GEOAI benchmark problems BM/AirportSoilProperties/2/2025

This paper presents benchmark problems of predicting soil properties for the purpose of probabilistic site characterization and their sample solutions. The benchmark dataset is derived from an extensive geotechnical survey conducted at a 10 km × 10 km landfill site of a large-scale offshore airport in Japan, comprising 2,922 boreholes and 67,760 records. A subset of the dataset, referred to as the “verification site” (300 m × 300 m), consists of 51 boreholes and 1,001 records and is the focal point of the soil property prediction. Two benchmark problems are introduced: (1) predicting the spatial variation of undrained shear strength (s u ); (2) predicting missing mechanical parameters. For both benchmark problems, it is required to incorporate Big Indirect Data (BID) to support the prediction. Sample solutions are used to demonstrate the prediction results using various types of BID, including global and local datasets. The sample solutions are relatively simple for demonstration purpose only. It is optional to adopt advanced methods to pre-process the site-specific data or BIDs, such as outlier detection, clustering, and dimensionality reduction.