FIRE and MUD Contents

Real-time Seismic Amplitude Measurement (RSAM) and Seismic Spectral Amplitude Measurement (SSAM) Analyses with the Materials Failure Forecast Method (FFM), June 1991 Explosive Eruption at Mount Pinatubo

By Reinold R. Cornelius¹ and Barry Voight^{1 2}

¹The Pennsylvania State University, Department of Geosciences, University Park, PA 16802.

²Also at U.S. Geological Survey.

ABSTRACT

Seismic and other unrest at Mount Pinatubo culminated in a major eruption that began on June 12, 1991, and peaked on June 15, when it generated ash plumes over 35 kilometers high. Precursory seismic activity was tracked by the scientists of the Philippine Institute of Volcanology and Seismology and the U.S. Geological Survey using the Real-time Seismic Amplitude Measurement (RSAM) and Seismic Spectral Amplitude Measurement (SSAM) systems, which provide consecutive averages of the absolute amplitudes of the seismic signal from several seismometers. RSAM and SSAM records are compared here for a variety of event types, involving combinations of volcano-tectonic seismicity, long-period swarms, tremor, and hybrid activity. Due to the strong signal and broad-band nature of the seismicity, RSAM patterns were generally similar to individual SSAM bands; unlike at Redoubt Volcano, Alaska, in 1990, at Pinatubo RSAM provided an authentic measure of overall seismic energy release throughout the precursory period. These data were used by the authors to test applicability of the Materials Failure Forecast Method (FFM), in which the time of failure (eruption) can be estimated by extrapolation of the inverse rate versus time curve toward the abscissa.

Graphical RSAM data for the period June 1 to 1600 on June 10 were sent to us via fax. Inverse RSAM, representing inverse seismic energy rates, was plotted against time and extrapolated according to conventional FFM procedures. The fit indicated impending failure about June 12 and allowed us to anticipate an explosive eruption in foresight for the first time with the FFM technique. In hindsight, inverse-rate analyses on June 7 would have suggested an eruption window from June 12 to 20, and analyses between June 12 and June 15 would have anticipated an eruption between June 14 and June 16.

Note to readers: Figures open in separate windows. To return to the text, close the figure's window or bring the text window to the front.

INTRODUCTION

The recognition of unrest at Mount Pinatubo began on April 2, 1991, with steam explosions from a 1.5-km-long line of vents high on the north flank of the volcano, following a 500-year-long period of quiescence (Newhall and others, this volume). The unrest culminated in a major explosive eruption, beginning on June 12 and peaking on June 15, that generated extensive pyroclastic flows and ash plumes over 35 km high. Estimates of ash and gas emission rank this eruption as probably the world's largest since the eruption of Novarupta (Katmai) in 1912. A joint effort by the Philippine Institute of Volcanology and Seismology (PHIVOLCS) and the U.S. Geological Survey Volcano Crisis Assistance Team (USGS VCAT) produced effective hazard evaluation and forecast advisories, enabling decisionmakers to manage massive evacuations that possibly saved tens of thousands of lives (Punongbayan and others, this volume; Wolfe and Hoblitt, this volume).

Here, we show how the Materials Failure Forecast Method (FFM) was used, and could have been used, to track precursory volcanic activity at Pinatubo and to forecast forthcoming activity by fitting data according to an empirical rate-acceleration relation (Voight, 1988, 1989). This method uses numerical or graphical rate extrapolation toward an expected failure rate to forecast time windows within which eruptions are expected (Cornelius and Voight, 1989, 1994, 1995; Voight and Cornelius, 1991). Some of the series of FFM analyses on Pinatubo data reported here were performed in foresight, and others in hindsight.

We use Real-time Seismic Amplitude Measurements (RSAM) (Murray and Endo, 1989) as a measure of seismic-energy release rate and compare these data with the spectrally filtered energy release rate from the Seismic Spectral Amplitude Measurement system (SSAM) (Power and others, this volume).

The onsite eruption forecasts by scientists of PHIVOLCS and the USGS VCAT were also based on RSAM, as well as earthquake hypocenter migration, and waveform characteristics (Harlow and others, 1991; this volume), changes in gas flux (Daag, Tubianosa, and others, 1991; this volume), and other observations.

Following a brief presentation of event chronology, FFM methodology, and RSAM and SSAM monitoring tools, we present our foresight analysis based on RSAM through June 10, prior to the eruptive cycle of June 12 through 15. Hindsight analyses based on the original data are then presented and compared with foresight analyses. We compare RSAM records with comparable records for selected SSAM bands. Finally, we demonstrate with analyses of data through June 13 and 14 how the seismic development could have been tracked and forecasts made during a lull of explosive activity prior to climactic ejection on June 15. Our goal is to use the Pinatubo experience to evaluate FFM as a forecasting tool, to recognize limitations and possible pitfalls in applying FFM with RSAM and SSAM, and to provide perspectives for future near-real-time foresight applications.

CHRONOLOGIC SUMMARY

The following brief chronology is included to provide context to the discussions on forecasting. More detailed information is available elsewhere (Hoblitt, Wolfe, and others, this volume; Wolfe and Hoblitt, this volume). All times and dates in this paper are in Philippine local time, unless otherwise specified.

The climatic eruption on June 15 was preceded by at least 10 weeks of unrest. In response, PHIVOLCS installed several portable seismometers in early April; subsequently, a seven-station radiotelemetered seismic network was established during late April and early May that used equipment from the USGS VCAT cache at the Cascades Volcano Observatory (CVO) (Lockhart and others, this volume). Two electronic tiltmeters were installed in late May (Ewert and others, this volume). The data were received and evaluated with the aid of a PC-based acquisition and analysis system established at Clark Air Base (Murray and others, this volume).

Numerous small, high-frequency volcano-tectonic (VT) earthquakes occurred during April and early May. Prior to late May, seismic activity was mainly located 4 to 8 km northwest of the phreatic vents of Pinatubo, at 3 to 6 km depth, with lesser activity under the summit (Harlow and others, this volume). The cause of this activity, whether due to hydrothermal, tectonic, or magmatic processes, was puzzling and ambiguous (Punongbayan and others, 1991; this volume). Epicentral clusters and seismic monitoring stations are shown in figure 1. The climb of SO₂ flux to a peak value of 5,000 t/d on May 28 suggested the rise of magma to a level that promoted gas exsolution and (or) drying and gas transmission through the hydrothermal system (Daag and others, 1991; Daag, Tubianosa, and others, this volume). After late May, the rate of VT seismicity increased markedly under the near-summit vents on Pinatubo. A pronounced decrease in SO₂ flux on June 5 was interpreted as near-surface sealing of the magma and led to speculation and concern about increasing gas pressurization (Daag, Tubianosa, and others, 1991; this volume).

Figure 1. Locations of telemetered seismic stations in the Pinatubo network between May 13 and June 16, 1991. Epicentral zone northwest of Mount Pinatubo particularly active in May; epicentral zone near summit gained in activity after June 1. Figure courtesy of Mori, Eberhart-Phillips, and Harlow, this volume.

