Monitoring the evolution of the Pasig–Potrero alluvial fan, Pinatubo Volcano, using a decade of remote sensing data

Abstract

Since the 1991 climactic eruption of Pinatubo in the Philippines, various hazards have affected areas surrounding the volcano. The most significant of these hazards involve the redeposition of pyroclastic flow and fall deposits as lahars, deposit-derived pyroclastic flows, and ash falls due to phreatic explosions. Many of these processes occurred in areas that are inaccessible for ground observation and monitoring. We describe here how sequential remote sensing data obtained over the period December 18, 1991, to November 1, 2001, from the SPOT, ERS, RADARSAT, SIR-C/X-SAR, AIRSAR, LANDSAT 7 ETM, and ASTER sensors provide a means of monitoring the decade-long development of the post-eruption Pinatubo landscape. This method represents an efficient and safe alternative to time-consuming, physically demanding and risky field campaigns. We apply principal component analysis, image subtraction, band ratioing, and density slicing to these data to track the changes in the post-eruption landscape, estimate volumes of deposition, and allow hazard vulnerability prediction along the timeline establish by the series of data sets. The maps derived from the remote sensing data agree well with the field derived maps for the first 5 years (1991–1995), provide important large-area coverage, and show details that are unobtainable from conventional ground-based mapping. The volume of lahars deposited during the first 6 months following the eruption is estimated between 0.045 and 0.075 km³, covering an area of ∼45 km². Moreover, changes in the settlement patterns of the local population, as well as in the construction and modification of the engineering structures for controlling the lahar hazards, can be identified in the multi-temporal scenes spanning the entire decade of observations. These types of information are crucial inputs for local decision- and policy-making in volcanic hazard mitigation.

Introduction

The June 1991 eruption of Pinatubo Volcano on the island of Luzon (Philippines) produced about 5.5 km³ (bulk volume) of pyroclastic flow deposits (ignimbrite) that partly buried deep valleys and covered low-lying topography within 12 km of the volcano (Scott et al., 1996b, Torres et al., 1996, Torres, 2001). Since that time, large debris flows and hyperconcentrated stream flows, which are here called lahars, have frequently been generated by typhoon and monsoon rains (Janda et al., 1996, Major et al., 1996). As much of the redeposition of 1991 deposits around Pinatubo has been into land-locked basins, the severe erosion of the ignimbrite fans upstream has been balanced by an equivalent magnitude of deposition downstream in the alluvial basin. These lahars have resulted in significant local hazards to life and property affecting large population centers and many lowland villages. Post-eruption lahar events heavily impacted the floodplains of the Sacobia–Bamban, Abacan, Pasig–Potrero, Marella–Sto. Tomas, Balin–Baquero–Bucao, and O'Donnell river systems (Fig. 1), with widespread encroachment and rapid build-up of the alluvial fans.

It has been necessary to understand the erosion and remobilization of the 1991 ignimbrite and the deposition of lahars for several reasons, including the prediction of how long this hazard would last and to identify the vulnerable areas. However, field-based determination of the accumulated volumes of volcaniclastic deposits would mean committing extensive manpower and resources to monitoring the flow discharge of major river channels around Pinatubo (Rodolfo et al., 1996, Tuñgol and Regalado, 1996, Martinez et al., 1996, Arboleda and Martinez, 1996) and in mapping the entire alluvial fan (Punongbayan et al., 1993) every time new deposition had taken place. It is for this reason that we explore the application of orbital remote sensing in this analysis.

Detailed field monitoring of the changes to the Pasig–Potrero drainage system by PHIVOLCS staff continued only until the end of 1995 when other pressing volcanological and seismological concerns required the diversion of people and resources to other areas in the Philippines. Here we demonstrate that multiple optical and microwave remote sensing data sets are well suited to providing observations suitable for long-term analysis of surface changes resulting from the erosion of pyroclastic deposits and cumulative deposition by lahars. We do this for the Pasig–Potrero River system but the methodology is applicable anywhere. We use several different data sets, including SPOT, ERS, SIR-C/X-SAR, RADARSAT, LANDSAT 7 ETM+ and ASTER scenes, and have reconstructed the sequential development of the post-eruption Pinatubo landscape during the time period from December 18, 1991, to November 1, 2001. These data sets were not specifically collected for the study of Pinatubo, nor does any one sensor provide complete spatial and temporal coverage during the decade of observation because of the differing methods of data acquisition that have existed over the time interval. As we will show, the satellite data represent a viable alternative to conventional ground monitoring and field mapping over difficult and dangerous landscapes, such as fresh ignimbrite sheets and lahar deposits. In addition, we provide guidelines for future studies specifically intended to detect surface changes on volcanoes using data from different satellite- or aircraft-borne sensors.

