Modeling and Prediction

Z Yu, College of Hydrology and Water Resources, Hohai University, Nanjing, China; and University of Nevada
Las Vegas, Las Vegas, NV, USA
Ó 2015 Elsevier Ltd. All rights reserved.

Synopsis

This article describes the general approach for hydrologic modeling and prediction. After the hydrologic system is concep-
tualized, mathematical model can be set up and numerical model can be developed. Then data such as precipitation, land
use–land cover, digital elevation, and soil texture need to be prepared for the hydrologic simulation. Before the model can be
used for the prediction, it needs to be calibrated for assessing how well the simulation can reproduce the observed hydrologic
variables.

Introduction crude and the scales of measurement for many hydrologic

parameters are incompatible with the scales used in the
Hydrologic models have become an indispensable tool to study models. Studies indicated that the uniform effective parameters
hydrologic processes and the impact of natural variability and (e.g., saturated hydraulic conductivity) cannot represent the
anthropogenic factors on the hydrologic system. Mathematical ensemble heterogeneity, resulting in a significant underesti-
models, which are governed by the laws for conservations of mation of hydrologic variables (e.g., streamflow). These phys-
mass and momentum, are used to describe the temporal and ically based distributed models can be referred to as conceptual
spatial variation on various physical processes in the hydrologic lumped models in some sense. The physically based models are
system along with information concerning climate, land use– more complex and require more computing time for solving
land cover, and hydrology. Modeling the hydrologic response to PDEs numerically and considerable effort to master all their
various natural and human-induced changes has the potential to intricacies, such as model calibration. For simulating the
contribute to the understanding of these physical processes such hydrologic response (e.g., soil moisture content, groundwater
as flow and solute transport in the surface and subsurface and level) to climate forcing (e.g., storms and human-induced
the atmosphere–land surface interaction. global warming), these models currently offer no advantage
Two types of hydrologic models have been used in most over the traditional conceptual lumped water-balance models.
applications: lumped conceptual models and physically based Physically based hydrologic models are an important
modes. A lumped model is generally applied in a single point or evolutionary step in representing hydrologic processes and
a region for the simulation of various hydrologic processes. The spatially distributed data. At the present time, the ability to
parameters used in the lumped model represent spatially aver- represent spatially varying processes is outstripping our ability
aged characteristics in a hydrologic system and are often unable to collect various data sets. The need for the research on the
to be directly compared with field measurements. In general, better representation of physical processes in space and time is
lumped models use simple bookkeeping procedures to quantify evident given the increasing availability of digital products
physical processes by simulating the temporal variation of (e.g., distributions of elevation, soil, vegetation) and remotely
various physical processes in a hydrologic system. The advantage sensed data (e.g., soil moisture, vegetation), along with new
of these models over physically based models is that the remote sensing technologies for measuring temporal and
conceptual parameterization in the models is simple and spatial variability in precipitation.
computation is efficient. With the availability of spatially Research on data assimilation and analysis, subgrid-scale
distributed digital and remotely sensed data sets such as variability in precipitation and hydraulic parameters, subgrid-
precipitation, elevation, vegetation, etc., many distributed lum- scale model calibration, prediction uncertainty analysis, and
ped models have been developed in recent years. These kinds of inclusion of fine-resolution surface and subsurface hydrology
models have been widely used in climate, meteorological, and in various hydrologic and climatic models are being conducted
hydrologic studies to simulate hydrologic processes. in various research institutes to facilitate an interactive link
Many physically based distributed-parameter models have between the hydrosphere and the atmosphere. Simulation and
been developed to facilitate various hydrologic and climatic prediction of various physical processes in the hydrologic
applications over recent years, especially for simulating system are among the principal areas of study in the current
spatiotemporal variation in hydrologic processes under climate fields of hydrology and climate.
changes such as global warming. These models represent
hydrologic processes in a physically rigorous manner because
they use process-based partial differential equations (PDEs) to General Approach of Hydrologic Modeling
describe the spatial variability of hydrologic processes over
time. Within this framework provided by these equations, it is Climate variability has substantial impact on hydrologic,
possible to estimate governing parameters through field biologic, and ecologic systems such as water availability and
studies. One disadvantage of such models is that the repre- quality, floods, and droughts. Such climate variability has been
sentation of physical processes in these models is often too intensified due to the human-induced climate changes and