A small steam explosion on June 3 marked increasing unrest, together with minor ash emission, heightening near-summit seismicity, tremor episodes, and outward tilt (Harlow and others, this volume; Ewert and others, this volume). In response, PHIVOLCS issued an Alert Level 3 notice on June 5, indicating "if trend of increasing unrest continues, eruption possible within 2 weeks" (Pinatubo Volcano Observatory Team, 1991; Wolfe, 1992; Punongbayan and others, this volume). The concern was for a "major pyroclastic eruption" (Wolfe and Hoblitt, this volume).

Outward tilt noticeably increased on June 6, and near-summit seismicity continued to grow until the afternoon on June 7, when enhanced steam and ash emission generated a column about 8 km high. Thereafter, seismicity and tilt decreased, and PHIVOLCS announced an Alert Level 4--"eruption possible within 24 h." The outward tilt and shallow seismicity suggested near-surface magma, and this was confirmed the following morning by observation of a 100-m-diameter dome on the northwest flank of the volcano. Shallow VT earthquake swarms under the dome, episodes of harmonic tremor, and weak ash emissions marked the period from June 8 through early June 12. At times "dilute ashy density currents ... seen from (a) distance... resembled the ash clouds associated with pyroclastic flows, but they neither coincided with distinct seismic explosion signals nor left recognizable flowage deposits" (Wolfe and Hoblitt, this volume). Nevertheless, observation of these currents added to existing apprehensions and perceptions of unrest and led PHIVOLCS at 1700 on June 9 to raise the Alert Level to 5--"eruption in progress" (Wolfe and Hoblitt, this volume). The radius of evacuation was extended to 20 km, with evacuations increased to 25,000 (Punongbayan and others, this volume; Wolfe and Hoblitt, this volume). On June 10 about 14,500 military personnel and dependents were evacuated from Clark Air Base, and only volcanological staff and 1,500 security personnel remained.

A major eruptive vent-clearing phase started at 0851 on June 12, and this produced a sequence of three plinian eruption clouds that reached altitudes at least 24 km (table 1), accompanied by pyroclastic flows (Hoblitt, Wolfe, and others, this volume). The eruptions at 2252 on June 12 and 0841 on June 13 were preceded by 2- to 4-h episodes of long-period (LP) seismicity, which enabled volcanologists to announce advance warning (Harlow and others, this volume). The evacuation radius was extended to 30 km, and total evacuees increased to more than 58,000 (Punongbayan and others, this volume; Wolfe and Hoblitt, this volume).

After a hiatus of about 28 h, explosions recommenced on June 14, producing mostly surge deposits (Hoblitt and others, this volume). These explosions were preceded by complex seismicity, including VT and LP seismicity, and continued for about 24 h. Sweeping ash clouds were observed at dawn on June 15 from Clark Air Base, and increased concern led to evacuation of remaining base personnel and volcanologists and seismic acquisition hardware. Although volcanologists returned later, a 3-h gap was left in seismic recording (Harlow and others, this volume). At 1342 on June 15, the episodic activity culminated in 9 h of continuous cataclysmic ejection that produced a tephra plume exceeding 35 km high and heavy ash and pumice falls.

THE MATERIALS FAILURE FORECAST METHOD

FFM for eruption forecasting is based on an empirical law describing failure of materials (Voight, 1988, 1989). The concept of "failure" is broadly interpreted and may involve diverse processes such as rupture of solid or fluid-saturated porous rock, or critical points of fluid pressurization affecting fluid transport and (or) crack opening. The governing equation (eq. 1) relates rates, , during accelerating activity, to their change in rate, , where the state variable stands for any of several possible precursor observations such as ground displacement or tilt or seismic parameters. Constants A and are empirical and is dimensionless.

= A       (1)

= [A(1-)(t-t_*) + _*^(1-)]^1/(1-)    (2)

Figure 2. Generic plots of curves governed by the Materials Failure Forecast Method equation for (A) rates and (B) inverse rates. Constants A and are empirical, as used in eq. 1 in text. Curves calculated for A = 0.1 and -0.5<<2.25. Initial displacement (_o) = 0.0 and initial rate(_*) = 1.0; both at time t_o = t_* = 0. Values of in increasing order are -0.5, 0.0, 0.25, 0.5, 0.75, 1.0, 1.5, 2.0, and 2.25.

The time t_s is an upper limit for the time of failure t_f, with the assumptions that the described dynamic complex is terminated by some type of rock mass-fluid system failure or instability and that the onset of an eruption follows a relatively high, yet finite, rate. The critical point in time (t_s) can be estimated by extrapolating an inverse-rate data set to the time axis. The point t_f may be estimated with a predetermined intercept that represents the expected finite failure rate,_f. As selection of this failure rate may be difficult in foresight, the default strategy is to solve for t_s. Both t_s and t_f may be used to estimate time of eruption.

The solution of equation 1 in terms of rate (eq. 2) leads to the convenient graphical procedure of "inverse-rate" plots. Inverse-rate trends are linear for the special case of = 2, which simplifies the extrapolation of data. For < 2, the inverse-rate curves are concave upward, and for > 2 convex upward (fig. 2B). The case = 1 implies exponential growth, and, in general, for <= 1 the inverse-rate curves are asymptotic to the time axis.

We distinguish between graphical and numerical applications of FFM. The graphical procedure relies on visual extrapolation of the inverse-rate data; extrapolation is not necessarily linear. In this paper the graphical approach is followed, commonly in conjunction with linear extrapolation that assumes = 2. Other regression procedures are also suitable with FFM (Cornelius and Voight, 1989; Cornelius, 1992). An overview of the method is given by Cornelius and Voight (1995), and software suitable for use with conventional PC hardware is available from the authors.

Rate-time data are commonly evaluated by using values averaged at selected uniform time increments. An alternative is to consider time-integrated data over approximately uniform "rate-magnitude" increments instead of uniform time increments. This procedure yields an increasingly higher frequency of rate data toward the end of an accelerating time series and results in an end-weighted rate calculation that emphasizes the latest precursory developments.

A continuous (but not necessarily linear) inverse-rate downward trend may be interrupted by a sudden decrease in rate (positive slope in the plot of inverse rate versus time; solid dots in fig. 3) which is then followed by a sudden increase in rates (steepened negative slope in the plot of inverse rate versus time; open circles in fig. 3). The combined effect of such a pair of jogs in the trend may or may not offset the overall curve. We refer to a short-time acceleration of the rate (open circles) as an accelerating jog and to a short-time deceleration of the rate (solid dots) as a decelerating jog. Accelerating jogs may be analyzed individually by FFM (dashed arrows in fig. 3), or the overall trend may be analyzed (heavy solid line in fig. 3).

Figure 3. Schematic inverse-rate curve. Open circles define downward (accelerating) inverse-rate trends; solid dots define decelerating jogs. The overall trend may be time shifted by a decelerating jog. Decelerating jogs may cyclically alternate with accelerating jogs (dashed arrows) steeper than the overall trend (heavy solid line). Inverse-rate trends can be fit numerically, and a data envelope (angled dotted lines) can be constructed at a specific confidence level. The data envelope can be intersected with a range of expected rates near time of eruption (horizontal dotted lines). The resulting eruption window is shown by vertically dashed lines. Time of eruption may be separated from the time of apparent peak rate by a delay interval; the rate at time of eruption may or may not be larger than the apparent peak rate. Time of singularity (t_s) for curves with > 1 is the time at which the curve fit indicates an expected infinite rate.

