STATISTICAL FORECASTING OF REPETITIOUS DOME FAILURES DURING THE WANING ERUPTION OF REDOUBT VOLCANO, ALASKA, FEBRUARY-APRIL 1990

The waning phase of the 1989-1990 eruption of Redoubt Volcano in the C ook Inlet region of south-central Alaska comprised a quasi-regular pat tern of repetitious dome growth and destruction that lasted from Febru ary 15 to late April 1990. The dome failures produced ash plumes hazar dous to airline traffic. In response to this hazard, the Alaska Volcan o Observatory sought to forecast these ash-producing events using two approaches. One approach built on early successes in issuing warnings before major eruptions on December 14, 1989 and January 2, 1990. These warnings were based largely on changes in seismic activity related to the occurrence of precursory swarms of long-period seismic events. Th e search for precursory swarms of long-period seismicity was continued through the waning phase of the eruption and led to warnings before t ephra eruptions on March 23 and April 6. The observed regularity of do me failures after February 15 suggested that a statistical forecasting method based on a constant-rate failure model might also be successfu l. The first statistical forecast was issued on March 16 after seven e vents had occurred, at an average interval of 4.5 days. At this time, the interval between dome failures abruptly lengthened. Accordingly, t he forecast was unsuccessful and further forecasting was suspended unt il the regularity of subsequent failures could be confirmed. Statistic al forecasting resumed on April 12, after four dome failure episodes s eparated by an average of 7.8 days. One dome failure (April 15) was su ccessfully forecast using a 70% confidence window, and a second event (April 21) was narrowly missed before the end of the activity. The ces sation of dome failures after April 21 resulted in a concluding false alarm. Although forecasting success during the eruption was limited, r etrospective analysis shows that early and consistent application of t he statistical method using a constant-rate failure model and a 90% co nfidence window could have yielded five successful forecasts and two f alse alarms; no events would have been missed. On closer examination, the intervals between successive dome failures are not uniform but ten d to increase with time. This increase attests to the continuous, slow ly decreasing supply of magma to the surface vent during the waning ph ase of the eruption. The domes formed in a precarious position in a br each in the summit crater rim where they were susceptible to gravitati onal collapse. The instability of the February 15-April 21 domes relat ive to the earlier domes is attributed to reaming the lip of the vent by a laterally directed explosion during the major dome-destroying eru ption of February 15, a process which would leave a less secure founda tion for subsequent domes.