Encyclopedia of Atmospheric Sciences 2nd Edition, Volume 3 [Link] 217

218 Hydrology, Floods and Droughts j Modeling and Prediction

Global climate model

Regional climate model

Surface runoff Fractional areas
recharge for each
lake column

Overland 1
1
2
flow 3
•
•
• 1
Infiltration N–
N
w
l flo
nne
Cha
er
at
w
nd
Water table ou
gr ion
e
n c t
zo ra
s e inte
do
Lake groundwater Channel groundwater Or Channel vadose zone Va
interaction interaction interaction

Hydrology model
Figure 1 Frame structure of coupled climate and hydrology models. Source: Yu, Z., Pollard, D., and Cheng, L., 2006. On continental-scale hydrologic
simulations with a coupled hydrologic model. Journal of Hydrology 331, 110–124.

results in many extreme flood and drought events in recent insufficient progress on the theoretical side, related to defi-
years. The effects of climate variability on economic vitality and ciencies in our abilities to couple atmospheric and hydrologic
the quality of life (e.g., 1988 North American drought and processes that occur over spatial scales ranging from micro-
1993 Mississippi flood in the USA) indicate that future climate scopic to global (specifically ranging from 10
4 to 106 m) and
change will be of considerable global, national, and regional over time scales ranging from 10
4 to 107 s. Various parame-
importance. The current emphasis on the climate impact terization schemes have been used to reconcile the discordant
analysis and hydrologic modeling requires improved specifi- scales of hydrologic and atmospheric models. The approach
cations of the interactions between the atmosphere and the involves numerical modeling to bridge the gap in scale between
hydrosphere. The difficulty in reconciling the discordant scales global circulation models (GCMs) and hydrologic models by
in climate change and hydrologic research hampers the embedding a regional climate model within a GCM so that the
understanding of how regional hydrologic systems respond to natural variability at the regional scale can be represented
climate forcing. Part of the difficulty can be attributed to within the system (Figure 1).
 Hydrology, Floods and Droughts j Modeling and Prediction 219

The nested climate models used for studying the effect of meteorological and hydrologic data for the hydrologic simu-
climate change on the hydrologic system indicate that changes lation. Additional data can be obtained through laboratory
in the hydrologic system induced by a climatic change could and field works, if necessary. After all required data being
lead to a further change in climate, ecosystems, agriculture, and integrated into the models, the models should be calibrated to
hydrology. Many comprehensive hydrologic models such as available field-observed data (i.e., streamflow, groundwater
the Biosphere–Atmosphere Transfer Scheme have been devel- level). So the calibrated models can then be used to predict the
oped to integrate hydrology into the climate system by hydrologic response to various external forcings (e.g., future
coupling the physical processes from different systems. Such an water usage, land use–land cover change, and climate change).
integrated hydrologic model system, including components
from soil hydrology, surface water hydrology, and groundwater
hydrology, is necessary to understand how perturbed climate
System Conceptualization, Mathematical
conditions can modify regional hydrologic systems and, in
Model Setup, and Model Development
turn, how the modified hydrologic conditions can modify the
climatic system.
There are many surface and subsurface hydrologic processes in
Traditionally most hydrologic models have been developed
the hydrologic system (Figure 1). Many conceptual and
to simulate individual hydrologic processes such as rainfall–
mathematical models have been developed to describe these
runoff partitioning, soil moisture flow, overland flow, and
physical processes in time and space and various numerical
groundwater flow while some comprehensive models are used
schemes (i.e., finite-difference and finite-element methods) are
to simulate all the processes within the hydrologic cycle. In
available to solve various PDEs for the simulation of hydro-
general, any use of hydrologic models for hydrologic and
logic processes. The following sections describe some of the
climate studies follows five main steps (Figure 2). After the
major flow processes such as soil water flow, overland flow, and
purpose of hydrologic modeling is well-defined, the conceptual
groundwater flow while other processes such as sediment and
model then can be constructed to describe the hydrologic
contaminant transport will not be described here.
system and its components that need to be simulated. In the
When the precipitation reaches the ground, it partitions into
second step, the various mathematical models with various
components of surface runoff and infiltration. After the water
PDEs can be used to describe individual processes within the
enters the soil column, the soil water flow in the unsaturated
hydrologic system and computer codes in various computer
zone can be described by the following form of Richards’
languages (e.g., C, Fortran) can be developed to solve these
equation:
PDEs in the mathematical models. In general, the numerical
solution of these PDEs must be verified by comparing to

