Abstract. Forecasting the Thermosphere (Atmosphere's uppermost layer, from about 90 to 800 km altitude) is crucial to many space-related applications, from space mission design, to re-entry operations, to space surveillance. Thermospheric dynamics is directly linked to the solar dynamics through the solar UV input, which is highly variable, and through the solar wind and plasma fluxes, impacting Earth's magnetosphere. The solar input is non-periodic and non-stationary, with long-term modulations from the solar rotation and the solar cycle, and impulsive components, due to magnetic storms. Proxies of the solar input exist and may be used to forecast the thermosphere, such as the F10.7 radio flux and the MgII EUV flux. They relate to physical processes on the Solar atmosphere. Other indices, such as the Ap geomagnetic index, connect with Earth's geomagnetic environment. We analyse the proxies' time series comparing them with in-situ density data from the ESA/GOCE gravity mission, operational from March 2009 to November 2013, therefore covering the full rising phase of solar cycle XXIV, exposing the entire dynamic range of the solar input. We use Empirical Mode Decomposition (EMD), an analysis technique appropriate to non-periodic, multi-scale signals. Data are taken at an altitude of 260 km, exceptionally low for a LEO satellite, where density variations are the single most important perturbation to satellite dynamics. We show that the synthesized signal from optimally selected combinations of proxies's basis functions, notably Mg II for the solar flux and Ap for the plasma component, shows a very good agreement with thermospheric data obtained by GOCE, during low and medium solar activity periods. In periods of maximum solar activity, density enhancements are also well represented. The Mg II index proves to be, in general, a better proxy than the F10.7 one, to model the solar flux, because of its specific response to the UV spectrum, whose variations have the largest impact over thermospheric density.