STATISTICAL PRE-PROCESSING AND RECONSTRUCTION METHODS FOR SUBSURFACE HYDRO-METEOROLOGICAL AND CRACK APERTURE TIME SERIES

Summary. This paper focuses on statistical methods for pre-processing, reconstructing and/or synthesizing hydro-meteorological time series and shrinkage crack aperture signals collected in the galleries of an Underground Research Laboratory in clay rock (Tournemire, France). 1 INTRODUCTION We focus here on the effects of hydro-meteorological disturbances on the dynamics of shrinkage crack apertures measured in the walls of subsurface galleries in an URL (Underground Research Laboratory) in Tournemire, France. This URL is dedicated to the study of radioactive waste disposal in a clay rock geological repository. The present work is part of more extensive ongoing study involving statistical analyzes and cross-analyses of pore pressure signals, hydro-meteorological signals (air pressure, relative humidity, temperature), shrinkage crack dynamics, and water contents at various distances from the gallery walls. These analyses are being conducted in order to characterize the clay rock behavior hydraulically and mechanically. However, the collected time series have data gaps, spurious values, and irregular time steps. In order to obtain the longest possible "clean" signals, the raw data need to be pre-processed. This implies homogenizing the time steps, detecting spurious values, and reconstructing at least some parts of the missing data on a regular time step grid. (Beyond pre-processing, we are also interested in complete synthesis of the signals based on their statistical properties). In the present work, typically, the longest continuous sequences are on the order of half a year or a bit more (30 weeks), the reference time step is on the order of 15 mn, and the longest data gaps which have been reconstructed were on the order of one week, using a residual Auto-Regressive reconstruction method. In fact, sequences longer than one month have also been