Planning the placement of observation points for interpolation of the Vs30 parameter by the Poisson disk sampling method

The article describes an approach to planning the placement of observation points for interpolating the Vs30 parameter based on Poisson-disk Sampling. A model geological map measuring 100×100 cells with four lithological zones was used as a test site, on which three placement schemes were compared for densities of 7 and 15 points: random, regular grid, and Poisson disc sampling. Interpolation was performed using the classical Kriging method and RF-bootstrap augmentation, as well as ML models (SVR, XGBoost, k-NN) with an extremely small amount of a priori data. The results show that Poisson disc sampling provides the most uniform coverage, reduces ‘gaps’ and artefacts in Vs30 maps, and demonstrates higher stability and accuracy of restoration according to MAE, RMSE and correlation coefficients compared to classical placement schemes.

Not yet translated