Peter Webley (AVO/UAF), January 2008 
|
|
Sensitivity Testing
Technical Report - Sensitivity of Puff: a volcanic ash tracking model (121 k)
REU Report - Modeling Volcanic Ash Plumes Using Puff and AVHRR (1.5 MB)
Input Parameter Analysis
As part of the UPOS project, we are examining the sensitivity of Puff. Because of the relatively large number of adjustable parameters in Puff, it is adventageous to know which parameters have the most influence over the outcome of a simulation. The figures listed below show the cumulative distribution function for several parameters after a 10,000-run sensitivity test was performed.
For this test, the Mount Spurr eruption on August 19, 1992 was used. The success of an individual simulation was determined by
comparing the model results to satellite imagery obtained using AVHRR. If the simulation predictions covered 90% of the plume
visible in the satellite image, the simulation was considered a success.
The more sensitive parameters will have a larger "span" between their Success and their Failure curves. For this test, the
horizontal diffusion was most important, followed by the
plume height. The time step and
eruption hours were relatively unimportant.
Repeatability
The first question for a Monte-Carlo-type simulation is: How many runs does it take to get an average? To answer this,
we ran puff in parallel 100 times, taking an average over n runs at each step and then comparing the difference
between the two parallel runs at each step. The results shown here
indicate that a reasonably good average occurs after about 40 simulations. We use this fact in the rest of the sensitivity analysis.
We THEN analyzed the sensitivity of puff results to changing the input parameters for an arbitrary eruption. A single parameter
was varied over a range of values. The 10-simulation average was compared against a similar average with the incremental change.
Variability was quantified by "bin"ing the ash particles in a 3-D grid and counting the number of particles that did not end up in
the same grid box. Simulation durations of 1, 6, 12, 24, 48 and 72 hours were analyzed and
the results can be seen here
Wind Model Comparison
As part of on-going diagnostics, we are beginning to investigate the effects of different weather models on Puff predictions.
We currently are investigating four different weather models: AVN via Unidata, AVN via NWS, 35-km regional Eta, and 1x1-degree MRF.
For a hypothetical eruption of Mt. Spurr on 01-09-2001, go here to see the
Puff predictions using the four models over a 24-hour period.
Wind-Field Resolution Analysis
A common question pertaining to all tracking models is "What is the effect of the input wind-field model resolution?".
To investigate this regarding Puff, we used the
10-km regional ETA model on Grid #249 over Alaska. A hypothetical eruption of Mt. Spurr on May 22, 2002 was simulated. The
resolution of the input wind-field was gradually decreased from 10-km to 400-km. Successive simulations were performed to determine
at what point the predictions begin to substantially deviate. Images of the 12-hour simulation can be seen at
3, 6,
9 and 12
hours after the eruption began.
In is interesting to note that the qualitative predictions do not seem to differ much until the grid resolution degrades from
10 km to almost 400 km. Furthermore, the differences only become evdent at the highest elevations. There does appear to be
some stantial differences on the lower-elevation ash near the volcano at 12 hours. However, the general trend appears to be unchanged
|
The information provided on this site is purely for educational purposes, and should not be used for actual volcanic monitoring and predictive purposes.
The University of Alaska and the Geophysical Institute are not responsible for content provided on these pages.
|
Email comments
|
|
|
