Ann Modeling With Rbf

ANN BASED RAINFALL RUNOFF MODELLING S. Saravanan* Civil Engineering, Indian Institute of Technology Roorkee, Roorkee, India *Corresponding author, email: [email protected]

K.S.Kasiviswanathan Water Resources Development and Management, Indian Institute of Technology Roorkee, Roorkee, India

Stream flow in the Himalayan Rivers is generated from rainfall, and glacier melt. The distribution of runoff produced from these sources is such that the stream flow may be observed in these rivers throughout the year, i.e. they are perennial in nature. Snow and glacier melt runoff contributes substantially to the annual flows of these rivers and its estimation is required for the planning, development and management of the water resources of this region. Water is the most abundant substance on earth, the principle constituent of all living things, and the major force constantly shaping the earth. Hydrology may be broadly considered to be the study of the life cycle of water. The most important section of the cycle is transformation of rainfall into runoff. The relationship between rainfall and runoff is an important issue in surface hydrology and usually represents a major challenge for hydrologists. It quantifies the hydrologic response of a catchment in terms of surface runoff yield and properties of flood waves. This information is necessary for the selection and design of the appropriate water management and flood protection structures and measures. The transformation of rainfall into runoff over a catchment is a complex hydrological phenomenon, as this process is highly nonlinear, time-varying and spatially distributed. Many hydrologists devote themselves to develop rainfall–runoff models to estimate runoff. The rainfall–runoff process, which involves many mechanisms, is known as a highly complicated and nonlinear phenomenon. Difficulties exist in the modeling of the rainfall–runoff process. on the complexities involved, these models are categorized as empirical, black-box, conceptual or physically-based distributed models. Thus, an accurate and easily-used rainfall–runoff model that can appropriately model the rainfall–runoff process is of strong demand. A new approach for designing the network structure in an artificial neural network (ANN)-based rainfall-runoff model is presented. The Artificial Neural Network (ANN) is a method of computation inspired by studies of the brain and nervous systems in biological organisms. A neural network method is considered as robust tools for modelling many of complex non-linear hydrologic processes. The method utilizes the statistical properties such as cross-, auto- and partial-auto-correlation of the data series in identifying a unique input vector that best represents the process for the basin, and a standard algorithm for training. The distinct advantage of an ANN is that it learns the previously unknown relationship existing between the input and the output data through a process of training, without a priori knowledge of the catchment characteristics. The ANN is also described as a mathematical structure, which is capable of representing the arbitrary complex nonlinear process relating the input and the output

of any system. Presently more and more researchers are utilizing ANNs because these models possess desirable attributes of universal approximation, and the ability to learn from examples without the need for explicit physics. Models are generally classified as black-box models, conceptual models and physically based models. Black-box models, which are fully based on observational data and on the calibrated input-output relationship without description of individual processes (examples: unit hydrograph, empirical regression approaches, transfer function models, etc). Earlier these models are so called as empirical equations, now with a new tool called ANN these models. Conceptual models, wherein the basic processes (snowmelt, infiltration, evaporation, etc.) are separated to some extent, but their algorithms are essentially calibrated input-output relationships (the most famous among these models are Stanford Watershed Model, Hydrologic Engineering Centre-1 Model, Hydrological Simulation Model). Physically based models, which are based, as much as possible, on the mathematicalphysics equations of mass and energy transfer in the river basin and are intended to minimize the need for calibration by using measurable watershed characteristics as the model parameters or reliable relationships between these characteristics and the parameters (examples: Systeme Hydrologique Europeen (SHE), Hydrological Cycle models of the Water Problems Institute of RAS, Institute of Hydrology Distributed Model (IHDM)). The accuracy of runoff prediction with the models of the first two types depends fully on the quantity and quality of runoff measurements available for their calibration. That is, the blackbox models as well as conceptual models cannot be successfully used for runoff predictions in Poorly Gauged Basins (PGBs). Conceptual watershed models are generally reported to be reliable in forecasting the most important features of the hydrograph, such as the beginning of the rising limb, the time and the height of the peak, and volume of flow [Sorooshian, 1983]. However, the implementation and calibration of such a model can typically present various difficulties [Duan et al., 1992], requiring sophisticated mathematical tools [Sorooshian et al., 1993], significant amounts of calibration data [Yapo et al., 1995], and some degree of expertise and experience with the model. Physically based models which are advanced version of the conceptual models involve solution of a system of partial differential equation that represents our best understanding of the flow processes within the catchment. For most of the problems, discretizing space and timespace dimensions into discrete set of nodes seek a numerical solution. This implies that such models work best when data on the physical characteristics of the catchment are available at the grid scale. The kind of data required is rarely available. These models suffer from problems such as identification, assimilability and uniqueness of parameter estimation (Jain and Prasad, 2003).