Heuristics‐enhanced geospatial machine learning (SaaS) of an ancient Mediterranean environment

Abstract

Raw soil core physical data used in machine learning algorithms with corresponding spatial remotely sensed data is an emerging science. Using data derived from soil core samples previously collected in Universal Transverse Mercator zone 50 (Western Australia) and remotely sensed data, a model that predicted ground movement (GM) was developed specific to Australian Standards manual AS 1726-2017. This is the first approach for Australian soils and first in the world for soils older than 200 million yr. The model developed reliably predicted GM with 91.1% accuracy. The error obtained from the prediction is within acceptable limits currently used by engineers in calculations concerning soil classification for engineering purposes. Concerning the remotely sensed data analyzed, accuracy of the Atterberg limits method might be improved if additional information about soil structure (layering and horizon) or other variables (seasonal data) are built into this model. This model can be used to save on construction material costs, reduce the potential for human error associated with data collection and sample manipulation, but also fast-track (by up to 6 wk based on current wait times) building approvals while ensuring compliance to the relevant legislation. This platform also reduces the environmental effects of invasive drilling techniques. A requirement within principles of sustainable building practices, and associated with current standards commonly used by structural engineers who may seek better understanding of soil properties in Australia as a software service (with application potential in North America).

Machine learning applications in vadose zone hydrology: A review

“…In particular, Kumar et al (2021) discovered that bidirectionalstacked-LSTM (BS-LSTM) forecasted soil movement more accurately (RMSE of 0.27) than LSTM (RMSE of 0.37) when assessed in regions of India. Svatos et al (2022) analyzed soil cores in Western Australia and applied ML to predict ground movement with an accuracy of 91.1%. This ML application also identified impactful soil physical properties for the ground movement, which are important for geologic studies and engineering designs.…”

“…Svatos et al. (2022) analyzed soil cores in Western Australia and applied ML to predict ground movement with an accuracy of 91.1%. This ML application also identified impactful soil physical properties for the ground movement, which are important for geologic studies and engineering designs.…”