth, and this is generally the only parameter used to evaluate damage
to plant material [14,115,117–119,133–136]. However, flood duration [127,129,137–139] and
flow velocity [120,129,139] are not always taken into account in classical “loss functions”.
In fact, flood simulations, including the analysis of flood duration and flow velocity, should
be employed in advance if “loss functions” which take into account this parameter are
suitable for the floods and areas needing to be assessed. For example, the necessity of
Agriculture 2022, 12, 484 17 of 23

considering flow velocity depends on the type of flood occurring. If a flood event is a
flash flood, it is extremely important to take into account this parameter [35]. Yet, in the
Han River Basin, for example, the most notable features are a flattened floodplain and a
long duration of excessive rainfall; flood duration is important in constructing appropriate
“loss functions” for local agriculture. Moreover, crop yield losses can also be impacted
by the crop types planted during the flood period [38]. Rice is the most abundant crop in
the Han River Basin, and it can form longitudinal interconnections for gas to realize the
internal aeration of shoots and roots, leading to it having a good tolerance to excess water
stress [140].
Since the flow velocity and flood duration can only be obtained from 2D hydraulic
flood modeling [68], which lead to a limited number of studies utilized flow velocity and
flood duration in crop loss assessment. The research that calculated the flood characteristics
by 2D hydraulic models were listed in Table 2 for a quantitative literature review. We
ranked the flood characteristics, flood types, crop types, and flood simulation models of
related research. Table 2 shows that the importance rank of flood duration, flow velocity,
and water depth varies greatly over different watersheds, flood types, and crop types. Thus,
more research is needed to determine the relationship between flood parameters and crop
yield losses in more basins.

Table 2. The importance rank of Flood Duration (FD), Flow Velocity (FV), and Water Depth (WD)
calculated by 2D hydraulic models in crop yield loss related studies.

Parameter Flood
Flood Flow
Location Water Depth Importance Flood Type Crop Type Simulation
Duration Velocity
Rank Model.
Fruit trees,
√ √ olive trees,
Greece [120]
× WD > FV flash flood tomatoes and Mike Flood.
130 km2
green
vegetables.
Germany √ √ Wheat, 2D hydrody-
Did not
[137] × flash flood barley, and namic
mention.
838 km2 corn. model.
Mexico [139] √ √ √ Long-
FD > FV Corn. Mike 21.
649 km2 duration
Indonesia √ √ √ WD > FV, Long- 2D hydraulic
[129] Rice.
WD > FD duration model.
16,000 km2
√ √ √ Soybean,
China [35] FV > WD > Long- 2D hydraulic
corn, and
2953 km2 FD duration model.
rice.
China √ √ √ FD > FV > Long- Multiple 2D hydraulic
159,000 km2
WD duration crops. model.
This research

5. Conclusions
In this work, a comprehensive and spatially explicit risk evaluation framework was
developed to investigate how floods impact crop production. A simple and effective index
was proposed for detecting flood disturbances in crop production against the backdrop
of natural fluctuations in long-duration floods and multi-cropping agricultural systems
on a pixel-by-pixel basis. A coupled hydrological and 2D hydraulic model was employed
to simulate flood flows in order to derive flood inundation patterns and processes over
the whole basin. The evaluation framework was tested in the MLYRB and the assessment
results indicated the following: (1) the new disturbance detection index is capable of
acquiring a spatially explicit crop damage map, which is verified by the governmental
statistics and related reports and coordinated with the spatial characteristics of river floods;
Agriculture 2022, 12, 484 18 of 23

(2) the coupled hydrological and 2D hydraulic model, with the help of GPU parallel
computing, can provide enough meaningful detail on time-varying and grid-based flood
depths, velocities, and durations over the whole basin, and the simulated surface water
areas show a favorable level of agreement with the HJ imagery; (3) flood duration and flow
velocity are positively correlated with crop damage extent, as expected, with flood duration
being the most important variable in predicting crop damage extent in the MLYRB.

Author Contributions: Conceptualization, H.C.; methodology, H.C.; literature search, X.W.; re-
sources, X.W. and Z.L.; writing—original draft preparation, H.C.; writing—review and editing, X.W.;
funding acquisition, H.C. and X.W. All authors have read and agreed to the published version of the
manuscript.
Funding: This work was supported by the National Basic Research Program of China (2015CB458900),
the National Science Foundation of China (51721006), Major Science and Technology Program for
Water Pollution Control and Treatment of China(2014ZX07203-008), and the open fund of the Ministry
of Education Laboratory for Earth Surface Processes.
Institutional Review Board Statement: Not applicable.
Conflicts of Interest: The authors declare no conflict of interest.

References
1. Schmidhuber, J.; Tubiello, F.N. Global food security under climate change. Proc. Natl. Acad. Sci. USA 2007, 104, 19703–19708.
