Contrôle de la représentativité d’échantillonnage des sols contaminés

Abstract

Anthropogenic sites can be considered from two complementary perspectives: on the one hand, as a limitation due to their legacy contamination; on the other hand, as a driver for redevelopment in a context where access to urban land is becoming increasingly scarce and where the inventory of neglected sites remains substantial because of pollution. The rehabilitation of these sites depends on environmental characterization, which is still affected by high uncertainty. In practice, sampling errors that are still observed can lead to decisions marked either by major additional costs or by insufficient control of environmental and social risks. Within this framework, the present doctoral research focused on identifying and quantifying the factors influencing sampling bias and variance, with particular emphasis on the effects of sampling method and moisture on the reproducibility, variance, and representativeness of analytical results. Three research axes, grounded in the heterogeneity of particulate matter, were examined. First, the representativeness of the particle size distribution of a crushed-stone and clay mixture was assessed as a function of moisture and sampling protocol, by applying non probabilistic grab sampling and probabilistic sampling to dry and moist lots. A significant decrease in variability was observed when probabilistic sampling was used, particularly under dry conditions, whereas increased sensitivity to moisture was observed for grab sampling. Second, performance comparisons conducted for two analytes (physical and chemical) at two concentration levels showed that the sampling method exerted the dominant effect on bias and relative variance; an improvement in representativeness by more than two orders of magnitude was attributed to the probabilistic approach. It was further shown that screening out coarse particles, although associated with reduced apparent variability, introduced substantial bias by modifying the primary sample. Third, moisture was shown to reduce bias and relative variance for both protocols, without, however, compensating for the intrinsic limitations of grab sampling. Overall, the adoption of probabilistic protocols inspired by the Theory of Sampling of Fragmented Materials (TEMM) is therefore supported as a preferred pathway to strengthen result representativeness and to support more reliable rehabilitation decisions. More specifically, this thesis demonstrates that durable improvements in representativeness for contaminated site characterization are achieved through adopting TEMM-inspired probabilistic protocols capable of robustly controlling sampling errors, including under variable moisture conditions, and on avoiding sample preparation practices that may bias the sample.

Date	1 May 2026
Original language	French
Awarding Institution	École de technologie supérieure
Supervisor	Jean-Sébastien Dubé (Supervisor)