Geomagnetic Survey Interpolation with the Machine Learning Approach

Abstract

This paper portrays the method of UAV magnetometry survey data interpolation. The method accommodates the fact that this kind of data has a spatial distribution of the samples along a series of straight lines (similar to maritime tacks), which is a prominent characteristic of many kinds of UAV surveys. The interpolation relies on the very basic nearest neighbourss algorithm, although augmented with a Machine Learning approach. Such an approach enables the error of less than 5 percent by intelligently adjusting the nearest neighbours algorithm parameters. The method was pilot tested on geomagnetic data with Borok Geomagnetic Observatory UAV aeromagnetic survey data.

Modeling the Horizontal Velocity Field of the Earth’s Crust in a Regular Grid from GNSS Measurements

“…Cross-validation is a method for evaluating a model to determine how successfully the applied statistical analysis in the model can perform on an independent dataset. Crossvalidation methods are successfully applied to assess modeling effectiveness in various Earth sciences [Agayan et al, 2022;Aleshin et al, 2022;Sun et al, 2022], including forecasting recent crustal movement fields [Bogusz et al, 2013].…”

“…They are universal for generating gridded data for movement and deformation fields regardless of the studied geological process or phenomenon. These methods include geostatistical methods [Bogusz et al, 2013;Ghiasi and Nafisi, 2015], distance-weighting methods [Bogusz et al, 2013;Shen et al, 1996Shen et al, , 2015, spline and polynomial methods [Bogusz et al, 2013;Sandwell, 1987], machine learning methods [Aleshin et al, 2022;Grishchenkova, 2017;Manevich et al, 2021;Manevich and Tatarinov, 2017;Tatarinov et al, 2018], and others.…”

“…Therefore, during the algorithm's training process, it adapts to models created by the author based on the original data, rather than creating new relationships between the data. Geophysical fields, such as results from seismic tomography, magnetic and gravitational anomalies, have proven efficiency in applying machine learning methods in Earth sciences as a whole [Agayan et al, 2022;Aleshin et al, 2022;Dzeboev et al, 2019;Gvishiani et al, 2022Gvishiani et al, , 2023Sun et al, 2022]. In the considered task, they can reflect the deep structure of the Earth's crust and serve as informative features during algorithm training.…”

“…To obtain regular (grid) data on surface deformations, it is necessary to employ methods of mathematical modeling of displacement fields. For instance, in [Aleshin et al, 2022;Allmendinger et al, 2011;Esikov, 1979], it is noted that due to the complexity of calculating deformation parameters, it is advisable to choose regions where more complex deformed conditions should be specified within the finite element (where deformation is typically assumed to be uniformly distributed), thereby enhancing the order of approximation of the data. There are numerous methods for modeling field movements (which will be discussed below).…”