[CrossRef]
2. Vervuren, P.; Blom, C.W.P.M.; De-Kroon, H. Extreme flooding events on the Rhine and the survival and distribution of riparian
plant species. J. Ecol. 2003, 91, 135–146. [CrossRef]
3. Youssef, M.S.; Mira, M.M.; Renault, S.; Hill, R.D.; Stasolla, C. Phytoglobin expression influences soil flooding response of corn
plants. Ann. Bot. 2016, 118, 919–931. [CrossRef]
4. Sun, L.; Chen, S.; Chao, L.; Sun, T. Effects of flooding on changes in Eh, pH and speciation of cadmium and lead in contaminated
soil. Bull. Environ. Contam. Toxicol. 2007, 79, 514–518. [CrossRef]
5. Urban, D.W.; Roberts, M.J.; Schlenker, W.; Lobell, D.B. The effects of extremely wet planting conditions on maize and soybean
yields. Clim. Change 2015, 130, 247–260. [CrossRef]
6. Christopher, B.F.; Vicente, B.; Thomas, F.S.; Qin, D. Managing the Risks of Extreme Events and Disasters to Advance Climate Change
Adaptation; Cambridge University Press: Cambridge, UK, 2012; pp. 109–230.
7. Blöschl, G.; Hall, J.; Parajka, J.; Perdigão, R.A.P.; Merz, B.; Arheimer, B.; Aronica, G.T.; Bilibashi, A.; Bonacci, O.; Borga, M.; et al.
Changing climate shifts timing of European floods. Science 2017, 357, 588–590. [CrossRef]
8. Jiabo, Y.; Shenglian, G.; Pierre, G.; Sylvia, C. Does the Hook Structure Constrain Future Flood Intensification Under Anthropogenic
Climate Warming? Water Resour. Res. 2021, 57, 2020.
9. Liu, J.; Shi, Z.W. Quantifying land-use change impacts on the dynamic evolution of flood vulnerability. Land Use Policy 2017, 65,
198–210. [CrossRef]
10. Liu, X.; Chen, L.; Zhu, Y.; Singh, V.P.; Qu, G.; Guo, X. Multi-objective reservoir operation during flood season considering spillway
optimization. J. Hydrol. 2017, 552, 554–563. [CrossRef]
11. Zhang, Q.; Xu, C.Y.; Zhang, Z.; Chen, Y.D.; Liu, C.L.; Lin, H. Spatial and temporal variability of precipitation maxima during
1960–2005 in the Yangtze River basin and possible association with large-scale circulation. J. Hydrol. 2008, 353, 215–227. [CrossRef]
12. Yin, H.; Li, C. Human impact on floods and flood disasters on the Yangtze River. Geomorphology 2001, 41, 105–109. [CrossRef]
13. Chen, Z.; Li, J.; Shen, H.; Wang, Z. Yangtze River of China: Historical analysis of discharge variability and sediment flux.
Geomorphology 2001, 41, 77–91. [CrossRef]
14. Yu, F.; Chen, Z.; Ren, X.; Yang, G. Analysis of historical floods on the Yangtze River, China: Characteristics and explanations.
Geomorphology 2009, 113, 210–216. [CrossRef]
15. Jiang, L.; An, P. Temporal and spatial distribution and grain risk assessment of natural disasters in China. J. Catastrophology 2011,
26, 48–53. (In Chinese)
16. Su, B.; Xiao, B.; Zhu, D.; Jiang, T. Trends in frequency of precipitation extremes in the Yangtze River basin, China: 1960–2003.
Hydrol. Sci. J. 2005, 50, 479–492.
17. Zhang, Q.; Peng, J.; Xu, C.Y.; Singh, V.P. Spatiotemporal variations of precipitation regimes across Yangtze River Basin, China.
Theor. Appl. Climatol. 2014, 115, 703–712. [CrossRef]
18. Hu, M.; Dong, M.; Tian, X.; Wang, L.; Jiang, Y. Trends in Different Grades of Precipitation over the Yangtze River Basin from 1960
to 2017. Atmosphere 2021, 12, 413. [CrossRef]
19. Li, X.; Zhang, K.; Gu, P.; Feng, H.; Yin, Y.; Chen, W.; Cheng, B. Changes in precipitation extremes in the Yangtze River Basin during
1960–2019 and the association with global warming, ENSO, and local effects. Sci. Total Environ. 2021, 760, 144244. [CrossRef]
Agriculture 2022, 12, 484 19 of 23

20. Su, B.; Kundzewicz, Z.W.; Tong, J. Simulation of extreme precipitation over the Yangtze River basin using Wakeby distribution.
Theor. Appl. Climatol. 2009, 96, 209–219. [CrossRef]
21. Wu, Y.; Wu, S.Y.; Wen, J.; Tagle, F.; Xu, M.; Tan, J. Future Changes in Mean and Extreme Monsoon Precipitation in the Middle and
Lower Yangtze River Basin, China, in the CMIP5 Models. J. Hydrometeorol. 2016, 17, 2785–2797. [CrossRef]
22. Zhang, Z.; Zhang, Q.; Xu, C.; Liu, C.; Jiang, T. Atmospheric moisture budget and floods in the Yangtze River basin, china. Theor.