vqðz; tÞ v vjðz; tÞ vqðz; tÞ vKðz; tÞ vqðz; tÞ
known or analytic solutions. The second step can also be
¼ Kðz; tÞ þ (1)
vz vz vq vz vq vz
completed by selecting existed computer codes that have been
verified in previous applications (e.g., MODFLOW). The third where q is the vertical moisture flux, q is the volumetric water
step, data preparation, is to compile and analyze available content, z is depth, t is time, K is the unsaturated hydraulic
conductivity, and j is the hydraulic potential. The Richards’
equation can be numerically solved using the Crank–
Nicholson numerical scheme and a finite-difference scheme of
Conceptualization forward in time and backward in space or other numerical
methods. Various schemes are available for relating K and j to
the volumetric water content because K at a given time is
Mathematical model a function of constant saturated hydraulic conductivity and soil
moisture status. Infiltration and evaporation can be treated as
either a source or a sink, respectively, in Richards’ equation or
the upper boundary condition of the soil profile. The calcula-
Model development
tion of evaporation on the bare soil and evapotranspiration on
the vegetation canopy can be completed with various methods
(e.g., the Penman–Monteith method).
Data preparation The overland flow will form soon after the rainfall–runoff
partitioning and can be formulated as a kinematic wave with
comparison with field data a flow-direction algorithm to account for the overland flow
delay and storage in each grid cell. Mathematically, one-
dimensional overland flow may be described as
Calibration
simulation vd vq
þ ¼ ie ; q ¼ adm (2)
vt vx
where d ¼ d(x,t) is the depth of overland flow, q ¼ q(x,t) is the rate
Prediction of overland flow, ie is the excess rainfall rate, a is conveyance,
simulation and m is surface roughness. The overland flow is discretized
into a large number of smaller segments to solve the kinematic-
Figure 2 Flow chart of hydrologic model application. wave flow and the equation can be solved numerically for
220 Hydrology, Floods and Droughts j Modeling and Prediction

