The current work focuses on improving volcanic ash forecasts by integrating satellite observations of ash into the Lagrangian transport and dispersion model, HYSPLIT. The accuracy of HYSPLIT output is dependent on the accuracy of the initialization: the initial position, size distribution, and amount of ash as a function of time. Satellite observations from passive infrared, IR, sensors are used both to construct the initialization term and for verification. Space‐based lidar observations are used for further verification. We compare model output produced using different initializations for the 2008 eruption of Kasatochi in the Aleutian Islands. Simple source terms, such as a uniform vertical line or cylindrical source above the vent, are compared to initializations derived from satellite measurements of position, mass loading, effective radius, and height of the downwind ash cloud. Using satellite measurements of column mass loading of ash to constrain the source term produces better long‐term predictions than using an empirical equation relating mass eruption rate and plume height above the vent. Even though some quantities, such as the cloud thickness, must be estimated, initializations which release particles at the position of the observed ash cloud produce model output which is comparable to or better than the model output produced with source terms located above and around the vent. Space‐based lidar data, passive IR retrievals of ash cloud top height, and model output agree well with each other, and all suggest that the Kasatochi ash cloud evolved into a complex three‐dimensional structure.