Appl. Climatol. 2009, 95, 331–340. [CrossRef]
23. Yin, R.S.; Xiang, Q.; Xu, J.T.; Deng, X.Z.; Yin, R. Modeling the driving forces of the land use and land cover changes along the
upper Yangtze River of China. Environ. Manag. 2010, 45, 454–465. [CrossRef]
24. Cong, X.; Xin, H.; Mu, H.; Wei, Y. Impacts of land-use changes on the lakes across the Yangtze floodplain in china. Environ. Sci.
Technol. 2017, 51, 3669.
25. Nakayama, T.; Watanabe, M. Role of flood storage ability of lakes in the Changjiang River catchment. Glob. Planet. Change 2008,
63, 9–22. [CrossRef]
26. Du, Y.; Xue, H.P.; Wu, S.J.; Ling, F.; Xiao, F.; Wei, X.H. Lake area changes in the middle Yangtze region of China over the 20th
century. J. Environ. Manag. 2011, 92, 1248–1255. [CrossRef]
27. Zhang, Q.; Liu, C.; Xu, C.Y.; Xu, Y.; Jiang, T. Observed trends of annual maximum water level and streamflow during past 130
years in the Yangtze River basin, china. J. Hydrol. 2006, 324, 255–265. [CrossRef]
28. Nakayama, T.; Shankman, D. Impact of the Three-Gorges Dam and water transfer project on Changjiang floods. Glob. Planet.
Change 2013, 100, 38–50. [CrossRef]
29. Meng, L.; Wang, C.; Zhang, J. Heat injury risk assessment for single-cropping rice in the middle and lower reaches of the Yangtze
River under climate change. J. Meteorol. Res. 2016, 30, 426–443. [CrossRef]
30. Lesk, C.; Rowhani, P.; Ramankutty, N. Influence of extreme weather disasters on global crop production. Nature 2016, 529, 84–87.
[CrossRef]
31. Lee, K.S.; Lee, S.I. Assessment of post-flooding conditions of rice fields with multi-temporal satellite SAR data. Int. J. Remote Sens.
2003, 24, 3457–3465. [CrossRef]
32. Pantaleoni, E.; Engel, B.A.; Johannsen, C.J. Identifying agricultural flood damage using Landsat imagery. Precis. Agric. 2007, 8,
27–36. [CrossRef]
33. Džubáková, K.; Molnar, P.; Schindler, K.; Trizna, M. Monitoring of riparian vegetation response to flood disturbances using
terrestrial photography. Hydrol. Earth Syst. Sci. 2015, 19, 195–208. [CrossRef]
34. Shrestha, R.; Di, L.; Eugene, G.Y.; Kang, L.; Shao, Y.Z.; Bai, Y.Q. Regression model to estimate flood impact on corn yield using
MODIS NDVI and USDA cropland data layer. J. Integr. Agric. 2017, 16, 398–407. [CrossRef]
35. Chen, H.; Liang, Z.; Liu, Y.; Liang, Q.; Xie, S. Integrated remote sensing imagery and two-dimensional hydraulic modeling
approach for impact evaluation of flood on crop yields. J. Hydrol. 2017, 553, 262–275. [CrossRef]
36. Zhang, X.; Friedl, M.A.; Schaaf, C.B.; Strahler, A.H.; Hodges, J.C.F.; Gao, F.; Reed, B.C.; Huete, A. Monitoring vegetation phenology
using MODIS. Remote Sens. Environ. 2003, 84, 471–475. [CrossRef]
37. Sakamoto, T.; Yokozawa, M.; Toritani, H.; Shibayama, M.; Ishitsuka, N.; Ohno, H. A crop phenology detection method using
time-series MODIS data. Remote Sens. Environ. 2005, 96, 366–374. [CrossRef]
38. Chen, H.; Liang, Q.; Liang, Z.; Liu, Y.; Xie, S. Remote-sensing disturbance detection index to identify spatio-temporal varying
flood impact on crop production. Agric. For. Meteorol. 2019, 269, 180–191. [CrossRef]
39. Samuels, P.; Klijn, F.; Dijkman, J. An analysis of the current practice of policies on river flood risk management in different
countries. Irrig. Drain. 2006, 55, S141–S150. [CrossRef]
40. Dinh, Q.; Balica, S.; Popescu, I.; Jonoski, A. Climate change impact on flood hazard, vulnerability and risk of the Long Xuyen
Quadrangle in the Mekong Delta. Int. J. River Basin Manag. 2012, 10, 103–120. [CrossRef]
41. Marti, J.; Savin, R.; Slafer, G.A. Wheat yield as affected by length of exposure to waterlogging during stem elongation. J. Agron.
Crop Sci. 2015, 201, 473–486. [CrossRef]
42. Kaur, G.; Zurweller, B.A.; Nelson, K.A.; Motavalli, P.P.; Dudenhoeffer, C.J. Soil waterlogging and nitrogen fertilizer management
effects on corn and soybean yields. Agron. J. 2017, 109, 97–106. [CrossRef]
43. Ploschuk, R.A.; Daniel, J.M.; Timothy, D.C.; Striker, G.G. Waterlogging differentially affects yield and its components in wheat,
barley, rapeseed and field pea depending on the timing of occurrence. J. Agron. Crop Sci. 2020, 206, 363–375. [CrossRef]
44. Meng, Y.; Yu, S.; Yu, Y.; Jiang, L. Flooding depth and duration concomitantly influence the growth traits and yield of rice. Irrig.
