Fires play an important role in the atmosphere. They release aerosol particles that can interact with solar radiation, cloud formation and precipitation processes, among others. Diverse fire emission inventories exist and are used in atmospheric models to account for them. However, they differ considerably among them up to a factor of 7 in some regions. To minimize the uncertainty of existent fire emission inventories, we developed an Ensemble Fire emission inventory (EnsemFire). We conducted global simulations using the UFS fully coupled (Atmosphere-Ocean-Wave-Ice-Aerosols) model during August 2016, August 2017 and March 2018. EnsemFire reduced the biases and errors (RMSE) of aerosol optical depth (AOD) at 550 nm when compared to GBBEPx, which is NOAA's current operational fire emission inventory. We further explored the prediction of fire emissions for operational forecasts. This is a challenging task due to the inherent erratic behavior of fires and the ignition sources (natural vs. human-made). For mid-range weather forecasts (5-7 days), a common approach is assuming that fires will remain constant over the next days. Nonetheless, this method is not suitable for longer range forecasts, such as subseasonal-to-seasonal (S2S) since, at these scales, the interaction of the aerosols particles with the Earth system components becomes more relevant. To address this, we produced two predicted fire emission datasets: (1) Ensem-FCST, which is based on statistical methods to predict fire emissions up to one month in advance, and (2) Ensem-FCST-AI, which utilizes a machine learning model (LightGBM) and considers a wide-range of meteorological variables from GFS to estimate fire radiative power (FRP) up to one month in advance. We conducted simulations with these two predicted fire emissions for the same periods as above. Results showed that Ensem-FCST tends to underestimate AOD in time, while Ensem-FCST-AI produces slight overestimations. Ensem-FCST-AI is a promissory tool to improve NOAA's S2S forecasts, since it captures remarkably well the day-to-day variations of fire emissions.