--> --> Using Artificial Intelligence (AI) to Develop a Predictive Landslide Susceptibility Model to Assist with Well Pad and Pipeline Planning & Design

2019 AAPG Eastern Section Meeting:
Energy from the Heartland

Datapages, Inc.Print this page

Using Artificial Intelligence (AI) to Develop a Predictive Landslide Susceptibility Model to Assist with Well Pad and Pipeline Planning & Design


A predictive landslide susceptibility model is being developed to streamline and focus the current approach commonly used to address planning and design data needs at proposed well pad sites and along the projected limit of disturbance for pipeline construction projects. This predictive model uses a weighted indicator parameter approach to evaluate landslide susceptibility at targeted areas. Artificial intelligence (AI) methods are being considered to determine region specific weights for each indicator parameter within the model with the goal of converging on a landslide susceptibility index value. The goal of this project is to provide a process where a landslide susceptibility index value can be determined from any study area where shallow test boring data from the National Coal Resources Database System (NCRDS) is available. Lithologic and stratigraphic data queried from the NCRDS is a primary component of this model. Data from the NCRDS consists of thousands of test borings compiled by the US Geological Survey for the purpose of evaluating coal reserves in the United States. A unique computer based work flow is being refined at Youngstown State University (YSU) to screen, compile, and extract information from this enormous database that is of specific interest to our effort. The output from YSU’s work flow application is then imported into a three-dimensional (3-D) gridding software. Data from published structure contour maps are then used to bias the gridding algorithm for the creation of a 3-D model that is interfaced with geographical information system (GIS) software. Other critical indicator parameters considered for this model include proximity to existing landslides as mapped by others, slope, slope aspect, presence of over-dips, and land cover. Our preliminary research indicates that the available landslide delineation efforts for our current area of interest are based on published studies completed prior to the development high resolution remote sensing methods such as topographic mapping using LiDAR or high-resolution aerial photography. Therefore, a component of this study also involves evaluating the feasibility of using a machine learning approach to detect previously undocumented landslides through examination of LiDAR and high resolution aerial photography. The output from the 3-D lithologic model was integrated with the other indicator parameter data using GIS software. Information provided by industry is currently being used as a basis for the development of an AI neural network to efficiently converge on a defensible region specific indicator parameter weighting process. This research is currently focused on southwestern Pennsylvania, the northern panhandle of West Virginia, and southeastern Ohio but our methods can be applied anywhere test boring data from the NCRDS is available.