Abstract
The kriging method is a promising response surface approach for reliability analysis of realistic geotechnical problems involving implicit performance functions. For an analytical example, kriging models with various trend functions all outperform the second order polynomial function and ANN in terms of mean square error. For three geotechnical reliability benchmark problems taken from GEOSNet, it is found that the infinite slope and consolidation settlement problems are not sensitive to the trend function used while more complex trend functions produce worse results for a square footing resting on sand subjected to inclined loading. In the three GEOSNet problems studied, the kriging model with constant trend function can produces accurate failure probability prediction when the number of calibration samples is sufficiently large.