As used here, the eruption window spans the time between the intercepts for a range of anticipated (guessed) rates at time of failure (fig. 3), usually with the upper and lower branches of a data envelope reflecting the 97.5% confidence level. It is important to recognize that this confidence level applies only to the mathematical extrapolation of a particular data set, and its use does not imply a forecast with the same confidence level. The envelope is derived from the linear least squares analysis, either by assuming = 2 or by iterative linearization of equation 2 (Cornelius and Voight, 1995). The envelope is a measure of the error due to data scatter, and it increases in width with extrapolation over greater time.

Expected values of failure rate,_f, may be estimated from previously recorded peak rates of seismic activity associated with the onset of earlier eruption events (for example, in RSAM units, as discussed below). A relatively conservative eruption window (dashed, heavy vertical lines in fig. 3) is established with the lower bound, defined as the intersection of the earlier (left) branch of the extrapolated data envelope with the lower expected finite failure rate, and with the upper bound, defined as the intersection of the later (right) branch of the data envelope with the higher expected finite failure rate.

A delay interval may separate the apparent peak rate from the time of eruption (Voight, 1988; fig. 3). A higher or lower rate than the previous apparent peak rate may occur at the onset of an eruption, but the intervening delay interval is defined by drastically reduced rates: it is a period of apparent relative calm just prior to the eruption. It may be possible to establish a critical "failure rate" based on previously observed apparent peak rates in a sequence of eruptions rather than on the reduced rates sometimes observed at the actual onset of previous eruption.

Table 1. Eruptive episodes at Mount Pinatubo, June 1991.

[Data from various sources. Alerts issued by the Philippine Institute of Volcanology and Seismology (PHIVOLCS). Alert Level 3: If trend of increasing unrest continues, eruption possible within 2 weeks. Alert Level 4: eruption possible within 24 h. Alert Level 5: eruption in progress]

RSAM AND SSAM

Seismic activity was recorded at Pinatubo with the recently developed Real-time Seismic Amplitude Measurement (RSAM) system (Murray and Endo, 1989; Endo and Murray, 1991). RSAM provides consecutive 1- or 10-min averages proportional to the absolute voltage output (amplitude) of a seismometer, representing ground velocity. The advantage of RSAM is that it gives a measure of seismic activity in near-real-time, even if separate seismic events overlap in time or if volcanic tremor is present (figs. 4A, 5A).

The RSAM can be adjusted for geometric spreading and instrument response to obtain "reduced displacement" (Aki and others, 1977), and RSAM can be directly proportional to such commonly measured earthquake parameters such as seismic moment rate or energy rate (Fehler, 1983). Even if the proportionality constants are not precisely known, RSAM counts can be used as with FFM without further data manipulation; that is, the "inverse rate" becomes equivalent to "inverse RSAM" (Voight and Cornelius, 1991; Cornelius and Voight, 1995). The precision of proportionality is highest if the RSAM is dominated by a single type of seismic event, because different event types may have different constants.

In addition to RSAM, scientists used the relatively new Seismic Spectral Amplitude Measurement system (SSAM) (Rogers, 1989; Power and others, this volume; Stephens and others, 1994). SSAM is based on the same principle as RSAM in that it monitors mean amplitudes of ground velocity, but it selectively filters the signal into 16 spectral bands. During the 1989-90 eruption of Redoubt Volcano, Alaska, SSAM was valuable for tracking swarms of LP events and associated tremor that preceded many of the tephra-producing eruptions (Stephens and others, 1994; Cornelius and Voight, 1994). At Pinatubo (figs. 4B-E, 5B-E), LP events and related tremor were observed to predominate the bands between 0.5 and 2.5 Hz, while VT events and related broad-band tremor exhibited peaks between 2.5 and 4.5 Hz, and explosion eruptive-generated peaks occurred between 0.5 and 1.5 Hz (Power and others, this volume). By focusing on particular frequency bands, SSAM is able to eliminate both signals and noise associated with other
frequencies. The result is to improve the signal-to-noise ratio for selected data sets and to enhance the suitability of the data for FFM analyses (Cornelius and Voight, 1994). SSAM was not used for foresight analyses at Pinatubo largely due to relatively "cumbersome processing" required at that time (Harlow and others, this volume; Power and others, this volume). More efficient processing routines are now available.

FORECASTING THE EXPLOSIVE ERUPTIONS OF JUNE 12-15, 1991

We present four groups of FFM analyses. The first group is the foresight analyses, in which we use the graphical RSAM data for the period June 1-10, which were received at Penn State University by fax from the USGS VCAT. The second group is the hindsight analyses of data over the same interval, but we use the digital RSAM data as recorded on a computer disk, with additional consideration of SSAM data. This period was the plinian buildup, from June 8 to 12. Third, we consider hindsight analyses during an earlier eruption phase, precursory explosions and dome emplacement, which was from June 3 to 7. All analyses of the first through third groups use basically the same analytical technique, which is inverse-rate plots with FFM data extrapolation in which a linear inverse-rate trend is assumed. A fourth group of analyses, representing in hindsight the cataclysmic buildup phase of June 12 to 15, considers nonlinear extrapolation techniques.

Foresight Analyses: RSAM Data Base to June 10

Graphical RSAM output from two Pinatubo seismic stations with data from June 1 to 1600 on June 10 (fig. 6) was received by us at Penn State University by fax at about noon June 11, Eastern Daylight Time. The graphs were digitized and plotted as inverse RSAM the same day.

USGS VCAT also provided an earthquake summary for June 1 to 10, including an epicenter map and an epicenter depth-time plot, and data from June 5 through June 10 for the upper-flank UBO tiltmeter. VCAT noted two clusters of epicenters, one under the vent and the other under an intersection of youthful faults, and a trend toward fewer higher frequency earthquakes and more LP's.

During the previous year at Redoubt Volcano, Alaska, a major increase in shallow LP seismicity preceded the explosive January 2 eruptions. The LP character of the seismic swarms was interpreted to reflect pressurization of fluid-filled cracks (Chouet and others, 1994), and associated seismic rate changes were of sufficient consistency and duration to enable effective FFM analyses (in hindsight) by use of either RSAM or SSAM data (Voight and Cornelius, 1991; Cornelius and Voight, 1994).

The inverse-RSAM plots for two Pinatubo stations are shown in figure 7, referenced to Philippine local time. The seismic energy correlated with the gas/ash explosion of June 7 occurs as a spike on both plots. We knew about the cause of this spike and we knew about Alert Levels 3 and 4, issued for Pinatubo on June 5 and 7, respectively. About a day after this event, the data display a downward trend for station PIEZ (solid dots, fig. 7), which suggested the possibility of a forthcoming eruptive event. A linear least squares fit indicated a possible failure about June 12. We recorded the analogy between these results and our results from Redoubt, which had been published about 6 weeks before (fig. 4 of Voight and Cornelius, 1991). Although exhibiting more noise than station PIEZ, the inverse-RSAM downward trend at UBOZ (fig. 7B) crudely corroborated the analysis for station PIEZ.

