Uncertainty characterization and settlement analyses: the importance of distribution types

In recent years, significant effort has been put into considering advanced probabilistic soil models that represent soil variability [see e.g.,

1

,

2

]. In particular, the sensitivity of probability results to changes in the mean and variance of properties, and to changes in the structure of spatial correlation of soil properties has received wide attention [

3

–

5

].

We study the influence of using different types of statistical distributions to characterize the Young’s modulus of the soil for computation of settlements on spatially random soil. We use a linear elastic finite element model with a rigid footing founded on elastic soil. Poisson’s ratio of the soil is considered constant, and Young’s modulus is characterized using random fields with two limiting extreme values of their scales of fluctuation. We perform a number of simulations in which finite element program IRIS is used to compute the settlement of the foundation for each realization of the Young’s modulus random field. The mean Young’s modulus of the soil was considered constant in all cases, and standard deviations were varied among different simulations.

As expected, computed settlements are observed to be very similar when lognormal and gamma distributions are similar. Similarly, an “averaging” effect is observed for small scales of fluctuation; such “averaging” effect makes settlement results to be similar among different realizations of the random field. The election between lognormal or gamma distributions is significant, however, for high values of σ

E

. Differences between settlements computed with both distributions are particularly significant for large values of the scale of fluctuation. That is, our results suggest that, in some cases, the type of distribution considered for characterization of the random field of Young’s modulus can have a significant impact on computed settlement results. Such observation should be taken into account when performing simulations of random fields in the context of settlement analyses.