Geostatistical learning for decision–oriented mapping in earth sciences

In this study, geostatistical and deterministic learning approaches have been applied to earth science problems. The thesis work has been conducted based on a methodological synthesis of statistical learning and spatial data analysis. The thesis was composed of under the two sections. Under the first heading, ground water levels have been appraised from a time and distance-based geo-statistical learning perspective. The spatial relationships among the ground water levels (GWL) of the water-wells placed in the mine site and its around have been measured and it is analyzed to change of search domains in connection with spatio-temporal variability. In the second heading, a new smoother, which is an alternative to inverse distance weighting (IDW) and k-nearest neighbour (k–NN) approaches, have been suggested. The new interpolation algorithm producing global and local solution has been developed based on the molecular graph theory. The thesis findings would serve decision making in different disciplines being the first place in earth and environmental science by using spatial relationships.