Geostatistical modeling of geochemical variables in 3D

Abstract

Geostatistical mapping of soil properties in 3D refers to the application of geostatisticalmethods to the soil data in order to produce maps of soil properties at different depths.Through two separate studies, this thesis elaborates on two different approaches for 3Dsoil mapping. At first, the well established Spline-Than-Krige approach for the mappingof soil pollutants atmospherically deposited from the copper smelting plant, was used. Inthe absence of the monitoring data, which can be used for a detailed characterization of theplume spreading process, this study was confined to the consideration of terrain exposureto explain spatial trend in arsenic distribution at different depths. This study aims toexplore the extent to which the commonly available information, such as the prevailingwind direction, or the location of the source of pollution, in combination with the digitalterrain model, can be used to quantify the terrain exposure, and hence to improve thespatial prediction of the arsenic concentration at several soil depths.Next, the innovative geostatistical approach to 3D mapping of soil properties, based onsoil profile data, was proposed. It provides the semi-automatic way for 3D modeling ofsoil variables, prediction over the regular grids (rasters) and also the evaluation of predictionaccuracy. Methodologically, this approach operates within the 3D regression krigingframework. 3D trend model is conceptualized as hierarchical or non-hierarchical linearinteraction model. This means that the model includes the interactions between the spatialcovariates and depth in the hiearchial or non-hierarchial manner. The trend modelingis based on the application of the penalized regression technique, lasso. The lasso usesa specific regularization penalty in a fitting procedure to enable the efficient parameterestimation and variable selection (including interaction terms) at the same time...

Geostatistiˇcko kartiranje zemljišta u 3D odnosi se na primenu geostatistiˇckih metoda nazemljišnim podacima u cilju izrade karata zemljišnih karakteristika jednog podruˇcja, kojese odnose na razliˇcite dubine zemljišta. U okviru dve nezavisne studije, ova doktorskadisertacija razmatra dva razliˇcita pristupa geostatistiˇckog modeliranja zemljišta u 3D. Uokviru prve studije, "Spline-Than-Krige" metod je koriš´cen za kartiranje koncentracijearsena u zemljištu, u blizini Rudarsko-topioniˇcarskog basena Bor, na tri razliˇcite dubine(0-5 cm, 5-15 cm i 15-30 cm). Dugogodišnje emitovanje nepreˇciš´cenih materija iz topionicerudnika u atmosferu, dovelo je do zagadjenja zemljišta u okolini, taloženjem štetnihmaterija nošenih vetrom. U odsustvu podataka kojima bi se detaljnije mogao opisati procesraspršivanja štetnih materija, ova studija se ograniˇcila na analizu izloženosti terenauticaju vetra, a time i procesu zagad¯enja. Predstavljen je inovativan pristup kvantifikacijiizloženosti terena izvoru zagad¯enja. Na osnovu opšte dostupnih podataka, kreirano jenekoliko parametara kojima se kvantifikuje geometrijska i topografska izloženost svaketacˇke terena izvoru zagad¯enja. Tako kreirani parametri, iskorišc´eni su za opisivanje prostornogtrenda koncentracije arsena na tri razliˇcite dubine. Definisani trendovi, koriš´ceni suu okviru regresionog kriginga, za prostornu predikciju. Na taj naˇcin pokušalo se odgovoritina pitanje, u kojoj meri, opšte dostupni podaci, kao što su pravac dominantnog vetraili poznavanje taˇcne lokacije izvora zagadjenja u kombinaciji sa digitalnim modelom terena,mogu biti iskoriš´ceni da bi se unapredila preciznost prostorne predikcije zemljišnihzagadjivaˇca, kako na površinskim slojevima tako i na ve´cim dubinama...