the one-dimensional flow. In general, a conservative numerical Data Compilation and Preparation
scheme is implemented in the model to maintain the numer- for Hydrologic Modeling
ical stability in some extreme cases.
The overland flow quickly reaches nearby streams to form Data compilation and preparation is a major fundamental part
channel flow. The channel flow can be simulated hydraulically of hydrologic modeling. The work involves compiling various
and hydrologically. The Muskingum–Cunge method or other available field and digital data sets and processing and inte-
channel routing schemes can be used for channel-flow routing grating these data sets into models for hydrologic simulations.
through the channel networks that are derived from digital Sources of data can be from various government agencies and
elevation models (DEMs). For instance, the extended finite- institutes. With the availability of new technology, ground-
difference form of the governing equation can be written in the based and remotely sensed data sets as well as the collected
Muskingum channel routing method: data through traditional field methods have been used for
hydrologic modeling. Some data are required to be reprojected
d
K ½3Qx þ ð1
3QÞxþ1  ¼ Qx
Qxþ1 (3) or reformatted for the specific models.
dt For topography-based hydrologic models, DEMs are
where Qx is the inflow to a given stream reach, Qxþ1 is the routinely used to describe the land surface topography that
outflow from a given stream reach, K is a travel time parameter, contains crucial information for the surface water flow and the
and 3 is a weighting factor. The routing equations are produced interaction of surface water and groundwater. After DEMs are
using a finite-difference form of the above equation. The conditioned for removing spurious sinks and peaks, DEMs can
outflow of a stream reach is calculated at each time step and be used to determine the flow direction of surface water at
routed to downstream reach along DEM-derived channel a grid cell by comparing the elevation at a grid cell to elevations
networks, eventually to the basin outlet. of neighboring grid cells within the simulation domain
The following second-order PDE is used to describe assuming that the water always flows from high level to low
groundwater flow in an aquifer: level. With DEMs, a flow accumulation data set can be obtained
    in which the contributing area that drains to a grid cell is
v vh v vh vh
T þ T ¼ S þ Qnet (4) determined for each grid cell. The DEMs and derived data sets
vx vx vy vy vt
of flow direction and flow accumulation can then be processed
where T is transmissivity, h is hydraulic head, S is storativity, t is to delineate drainage and stream networks, overland flow
time, and Qnet is net groundwater withdrawal rate. Qnet paths, watershed and subwatershed boundaries, and other
includes groundwater recharge from infiltration, evaporation features. Derived stream networks facilitate the channel–
of shallow groundwater, withdrawal of groundwater from shallow groundwater interaction while the derived overland
wells, and other sources or sinks, such as induced infiltration of flow paths provide accurate prediction of the rainfall–runoff
groundwater from the stream network. partitioning and soil moisture status over overland flow paths.
In general, the simulation domain for a basin is discretized Simulations of various hydrologic processes require
into a set of rectangular cells with dimensions Dz, Dx, and Dy. spatially distributed information on land use–land cover, soil
Dz is the vertical thickness of the aquifer, and Dx and Dy are type, and hydraulic properties besides the topography. Many
the cell dimensions in the x- and y-coordinate directions digital soil data sets are available from various sources. For
(Figure 2). A finite-difference or finite-element scheme can be instance, the Natural Resources Conservation Service, U.S.
used to discretize the hydrologic system for the numerical Department of Agriculture, compiled the State Soil Geographic
solution of the flow domain. Many iterative methods such as (STATSGO) soils data. In general, these digital data sets are
the method in MODFLOW (developed by U.S. Geological stored in a Geographical Information System format. The
Survey (USGS)) are available for obtaining solutions for the hydraulic parameters (i.e., the average saturated hydraulic
above PDE. conductivity and the average capillary suction) data sets can be
The interaction between a stream system and groundwater derived by assigning a value for each grid cell based on various
can be simulated with Darcy’s law. The calculated inflow/ parameterization schemes of hydraulic variables and the soil
outflow is a function of the hydraulic head difference between texture from STATSGO soil data. Then, these derived data sets
the groundwater and the river stage at a given stream reach. of various hydraulic parameters can be integrated into models
When the groundwater level is lower than the streambed, water for the hydrologic simulation.
flows from the stream to the groundwater. When the ground- With the development of remote sensing and satellite
water level is lower than the streambed, the stream will dry out technology, these data sets hold great potential for practical
and eventually there will be no channel groundwater application to regional ecology, hydrology, and planning. The
interaction. satellite data can be used to derive a variety of surface param-
After the implementation of numerical schemes for solving eters, such as radiant surface temperature and vegetation
various PDEs in the hydrologic system, the developed codes for fraction; these variables can be used for both spatial and
hydrologic models need to be tested against to known and temporal comparisons and for regional hydrologic modeling of
analytic solutions before they can be used for modeling important hydrologic processes. The remotely sensed data such
applications. For the specific applications, the simulation as vegetation classification can be obtained from the AVHRR
domain must be set up with defined boundary conditions. So and Landsat TM images. Spatial distributed data of meteoro-
the given forcings (i.e., meteorological data and water usage) logical variables (e.g., precipitation, temperature, wind speed)
can be used to drive the simulation of various hydrologic are available from NEXRAD, radar, and satellite images at
processes. a time resolution from minutes to days. With the improved
 Hydrology, Floods and Droughts j Modeling and Prediction 221