Drain. 2021, 9, 94–107. [CrossRef]
45. Yang, J.; Huo, Z.; Wu, L.; Wang, T.; Zhang, G. Indicator-based evaluation of spatiotemporal characteristics of rice flood in
southwest china. Agric. Ecosyst. Environ. 2016, 230, 221–230. [CrossRef]
46. Li, S.; Tompkins, A.M.; Lin, E.; Ju, H. Simulating the impact of flooding on wheat yield–Case study in East China. Agric. For.
Meteorol. 2016, 216, 221–231. [CrossRef]
47. Chipanshi, A.; Zhang, Y.; Kouadio, L.; Newlands, N.; Davidson, A.; Hill, H.; Warren, R.; Qian, B.; Daneshfar, B.; Bedard, F.; et al.
Evaluation of the integrated Canadian crop yield forecaster (ICCYF) model for in-season prediction of crop yield across the
Canadian agricultural landscape. Agric. For. Meteorol. 2015, 206, 137–150. [CrossRef]
48. Biancamaria, S.; Bates, P.D.; Boone, A.; Mognard, N.M. Large-scale coupled hydrologic and hydraulic modelling of the Ob river
in Siberia. J. Hydrol. 2009, 379, 136–150. [CrossRef]
Agriculture 2022, 12, 484 20 of 23

49. Caviedes-Voullième, D.; García-Navarro, P.; Murillo, J. Influence of mesh structure on 2D full shallow water equations and SCS
Curve Number simulation of rainfall/runoff events. J. Hydrol. 2012, 448–449, 39–59. [CrossRef]
50. Cea, M.; Rodriguez, M. Two-dimensional coupled distributed hydrologic–hydraulic model simulation on watershed. Pure Appl.
Geophys. 2016, 173, 909–922. [CrossRef]
51. Fernández-Pato, J.; Caviedes-Voullième, D.; García-Navarro, P. Rainfall/runoff simulation with 2d full shallow water equations:
Sensitivity analysis and calibration of infiltration parameters. J. Hydrol. 2016, 536, 496–513. [CrossRef]
52. Liu, Z.; Merwade, V.; Jafarzadegan, K. Investigating the role of model structure and surface roughness in generating flood
inundation extents using one-and two-dimensional hydraulic models. J. Flood Risk Manag. 2019, 12, e12347. [CrossRef]
53. Shen, D.; Qian, T.; Xia, Y.; Zhang, Y.; Wang, J. Micro-scale Flood Hazard Assessment Based on Catastrophe Theory and an
Integrated 2-D Hydraulic Model: A Case Study of Gongshuangcha Detention Basin in Dongting Lake Area, China. ISPRS Int. J.
Geo-Inf. 2020, 9, 206. [CrossRef]
54. Yamazaki, D.; Kanae, S.; Kim, H.; Oki, T. A physically based description of floodplain inundation dynamics in a global river
routing model. Water Resour. Res. 2011, 47, W04501. [CrossRef]
55. Schumann, J.P.; Neal, J.C.; Voisin, N.; Andreadis, K.M.; Pappenberger, F.; Phanthuwongpakdee, N.; Hall, A.C.; Bates, P.D. A first
large-scale flood inundation forecasting model. Water Resour. Res. 2013, 49, 6248–6257. [CrossRef]
56. Kalyanapu, A.J.; Shankar, S.; Pardyjak, E.R.; Judi, D.R.; Burian, S.J. Assessment of GPU computational enhancement to a 2D flood
model. Environ. Model. Softw. 2011, 26, 1009–1016. [CrossRef]
57. Vacondio, R.; Dal Palù, A.; Ferrari, A.; Mignosa, P.; Aureli, F.; Dazzi, S. A non-uniform efficient grid type for GPU-parallel Shallow
Water Equations models. Environ. Model. Softw. 2017, 88, 119–137. [CrossRef]
58. Wu, Q.; Chen, Y.; Wilson, J.; Liu, X.; Li, H. An effective parallelization algorithm for DEM generalization based on CUDA. Environ.
Model. Softw. 2019, 114, 64–74. [CrossRef]
59. Morales-Hernández, M.; Sharif, B.; Alfred, J.K.; Sheikh, K.G.; Tigstu, T.D.; Gangrade, S.; Kao, S.; Matthew, R.N.; Katherine, J.E.
TRITON: A Multi-GPU open source 2D hydrodynamic flood model. Environ. Model. Softw. 2021, 141, 105034. [CrossRef]
60. Carlotto, T.; Chaffe, P.L.B.; Santos, C.I.D.; Lee, S. SW2D-GPU: A two-dimensional shallow water model accelerated by GPGPU.