Figure 4. Cumulative RSAM and SSAM data for station PIEZ (10-min-interval data) from June 10 until the station was destroyed: A, RSAM; B, SSAM band 4 (3.5-4.5 Hz); C, SSAM band 3 (2.5-3.5 Hz); D, SSAM band 2 (1.5-2.5 Hz); E, SSAM band 1 (0.5-1.5 Hz). RSAM data represent seismic-energy release rates integrated over a broad frequency spectrum; low-frequency SSAM (0.5-2.5 Hz) is characteristic of long-period events and tremor buildup; higher-frequency SSAM (2.5-4.5 Hz) is characteristic of swarms of volcano-tectonic earthquakes or quasicontinuous volcano-tectonic earthquake activity. Explosive pyroclastic eruptions began on June 12 and culminated in the cataclysmic eruption on June 15; sequence is marked in A by arrows with lengths indicating plume altitude: short, 10-19 km; medium, 20-24 km; and long, 25-40 km (table 1).

Figure 5. Cumulative RSAM and SSAM for station CABZ (10-min-interval data) from June 12 until station failure: A, RSAM; B, SSAM band 4 (3.5-4.5 Hz); C, SSAM band 3 (2.5-3.5 Hz); D, SSAM band 2 (1.5-2.5 Hz); E, SSAM band 1 (0.5-1.5 Hz). Explosive, pyroclastic eruptions began on June 12 and culminated in the cataclysmic eruption on June 15; sequence is marked in A by arrows with lengths indicating plume altitude: short, 10-19 km; medium, 20-24 km; and long, 25-40 km (table 1). Eruptions with plume altitudes less than 10 km, or with unconfirmed plume heights, are not shown.

Figure 6. Cumulative RSAM energy release from two stations at Mount Pinatubo, UBOZ and PIEZ. Data shown through 1600 on June 10. Graphs were received at Penn State via fax from the U.S. Geological Survey Volcano Crisis Assistance Team.

Figure 7. Inverse RSAM derived manually from graphs shown in figure 6, in foresight to the June 12 eruption. Data end at 1600 on June 10. RSAM data represent seismic-energy release rates, and inverse RSAM is expected to follow a downward trend prior to an eruption. Linear fit in A (PIEZ) to data between June 9 and 10 (solid dots) extrapolates the trend toward June 12 as the expected eruption time. In B (UBOZ), the trend is less distinct, but the overall trend between June 2 and 10 corroborates the analysis in A. The inverse-RSAM peak (low-point) on June 7 correlates with the high seismic energy release during a gas/ash explosion.

The results were sent to USGS VCAT when our fax facility opened the next morning, but by then the eruption in fact had already begun. Therefore, our analysis, though made in foresight, did not contribute to hazard management at Pinatubo.

Hindsight Analyses: RSAM and SSAM Data Base to June 10

Next we recreate the foresight analysis of June 11 with the original digital data as stored on a computer disk. The "raw data," representing 10-min averages of the absolute voltage output of the seismometer, are averaged over 3-h intervals (fig. 8). Spikes in the RSAM output can be correlated to the phreatic explosions of June 3, 4, and 7 (compare table 1). The short data window from 2310 on June 8 to 1600 on June 10 (solid dots) allows a linear fit at both stations (figs. 8A,B). The time (t_s) when extrapolated rates become infinite--which represents a probable upper bound to the forecast eruption time--is 1830 on June 12 for station PIEZ and 2100 on June 12 for station UBOZ (fig. 8). The eruption windows (given by data extrapolation at the 97.5% confidence level) are large in both cases, but a lower bound is established for the morning of June 11--about 16 to 19 h after the last datum in the analysis. (As indicated previously, this confidence level reflects extrapolation statistics, not forecast probabilities.) For comparison, figures 8C and D show conventionally plotted RSAM data, together with fits based on the extension of the inverse-RSAM data for figures 8A and B.

Additional insight into physical processes during this time period is gained by comparing RSAM and SSAM records over a 1-day period from June 9 to June 10 (fig. 9). The data comprise part of the precursory trend investigated in figures 6 and 8. The sequence is characterized by repetitive swarms of broad-band VT earthquakes over intervals of several hours; during these cycles, energy also occurred in the low-frequency bands. A strong 3-h signal starting at about 1100 on June 9 (June 9.46) is predominant in low-frequency energy and was caused by LP seismicity including volcanic tremor (Power and others, this volume). A similar but less pronounced period of tremor occurred early on June 10 (fig. 9). Thus, the narrow frequency bands of SSAM allow the interpreter to distinguish between the buildup of energy release from different source mechanisms. These bands can be treated individually with FFM.

Figure 10 shows inverse SSAM calculated as 3-h averages from original 1-min PIEZ data files from June 1 to 1600 on June 10. The graphs can be compared to similar inverse-RSAM analyses for the same station (fig. 8A). The data involving higher frequency in band 4 (fig. 10A) reflect energy from VT activity that dominated the seismicity during this period. The inverse trend appears better defined for these data than for those of figure 10B, which includes energy from broad-banded VT events and for tremor. Comparison with the RSAM analysis of figure 8A suggests that if there is any advantage to evaluating the separate frequency band from 3.5 to 4.5 Hz, it is slight. In this particular case, RSAM was able to provide a realistic real-time "summary" of seismic development. This outcome was fortunate at Pinatubo, but such an outcome is not assured. At Redoubt Volcano after January 2, 1990 (Cornelius and Voight, 1994), where signal-to-noise ratio was low, the FFM results were sensitive to the source process evaluated. During this period at Redoubt, SSAM data in low-frequency bands were useful, but RSAM data were generally misleading.

Figure 8. Inverse RSAM calculated as 3-h averages from original 10-min-interval digital data. Data end at 1600 on June 10, and the graphs are equivalent to the relatively crude digitized plots constructed in foresight and shown in figure 7. Linear fits to inverse-RSAM trends between 2310 on June 8 and 1600 on June 10 (solid dots) for A (PIEZ) and B (UBOZ) are shown as solid lines with error envelopes at the 97.5% confidence level (dashed lines). Explosive eruptions beginning on June 12 and culminating in the cataclysmic eruption on June 15 are marked by downward-pointing arrows. Short upward-pointing arrows mark ash/gas explosions. C and D show conventional, non-inverted RSAM together with the solid-line curve fits and dashed envelopes developed in A and B. Continuation of data is shown as dotted lines.

Figure 9. Comparison of RSAM record (solid line) with SSAM bands 1 (open circles) and 4 (x's), both from station PIEZ, within the plinian buildup phase of seismicity as defined by Power and others (this volume), from 0224 on June 9 to 0224 on June 10. Note difference in RSAM and SSAM scales. The sequence is characterized by cyclic swarms of volcano-tectonic (VT) earthquakes at intervals of several hours. A strong, 3-h signal starting about 1100 on June 9 has predominantly lower frequencies and was caused by volcanic tremor (Power and others, this volume).

