Environmental Engineering Reference
(GA) in forecasting groundwater level in an individual well. Das et al. ( 2010 )
attempted to develop a hybrid neural model (ANN-GA) employing an ANN model
in conjunction with famous optimization strategy called genetic algorithms (GA)
for accurate prediction of groundwater levels in the lower Mahanadi river basin of
Orissa State, India. Jain and Kumar ( 2009 ) used a hybrid ANN regression model
structure for knowledge extraction from trained neural network hydrologic models.
In this paper, hybrid model, which comprises ANN and non-linear regression
model, was developed to predict the GWL fluctuations.
The study area is located in the Sagar city, Sagar district of Madhya Pradesh, India.
The well is located in the catchment of Sagar lake of Sagar city (Fig. 1 ). The Sagar
city is located at the latitude of 23° 50′ N and longitude of 78° 45′ E. The catchment
area of Sagar lake is 18 km 2 and the land use pattern of the area is 40.9 % barren
land, 20.9 % agriculture, 18.7 % settlement, 11.5 % open forest and 8.1 % water
body. The soils of this area are of two types: the red or reddish brown lateritic soil
on hill tops and the black soil at the foothills.
The geological formation of the area mainly comprises quartzite sandstone of
Vindhyan age and Deccan traps. The Deccan traps are basaltic in nature having
vertical, polygonal and columnar joints. The Vindhyan quartzite sandstone is hard
and compact with nearly vertical joints.
The data acquired from the area consisted of rainfall and GWL. The GWL time
series data were measured at a particular location (bus stand) in the Sagar city and
the precipitation data were collected from the meteorological station located in the
Sagar city. The rainfall and GWL data were available for the period from June 1999
to July 2000. The data were divided into two sets: a training data set consisting of
daily rainfall and GWL data for 9 months (June to February), and a testing data set
of 5 months (March to July).
Development of Hybrid Model
A hybrid model is comprised of both ANN and regression modelling techniques. In
the hybrid model, the output layer was replaced by a non-linear regression model
(NLRM). The structure of the hybrid model is depicted in Fig. 2 . The input was fed
through input layers and the intermediate output i.e. hidden neurons output was
calculated at the hidden layer using ANN. The outputs from hidden neurons were
the inputs to the NLRM of the hybrid model. Jain et al. ( 2004 ) used the output of
hidden layer for identification of physical processes inherent in artificial neural net-
work rainfall runoff models.