Environ. Model. Softw. 2021, 145, 105205. [CrossRef]
61. Armstrong, W. Aeration in higher plants. Adv. Bot. Res. 1980, 7, 225–332.
62. Collaku, A.; Harrison, S.A. Losses in wheat due to waterlogging. Crop Sci. 2002, 42, 444–450. [CrossRef]
63. Kaur, G.; Zurweller, B.; Motavalli, P.P.; Nelson, K.A. Screening Corn Hybrids for Soil Waterlogging Tolerance at an Early Growth
Stage. Agriculture 2019, 9, 33. [CrossRef]
64. Gautam, P.; Lal, B.; Raja, R.; Tripathi, R.; Shahid, M.; Baig, M.J.; Puree, C.; Mohanty, S.; Nayak, A.K. Effect of simulated flash
flooding on rice and its recovery after flooding with nutrient management strategies. Ecol. Eng. 2015, 77, 250–256. [CrossRef]
65. Ding, J.; Liang, P.; Guo, D.; Liu, D.; Yin, M.; Zhu, M.; Li, C.; Zhu, X.; Guo, W. Remedial Application of Urea Eliminates Yield
Losses in Wheat Waterlogged during Stem Elongation. Agriculture 2020, 10, 23. [CrossRef]
66. Ciancio, N.; Daniel, J.M.; Striker, G.G.; Abeledo, L.G. Plant growth rate after, and not during, waterlogging better correlates to
yield responses in wheat and barley. J. Agron. Crop Sci. 2021, 207, 304–316. [CrossRef]
67. Zhang, X.; Zhang, Q. Monitoring interannual variation in global crop yield using long-term AVHRR and MODIS observations.
ISPRS J. Photogramm. Remote Sens. 2016, 114, 191–205. [CrossRef]
68. Rahman, M.S.; Di, L. A Systematic Review on Case Studies of Remote-Sensing-Based Flood Crop Loss Assessment. Agriculture
2020, 10, 131. [CrossRef]
69. Benedetti, R.; Rossini, P. On the use of NDVI profiles as a tool for agricultural statistics: The case study of wheat yield estimate
and forecast in Emilia Romagna. Remote Sens. Environ. 1993, 45, 311–326. [CrossRef]
70. Funk, C.; Budde, M.E. Phenologically-tuned MODIS NDVI-based production anomaly estimates for Zimbabwe. Remote Sens.
Environ. 2009, 113, 115–125. [CrossRef]
71. Esquerdo, J.C.D.M.; Zullo Júnior, J.; Antunes, J.F.G. Use of NDVI/AVHRR time-series profiles for soybean crop monitoring in
Brazil. Int. J. Remote Sens. 2011, 32, 3711–3727. [CrossRef]
72. Son, N.T.; Chen, C.F.; Chen, C.R.; Minh, V.Q.; Trung, N.H. A comparative analysis of multitemporal MODIS EVI and NDVI data
for large-scale rice yield estimation. Agric. For. Meteorol. 2014, 197, 52–64. [CrossRef]
73. Bernardes, T.; Moreira, M.A.; Adami, M.; Giarolla, A. Monitoring biennial bearing effect on coffee yield using MODIS remote
sensing imagery. Remote Sens. 2012, 4, 3760–3763. [CrossRef]
74. Zhang, J.; Feng, L.; Yao, F. Improved maize cultivated area estimation over a large scale combining MODIS–EVI time series data
and crop phenological information. ISPRS J. Photogramm. Remote Sens. 2014, 94, 102–113. [CrossRef]
75. Huete, A.; Didan, K.; Miura, T.; Rodriguez, E.P.; Gao, X.; Ferreira, L.G. Overview of the radiometric and biophysical performance
of the MODIS vegetation indices. Remote Sens. Environ. 2002, 83, 195–213. [CrossRef]
76. Rocha, A.V.; Shaver, G.R. Advantages of a two band EVI calculated from solar and photosynthetically active radiation fluxes.
Agric. For. Meteorol. 2009, 149, 1560–1563. [CrossRef]
77. Mildrexler, D.J.; Zhao, M.; Running, S.W. Testing a MODIS global disturbance index across North America. Remote Sens. Environ.
2009, 113, 2103–2117. [CrossRef]
Agriculture 2022, 12, 484 21 of 23

78. Woodward, D.; Hawkings, R.; Hjelmfelt, A.; Van Mullen, J.; Quan, Q. Curve Number Method: Origins, Applications and
Limitations. In Proceedings of the US Geological Survey Advisory Committee on Water Information—Second Federal Interagency
Hydrologic Modeling Conference, Las Vegas, NV, USA, 28 July–1 August 2002.
79. Bartlett, M.S.; Parolari, A.J.; McDonnell, J.J.; Porporato, A. Beyond the SCS-CN method: A theoretical framework for spatially
lumped rainfall-runoff response. Water Resour. Res. 2016, 52, 4608–4627. [CrossRef]
80. Mishra, S.K.; Singh, V.P. Soil Conservation Service Curve Number (SCS-CN) Methodology (Volume 42), 1st ed.; Kluwer Academic
Publisher: New York, NY, USA, 2003; pp. 88–108.