Figure 10. Inverse SSAM calculated as 3-h averages from original 1-min-interval data files: A, Band 4 (3.5-4.5 Hz); B, Band 1 (0.5-1.5 Hz). Data end at 1600 on June 10. Compare RSAM analysis, figure 8A. Linear fits are solid lines, and error envelopes at the 97.5% confidence level are dashed lines. Data of higher frequency, band 4 are dominated by VT energy. Tremor at midday on June 9 is marked as T (compare fig. 9). Explosive eruptions beginning on June 12 and culminating in the cataclysmic eruption on June 15 are marked by downward-pointing arrows. Upward-pointing arrows mark ash/gas explosions.

Figure 11. Inverse RSAM calculated as 3-h averages from original 10-min-interval data files between May 2 and June 7. Linear fits to inverse-RSAM trends between 0010 on May 31 and 0900 on June 7 (solid dots) for (A) PIEZ and (B) UBOZ are shown as solid lines and with error envelopes at the 97.5% confidence level (dashed lines). Inverse peak marked H was caused by helicopter noise (compare Power and others, this volume, fig. 3). Explosive eruptions beginning on June 12 and culminating in the cataclysmic eruption on June 15 are marked by arrows. C and D show conventional, non-inverted RSAM together with the solid-line curve fits and dashed envelopes developed in A and B. Continuation of data shown as dotted lines.

Hindsight Analyses: RSAM Data Base to June 7

The previous analyses represent the situation during the plinian buildup phase, when the possibility of an eruption was already clearly recognized (Harlow and others, this volume). The following analyses explore how FFM analyses might have supported forecasts at an earlier time.

Following the metastable phase of May, during which no trend in seismicity developed, a clear inverse trend could have been detected in early June, during the precursory explosions and dome emplacement phase, as shown in figures 11A and B for a data window chosen between 0010 on May 31 and 0900 on June 7 (solid dots). The trend is apparently suitable for a linear least squares fit. The resulting data envelope is better constrained (narrower) than in the analyses previously illustrated for June 10, chiefly because of the longer data window. Infinite rates are forecast at 0745 on June 16 for PIEZ and at 0030 on June 14 for UBOZ. The eruption window combining results from the two stations falls between 0845 on June 12 and 1945 on June 20. For comparison, figures 11C and D show conventionally-plotted RSAM data, together with conventional curve fits based on parameters derived from inverse-RSAM analyses. The gas/ash emission and dome extrusion of June 7 occurred just after the close of this data window; the graphs and fits of figure 11 suggest that it would not have been anticipated by FFM for RSAM-data averaged over 3-h intervals.

However, more detailed representation with 10-min data on June 7 (fig. 12) suggests that recognizable precursory seismicity occurred prior to dome extrusion and gas emission. The inverse trend suggests event occurrence about 1648 on June 7 (June 7.7), near the time of the observed explosive event, and possibly reflecting the time of dome emergence. Due to weather conditions, the dome was not actually observed until the following day. The seismicity during this period was dominated by bursts of VT swarm activity, as reflected by a fairly even distribution of energy throughout several frequency bands, with a slight maximum between 2.5 and 3.5 Hz. In such cases, with noise relatively low, RSAM provides an accurate overall measure of seismicity.

Cumulative RSAM based on 10-min interval data, from May 2 to mid-June, is shown in figure 13. The dashed curve is based on the inverse-RSAM linear fit for the data base to 1600 on June 10 (fig. 8); the dotted curve is based on the data base to 0900 on June 7 (fig. 11). The dotted and dashed vertical lines bracket the data bases used for June 7 and June 10 analyses, respectively. Observed values may be compared with extrapolated values beyond the data base limits.

These figures illustrate the advantage of the inverse-rate technique of FFM over conventional plots in amplifying changes in rates for data at subdued rate levels and in facilitating the extrapolation of accelerating trends, because the inverse trends are commonly nearly linear. Graphical curve fitting to conventionally displayed data is usually more ambiguous than curve fitting to inverse-rate data.

Figure 12. Comparison of station PIEZ RSAM record (solid line) of June 7 with SSAM bands 1, 2, 3, and 4 (symbols). A, conventional RSAM and SSAM; B, inverse RSAM and inverse SSAM. Note difference in RSAM and SSAM scales. A swarm of volcano-tectonic (VT) earthquakes occurs during the dome building phase (compare Power and others, this volume, fig. 2). Separate events overlap on helicorder records from 1600 onward, and the seismicity was classified as continuous VT (Power and others, this volume).

Figure 13. Cumulative RSAM (10-min-interval data) from May 2 to mid-June for (A) PIEZ and (B) UBOZ. Sequence of explosive eruptions beginning on June 12 and culminating in the cataclysmic eruption on June 15 is marked by arrows. Dashed curve is based on linear fits to inverse-RSAM data at 1600 on June 10 (fig. 8), and dotted curve is derived from linear fits at 0900 on June 7 (fig. 11). Dotted and dashed vertical lines mark the data window of the June 7 and 10 analyses, respectively.

Hindsight Analyses: RSAM and SSAM Data Base beyond June 10

Next we consider analyses for the later stages of the mid-June eruption. The major eruption had begun with a vent-clearing phase on June 12, which produced plinian columns exceeding 20 km in altitude. A 28-h break then occurred in explosive events, until 1309 on June 14. During this period the continuously updated RSAM and SSAM data might have been evaluated by FFM prior to the cataclysmic eruption phase of June 15, and we consider what might have been accomplished by this evaluation.

Figures 14A and 15A show 3-h averages of inverse RSAM from two stations from June 10 to 15 (compare figs. 4, 5). Data from distal station CABZ delineate an approximately continuous inverse trend beyond June 11. Data at station PIEZ exhibit a step function, with a prominent energy rate increase beyond the onset of the June 12 explosion. SSAM data for four bands are shown with 3-h-averaged data during this same period (figs. 14B-E, 15B-E). A significant precursory inverse-rate trend developed at PIEZ over the half-day preceding resumption of explosive activity on June 14 (fig. 14); this trend is evident both on RSAM and all bands of SSAM and indicates the continued escalation of both VT swarms and LP seismicity that occurred during this period. SSAM spectrograms show enhanced peaks at 2.5 to 4.5 Hz, correlated with VT swarms and continuous VT seismicity, and also at 0.5 to 1.5 Hz, correlated with LP swarms and tremor (Power and others, this volume). Some of the tremor (noise) during this period (at 2-3 Hz) may be correlated with gas and ash emissions (low-level eruptive activity) rather than shallow hydrothermal events (Harlow and others, this volume). Relative band strength is also influenced by station location as well as by source characteristics; thus band 4 at CABZ is relatively weaker in comparison to other bands than band 4 at PIEZ (figs. 4, 5), and this difference probably reflects the selective attenuation of high-frequency signals.

The excellent match of patterns for SSAM bands 1 to 4 suggests that VT swarm seismicity may have dominated the patterns during this period, at least on the scale of 3-h-averaged data of figures 14 and 15. For example, there are no peaks or troughs in low-frequency band 2 not matched by those in higher frequency bands 3 and 4, an observation that suggests broad band activity. Yet from the perspective of helicorder records, a dominating aspect of this period was the buildup of LP activity from minor swarms to a series of increasingly large events, with the largest of these approximately equivalent to magnitude 4 earthquakes (Harlow and others, this volume). It is possible that a significant proportion of the seismicity of this period may have been hybrid, reflecting source processes involving both VT and LP endmember types; this hybrid nature could help to explain the combination of broad band activity and multiple spectral peaks indicated by SSAM.

