Multiscalar Integration of Dense and Sparse Spatial Data: an Archaeological Case Study with Magnetometry and Geochemistry

Integration of different kinds of data is an important issue in archaeological prospection. However, the current methodological approaches are underdeveloped and rarely use the data to their maximum potential. Common approaches to integration in the geophysical sciences are mostly just various forms of comparison. We argue that true integration should involve the mathematical manipulation of input data such that the original values of the input data are changed, or that new variables are produced. To address this important research gap, we present an innovative approach to the analysis of geochemical and geophysical datasets in prospection-focused disciplines. Our approach, which we refer to as “multiscalar integration” to differentiate it from simpler methods, involves the application of mathematical methods and tools to process the data in a unified way. To demonstrate our approach, we focus on integrating geophysical data (magnetometry) with geochemical data (elemental content). Our approach comprises three main stages: Quantification of the data deviation from random distributions, linear modelling of geophysical and geochemical data and integration based on weighting of the different elements derived in previous steps. All the steps of the workflow can be also applied separately and independently as needed or preferred. Our approach is implemented in the R environment for statistical computing. All data, functions and scripts used in the work are available from open access repositories (Zenodo.org and Github.com) so that others can test, modify and apply our proposed methods to new cases and problems. Our approach has the following advantages: (1) It allows the rapid exploration of multiple data sources in an unified way; (2) it can increase the utility of geochemical data across diverse prospection disciplines; (3) it facilitates the identification of links between geochemical and geophysical data (or generally, between point-based and raster data); (4) it innovatively integrates various datasets by weighting the information provided by each; (5) it is simple to apply following a step-by-step framework; (6) the code and workflow is fully open to allow for customization, improvements and additions.