Towards spatial machine learning to reveal hidden patterns and relationships in national and international geochemical databases

Environmental geochemistry is playing an increasingly important role in mineral exploration, environmental management, agricultural practices as well as links with health. With rapidly growing databases available at regional, national, and global scales, environmental geochemistry is facing the challenges in the “big data” era. One of the main challenges is to find out useful information hidden in a large volume of data, with the existence of spatial variation found at all the sizes of global, regional (in square kilometers), field (in square meters) and micro scales (in square centimeters). Meanwhile, the rapidly developing techniques in machine learning become useful tools for classification, identification of clusters/patterns, identification of relationships and prediction. This presentation demonstrates the potential uses of a few practical spatial machine learning techniques (spatial analyses) in environmental geochemistry: neighborhood statistics, hot spot analysis and geographically weighted regression.

 

Neighborhood (local) statistics are calculated using data within a neighborhood such as a moving window. In this way, spatial variation at the local level can be quantified and more details are revealed. Hot spot analysis techniques are capable of revealing hidden spatial patterns. The techniques of hot spot analysis including local index of spatial association (LISA) and Getis Ord Gi* are investigated using examples of geochemical databases in Ireland, China, the UK and the USA. The geographically weighted regression (GWR) explores the relationships between geochemical parameters and their influencing factors at the local level, which is effective in identifying the complex spatially varying relationships. Machine learning techniques are expected to play more important roles in environmental geochemistry. Challenges for more effective “data analytics” are currently emerging in the era of “big data”.