81. Liang, Q.; Marche, F. Numerical resolution of well-balanced shallow water equations with complex source terms. Adv. Water
Resour. 2009, 32, 873–884. [CrossRef]
82. Song, L.; Zhou, J.; Guo, J.; Zou, Q.; Liu, Y. A robust well-balanced finite volume model for shallow water flows with wetting and
drying over irregular terrain. Adv. Water Resour. 2011, 34, 915–932. [CrossRef]
83. Hou, J.; Liang, Q.; Zhang, H.; Hinkelmann, R. An efficient unstructured MUSCL scheme for solving the 2D shallow water
equations. Environ. Model. Softw. 2015, 66, 131–152. [CrossRef]
84. Xia, X.; Liang, Q.; Ming, X.; Hou, J. An efficient and stable hydrodynamic model with novel source term discretization schemes
for overland flow and flood simulations. Water Resour. Res. 2017, 53, 3730–3759. [CrossRef]
85. Su, B.; Huang, H.; Zhu, W. An urban pluvial flood simulation model based on diffusive wave approximation of shallow water
equations. Hydrol. Res. 2019, 50, 138–154. [CrossRef]
86. Pilar, G.; Javier, M.; Fernández-Pato, J.; Isabel, E.; Mario, M. The shallow water equations and their application to realistic cases.
Environ. Fluid Mech. 2019, 19, 1235–1252.
87. Vreugdenhil, C.B. Numerical Methods for Shallow-Water Flow (Volume 13), 1st ed.; Kluwer Academic Publisher: New York, NY, USA,
1994; pp. 15–41.
88. Dehghan, M.; Abbaszadeh, M.; Dehghan, M.; Abbaszadeh, M. The use of proper orthogonal decomposition (POD) meshless
RBF-FD technique to simulate the shallow water equations. J. Comput. Phys. 2017, 351, 478–510. [CrossRef]
89. Francesca, A.; Federico, P.; Vacondio, R.; Susanna, D.; Alessia, F. A GPU-Accelerated Shallow-Water Scheme for Surface Runoff
Simulations. Water 2020, 12, 637.
90. NVIDIA CUDA. NVIDIA CUDA C++ Programming Guide v11.4. 2021. Available online: https://docs.nvidia.com/cuda/pdf/
CUDA_C_Programming_Guide.pdf (accessed on 24 January 2022).
91. Schumann, G.J.P.; Bates, P.D.; Neal, J.C.; Andreadis, K.M. Technology: Fight floods on a global scale. Nature 2014, 507, 169.
[CrossRef]
92. Yamazaki, D.; Ikeshima, D.; Tawatari, R.; Yamaguchi, T.; O’Loughlin, F.; Neal, J.C. A high-accuracy map of global terrain
elevations. Geophys. Res. Lett. 2017, 44, 5844–5853. [CrossRef]
93. Chen, H.L.; Liang, Q.H.; Liu, Y.; Xie, S. Hydraulic correction method (HCM) to enhance the efficiency of SRTM DEM in flood
modeling. J. Hydrol. 2018, 559, 56–70. [CrossRef]
94. Song, S.; He, R.; Shi, Z.; Zhang, W. Variable Importance Measure System Based on Advanced Random Forest. CMES-Comput.
Modeling Eng. Sci. 2021, 128, 65–85. [CrossRef]
95. Breiman, L. Random forests. Mach. Learn. 2001, 45, 5–32. [CrossRef]
96. Welling, S.H.; Refsgaard, H.H.F.; Brockhoff, P.B.; Clemmensen, L.H. Forest Floor Visualizations of Random Forests. 2016. Available
online: https://arxiv.org/abs/1605.09196 (accessed on 4 July 2016).
97. Liaw, A.; Wiener, M. Classification and Regression by random Forest. R News 2002, 2, 18–22.
98. R Core Team. A Language and Environment for Statistical Computing; R Foundation for Statistical Computing: Vienna, Austria, 2015;
Available online: http://www.R-project.org/ (accessed on 24 January 2022).
99. Ding, Y.; Wang, W.; Song, R.; Shao, Q.; Jiao, X.; Xing, W. Modeling spatial and temporal variability of the impact of climate change
on rice irrigation water requirements in the middle and lower reaches of the Yangtze River, China. Agric. Water Manag. 2017, 193,
89–101. [CrossRef]
100. Sun, Z.; Zhu, X.; Pan, Y.; Zhang, J. Assessing Terrestrial Water Storage and Flood Potential Using GRACE Data in the Yangtze
River Basin, China. Remote Sens. 2017, 9, 1011. [CrossRef]
101. Chinese Office of State Flood Control and Drought Relief Headquarters. Flood disaster of China in 2010. China Flood Drought
Manag. 2011, 21, 1–3.