Analyses for RSAM from station CABZ were made in hindsight at intervals of 7 h for data windows beginning on June 11 and ending at 1930 on June 13 and at 0230 and 0930 on June 14 (fig. 16). Using the end-weighted FFM procedure as previously described, we calculated rates from integrated RSAM time series by linear interpolation over approximately constant RSAM increments. These constant RSAM increments were arbitrarily chosen to be 10% of the total encountered RSAM value, so each resulting rate series has nine data points, which occur with an increasing frequency as RSAM climbs toward the end of the data window. Spikes at times of the three eruptions within the data set (arrows, fig. 16) were removed prior to rate calculation so as not to contaminate precursory signal with syneruptive activity.

The first fit (fig. 16B) yields =1.70 (A=0.0486) and gives 0400 on June 15 for the time of singularity with an eruption window from 0930 on June 14 to 2115 on June 16. The point t_f may also be estimated with a predetermined intercept that represents the expected finite failure rate. This failure value is unknown in this case, but to test the sensitivity of the solution, we arbitrarily have taken a rate one order of magnitude larger than the last data point in the data set. With this assumption, the time of failure is indicated as 1915 on June 14, and the eruption window is from 0315 on June 14 to 0700 on June 16. These extrapolations are consistent and in good agreement with the onset of the next explosive pulse at 1309 on June 14 and the cataclysmic phase at 1027 on June 15 (table 1).

Some short-term deceleration followed the accelerating jog correlated with the first eruption on June 12 (fig. 15). Also, the last two rate points of figure 16B suggest some deceleration toward the end of the data window, with respect to the curve fit. This deceleration trend continued through 0230 on June 14, the time of the second analysis (fig. 16C). The fit yields a trend with =0.20 (A=7.988). This is a low value, because values recorded elsewhere typically fall between 1.5 and 2.2 (Cornelius, 1992; Voight, 1988). Interpretive concerns might arise with low in a foresight situation, and an accurate failure rate would be required for event forecasting. Periods of deceleration near the end of a data window can cause such a shift toward extreme values. If an accelerating jog occurred soon thereafter, the overall trend might continue as before; but the interpreter needs to be vigilant about an alternative possibility, namely, that a deceleration in seismicity commonly precedes an eruption.

However, such an acceleration was observed by 0930 the same day (fig. 16D); the overall trend at this time was =1.63 (A=0.0598), and forecast times returned to values similar to the first analysis: singularity at 1330 on June 15 with a window from 2315 on June 14 to 1600 on June 16. Assuming a finite failure rate as before yields 0800 on June 15, with the eruption window from 1845 on June 14 to 0815 on June 16, consistent with final phases of the eruption.

For perspective, we reemphasize that we view the FFM approach as one tool among several, a convenient method to track systematic, progressively accelerating energy release or deformation. Statements based on application of the method have to be interpreted within the context of volcano observations and other acquired information. At the time indicated for the above analysis, the morning of June 14, three plinian eruptions had already occurred (between 0851 on June 12 and 0840 on June 13) within the past 48 h. The continuing overall downward inverse trend within a quiet interval suggested that a very high rate (perhaps an eruption) was to be expected on June 15, and this turned out to be a correct inference. However, this did not exclude the possibility of eruptive activity at an earlier time, especially as the interpretation should consider the accelerating jog, which can be seen at the end of the last data window (fig. 16D, last three data points). In fact, the next eruptive pulse at 1309 on June 14 was heralded by a renewed spurt of seismic acceleration.

Details of trends on the morning of June 14 are shown in figure 17; error envelopes are shown at the 97.5% confidence level. The data are from an accelerating jog on the morning of June 14 (until 0830, dashed vertical line). The nonlinear analyses give somewhat varied results, implying imminent failure with RSAM or band 4 SSAM but failure later that day with band 1 SSAM. Overall, the trend extrapolations provided a fair indication for the renewed eruptive activity during the early afternoon (vertical arrow). A nonlinear curve fit to data from band 3 results in < 0 but is not accepted, because a roughly linear trend seems visually supportable for this time series; a linear fit is shown.

Figure 14. Inverse RSAM and inverse SSAM at station PIEZ calculated as 3-h averages from June 10 until the station was destroyed: A, RSAM; B, SSAM band 4 (3.5-4.5 Hz); C, SSAM band 3 (2.5-3.5 Hz); D, SSAM band 2 (1.5-2.5 Hz); E, SSAM band 1 (0.5-1.5 Hz). Open circles are inverse RSAM and inverse SSAM multiplied by a factor of 10. Many of the data beginning midday on June 14 were electronically clipped (dotted lines) and should not be used with FFM because of unknown distortions. Explosive eruptions beginning on June 12 are marked in A by arrows whose lengths indicate plume altitude: short, 10-19 km; medium, 20-24 km; and long, 25-40 km (table 1). Conventionally plotted cumulative data are shown in figure 4, at the same time scale.

Figure 15. Inverse RSAM and inverse SSAM at station CABZ calculated as 3-h averages from June 10 until station failure. A, RSAM; B, SSAM band 4 (3.5-4.5 Hz); C, SSAM band 3 (2.5-3.5 Hz); D, SSAM band 2 (1.5-2.5 Hz); E, SSAM band 1 (0.5-1.5 Hz). Many of the data beginning midday on June 15 were electronically clipped (dotted lines at tail end of the time series) and should not be used with FFM because of unknown distortions. Inverse-SSAM trends of all bands (and to a lesser extent also inverse RSAM) define a decelerating jog in the aftermath of the June 12 eruption and an accelerating jog prior to renewed eruptions on June 14. Explosive eruptions beginning on June 12 are marked in A by arrows whose lengths indicate plume altitude: short, 10-19 km; medium, 20-25 km; and long, 25-40 km (table 1). Conventionally plotted cumulative data are shown in figure 5, at the same time scale.

Figure 16. Nonlinear FFM analyses of RSAM data from station CABZ. 10-min-interval data of RSAM are shown in A. Spikes on June 12 and 13 (vertical arrows in A) reflect pyroclastic eruptions, and RSAM data during the eruptions are omitted to emphasize precursory changes. Data bases start at 2000 on June 10 and end at (B) 1930 on June 13, (C) 0230 on June 14, and (D) 0930 on June 14. Rates were calculated over constant RSAM increments (10% of total encountered RSAM value). B-D show best-fit curves (solid and dashed lines); B and D show error envelopes at the 97.5% confidence interval (dotted lines). a is empirical curve-fitting constant, as in equation 1 in text. Sequence of eruptions following the analyses is marked with arrows in D; lengths of arrows indicate altitude reached by eruption cloud, as in figure 4.

