Generative AI for hydro-meteorological time series

The Permanent Environmental Observatory (OPE) is located around the village of Bure, on the border of the Meuse and Haute-Marne departments, in a sparsely populated area. The observatory's purpose is to assess water quality (surface and groundwater) using regular sampling that provides information on a large number of parameters. Here, we will focus on the processing of data collected by six automatic stations that continuously measure various parameters (conductivity, temperature, dissolved oxygen, PAHs, pH, dissolved organic carbon, nitrates, and water level), thus generating multi-site and multi-variable time series of surface water flow and quality.

The objective of this project is to propose a joint model of flow rates and water quality in the OPE zone. The model under development follows a generative AI approach. It is a diffusion model that is fitted to the case of multi-site and multi-variable time series, and adapted to allow training on incomplete data and stochastic simulation in the following contexts: gap-filling, nowcasting, synthetic series generation, and conditional simulation with covariates.

The outputs of this diffusion model will then be used to study the impact of different climate and/or anthropogenic change scenarios on headwater watershed hydro-systems, and to emulate certain hydro-meteorological parameters used as inputs for impact models (hydrogeology, stream ecology, etc.).