Although central to the sustainable development of human societies, freshwater resources are threatened globally by combined effects of climate change, population growth and pollution. In the context of increased withdrawals from increasingly scarce resources, the sustainability of water supply relies on the understanding of water distribution between the different compartments in continental hydrosystems. Located at the intersection between surface and subsurface compartments, the stream-aquifer interface is therefore of particular interest for the management of both surface and subsurface water resources.
Within the stream-aquifer interface, the hyporheic zone (HZ), defined as the zone below the streambed where hydrological exchanges are controlled by surface water pressure gradients as well as subsurface properties, is extremely chemically and biologically active. While HZ processes are governed by water fluxes at the local scale, they integrate over the hydrological river network and impact water availability and quality up to the catchment scale. However, characterized by complex nested spatial and temporal patterns, water fluxes in the HZ remain poorly understood.
Quantifying water movement in the subsurface relies on the interpretation of collected environmental data, and is usually hindered by a general lack of information about subsurface properties due to low data availability as well as spatial and temporal variability. Stochastic hydrogeology provides a framework for assimilating information contained in sparse environmental data by posing problems in a probabilistic yet physical framework, where physical properties in the subsurface are considered as random variables defined by their probability density function (pdf) and are linked to observations by means of physically-based models.
This dissertation aims at assimilating ex-situ and collected in-situ data for estimating spatially-distributed water exchanges at the stream-aquifer interface. An integrated approach is proposed, combining experimental monitoring developments, physically-based numerical modeling and stochastic hydrogeology concepts. The approach is illustrated on an agricultural and heavily anthropized sedimentary case study, the Avenelles basin, France.
Foundational aspects related to ex-situ and in-situ data assimilation were investigated. The issue of in-situ data scarcity and generally sparse, infrequent and expensive measurements was addressed with the development of a low-cost, easy to construct and robust experimental sensor, the LOMOS-mini. Uncertainty-quantified estimation of subsurface properties from collected in-situ data was addressed by adopting a Bayesian approach, where the identifiability of each property is studied via the update of a non-informative prior distribution into a posterior distribution conditioned on measurements. The assimilation of ex-situ data from similar sites was enabled with the introduction of regionalized priors, defined as informative pdfs resulting from the assimilation of ex-situ data of possibly multiple types and recognizing inter-site variability. These developed concepts and tools were then applied to the Avenelles basin case study, where LOMOS-mini data collected during field campaigns supported the estimation of local-scale and spatially-distributed hydrogeological properties and water exchanges in the HZ and the investigation of their spatial distribution.