28
S. Mandal and S. Biswas
1 Introduction
Rainfall-runoff (RR) models are of paramount importance for the strategic planning of water resources and its development. This issue becomes more crucial and
challenging as the population grows in a particular area and makes the basin more
heterogeneous from the landcover perspective. The difficulty in the analysis exists
mainly due to the complex nonlinear relationships in the transformation of rainfall
into runoff. Inconsistent watershed characteristics and overland, as well as inland
flow generation processes, make this transformation more complex in modeling.
Nonuniform precipitation patterns, along with the uneven distribution of soil moisture, influence the infiltration rates and affect the runoff generation process [1].
The analyzed outcomes are essential in the design and operation of works in water
resources management sectors.
In the recent scenario, the role of climate change also imparts importance in
the design of RR modeling [2]. Intensification of the global hydrologic cycle due
to changed climatic factors [3] such as rainfall distribution, increased temperature
makes it necessary for preparing an accurate RR modeling for predicting future
scenarios toward the utilization of the available resources in an optimized way.
Anthropogenic activities such as groundwater exploitation, inefficient, excessive irrigation alter the river characteristics, which also affect the basin hydrology and related
ecosystem. In this context, understanding the interdependent processes and analyzing
the cause and effect study, an RR modeling with minimum uncertainty is significant.
Since the last few decades, various methods with different prediction models have
been developed to increase the efficiency of RR models for forecasting the accurate
runoff in scarce data situation. Among them, statistical methods [4] combined with
a GIS-based approach, an analytical approach based on dynamic stochastic models
with different algorithms [5] is popular. However, the uses of those models require a
range of different data sets, which is also a rigorous and time-consuming operation.
Recently, the evolution of machine learning and artificial intelligence [6] makes that
scenario different and exciting due to their variety of broad approaches with optimized nonlinear algorithms. Among the machine learning techniques, artificial neural
network is a relatively new concept used in hydrologic and water resources systems
modeling [7]. Artificial neural networks can be considered as black-box models.
ANN is a nonlinear mathematical structure, competent in analyzing the complexity
in nonlinear relationships between raw and processed data of a system [8]. ANN
models are particularly helpful as well as most useful to use as pattern-recognition
tools for generalization of input–output relationships. Thus, it is mostly used as a quite
common approach in hydrologic problems such as streamflow prediction [9], flood
forecasting [10], prediction of water quality parameters [11], and runoff forecasting
[12–14] which is the main concerned area of focus.
Runoff is affected by various factors such as the area of the catchment, land use
pattern, soil characteristics, slope distribution of the area, and initial abstraction rate
of that catchment. To construct a hydrologic model, predicting runoff is one of the
essential hydrologic variables in water resources applications. For the prediction of
S. Mandal and S. Biswas
1 Introduction
Rainfall-runoff (RR) models are of paramount importance for the strategic planning of water resources and its development. This issue becomes more crucial and
challenging as the population grows in a particular area and makes the basin more
heterogeneous from the landcover perspective. The difficulty in the analysis exists
mainly due to the complex nonlinear relationships in the transformation of rainfall
into runoff. Inconsistent watershed characteristics and overland, as well as inland
flow generation processes, make this transformation more complex in modeling.
Nonuniform precipitation patterns, along with the uneven distribution of soil moisture, influence the infiltration rates and affect the runoff generation process [1].
The analyzed outcomes are essential in the design and operation of works in water
resources management sectors.
In the recent scenario, the role of climate change also imparts importance in
the design of RR modeling [2]. Intensification of the global hydrologic cycle due
to changed climatic factors [3] such as rainfall distribution, increased temperature
makes it necessary for preparing an accurate RR modeling for predicting future
scenarios toward the utilization of the available resources in an optimized way.
Anthropogenic activities such as groundwater exploitation, inefficient, excessive irrigation alter the river characteristics, which also affect the basin hydrology and related
ecosystem. In this context, understanding the interdependent processes and analyzing
the cause and effect study, an RR modeling with minimum uncertainty is significant.
Since the last few decades, various methods with different prediction models have
been developed to increase the efficiency of RR models for forecasting the accurate
runoff in scarce data situation. Among them, statistical methods [4] combined with
a GIS-based approach, an analytical approach based on dynamic stochastic models
with different algorithms [5] is popular. However, the uses of those models require a
range of different data sets, which is also a rigorous and time-consuming operation.
Recently, the evolution of machine learning and artificial intelligence [6] makes that
scenario different and exciting due to their variety of broad approaches with optimized nonlinear algorithms. Among the machine learning techniques, artificial neural
network is a relatively new concept used in hydrologic and water resources systems
modeling [7]. Artificial neural networks can be considered as black-box models.
ANN is a nonlinear mathematical structure, competent in analyzing the complexity
in nonlinear relationships between raw and processed data of a system [8]. ANN
models are particularly helpful as well as most useful to use as pattern-recognition
tools for generalization of input–output relationships. Thus, it is mostly used as a quite
common approach in hydrologic problems such as streamflow prediction [9], flood
forecasting [10], prediction of water quality parameters [11], and runoff forecasting
[12–14] which is the main concerned area of focus.
Runoff is affected by various factors such as the area of the catchment, land use
pattern, soil characteristics, slope distribution of the area, and initial abstraction rate
of that catchment. To construct a hydrologic model, predicting runoff is one of the
essential hydrologic variables in water resources applications. For the prediction of