Optimal evaluation of aquifer parameters in Dhaka city by kriging

Abstract

A geostatistical technique, kriging, has been applied in evaluating the groundwater flow parameter, transmissivity in Dhaka City. A second parameter, specific capacity, has also been used in the estimation procedure. Dhaka City has been chosen considering its importance from the view point of its ever increasing problem of groundwater shortage due to huge abstraction of groundwater by around 600 deep tubewells. A grid system of the metropolitan area of the city has been selected for generating transmissivity data at the node points. Groundwater parameter data are often scarce but have to be given at every node of a groundwater model grid. Kriging is the geostatistical interpolation method used most frequently for generating such data. Kriging considers the spatial structure of the variable in question and provides a best linear unbiased estimate with minimum variance of estimation error. The error is obtained in the form of standard deviation of the kriged values which are needed while assigning plausible ranges of parameter values prior to model calibration. Three kriging methods have been selected, viz., ordinary kriging, kriging combined with linear regression and cokriging. Ordinary kriging employs only the measurements of the main variable, i.e., transmissivity, while the other two methods use data of more than one variable where the variables are strongly correlated. A number of 50 transmissivity data and 200 specific capacity data have been used. A correlation coefficient of 0.84 was obtained between the two thus advocating the use of the latter two methods. A general computer program has been developed in FOR TRAN77 to solve the kriging methods. The transmissivity data have been generated at the 457 node points of the selected grid system for each case. The three kriging methods have been compared and it has been inferred that the three methods provide data with almost identical ranges, but use of specific capacity data improves the estimates of the transmissivity and, in general, reduces the estimation error. Both kriging combined with linear regression and cokriging work almost equally well for Dhaka City. However, performance of cokriging seems to be slightly better. With a view to clarifying the measurement network optimization procedure, two cases have been considered. In the first case, different densities of data measurement network have been studied which reveal that increase of measurement points reduces the standard deviation of estimation error or uncertainty. The need for judgement regarding the optimum number of measurement points has been highlighted. In the second case, the effect of an additional measurement point upon the variance of estimation error has been tested and a procedure for locating new measurement site in the existing network has been elucidated. A sensitivity analysis has been performed, by following an indirect approach, to see the sensitivity of pumping water level to the measured transmissivity values. Lastly, two possible future scenarios of transmissivity field have been studied by considering reduced specific capacity data.