Data assimilation methods have been used to improve the performances of land surface models by integrating remote sensing and in situ measurements. However, the impact of data assimilation on improving the forecast of land surface variables has not been well studied, which is essential for weather and hydrology forecasting. In this study, a multi-pass land data assimilation scheme (MLDAS) based on the Noah-MP model was used to predict short-term land surface variables (e.g., sensible heat fluxes (H), latent heat fluxes (LE), and surface soil moisture (SM)) by jointly assimilating soil moisture, leaf area index (LAI) and solar-induced chlorophyll fluorescence (SIF). The test was conducted at the Mead site during the growing season (1 May to 30 September) in 2003, 2004, and 2005. Four assimilation-prediction scenarios (assimilating for 15 days, 45 days, 75 days, and 105 days from 1 May, then predicting one future month) are adapted to evaluate the influence of assimilation on subsequent prediction against Noah-MP open-loop simulation (OL). On average, MLDAS produces 28.65%, 27.79%, and 19.15% lower root square deviations (RMSD) for daily H, LE, and SM prediction compared to open-loop run, respectively. The influence of assimilation on prediction can reach around 60 days and 100 days for H (LE) and SM, respectively. Our findings indicate that data assimilation can improve the accuracy of land surface variables in a short-term prediction period.