This is Part II of the themed issue focusing on flow and other related transport problems in porous media with emphasis on how uncertainties in hydraulic properties are characterized and how these uncertainties impact engineering decisions. The Editorial for Part I has described the research challenges and the management of uncertainties as one step towards the broader goal of digitalization (Phoon, 2019). Data lies at the core. But deeper insights must be gleaned, beyond reducing data to deterministic inputs in a physical model to predict deterministic responses or as direct measurements of responses to support the observational approach, to produce sufficient value for decision making so that data can be viewed as assets in themselves. The Guest Editor ventures to suggest that the perceived need to cope with uncertainties is predicated on a debatable belief that big noisy data of indirect relevance to the problem at hand are less useful than high quality precise input data directly useful to a physical model (usually “small” data). The conventional coping mechanism to force-fit real world big noisy data into a few ideal noise-free inputs mandated by the prevailing deterministic modelling paradigm is to determine a single ‘characteristic’ value for each input in the form of a cautious estimate, worst credible estimate or a fractile. There are many well-known examples in other industries that demonstrated the value of exploiting big indirect data (BID) that accepts the characteristics of real world data ‘as it is’ without the mediation of physical models and without imposing artificial determinism. Such examples have yet to emerge in geotechnical engineering, but a recent report by the Boston Consulting Group (Gerbert et al., 2016) pointed out that the engineering and construction industry is ripe for change.
In Part II of this themed issue, Feng and Vardanega (2019) presented a simple regression model for predicting the saturated hydraulic conductivity using the water content ratio. The calibration database (FG/KSAT-1358) is exceptionally large, consisting of 1358 experimental values of saturated hydraulic conductivity representative of a wide range of fine-grained soils. This generic database is an example of BID. Other generic univariate and multivariate databases on soil/rock properties are reported in Phoon et al. (2016) and Ching et al. (2016). A comprehensive survey of performance databases for geotechnical structures (including piles) was carried out by Phoon and Tang (2019). It is often mentioned that ‘data is the new oil’. If this were to be true, one would need to find it (literature, design offices, etc.), extract it (digitized records), refine it (compilation of meaningful databases), distribute it (open sharing of databases such as ISSMGE TC304 database open sharing initiative, 304dB – ISSMGE, 2019), and monetize it such as the application of probabilistic methods to manage the impact of uncertainty on the stability of embankment or slope described below.
Two of these papers (Feng et al., 2019; Lanzafame and Sitar, 2019) modeled the uncertain soil properties as random variables. The other two papers (Liu et al., 2019; Zhu and Witt, 2019) discussed emergent phenomena arising from spatial variation that are not observable in conventional homogeneous analyses. Feng et al. (2019) investigated the effect of precipitation extremes on the stability of a typical Xuzhou–Shanghai high-speed railway embankment section. Infiltration and stability analyses were conducted for three extreme storm scenarios upscaled from a base storm which occurred in July 2016 in Nanjing. The fill materials and the foundation soil are assumed to be silty soils with a saturated hydraulic conductivity of 41 mm/h. This parameter (ks) is the governing parameter and its uncertainty is characterised by a coefficient of variation of 100%. It was found that an extreme storm magnifies the effects of the uncertainty of ks and the uncertainty bounds of factor of safety are wider under a more severe storm. Lanzafame and Sitar (2019) studied the probability of failure of a flood defence sandy embankment situated on top of a low hydraulic conductivity blanket layer of silt and/or clay, and a high hydraulic conductivity aquifer layer of sandy material. The random variables considered are (a) thickness, unit weight, effective cohesion and friction angle of the blanket layer; (b) the unit weight and friction angle of the embankment; and (c) hydraulic conductivity ratio. It was shown that the factor of safety and reliability for two failure modes (underseepage and slope stability) were dominated by the blanket thickness followed by the blanket unit weight. Liu et al. (2019) studied the local velocity distribution under an earth embankment, induced by the spatial variability of the foundation hydraulic conductivity, and considered its influence on the potential for piping. The hydraulic conductivity is modelled as a lognormal anisotropic random field with a mean of 10−6 m/s, coefficient of variation (COV) of 0·1, 0·5, 1·0, 2·0, 3·0, 4·0, 5·0 and 6·0, and the ratio of horizontal to vertical scale of fluctuation was 1, 8 and 20. The results showed that the exit velocity gradient for the spatially variable case is generally higher than that of the classical homogeneous case. In addition, when the COV is low, the location of the maximum local velocity is typically near the downstream toe of the embankment, following conventional wisdom from homogeneous analysis. In contrast, this location is more scattered when the COV is high. Zhu and Witt (2019) conducted a stochastic infinite slope stability analysis during infiltration in a medium with spatially variable saturated hydraulic conductivity that increases linearly in the mean sense. It was found that, during extended infiltration resulting in a build-up of positive pore water pressures, a linear mean trend has a significant effect on the pressure head profile and the factor of safety profile. A larger linear rate of increase in the mean produces a larger pressure head and a lower factor of safety. In addition, the probability of failure can be highly underestimated if the actual linear trend is assumed to be uniform.
The Editorial for Part I pointed to one critical future research need to characterize and exploit geotechnical information in its original form, Multivariate, Uncertain and Unique, Sparse, Incomplete, and potentially Corrupted (MUSIC). To guide the exploitation of MUSIC in a manner that will be of value to practice, robust, maintains alignment with current knowledge and experience, and engages engineering judgment in a meaningful way, it is useful to bear the following ‘Six Es’ in mind.
Economic value: Focus on monetizing data. Remember the adage: ‘all models are wrong, but some are useful’.
Essence: Data is the essence and therefore algorithms must be data-centric besides value-centric.
Extremes: Pay attention to potential presence of corrupted data. Algorithms should include identification of outliers and/or made robust against outliers.
Errors: Biased and imprecise data will produce biased and imprecise outcomes. It is not sufficient to provide the most likely outcomes because an engineer needs to manage risks. Responsible risk management is a core element of our professional ethics founded on protection of human lives and communities.
Extrapolation: Need to watch out for over-fitting and to caution users when extrapolation occurs.
Existing practice: Correlation is not the same causality. Engineers cannot ‘understand’ outcomes delivered purely by a black-box algorithm and cannot meaningfully ‘agree’ or ‘disagree’ with such outcomes. It is judicious to establish a degree of connection with the existing body of knowledge and experience.
The Guest Editor would like to extend his gratitude to the invited authors for their technical contributions that further our understanding of this challenging class of problems.
