Spread Selection in Rbf Neural Network for Groundwater Flow Simulation

Effectiveness of FFNN-LMBP Algorithm in Groundwater Flow Modelling

Roshni T.a, Pramada S. K.b, Saliha Jaleel, Anza Thasneem, Shweta Panwar, Yugal Badlani, M. Bharat c

a Assistant Professor, National Institute of Technology Patna, e mail: roshni@nitp.ac.in; b Assistant Professor, National Institute of Technology Calicut, e mail: pramada@nitc.ac.in; c B.Tech Students, National Institute of Technology Calicut.

Abstract

Forecasting the ground water level fluctuations is an important requirement for planning conjunctive use in any basin. In this paper, the Artificial Neural Network (ANN) approach is applied for forecasting groundwater level fluctuation of an observation well in Peringolam Watershed, Calicut. The most appropriate set of input variables to the model are selected through a combination of domain knowledge and statistical analysis of the available data series. Various ANN models are developed that predicts the water level of the observation well. The results suggest that the model predictions are reasonably accurate as evaluated by various performance indicators. The ANN is set up using monthly groundwater time series data recorded between 2006 and 2009 and employing three independent predictor variables, namely, previous months’ groundwater level, monthly mean rainfall and monthly mean temperature. For this a Feed Forward Neural Network (FFNN) -Levenberg Marquardt (LM) algorithm is used. The results show that the developed FFNN can accurately reproduce groundwater depths of the shallow aquifer for up to three months. The study suggests that artificial neural networks can be used as a viable alternative to physical-based models for predicting groundwater levels.

1. Introduction

Water scarcity in non-monsoon season and water logging during monsoon season are the two major problems in Calicut city. The climate of the basin is sub-humid with average annual rainfall of 320 mm; two-thirds of which concentrates during monsoon season (June to September). Therefore, monitoring of groundwater levels is extremely important, especially in coastal aquifers. Network design and monitoring of groundwater levels spatially and temporally depends on hydro-geological conditions and available logistical resources. The prediction of groundwater levels in a well, based on continuous monitoring of few nearby wells, is of immense importance in the management of groundwater resources in coastal region.

Modelling groundwater level with lead-time at different seasons of a year would help in alleviation of problem of water logging during monsoon season and water scarcity during non-monsoon season. The conceptual and physically based models are main tools for depicting hydrological variables and understanding the physical processes that are taking place in a system. In these models, the