102. Li, S.; Zhang, Q. Geochemistry of the upper Han River basin, China: 2: Seasonal variations in major ion compositions and
contribution of precipitation chemistry to the dissolved load. J. Hazard. Mater. 2009, 170, 605–611. [CrossRef]
103. Liu, W.; Liu, G.; Zhang, Q. Influence of vegetation characteristics on soil denitrification in shoreline wetlands of the Danjiangkou
Reservoir in China. Clean-Soil Air Water 2011, 39, 109–115. [CrossRef]
104. Huete, A.; Didan, K.; Leeuwen, W.V.; Miura, T.; Glenn, E. MODIS Vegetation Indices. In Land Remote Sensing and Global
Environmental Change; Ramachandran, B., Christopher, O.J., Michael, J.A., Eds.; Springer: New York, NY, USA; Dordrecht, The
Netherlands; Heidelberg, Germany; London, UK, 2011; Volume 11, pp. 579–602.
105. Ticehurst, C.; Guerschman, J.P.; Chen, Y. The strengths and limitations in using the daily MODIS open water likelihood algorithm
for identifying flood events. Remote Sens. 2014, 6, 11791–11809. [CrossRef]
106. Huete, A.; Justice, C.O.; Van Leeuwen, W. MODIS Vegetation Index (MOD13): Algorithm Theoretical Basis Document. 2011.
Available online: http://modis.gsfc.nasa.gov/data/atbd/atbd_mod13.pdf (accessed on 8 February 2011).
Agriculture 2022, 12, 484 22 of 23

107. Arino, O.; Perez, J.J.R.; Kalogirou, V.; Bontemps, S.; Defourny, P.; Van Bogaert, E. Global Land Cover Map for 2009 (GlobCover
2009). European Space Agency (ESA) & Université Catholique de Louvain (UCL), PANGAEA. 2012. Available online: https:
//doi.org/10.1594/PANGAEA.787668 (accessed on 24 January 2022). [CrossRef]
108. Nachtergaele, F.O.; Velthuizen, V.L.; Batjes, N.H.; Dijkshoorn, J.A.; van Engelen, V.W.P.; Fischer, G.; Jones, A.; Montanarella, L.;
Petri, M.; Prieler, S.; et al. Harmonized World Soil Database (Version 1.2). 2012. Available online: https://research.wur.nl/en/
publications/harmonized-world-soil-database-version-12 (accessed on 24 January 2022).
109. Lehner, B.; Verdin, K.; Jarvis, A. New global hydrography derived from spaceborne elevation data. Eos Trans. Am. Geophys. Union
2008, 89, 93–94. [CrossRef]
110. Zhang, Q.; Gu, X.; Singh, V.P.; Liu, L.; Kong, D. Flood-induced agricultural loss across China and impacts from climate indices.
Glob. Planet. Change 2016, 139, 31–43. [CrossRef]
111. Tehrany, M.S.; Pradhan, B.; Mansor, S.; Ahmad, N. Flood susceptibility assessment using GIS-based support vector machine
model with different kernel types. Catena 2015, 125, 91–101. [CrossRef]
112. Kazakis, N.; Kougias, I.; Patsialis, T. Assessment of flood hazard areas at a regional scale using an index-based approach and
Analytical Hierarchy Process: Application in Rhodope–Evros region, Greece. Sci. Total Environ. 2015, 538, 555–563. [CrossRef]
[PubMed]
113. Samela, C.; Troy, T.J.; Manfreda, S. Geomorphic classifiers for flood-prone areas delineation for data-scarce environments. Adv.
Water Resour. 2017, 102, 13–28. [CrossRef]
114. Tang, Z.; Yi, S.; Wang, C.; Xiao, Y. Incorporating probabilistic approach into local multi-criteria decision analysis for flood
susceptibility assessment. Stoch. Environ. Res. Risk Assess. 2018, 32, 701–714. [CrossRef]
115. Chau, V.N.; Cassells, S.; Holland, J. Economic impact upon agricultural production from extreme flood events in Quang Nam,
central Vietnam. Nat. Hazards 2015, 75, 1747–1765. [CrossRef]
116. Dao, P.; Liou, Y. Object-based flood mapping and affected rice field estimation with Landsat 8 OLI and MODIS data. Remote Sens.
2015, 7, 5077–5097. [CrossRef]
117. Sande, C.J.V.D.; Jong, S.M.D.; Roo, A.P.J.D. A segmentation and classification approach of ikonos-2 imagery for land cover
mapping to assist flood risk and flood damage assessment. Int. J. Appl. Earth Obs. Geoinf. 2003, 4, 217–229. [CrossRef]
118. Jonkman, S.N.; Bočkarjova, M.; Kok, M.; Bernardini, P. Integrated hydrodynamic and economic modelling of flood damage in the
Netherlands. Ecol. Econ. 2008, 66, 77–90. [CrossRef]
119. Vozinaki, A.E.K.; Kourgialas, N.N.; Karatzas, G.P. Estimation of agricultural flood loss in the Koiliaris river basin in Crete, Greece.
Eur. Water. 2012, 39, 53–63.
