Καμπάνης, ΝίκοςΠαρασύρης, Αντώνιος2016-11-18In sparsely monitored basins, accurate mapping of the spatial variability of groundwater level re¬quires the interpolation of scattered data. The methodology that is presented here is Ordinary Kriging, a method that is called the best exact interpolator, because in the absence of a nugget term, Kriging is an exact interpolator at the measurement points. In addition, Kriging allows the estimation of interpolation uncertainties which is also presented. Then, this work tackles the prob¬lem of deficient sampling of an area, due to budget constraints. To that end, the Adaptive Genetic Algorithm is being introduced, that is an Evolutionary Algorithm used for minimizing errors, and is coupled with the geo-statistical methodology to optimize the monitoring network. To do that, three different errors are defined and optimized for a constant number of measurement removals (called herein scenarios). The errors that are presented, are based either on the difference of the initial mapping with each of the reduced networks that the genetic algorithm will evaluate and evolve (RMSD, RMSE), or based on the Akaike criterion, which finds the best set of data that minimizes the error of the variogram. The described method is applied successfully to two test cases, in Mires and in Drama basin. In the first case, the initial dataset is consisted of 70 boreholes, and the method concluded that in some cases even 40 measurements could be neglected and still have an accurate mapping of the underground water level, but the safer choice would be to stop at 30 removals, because in that case, the uncertainty is much lower. Lastly, in Drama, there were 250 measurements, and the interest was to investigate the robustness of the kriging based optimization tool, and its applicability to different test cases. There, because of the bigger dataset, the RMSD was outperformed by the RMSE which only evaluates on the missing wells, instead of make the predictions in the entire grid. So a 150 removal or even 200 in some cases, where the RMSE error was more practical and Akaike was focusing more on the variogram fit. RMSD was in almost every instance slightly more accurate than RMSE except the last case when surprisingly RMSE outper¬formed RMSD error. So the conclusion that this work has reach is that using a genetic algorithm, and defining properly the fitness function and the succesive errors leads to a significant reduction in data measurements needed for an accurate kriging mapping. The scenario number of removals are proposed here for the two test cases, but in the end, it is a management decision of how high the uncertainty growth is allowed , or the degree of similarity of the reduced network mapping with the original dataset mapping61 σ. ; : ; 30 εκ.http://elocus.lib.uoc.gr:443/dlib/3/f/8/metadata-dlib-1482308368-41844-2033.tkl000404560engΣχολή/Τμήμα--Σχολή Θετικών και Τεχνολογικών Επιστημών--Τμήμα Μαθηματικών και Εφαρμοσμένων Μαθηματικών--Μεταπτυχιακές εργασίες ειδίκευσηςΣπαρτιάτικο ημι-βαριόγραμμαΤριγωνισμόςΑπεικόνιση GrigingSpartan variogramGroundwater monitoring networks management using genetic algorithmsΔιαχείρηση υπόγειων υδάτων με χρήση γενετικών αλγόριθμωνtextΤύπος Εργασίας--Μεταπτυχιακές εργασίες ειδίκευσηςΜεταπτυχιακή εργασίαMaster thesisΠανεπιστήμιο ΚρήτηςGreek Aggregator OpenArchives.gr | Nationalother