REAL (UNKNOWN) Measured

FLOW SYSTEM output Acceptable CALIBRATED
error MODEL

Available measured
input
ERROR
ANALYSIS

Unacceptable PARAMETER
Initial estimates error ADJUSTMENT
NUMERICAL Computed
of parameters MODEL output

New parameters

Figure 3 Procedure of model calibration. Source: Freeze, R.A., and Cherry, J.A., 1979. Groundwater. Prentice-Hall, pp. 604.

resolution on these data sets, the more accurate simulation of numerical models are now equipped with a semi-automated or
various hydrologic processes is expected in both temporal and automated procedure to optimize one or multiple parameters.
spatial scales. The range of adjustment to values of hydrologic and hydraulic
Different types of field-observed data are available for the parameters must be constrained by plausible site-specific field
hydrologic simulation and model calibration that will be data such as streamflow, water levels, hydraulic conductivity,
discussed in the next section. There are networks of rain gages and so on. The difficulty in achieving a good calibration is that
over the world. At outlets of many river systems, gage stations are boundary conditions and values of hydrologic and hydraulic
available to measure daily streamflow rates and other hydrologic parameters are always known with uncertainty.
variables. Groundwater levels are also available from many wells Goodness-of-fit calibration can be evaluated through visual
and many other data such as evapotranspiration and moisture comparison and statistical measures. Visual comparison
content are obtainable from local specific regions. For instance, includes scatterplot of simulated versus measured variables,
USGS provides infrastructure for monitoring stations and simulated and field-based temporal and spatial distribution,
collecting measured data of streamflow and groundwater level and spatial distribution of residuals. Statistical measures
while the National Oceanic and Atmospheric Administration consist of mean error, absolute mean error, and root mean-
maintains the collection of meteorological data. squared error, between simulated and observed variables.
The trial-and-error calibration involves with manually
adjusting parameters to match the simulated results
Model Calibration (e.g., streamflow and groundwater level hydrographs) with
field-observed historical records. The fitness between the
Complexity in the hydrologic modeling over large space and simulation and observation can be evaluated visually and
long times has prompted a significant need for model statistically after each simulation with changed parameters.
calibration or parameter optimization. Model calibration is Knowledgeable modelers are able to conduct acceptable model
a demonstration that the model is capable of reproducing field- calibration with this approach. On the other hand, the cali-
observed values of various hydrologic variables (e.g., stream- bration process could be a very frustrating and time-consuming
flow, soil moisture, and well-observed groundwater level) practice for many untrained modelers and sometimes even for
(Figure 3). Prediction of various hydrologic variables based on experienced modelers when many parameters are changed
an uncalibrated flow model are sterile and undefensible. simultaneously. The disadvantage for such approach is the low
Generally, the goodness-of-fit between simulated and efficiency and the lack of standard measures in evaluating the
measured variables is not satisfactory based on the initial performance of model calibration.
values of hydrologic and hydraulic parameters used in the While the trial-and-error calibration is still the most
model. The goodness-of-fit can be improved by the adjustment common method for calibrating models of all kinds, many
or optimization of these parameter values until the difference automatic or semi-automatic direct or indirect techniques are
between simulated and measured variables is satisfactory available to solve an inverse problem in various hydrologic
during this model practice. The adjustment process most systems. Automatic direct or indirect model calibrations
commonly is based on trial-and-error changes in a parameter require setting up objective functions, developing optimiza-
while other parameter values are held constant. Some tion algorithms, setting up termination criteria, and of course,
222 Hydrology, Floods and Droughts j Modeling and Prediction

