Abstract. Characterization of spatial variability of hydraulic properties of groundwater systems at high resolution is essential to simulate flow and transport phenomena. This paper investigates two schemes to invert transient hydraulic head data resulting from multiple pumping tests for the purpose of estimating the spatial distributions of the hydraulic conductivity, K, and the specific storage, Ss, of an aquifer. The two methods are centralized fusion and decentralized fusion. The centralized fusion of transient data is achieved when data from all pumping tests are processed concurrently using a central inversion processor, whereas the decentralized fusion inverts data from each pumping test separately to obtain optimal local estimates of hydraulic parameter, which are consequently fused using the Generalized Millman Formula, an algorithm for merging multiple correlated or uncorrelated local estimates. For both data fusion schemes, the basic inversion processor employed is the Ensemble Kalman Filter, which is employed to assimilate the temporal moments of the transient hydraulic head measurements resulting from multiple pumping tests. Assimilating the temporal moments instead of the hydraulic head transient data themselves is shown to provide a significant improvement in computational efficiency. Additionally, different assimilation strategies to improve the estimation of Ss are investigated. Results show that estimation of the K and Ss distributions using temporal moment analysis is fairly good; however, the centralized inversion scheme consistently outperforms the decentralized inversion scheme. Investigations on the sensitivity of the inversion estimates to errors in geostatistical parameters of the random fields of K and Ss reveal that the estimates are not sensitive to errors in the correlation length and the variance of hydraulic properties, but are noticeably sensitive to errors in the stationary mean. The proposed inversion schemes are expanded to estimate the geostatistical parameters of the K and Ss fields. The results show that the estimation of the true stationary mean of the K field and, to a lesser degree, the stationary mean of the Ss field can be successfully achieved, while the estimation of correlation length and standard deviation for both the K and Ss fields are not as effective.
ESTIMATION OF HYDRAULIC PROPERTIES FROM PUMPING TESTS DATA OF NAIROBI AREA, KENYA
