New York State Gas Potential

We used kriging analysis to find spatial correlation for new gas drilling in New York State. By combining geological data with records from existing gas wells, the project leverages kriging, a method that estimates values between known data points, to map high-potential areas for natural gas. This approach helps visualize promising new drilling sites, supporting smarter resource management and decision-making in gas exploration.

To identify optimal drilling locations for gas wells, I used a combination of spatial data analysis and kriging techniques. The first step involved gathering and cleaning the data, which included geographic survey and gas production data for existing wells. After ensuring the data was prepared for analysis, I employed various kriging methods to make predictions about gas production potential in areas not yet drilled.

I experimented with several kriging techniques over the detrended data, including Ordinary Kriging, Universal Kriging, and CoKriging, each offering a slightly different approach to spatial predictions. Due to time constraints, I focused on Ordinary Kriging, which provided the most reliable results. This model, implemented through the SKgstat package, was fine-tuned using cross-validation and hyperparameter optimization techniques. I performed both randomized and grid searches to find the best parameters, including choices for the variogram model and anisotropy scaling, which are critical in spatial data analysis.

To evaluate the model’s performance, I used Root Mean Squared Error scores. The final model’s predictions were visualized using interpolation plots, which not only showed the predicted gas production but also highlighted areas with the highest potential through a heatmap-style representation. These visualizations helped pinpoint the most promising drilling locations, providing actionable insights based on the current gas well data.

The entire process, from data cleaning and model training to visualization, was designed to be efficient and scalable, ensuring that the results could be easily reproduced and adapted for future analyses.

Duis euismod curabitur leo justo risus sollicitudin in mus, diam nascetur cursus aliquam fermentum tincidunt felis, sodales nulla primis pretium class nec blandit. Hendrerit id ante egestas mollis habitasse eleifend interdum tristique, suspendisse sem quisque condimentum felis tortor eu varius, facilisis molestie himenaeos elementum eros leo porttitor.

ulputate cursus primis scelerisque nisl ut, nibh sociosqu felis platea enim diam habitant, eget tempor morbi turpis sociis accumsan quam. Vitae sed aliquet urna conubia fermentum class risus phasellus nullam dignissim ornare dui sociis, leo posuere ligula porttitor himenae