Although not the main focus of this study, we include here a brief account of the post-June 15, 1991, events at Pinatubo to put the significance of the remote sensing data in perspective. After the climactic eruption, the surrounding area within 12 km of the Pinatubo's vent region was covered by non-welded ignimbrite that in some places attained a thickness in excess of 200 m along the axis of the steep-walled pre-eruption valleys (Scott et al., 1996b). Erosion and remobilization of the valley-ponded ignimbrites and pyroclastic materials on the interfluves (Torres et al., 1996) occurred mainly as a series of short-lived, intense events that peak at the passage of tropical storms and typhoons. Perhaps significantly, the proximity of Typhoon Yunya and its associated heavy rainfall at the time of the climactic eruption (Oswalt et al., 1996) established the initial drainage channels that were enlarged and developed by later typhoon-induced surface runoff (Pierson et al., 1996). The Pinatubo ignimbrite sheet has eroded much faster (60% remobilized in the first 6 years) than comparable examples, such as the 1912 Valley of Ten Thousand Smokes (VTTS) ignimbrite (S. Self, personal observation), a difference that cannot be solely attributed to the non-welded nature of the ignimbrite at Pinatubo.

Headward erosion and gullying during torrential rain was the most significant trigger of lahar generation at Pinatubo (Pierson et al., 1996) and most of the major lahar events were initiated in this way. A series of spectacular mass movements in the easily erodable, hot, “fluffy” ignimbrite formed large scarps and led to a series of deposit-derived flows or secondary pyroclastic flows, redepositing the pyroclastic materials further downslope than the vent-derived or primary-deposited ignimbrite fan (Torres et al., 1996, Torres, 2001). Although decreasing in volume and frequency with time, the remobilization of hot ignimbrite from cliffs and steep channel walls persisted for several years after deposition of the vent-derived flows on June 15. Cumulatively, these events have delivered great amounts of material into the valleys and supplied the materials for lahars that eventually built up the alluvial fan. Some of these secondary pyroclastic flow events involved materials with volumes of several million cubic meters, and their deposition rapidly aggraded the valley floor by a thickness of several meters (e.g., 5–7 m in Sacobia during the April 4, 1992, event; Torres et al., 1996). Most secondary flow events coincided with major typhoons and rainstorms, causing the channel-confined deposit-derived pyroclastic flows to bulk up almost instantly into lahars. In instances when extremely large deposit-derived flows occurred, the entire valley floor was overwhelmed by aggradation of dry pyroclastic deposits, locally preventing the bulking and generation of lahars. Deposit-derived pyroclastic flows sometimes temporally blocked tributary channels, which created localized ponded water that subsequently generated lake-breakout lahars. Meanwhile, the removal of thick ignimbrite sections from sites at the interfluves of river valleys sometimes resulted in stream piracy. One notable example led to the capture of the upper Sacobia watershed by the Pasig–Potrero River as an aftermath of the October 5, 1993, event, which was accompanied by intense typhoon-borne rains and continuous lahar generation (GVN, 1993).

Post-depositional processes on the scale of those described here are difficult to document in the field by virtue of the size of the affected area (i.e., ∼2000 km²), remoteness and roughness of the terrain, and the level of exposure to dangerous phreatic explosion, avalanche, and lahar hazards. Moreover, frequent cloudiness hampers conventional aerial observation and photography at Pinatubo as with many tropical volcanoes. Overall, lahar events have caused greater devastation to populated areas and have been responsible for more damage to life and property than the eruption itself. Thus, it is significant that remote sensing techniques have yielded important complementary information to field observations and, as in some cases, provide the primary source of data in areas that are inaccessible from ground observations (Mouginis-Mark et al., 1993). In particular, radar data are useful for imaging the landscape irrespective of weather conditions or time of day.

Past efforts to characterize the changes on the Pinatubo landscape have been conducted on various regions of the volcano within shorter time periods. Quantification of up-slope erosion at Pinatubo has also been attempted as an indirect way to estimate the volume of lahar deposits in the Sacobia drainage system (Daag, 2003). Daag and Van Westen (1996) studied geomorphic changes in the Sacobia watershed region and examined the sediment budget in the Sacobia River system from 1991 to 1993 by using a series of aerial photographs and constructing Digital Elevation Models (DEMs). Chorowicz et al. (1997) used ERS-1 radar images to characterize lahar deposit surfaces in the depositional fan of Balin–Baquero drainage on the western flank. In their work, Chorowicz et al. (1997) examined two ERS-1 images that were obtained in the summer of 1993, including a day when lahar channels were active. Radar backscatter characteristics for both active and inactive lahars were identified, but because the radar data were only obtained at a single incidence angle and wavelength, there were a number of non-unique surface morphologies that were identified using ratio images. Recent lahar deposits were distinguished from unaffected areas by their distinctive dark tones in radar scenes (Fig. 2A) caused by water saturation, surface humidity, and characteristic roughness, which all tend to yield weak backscatter signals. However, because Chorowicz et al. (1997) studied ERS-1 data that had a wavelength of 5.6 cm, the strength of the radar return was dominated by the influence of topography (roughness) at this scale, rather than moisture. Thus, their analysis provided mainly textural information on fresh lahar deposits.

