Large-scale climate drivers of groundwater level variations in northern France over the last century

Groundwater levels (GWL) are often affected by low-frequency variations, i.e., on timescales greater than a year. In northern France, this low-frequency variability (LFV) generally explains more than half of total GWL variability. Such LFV in GWL is thus often responsible for the emergence of groundwater droughts and floods. It is, herefore, essential to explore the large-scale climate drivers of such LFV to improve GWL prediction system and better forecast and manage groundwater droughts and floods. In this study, we investigate the large-scale climate drivers of the LFV of northern France GWL. We use a century long timeseries of GWL and precipitation, and examine their links to different large-scale ocean atmospheric patterns, using sea-surface temperature [SST] and geopotential heights at 500mb [Z500]. We first assessed the multi-timescale variability of GWL and precipitation using continuous and discrete wavelet transforms. Besides annual variability, interannual (IV: ~7-yr) and decadal (DV: ~14-yr) variability better explain GWLs’ total variance. Then, we identify atmospheric and oceanic patterns, which are associated with each timescale of LFV using composite analyses. IV is associated with a “wave train” pattern over the Euro-Atlantic sector with a low (high) pressure system over northernwestern Europe favoring wet (dry) conditions over northern France during interannual high (low) GWL. Such pattern seems quite similar to a ridge regime (during high GWL) and Scandinavia blocking regime (during low GWL). IV is also associated with a SST pattern in North Atlantic that is typical of an Atlantic ridge regime. DV seems to be associated with a low (high) pressure system covering Great Britain and extending through the Gulfstream front favoring wet (dry) conditions and leading to decadal high (low) GWL in northern France. The SST in North Atlantic display a very similar pattern than atmospheric circulation. Z500 pattern associated to IV is close to the Scandinavian Pattern (SCAND) index, whereas that associated to DV diverges quite significantly from a standard climate index. Using a multiresolution Empirical Statistical Downscaling model, we show that low-frequency variations of GWL can be correctly simulated using the large-scale low-frequency information in atmospheric fields (Z500), which enhance the simulation of total GWL too. This result highlights the potential of largescale atmospheric patterns found by composite analysis to be used as predictors in statistical models for reconstructing or predicting GWL.