Résumé : Realistic models of the Galactic double white dwarf (DWD) population are crucial for testing and quantitatively defining the science objectives of the Laser Interferometer Space Antenna (LISA), a future European Space Agency's gravitational-wave observatory. In addition to numerous individually detectable DWDs, LISA will also detect an unresolved confusion foreground produced by the underlying Galactic population, which will affect the detectability of all LISA sources at frequencies below a few mHz. So far, the modelling of the DWD population for LISA has been based on binary population synthesis (BPS) techniques. The aim of this study is to construct an observationally driven population. To achieve this, we employ a model developed by Maoz, Hallakoun and Badenes (2018) for the statistical analysis of the local DWD population using two complementary large, multi-epoch, spectroscopic samples: the Sloan Digital Sky Survey (SDSS), and the Supernova Ia Progenitor surveY (SPY). We calculate the number of LISA-detectable DWDs and the Galactic confusion foreground, based on their assumptions and results. We find that the observationally driven estimates yield 1) 2 - 5 times more individually detectable DWDs than various BPS forecasts, and 2) a significantly different shape of the DWD confusion foreground. Both results have important implications for the LISA mission. A comparison between several variations to our underlying assumptions shows that our observationally driven model is robust, and that the uncertainty on the total number of LISA-detectable DWDs is in the order of 20 per cent.