Figure 17. Nonlinear analyses of (A) RSAM and (B-E) SSAM data on June 14 from station CABZ with FFM. SSAM data shown for (B) band 4, (C) band 3, (D) band 2, and (E) band 1. Data window is from 0000 to 0830 (vertical dashed line), a period of accelerating seismicity. Inverse rates (solid dots) were calculated over constant RSAM or SSAM increments (10% of total encountered values). Inverse RSAM and inverse SSAM averaged over 10 min shown as dotted lines for comparison. Best-fit curves are solid lines; error envelopes are shown at the 97.5% confidence interval (dashed lines). is empirical curve-fitting constant, as in equation 1 in text. Arrow in A marks eruption at 1309 on June 14.

DISCUSSION

Pinatubo provided the first opportunity to compare RSAM and SSAM records in detail for a variety of event types. Previously, only data for the 1989-90 eruption of Redoubt Volcano, Alaska, had been thoroughly analyzed, and in that eruption precursory LP seismicity was dominant (Stephens and others, 1994). For the January 2 eruption at Redoubt, SSAM and RSAM provided comparable data; after January 2, when signal strength diminished relative to noise, RSAM data were typically misleading, although SSAM in low-frequency bands could detect eruption precursors (Cornelius and Voight, 1994; Stephens and others, 1994).

In contrast, at Pinatubo, RSAM was relied on throughout as a monitoring tool, and, unlike at Redoubt, the signal- to-noise ratio was large throughout the monitoring period. Up to the time of dome emplacement, VT seismicity was dominant, and the individual SSAM bands virtually replicated the RSAM pattern (fig. 13). From June 8 to June 12, seismicity was characterized by cyclic swarms of VT seismicity, with a gradual increase in tremor. Most SSAM bands simulated the RSAM pattern, with some differences in low-frequency bands that reflect LP energy in addition to VT broad-band components (figs. 9, 10). From June 12 to 15 the seismicity increased in complexity, involving combinations of VT seismicity, LP swarms, and tremor. That many swarms may have been hybrid would help to explain the similar patterns displayed by RSAM and SSAM bands 1-4, despite two distinctive spectral peaks correlative with VT and LP seismicity. However, resolution of the question of hybrid seismicity will require detailed analyses of digital data and helicorder records.

On the whole, due to the broad-band nature of dominant seismicity, RSAM patterns were broadly simulated by each of the SSAM bands analyzed. Thus, unlike at Redoubt in 1990, RSAM at Pinatubo was able to provide an authentic measure of overall seismic energy release. This was fortunate, as the decisions based on seismic monitoring at Pinatubo, particularly after June 12, indeed relied on RSAM for this purpose (Harlow and others, this volume). SSAM played an important supporting role in pointing to source processes (LP and hybrid seismicity) suggestive of high eruptive potential.

In this paper we have recreated our foresight analyses on the basis of graphical RSAM data, and we presented several more FFM analyses to simulate what might have been done in foresight with digital data. By June 7 an eruption might have been anticipated around June 14-16, with an eruption window from June 12 to 20. At the time (June 5-7), the possibility of a major eruption was recognized by monitoring scientists from a multitude of pertinent volcanological observations; analyses using FFM could have been integrated with these data. Updated FFM analyses, such as those presented for June 10-14, provided information consistent with actual event occurrence and could have been integrated with the cumulative information base utilized for onsite eruption forecasts.

However, we recognize that the hazard management outcome of the Pinatubo eruption could not have been much improved by FFM, or, for that matter, by any additional technological device or methodology. The precursory phenomena were strong, and despite inevitable interpretive uncertainties, scientific advisories were effective in influencing prudent decisionmaking by public officials, who, to their credit, were willing to act decisively on the uncertain information available to them. In hindsight, hazard managers made virtually all the right moves.

As with all tools, experience is needed for a skillful application of the FFM approach to eruption prediction. The potential advantages of viewing accelerating processes with inverse-rate curves are significant. Small changes at low rates are amplified at the onset of any new development, and complex curves of conventionally plotted graphs may be nearly linearized by data inversion. However, irregularities involving precursory phenomena, such as decelerating and accelerating jogs (reflecting cyclic swarms), may complicate the interpretation. Perhaps not infrequently, profound departures from systematic progressive rate growth may render the FFM procedure inapplicable for certain cases. Under favorable circumstances a skilled observer may be able to use FFM to anticipate future developments and will make use of constantly renewed updates. Reliance of FFM on a feedback system between trend recognition and forecasting makes the procedure a potentially useful tracking method for physical processes.

Certain "conventions" of data treatment may help to define experience. Three such conventions have been demonstrated with the aforementioned analyses. First, eruption windows are based on data envelopes constructed at a specified confidence level to the linearized inverse-rate fit. The windows thus reflect data scatter, length of data window, and the time interval between the time of analysis and time of calculated failure. Second, in some sensitivity analyses used here, failure rates are arbitrarily assumed to be an order of magnitude larger than the last previously calculated rate. This alternative treatment is generally conservative, and its main effect is to counterbalance too-late forecasts founded on the simplifying assumption of infinite failure rates, especially for concave-upward inverse-rate curves (<2.0). The sensitivity of the approach can be judged by comparing values obtained using t_f and t_s. Third, rates may be calculated over constant data increments instead of constant time increments. Data increments of about 10% of the total encountered value produce a smoothed time series of rates that is favorably biased toward the latest developments (end-weighted).

Calculated eruption windows should be qualified by statements describing the observed situation. Ideally, eruption windows should combine analyses from several stations, and, where feasible, several data types. Forecast sensitivity should be tested in relation to choice of data-averaging intervals. In systems for which systematic behavior is dominant, FFM may provide the opportunity to quantify observed accelerating processes. However, it would be unwise to interpret inverse-rate curves without simultaneous and full consideration of other volcanologic information.

ACKNOWLEDGMENTS

We thank colleagues and U.S. Geological Survey staff for their assistance. We especially thank Dave Harlow, Dick Janda, Tom Murray, John Power, and Ed Wolfe for access to and assistance with data. Advance drafts of Pinatubo manuscripts were kindly provided by Dave Harlow, Chris Newhall, John Power, and Randy White. This work was conducted under a cooperative project involving Penn State University and the Cascades Volcano Observatory (U.S. Geological Survey), with support by the National Science Foundation. Colleague review was provided by William Chadwick, Chris Newhall, and Don Swanson. Voight's involvement with Pinatubo was kindled during April 1991 in deliberations with Dick Janda on the cobblestoned streets of rococo Popayán, Colombia, and at a paramo tent camp on the shoulder of Cotopaxi, Ecuador, as an early VCAT overseas mobilization was being choreographed by Janda. With voluminous pumiceous pyroclastic flows hundreds of years old at an otherwise little-known and long-dormant volcanic complex, our concern for this densely populated region was that an event resembling but exceeding El Chichón might be in developmental stages. Janda would play a vital role in the Pinatubo crisis response. He is gone now.

The sea and the earth are unfaithful to their children: a truth, a faith, and a generation of men goes--and is forgotten, and it does not matter! Except, perhaps, to the few of those who believed the truth, confessed the faith--or loved the men.

References Cited

Aki, K., Fehler, M., and Das, S., 1977, Source mechanism of volcanic tremor: Fluid-driven crack models and their application to the 1963 Kilauea eruption: Journal of Volcanology and Geothermal Research, v. 2, no. 3, p. 259-287.

