Evaluating the Split Window Volcanic Ash Detection Method Using the Numerical Dispersion Model Puff

Use of remote sensing data is one of the few techniques for monitoring the movement of volcanic ash clouds during and after an eruption. The widely used "split window", or band 4 (10.3-11.3 um) minus band 5 (11.5-12.5 um) technique has proven capable of discriminating volcanic clouds following an eruption. However, the detection limits of this technique are difficult to assess due to the limited availability of corroborating data. As part of the Solid Earth Geophysics Summer REU program at the Geophysical Institute, University of Alaska, Fairbanks, the volcanic ash dispersion model Puff was used to quantify the ability of the split window technique to monitor the ash cloud produced by the September 17, 1992, eruption of Mt. Spurr, Alaska. Eight split-window images of the ash cloud were created using Advanced Very High Resolution Radiometer (AVHRR) imagery from 09-17-1992 12:40Z to 09-20-1992 17:04Z. Correspondingly, a 78-hour simulation of the eruption was produced using the Puff volcanic ash dispersion model. Simulation results typically overlap satellite images only in the more dense regions, which leads to the hypothesis that there is a concentration limit to the split-window technique below which the technique fails to discriminate ash clouds from normal clouds. This limit was quantified by systematically filtering out low concentration areas from the simulation results and comparing with the satellite imagery. Results indicate that masking areas with 3% or less of the maximum cloud density produce model results that best resemble satellite imagery. Furthermore, this technique permitted assessing different model input parameters for best simulating this eruptive event. Results indicate that initializing the model with a linear distribution of material above the vent produced better results than using a non-linear distribution more heavily weighted at higher elevations. The implications of these results for using remote sensing data and dispersion model results for monitoring future volcanic eruptions will be presented.