Résumé
We used a high-resolution delta-O18 record from the North Greenland Ice Core Project (NGRIP) to derive a low-dimensional inverse stochastic model of this record. The model reproduces well the record's visual features, its probability density function and its lag-correlation function. The model's deterministic part needs to include explicitly several lags in order to achieve these properties of the delta-O18 variations. The relevance of the memory effects captured by these lagged dependencies, along with their physical interpretation, will be discussed from a theoretical perspective based on the Mori-Zwanzig formalism of statistical mechanics.