using observed data or derived variables as calibration targets. In a matrix form, the above equation is written as
The direct approaches use an inverse operator to solve the Jdp ¼ r (8)
problems directly, while the indirect approaches use an
iterative scheme. For the indirect approaches, various where J is a rectangular Jacobian matrix, p is the parameter
sampling procedure schemes (e.g., Monte Carlo and Latin vector to be inverted, dp is the parameter correction vector,
Hypercube samplings) are available to sample data that cover and r is the residual vector between the measured and
the range of adjustable variables systematically. In effect, simulated hydraulic heads. In order to solve the above
the simulation problem is solved many times over with the equation, the transpose of matrix J multiplies both sides of
coverage of the potential solution space. Ultimately, the the equation. The result is a system of equations with m
process finds the optimum solution, providing a set of unknowns. A problem occurs because the solution of matrix
parameters that yield the best fit between the observed data system is unstable. Marquardt’s approach, which involves
and simulated results. adding terms to the diagonals of the matrix JTJ, can over-
For any model calibration, objective functions need to be come this problem. The Levenberg–Marquardt method or
set up for variables (e.g., streamflow, groundwater level) that Marquardt method (iterative procedures) can be used to
will be optimized during the calibration process. The least- solve this system.
squares method, a common method in the model calibration, After the model calibration, the sensitivity analysis is
is equivalent to the minimization of the sum of squares of the needed to investigate the uncertainty in the calibrated model
residuals (also called objective function or fitness measure). caused by the uncertainty in the estimation of various param-
The general form for a steady state is given by eters. The sensitivity analysis can be conducted by evaluating
the hydrologic flow system in response to various parameter
X
n
disturbances based on the calibrated hydraulic parameters. The
Obj ¼ wi ½soi
sci 2 (5)
i¼1
analysis involves perturbing the values of the hydraulic
parameters with respect to the best estimates and examining
where wi is a weighted factor, soi is the observed data, sci is how the simulated hydrologic variables change. The common
a simulated result, i is the index for location points, and n is the approach is to evaluate one parameter at a time; all other
number of location points. parameters are kept constant while a given parameter is being
The weighted factor represents the percentage of importance evaluated. Then various parameters are varied simultaneously
allocated to a variable at particular locations. If the sum of all to examine the behavior of the hydrologic flow system under
weighted factors is 1.0, the above equation becomes the simple different situations. By doing this, the nonuniqueness of
least-squares function. It is feasible to zone spatially for both solution of the parameter estimations for the hydrologic flow
steady state and transient state, although execution times will system can be evaluated and other issues along with the model
increase significantly, especially for the transient state, as the uncertainty can be assessed.
degrees of freedom in parameter values increase. The objective
function can be constructed for hydrologic variables that can be
used as calibration targets (e.g., streamflow and groundwater Model Prediction
level). Objective functions, based on a weighted least-square
error criterion, can then be used to drive the calibration process. With the set of optimized hydraulic parameters from the model
Different combinations of various changes in different opti- calibration, the calibrated model can be used to conduct
mizing parameters are selected for each simulation run. The predictive simulations. The predictive simulations can be
objective is to minimize the difference between measured data driven by external forcings (e.g., climate change scenarios) to
and simulated results. The objective function for a transient derive the system response to future events. Some climate and
simulation can also be formed as hydrology problems require the hydrologic modeling to
m X
X n predict the system response for as many as thousands of years.
Obj ¼ wi; j ½soi; j
sci; j 2 (6) For instance, the risk assessment of contaminant transport for
j¼1 i¼1 storing low-level nuclear wastes in arid regions requires pre-
dicting the water flow in the unsaturated and saturated zones
where m is the number of time steps, j is the index for time
for a time scale of 10 000 years.
steps, and other variables are as defined previously. The tran-
How good predictions are depends on how well the
sient calibration requires much more computational time than
model is calibrated assumed mathematical and numerical
the steady-state calibration due to the additional time dimen-
models are properly established and how creditable future
sion. There are other alternatives to the least-squares method
external forcings are, which are used to drive predictive
such as the maximum-likelihood approach for objective
simulations. Because of the nonlinearity of natural hydro-
functions.
logic systems and simplified model structures or assump-
The direct approaches use an inverse operator to solve the
tions as compared to the actual complex natural physical
problems directly. Let ho be the observed hydrologic variables
processes, the calibration process cannot guarantee that the
and hc be the simulated hydrologic variables. Then, according
global minimum can be found and so the calibrated model
to Taylor’s series expansion, the equation can be written as
could produce unrealistic results beyond the confident
X
m
vhc period. In general, the calibrated model should not be used
i
dpj ¼ hoi ðp þ dpÞ
hci ðpÞ i ¼ 1; .; n (7) to predict the system in the future longer than twice the
j¼1
vpj
period of model calibration with available field-observed
 Hydrology, Floods and Droughts j Modeling and Prediction 223

