Monthly River Flow Forecasting by Data Mining Process

Abstract

In design, plan, project, construction, maintenance, and especially management of water resources, surface water input and output must be calculated based on measurements. One of the priority parameter is surface flow in the studies. The flow data measured in the past is required for design of the water structure be built in the future and calculation of natural disasters such as flood and drought according to the pre-specified risk level (Sen, 2003). Stochastic models and artificial intelligence techniques (artificial neural networks, fuzzy logic and adaptive neuro fuzzy inference systems etc.) on flow predicting are commonly used by many researchers while data mining (DM) process is not yet widely used in the hydrology area. Russo et al. (2006) fitted a stochastic rainfall model to rainfall radar data in order to produce a realistic representation of the distribution of rainfall in space and time. The results show that the model, calibrated on the study area, is able to forecast satisfactorily the rain field in space and time. Archer & Fowler (2008) investigated the links between climate and runoff for eight gauging stations in the Jhelum catchment but then concentrated on seasonal forecasting of spring and summer inflows to Mangla Dam. They are used precipitation and temperature variables to forecast summer season flows at stations upstream from the reservoir with a lead time of up to three months based on multiple linear regression models. The analysis demonstrates that good forecasts within 15% of observed flows for 92% of years can be achieved for summer season flows from April to September. For spring flows from April to June, excellent forecasts can be provided within 15% of observed flows for 83% of years. Lin & Chen (2004) used the radial basis function network (RBFN) to construct a rainfall-runoff model, and presented the fully supervised learning algorithm for the parametric estimation of the network. The proposed methodology has been applied to an actual reservoir watershed to forecast the oneto three-hour ahead runoff. The result shows that the RBFN can be successfully applied to build the relation of rainfall and runoff. Rajurkar et al. (2004) presented an approach for modeling daily flows during flood events using ANN. They showed that the approach produces reasonably satisfactory results for data of catchments from different geographical locations. Nayak et al. (2004) suggested that performance of ANFIS model is capable of preserving the statistical properties of the time series and it is viable for modeling river flow series. Keskin et al. (2006) developed a flow prediction model, based on the adaptive neural-based fuzzy inference system (ANFIS) coupled with stochastic hydrological models. An ANFIS is applied to river flow prediction in Dim Stream in the southern part of Turkey. Synthetic