An AI driven, mechanistically grounded geospatial liquefaction model for rapid response and scenario planning

Geospatial models for predicting soil liquefaction infer subsurface traits via satellite remote sensing and mapped information, rather than directly measure them with tests. Field tests of such have demonstrated both promising potential severe shortcomings. Informed by these tests, this paper develops geospatial that are driven algorithmic learning but pinned to a physical framework, thereby benefiting from machine deep learning, or ML/DL, the knowledge mechanics developed over last 50 years. With approach, cone penetration test (CPT) measurements predicted remotely within framing popular CPT model ground failure. This has three advantages: (i) mechanistic underpinning; (ii) significantly larger training set, principally trained on in-situ data, failures; (iii) insights greater data be exploited. While is phenomenon best mechanics, lack theoretical links above-ground parameters, correlate in complex, interconnected ways - prime problem ML/DL. Preliminary using ML/DL modest U.S. dataset CPTs predict liquefaction-potential-index values 12 variables. The tested recent earthquakes shown – statistically significant degree perform as well as, better than, current leading model. coded free, simple-to-use Windows software. only input ground-motion raster, downloadable minutes after an earthquake available countless future scenarios. Ultimately, proposed approach models, which warrant further application evaluation, could improved upon additional new predictor Users should understand key limitations, discussed detail herein.