The decade-long evolution of post-eruption landscape at Pinatubo took place at a time when innovative technologies in satellite and airborne remote sensing were just being introduced and made publicly accessible. Some spacecraft only operated for part of the time period, while data from other platforms were available only from commercial systems and so had a high purchase price. Only when special research opportunities were available for the free access to these data was it possible to include these data sets in our analysis. In addition, the manner in which data could be obtained on the ground varied during the decade, so that not all observations from a particular spacecraft were recorded on the ground. A more complete discussion of the problems associated with building an archive of satellite data sets such as the one used here is provided by Mouginis-Mark and Domergue-Schmidt (2000). Information on the data sets used in our analysis, including their spectral coverage, spatial resolution, and the acquisition date for each image, is given on Table 1. Further details on the performance of the sensors, and their use in volcanic terrain, can be found in Mouginis-Mark et al. (1993), Stofan et al. (1995), Mouginis-Mark (1995), Hess et al. (1995), MacKay and Mouginis-Mark (1997), Rowland et al. (1999), Abrams (2000), and Arvidson et al. (2001).

The evolving landscapes being formed by lahar deposition are visible from space-borne monitoring because of the size of the encroached area, tonal contrast of lahar deposits with surrounding albedo, and the development of man-made structures around it. Lahar deposits rapidly built up the alluvial fans in response to the erosion of pyroclastic flow deposits, which exponentially decrease to near ambient level within a decade. The surfaces of lahar deposits are generally flat, but form a topography with terraces and braided stream landforms, the extent of which depends on the duration of erosion or deposition in the adjacent channel. Lahar terraces exhibit a smooth surface consists of moderately sorted ash-derived sand with occasional pebble-to gravel-sized pumice clasts. Areas in alluvial fan where water had locally ponded are veneered with silt and mud deposits when the water has evaporated. On the other hand, braided lahar landforms display greater surface roughness with the formation of gravelly channel bars. During daytime, an alluvial fan experiences a rapid loss of soil moisture due to the characteristic porosity and permeability of unconsolidated lahar deposits, allowing the regions of the fan away from the active channel to quickly develop a dry dusty surface. Meanwhile, older lahar deposits are increasingly covered by vegetation as wild tall grasses and woody shrub growths colonized the landscape. Fig. 2B shows the surface conditions of lahar deposits at various locations (see Fig. 2A) in the Pasig–Potrero alluvial fan.

Several alluvial fans are visibly outlined or shaped by man-made structures, which were constructed to mitigate lahar hazards and control sediment disaster. The largest of these structures was emplaced at Pasig–Potrero River and locally known as “Megadike”, which was designed as a large sediment trap. The construction of the sediment control structure at Pasig–Potrero has evolved through the years (Fig. 3). Earlier structures have briefly contained the lahar deposition and trained the fan shape, but later dike alignment, including the “Megadike” project, apparently responded to the direction of fan encroachment. Other forms of human activities, such as settlement build-up and farming, introduce landscape alteration that provides contrast with the naturally evolving surface of the alluvial fan.

Different sensors create different levels of contrast that have important applications in mapping lahar deposits. The visible and near-infrared bands in LANDSAT TM, SPOT, and ASTER are useful for delineating the barren regions of the alluvial fan and the areas covered with vegetation, which hints at the distinction between young and old lahar deposits. These data sets also provide reliable criteria for mapping the active channels and for distinguishing the wet and relatively dry deposits. ASTER and SPOT provide higher spatial resolution images at 15 and 20 m, respectively, allowing man-made modification in the landscape, such as concrete dikes, roads, farm lots, and house clusters, to be directly mapped. On the other hand, radar bands detect surface roughness contrast created by sandy vs. gravelly, planar vs. rilled, and barren vs. vegetated surfaces. Moisture-laden young lahar deposits generate significant tonal contrasts with dry lahar deposits, such that newly emplaced lahar deposits may be better observed in radar bands than in higher resolution VNIR spectral bands. Moreover, man-made structures and vegetation produce greater backscatter, appearing with bright pixel qualities.