Using a multiple variogram approach to improve the accuracy of subsurface geological models

Abstract

Subsurface geological models are often used to visualize and analyze the nature, geometry, and variability of geologic and hydrogeologic units in the context of groundwater resource studies. The development of three-dimensional (3D) subsurface geological models covering increasingly larger model domains has steadily increased in recent years, in step with the rapid development of computing technology and software, and the increasing need to understand and manage groundwater resources at the regional scale. The models are then used by decision makers to guide activities and policies related to source water protection, well field development, and industrial or agricultural water use. It is important to ensure that the modelling techniques and procedures are able to accurately delineate and characterize the heterogeneity of the various geological environments included within the regional model domain. The purpose of this study is to examine if 3D stratigraphic models covering complex Quaternary deposits can be improved by splitting the regional model into multiple submodels based on the degree of variability observed between surrounding data points and informed by expert geological knowledge of the geologicaldepositional framework. This is demonstrated using subsurface data from the Paris Moraine area near Guelph in southern Ontario. The variogram models produced for each submodel region were able to better characterize the data variability, resulting in a more geologically realistic interpolation of the entire model domain as demonstrated by the comparison of the model output with preexisting maps of surficial geology and bedrock topography as well as depositional models for these complex glacial environments. Importantly, comparison between model outputs reveals significant differences in the resulting subsurface stratigraphy, complexity, and variability, which would in turn impact groundwater flow model predictions.Résumé : Des modèles de géologie de subsurface sont souvent utilisés pour visualiser et analyser la nature, la géométrie et la variabilité d'unités géologiques et hydrogéologiques dans le contexte d'études sur les ressources d'eau souterraine. Le dével-oppement de modèles géologiques de subsurface en trois dimensions (3D) couvrant des domaines modélisés de plus en plus grands s'est accéléré ces dernières années, parallèlement à l'essor rapide de la technologie informatique et des logiciels et au besoin accru de comprendre et de gérer les ressources d'eau souterraine à l'échelle régionale. Les modèles sont ensuite utilisés par les décideurs pour orienter les activités et politiques touchant à la protection des sources d'eau, la mise en valeur de champs de captage et l'utilisation industrielle ou agricole de l'eau. Il importe de s'assurer que les techniques et procédures de modéli-sation puissent délimiter et caractériser de manière exacte l'hétérogénéité des divers milieux géologiques contenus dans le domaine régional modélisé. Le but de l'étude est d'évaluer si les modèles stratigraphiques 3...

Evaluating spatially enabled machine learning approaches to depth to bedrock mapping, Alberta, Canada

“…The former approach is more suitable in mountainous and upland terrain where bedrock elevation is typically highly correlated with land surface elevation, while the latter is preferred in regions where the bedrock topography is uncorrelated from land surface topography, typical of many lowland regions. This is problematic in regions of mixed physiography, and most studies tend to either generate multiple separate models [ 9 , 13 , 14 ] or resort to an iterative process where the initial results are repeatedly re-interpolated with synthetic data that has been generated to resolve problematic areas [ 14 , 51 ]. In either case, this makes the modelling process cumbersome and impossible to automate.…”

An Automated Approach to Groundwater Quality Monitoring—Geospatial Mapping Based on Combined Application of Gaussian Process Regression and Bayesian Information Criterion

“…When analysing the RMSE statistics, a small RMSE value indicates that the interpolated values for the output model are more similar to the observed data point values, whereas a large RMSE value suggests that the interpolated model values are less similar to the observed data points. Thus, RMSE was used here to determine how well the model fits the observed data values, with low RMSE values indicting a high degree of model accuracy [50,51].…”

Dynamic Update Method of Working Face Geological Model Driven by Multi-Source Data

“…Constrained interpolation is applied with non-uniform discrete geological data points to construct a high-precision geological model (Le et al 2014;MacCormack et al 2018). The discrete smooth interpolation (DSI) algorithm creates a grid of interconnected nodes by constructing a discretized natural body model.…”