data. In some cases, due to the lack of available long-period Further Reading
historical data for model calibration and the nature of some
applications that require an extended period prediction, Anderson, M.P., Woessner, W.W., 1992. Applied Groundwater Modeling. Academic
Press, San Diego, CA, pp. 381.
there will be always uncertainty in predictive simulations. So Beven, K.J., Moore, I.D., 1993. Terrain Analysis and Distributed Modeling in Hydrology.
users must conduct a detailed assessment of the uncertainty John Wiley & Sons, New York, NY, pp. 249.
on any future prediction of system response as well as the Freeze, R.A., Cherry, J.A., 1979. Groundwater. Prentice-Hall, Upper Saddle River, NJ,
uncertainty associated with model calibration and future pp. 604.
Maidment, D.R., 1993. Handbook of Hydrology. McGraw-Hill, New York, NY.
external forcings.
Neuman, S.P., 1973. Calibration of distributed parameter groundwater flow models
The hydrologic cycle plays an important role in the climate viewed as a multiple-objective decision process uncertainty. Water Resources
system. The two-way interaction between the hydrologic and Research 9, 1006–1021.
climate is a crucial element in our current and future hydrology Singh, V.P., 1995. Computer Models of Watershed Hydrology. Water Resource
and climate modeling to understand how perturbed climate Publications, Littleton, CO, pp. 1130.
Yang, C., Lin, Z., Yu, Z., Hao, Z., Liu, S., 2010. Analysis and simulation of human
conditions can modify regional hydrologic systems and, in activity impact on streamflow in the Huaihe river basin with a large-scale hydrologic
turn, how the modified hydrologic conditions can influence the model. Journal of Hydrometeorology 11, 810–821.
climatic system. Data assimilation and analysis of remotely Yu, Z., Schwartz, F.W., 1999. Automated calibration applied to constrained ground-
sensed and satellite-derived information should be integrated water flow modeling. Hydrological Processes 13, 191–209.
Yu, Z., Lakhtakia, M.N., Yarnal, B., Johnson, D.L., White, R.A., Miller, D.A.,
into hydrologic models to better represent subgrid-scale spatial
Barron, E.J., Duffy, C., Schwartz, F.W., 1999. Simulation of the hydrologic
and temporal variability in meteorological and hydrologic response to atmospheric forcing in large river basins: linking a mesoscale mete-
parameters. The model calibration and subgrid-scale model orological model and a hydrologic model system. Journal of Hydrology 218, 72–91.
calibration remain as an important step toward the better Yu, Z., Barron, E.J., Schwartz, F.W., 2000. Retrospective simulation of a storm event:
simulation of various hydrologic components. In addition, a first step in coupled climate/hydrologic modeling. Geophysical Research Letters
27, 2561–2565.
prediction of uncertainty analysis and inclusion of fine- Yu, Z., Pollard, D., Cheng, L., 2006. On continental-scale hydrologic simulations with
resolution surface and subsurface hydrology are some impor- a coupled hydrologic model. Journal of Hydrology 331, 110–124.
tant topics in such an interactive link.