Technical University of Crete (TUC)
Geostatistical analysis and space-time models of aquifer levels: application to Mires hydrological basin in the prefecture of Crete
Abstract
dc:descriptionThe accurate representation of groundwater levels in an aquifer is very important for groundwater modeling and effective groundwater resources management. However, the number and spatial distribution of monitoring sites in a given aquifer are not always sufficient to accurately represent the water table. Predictions of groundwater level at unvisited locations of an aquifer can be obtained by applying geostatistical methods on the available groundwater level data and thus the free surface of the aquifer can be reliably mapped. In sparsely monitored basins, accurate mapping of the spatial variability of groundwater level requires the interpolation of scattered data. This thesis aims to present new modeling tools that help to better monitor and predict the groundwater level in sparsely gauged basins. The specific area of focus is the Mires basin of the Mesara valley in the island of Crete (Greece). The study area is a sparsely sampled basin that has limited groundwater resources which are vital for the area’s welfare; spatiotemporal variations of groundwater level are important for developing management and monitoring strategies. Efficient groundwater management in the basin is crucial in light of regional climate change model estimates showing a substantial risk of desertification for Crete. Our goal is to construct accurate spatial and spatiotemporal models of the basin’s groundwater level. Therefore, spatial and spatiotemporal models for the accurate representation of the groundwater level variability in already vulnerable areas with low groundwater resources, like Mires basin, need to be developed in order to identify the susceptible locations, to estimate the groundwater level distribution spatially and spatiotemporally and to provide input for potential groundwater resources management plans. The main data used in this research consist of seventy hydraulic head measurements (wet period of 2002-2003 hydrological year) which are unevenly distributed and mostly concentrated along a temporary river and time series performance consisting of biannual groundwater level data from tenboreholes (1981-2003). After the year 2003 observations of a shorter number of wells are available biannually. This thesis initially presents a comparison of deterministic interpolation methods, i.e., Inverse Distance Weight (IDW) and Minimum Curvature (MC), with stochastic methods, i.e., Ordinary Kriging (OK), Universal Kriging (UK) and Kriging with Delaunay Triangulation (DK). We evaluate the performance of the interpolation methods with respect to different statistical cross validation measures. The Spartan variogram family is implemented for the first time to hydrological data and is shown to be optimal with respect to the stochastic interpolation methods such as OK, UK, DK and Residual Kriging (RK) applied in this dataset. The three stochastic methods (OK, DK, UK) perform overall better than the deterministic counterparts (IDW, MC). DK, which is herein for the first time applied to hydrological data, yields the most accurate cross validation estimate for the lowest value in the dataset. OK and UK lead to smooth isolevel contours, while DK and IDW generate more edges. The stochastic methods also deliver estimates of prediction uncertainty. The present research study also investigates the application of non-linear normalizing data transformations in conjunction with Ordinary Kriging (OK) for the accurate prediction of groundwater level spatial variability in a sparsely gauged basin. We investigate three established normalizing methods, Gaussian Anamorphosis, Trans-Gaussian Kriging and Box-Cox to improve the estimation accuracy. The first two are for the first time applied to groundwater level data. All three methods improve the mean absolute prediction error compared with the application of OK to the non-transformed data. In addition, a Modified Box-Cox (MBC) transformation is proposed and applied to normalize the hydraulic heads. MBC in conjunction with OK is found to be the optimal spatial model based on leave-one-out cross-validation. The recently established Spartan semivariogram family provides the optimal model fit to the transformed data. Trend functions, as previous studies have shown, improve the accuracy of interpolation. Therefore, we propose that the prediction of the hydraulic head spatial variability in Mires basin can be improved by incorporating in the trend function local properties. Firstly the distance of the prediction points from the temporary river crossing the basin is incorporated in the trend function and secondly a component based on the generalized Thiem’s equation for multiple wells. Residual Kriging is performed based on these two spatial trend models as well as using the novel MBC transformation to normalize the residuals and the flexible Spartan semivariogram family to optimally determine their spatial correlation. Both proposed spatial models’ improve significantly the cross validation measures compared to the other Krigingbased methods tested. We also present maps of the groundwater level spatial variability and the estimation variance in Mires basin obtained by means of the optimal spatial models. We determine that the optimal spatial interpolation approach is based on Residual Kriging using the non-differentiable Spartan semivariogram model for the normalized (MBC) fluctuations. Our proposal is supported by the results of cross validation analysis. The suggested methodology is applicable to other unconfined aquifers as well. The non-differentiable property of the Spartan semivariogram model is interpreted herein as the result of a deposition-removal process that leads to a fractional Brownian motion, fBm-like, behavior of the groundwater level surface. The overall goal of this dissertation is to use stochastic methods for the monitoring and prediction of the groundwater level in space and time. Following the spatial interpolation, first we model the temporal variation of groundwater level with a discrete time autoregressive exogenous variable model (ARX) model. In this study pumping data are used in addition to precipitation measurements. The ARX model is embedded in a discrete-time Kalman filter to estimate the model parameters andpredict the optimal mean annual groundwater level. The ARX model is calibrated for the years 1981 to 2006 and is then used to predict the mean annual groundwater level in the basin for recent years (2007-2010). The predictions are validated with the available annual averages reported by the local authorities. Secondly, we use a spatiotemporal geostatistical analysis of the groundwater level using space-time Residual Kriging (STRK). The space-time trend is calculated using the product function of the estimated temporal trend from a weighted moving average filter and the spatial trend determined from the closest distance of the measurement locations from the river bed. A space-time experimental semivariogram is determined from the biannual (wet and dry period) groundwater level fluctuations time series between the years 1981 and 2003 at the ten sampling stations. We model the semivariogram with separable and non-separable theoretical spatiotemporalsemivariogram functions. STRK is used to predict the groundwater level for selected hydrological periods at each sampling station in the time period (2004-2010) biannually. Maps of groundwater level predictions and of prediction accuracy are desirable and significant in order to assess the groundwater level spatiotemporal variability, whether observed changes in water-table levels are statistically significant and finally to identify additional locations where further monitoring is needed to increase the accuracy of the maps. All the methodologies and tools presented in this thesis are implemented by original code developed by the author and run in Matlab® programming environment (Matlab v.7.5 on Microsoft Windows XP).
Degree
thesis:*- Grantor dc:publisher
- Technical University of Crete (TUC)
- Year dc:date
- 2012
Author and committee
dc:creator, dc:contributor.*- Authors dc:creator
-
- Varouchakis, Emmanouil
- Βαρουχάκης, Εμμανουήλ
Subjects
dc:subject × 12Rights
- Language dc:language
- eng
Identifiers
dc:identifier.*- Identifier
- 10.12681/eadd/31977