Chouet, B.A., Page, R.A., Stephens, C.D., Lahr, J.C., and Power, J.A., 1994, Precursory swarms of long-period events at Redoubt Volcano (1989-1990), Alaska: Their origin and use as a forecasting tool, in Miller, T.P., and Chouet, B.A., eds., The 1989-1990 eruptions of Redoubt Volcano, Alaska: Journal of Volcanology and Geothermal Research, v. 62, p. 95-135.

Cornelius, R.R., 1992, Feasibility study for the materials science approach to volcano eruption prediction: University Park, Pa., The Pennsylvania State University, Ph.D. thesis, 470 p.

Cornelius, R.R., and Voight, B., 1989, Determination of eruption-prediction constants for accelerated creep or seismicity [abs.]: Geological Society of America Abstracts with Programs, v. 21, p. 13.

------1994, Seismological aspects of the 1989-1990 eruption at Redoubt Volcano, Alaska: The Materials Failure Forecast Method (FFM) with RSAM and SSAM seismic data, in Miller, T.P., and Chouet, B.A., eds., The 1989-1990 eruptions of Redoubt Volcano, Alaska: Journal of Volcanology and Geothermal Research, v. 62, nos. 1-4, p. 469-498.

------1995, Graphical and PC-software analysis of volcano eruption precursors according to the Materials Failure Forecast Method (FFM): Journal of Volcanology and Geothermal Research, v. 64, p. 295-320.

Daag, A.S., Tubianosa, B.S., and Newhall, C.G., 1991, Monitoring sulfur dioxide emissions at Pinatubo Volcano, central Luzon, Philippines [abs.]: Eos, Transactions, American Geophysical Union, v. 72, no. 44, Fall Meeting Supplement, p. 61.

Daag, A.S., Tubianosa, B.S., Newhall, C.G., Tuñgol, N.M, Javier, D., Dolan, M.T., Delos Reyes, P.J., Arboleda, R.A., Martinez, M.L., and Regalado, M.T.M., this volume, Monitoring sulfur dioxide emission at Mount Pinatubo.

Endo, E.T., and Murray, T.L., 1991, Real-time Seismic Amplitude Measurement (RSAM): A volcano monitoring and prediction tool: Bulletin of Volcanology, v. 53, p. 533-545.

Ewert, J.W., Lockhart, A.B., Marcial, S., and Ambubuyog, G., this volume, Ground deformation prior to the 1991 eruptions of Mount Pinatubo.

Fehler, M., 1983, Observations of volcanic tremor at Mount St. Helens volcano: Journal of Geophysical Research, v. 88, p. 3476-3484.

Harlow, D.H., Power, J.A., Laguerta, E.P., Ambubuyog, G., White, R.A., and Hoblitt, R.P., this volume, Precursory seismicity and forecasting of the June 15, 1991, eruption of Mount Pinatubo.

Harlow, D.H., Punongbayan, R.S., Ambubuyog, G., Laguerta, E., Power, J.A., Newhall, C.G., Hoblitt, R.P., Lockhart, A.B., Murray, T.L., Wolfe, E.W., and Ewert, J.E., 1991, Seismic activity and forecasting of the climactic eruption of Pinatubo Volcano, Luzon, Philippines on June 15, 1991 [abs.]: Eos, Transactions, American Geophysical Union, v. 72, no. 44, p. 61.

Hoblitt, R.P., Wolfe, E.W., Scott, W.E., Couchman, M.R., Pallister, J.S., and Javier, D., this volume, The preclimactic eruptions of Mount Pinatubo, June 1991.

Lockhart, A.B., Marcial, S., Ambubuyog, G., Laguerta, E.P., and Power, J.A., this volume, Installation, operation, and technical specifications of the first Mount Pinatubo telemetered seismic network.

Murray, T.L., and Endo, E.T., 1989, A Real-time Seismic Amplitude Measurement system (RSAM): U.S. Geological Survey Open-File Report 89-684, 21 p.

Murray, T.L., Power, J.A., March, G.D., and Marso, J.N., this volume, A PC-based realtime volcano-monitoring data-acquisition and analysis system.

Newhall, C.G., Daag, A.S., Delfin, F.G., Jr., Hoblitt, R.P., McGeehin, J., Pallister, J.S., Regalado, M.T.M., Rubin, M., Tamayo, R.A., Jr., Tubianosa, B., and Umbal, J.V., this volume, Eruptive history of Mount Pinatubo.

Pinatubo Volcano Observatory Team, 1991, Lessons from a major eruption: Mt. Pinatubo, Philippines: Eos, Transactions, American Geophysical Union, v. 72, p. 545-555.

Power, J.A., Murray, T.L., Marso, J.N., and Laguerta, E.P., this volume, Preliminary observations of seismicity at Mount Pinatubo by use of the Seismic Spectral Amplitude Measurement (SSAM) system, May 13-June 18, 1991.

Punongbayan, R.S., Ambubuyog, G., Bautista, L., Daag, A.S., Delos Reyes, P.J., Laguerta, E., Lanuza, L., Marcial, S., Tubianosa, B., Newhall, C.G., Power, J., Harlow, D.H., Ewert, J.W., Gerlach, T., Hoblitt, R.P., Lockhart, A.B., Murray, T.L., and Wolfe, E.W., 1991, Eruption precursors or a false alarm? A process of elimination [abs.]: Eos Transactions, American Geophysical Union, v. 72, no. 44, p. 61.

Punongbayan, R.S., Newhall, C.G., Bautista, M.L.P., Garcia, D., Harlow, D.H., Hoblitt, R.P., Sabit, J.P., and Solidum, R.U., this volume, Eruption hazard assessments and warnings.

Rogers, J.A., 1989, Frequency-domain detection of seismic signals using a DSP coprocessor board, in Lee, W.H.K., ed., Toolbox for seismic data acquisition, processing, and analysis: International Association of Seismology and Physics of the Earth's Interior in collaboration with Seismological Society of America, El Cerrito, Calif., p. 151-172.

Stephens, C.D., Chouet, B.A., Page, R.A., Lahr, J.C., and Power, J.A., 1994, Seismological aspects of the 1989-1990 eruptions at Redoubt Volcano, Alaska: The SSAM perspective, in Miller, T.P., and Chouet, B.A., eds., The 1989-1990 eruptions of Redoubt Volcano, Alaska: Journal of Volcanology and Geothermal Research, v. 62, p. 153-182.

Voight, B., 1988, A method for prediction of volcanic eruptions: Nature, v. 332, p. 125-130.

------1989, A relation to describe rate-dependent material failure: Science, v. 243, p. 200-203.

Voight, B., and Cornelius, R.R., 1991, Prospects for eruption prediction in near real-time: Nature, v. 350, p. 695-698.

Wolfe, E.W., 1992, The 1991 eruptions of Mount Pinatubo, Philippines: Earthquakes & Volcanoes, v. 23, no. 1, p. 5-35.

Wolfe, E.W. and Hoblitt, R.P., this volume, Overview of the eruptions.

FIRE and MUD Contents

PHIVOLCS | University of Washington Press | U.S.Geological Survey
This page is <https://pubs.usgs.gov/pinatubo/corn/>
Contact: Chris Newhall
Last updated 06.10.99