STOCHASTIC MODELING OF• GROUND WATER LEVELS OF ROORKEE (U.P.)

Abstract

This dissertation aims at analysis of stochastic nature of ground water levels of shallow and deep tube wells installed in the campus of Department of Hydrology, University of Roorkee, Roorkee. Univariate stochastic AR (P) models have been used for short term lead forecast of the ground water levels. The forecasted water levels have been matched with observed water levels. The present work starts with discritization of 11 years record of daily ground water levels and rainfall. The shallow tube well is 6 metres deep tapping unconfined aquifer while deep tube well is 200 metres deep tapping confined aquifers. The aquifers are part of the alluvial plain in the north of Ganga, characterised by high transmissivity and specific yield. Rainfall is the major source of ground water recharge. Monthly depth to water levels have been found to vary from 4.103 m to 1.135 m below ground level in shallow tube well and 6.689 to 3.45 m in deep tube well. The area receives annual normal rainfall of 1164.6 mm, 85% of which is brought by south-west monsoon. Ground water level fluctuation observed in shallow tube well is between .13 and 1.73 metres and in deep tube well between .56 and 2.44 metres across the monsoon. Stochastic modelling has been attempted in both monthly and daily data. The data were found free from long term trend. Periodicities in mean of monthly data were explained by 3 and 2 harmonics of shallow an'd deep tube wells respectively. Periodicities in standard deviation were explained by 5 and 6 harmonics in deep and shallow tube wells respectively. The dependent stochastic component was explained by AR (1) for the data of shallow tube well while AR (3) explained for the data of deep tube well. Stochastic modelling of daily ground water level data showed that periodicity in mean was explained by 2 harmonics in both shallow and deep tube wells. Periodicity in standard deviation was explained by 6 harmonics in case of shallow tube well while in case of deep tube well the standard deviation was almost static. AR (1) model explained the dependent stochastic components of both the wells. One month lead forecast of monthly ground water levels for period January-October 1997 showed close agreement with observed data. The forecast error remained below 16 cm on average in shallow tube well and 30 cm in case of deep tube well. One day lead forecast was done with the help of AR (1) models fitted to the daily data of both the tube wells. The forecast error was found to be within 15 centimetres in case of both shallow and deep tube wells. The above univariate forecast models fitted to daily data were found better than those fitted to monthly data. Autoregressive models closely explain the stochastic nature of ground-water levels of both the wells. The model has been built under the assumption that the ground-water levels are free from influences of pumping etc. in the surroundings. In case of change in existing scenario the present models needs be updated with the recent data.