120. Vozinaki, A.E.K.; Karatzas, G.P.; Sibetheros, I.A.; Varouchakis, E.A. An agricultural flash flood loss estimation methodology: The
case study of the Koiliaris basin (Greece), February 2003 flood. Nat. Hazards 2015, 79, 899–920. [CrossRef]
121. Juan, P.; Luis, T. Performance assessment of two-dimensional hydraulic models for generation of flood inundation maps in
mountain river basins. Water Sci. Eng. 2019, 12, 11–18.
122. Jia, D.W.; Zhong, S.Q.; Li, X.; Peng, B. Extracting water information from the HSI data of HJ-1A/B satelite. Sci. Surv. Mapp. 2011,
36, 128–130.
123. Wang, Q.Y.; Chen, R.X.; Xu, J.; Chen, X.L. Research on methods for extracting water body information from HJ-1A/B data. Sci.
Technol. Eng. 2012, 12, 3051–3056.
124. Liu, H.; Zheng, L.; Jiang, L.; Liao, M.W. Forty-year water body changes in Poyang Lake and the ecological impacts based on
Landsat and HJ-1 A/B observations. J. Hydrol. 2020, 589, 125–161. [CrossRef]
125. Mohammadi, A.; Costelloe, J.F.; Ryu, D. Application of time series of remotely sensed normalized difference water, vegetation
and moisture indices in characterizing flood dynamics of large-scale arid zone floodplains. Remote Sens. Environ. 2017, 190, 70–82.
[CrossRef]
126. Singh, D.S.; Prajapati, S.K.; Singh, P.; Singh, K.; Kumar, D. Climatically induced levee break and flood risk management of the
Gorakhpur region, Rapti River basin, Ganga Plain, India. J. Geol. Soc. India 2015, 85, 79–86. [CrossRef]
127. Shrestha, B.B.; Kawasaki, A. Quantitative assessment of flood risk with evaluation of the effectiveness of dam operation for flood
control: A case of the Bago River Basin of Myanmar. Int. J. Disaster Risk Reduct. 2020, 50, 101707. [CrossRef]
128. Brémond, P.; Grelot, F.; Agenais, A.L. Review article: Economic evaluation of flood damage to agriculture—Review and analysis
of existing methods. Nat. Hazards Earth Syst. Sci. 2013, 13, 2493–2512. [CrossRef]
129. Vempi, S.A.H.; Daisuke, K. Developing flood vulnerability curve for rice crop using remote sensing and hydrodynamic modeling.
Int. J. Disaster Risk Reduct. 2021, 54, 102058.
130. Yin, H.; Liu, G.; Pi, J.; Chen, G.; Li, C. On the river–lake relationship of the middle Yangtze reaches. Geomorphology 2007, 85,
197–207. [CrossRef]
131. Li, X.T.; Yu, C.Z.; Peng, L.C.; Zhang, F.F.; Li, J.; Yan, F.; Dong, Y.; Feng, B.L. How Does the Waterlogging Regime Affect Crop
Yield? A Global Meta-Analysis. Front. Plant Sci. 2021, 12, 634898–634907.
132. McDonald, G.T. Agricultural flood damage assessment: A review and investigation of a simulation method. Rev. Mark. Agric.
Econ. 1970, 38, 105–120.
133. Lacewell, R.D.; Eidman, V.R. A General Model for Evaluating Agricultural Flood Plains. Am. J. Agric. Econ. 1972, 54, 92–101.
[CrossRef]
Agriculture 2022, 12, 484 23 of 23

134. Hoes, O.; Schuurmans, W. Flood standards or risk analyses for polder management in the Netherlands. Irrig. Drain. 2006, 55,
113–119. [CrossRef]
135. Lacewell, R.D.; Freeman, R.; Petit, D.; Rister, E.; Sturdivant, A.; Ribera, L.; Zinn, M. Update of Estimated Agricultural Benefits
Attributable to Drainage and Flood Control in Willacy County, Texas. Texas Water Resources Institute. 2006. Available online:
https://hdl.handle.net/1969.1/6084 (accessed on 24 January 2022).
136. Dutta, D.; Herath, S.; Musiake, K. A mathematical model for flood loss estimation. J. Hydrol. 2003, 277, 24–49. [CrossRef]
137. Förster, S.; Kuhlmann, B.; Lindenschmidt, K.E.; Bronstert, A.L. Assessing flood risk for a rural detention area. Nat. Hazards Earth
Syst. Sci. 2008, 8, 311–322. [CrossRef]
138. Kwak, Y.; Shrestha, B.B.; Yorozuya, A.; Sawano, H. Rapid damage assessment of rice crop after large-scale flood in the Cambodian
floodplain using temporal spatial data. IEEE J. Sel. Top. Appl. Earth Obs. Remote Sens. 2017, 8, 3700–3709. [CrossRef]
139. Vega-Serratos, B.E.; Domínguez-Mor, R.; Posada-Vanegas, G. Seasonal flood risk assessment in agricultural areas. Tecnol. Cienc.
Agua. 2018, 9, 92–127. [CrossRef]
140. Nishiuchi, S.; Yamauchi, T.; Takahashi, H.; Kotula, L.; Nakazono, M. Mechanisms for coping with submergence and waterlogging
in rice. Rice 2012, 5, 2. [CrossRef]