Remote Sensing at Pisco, Peru | Center for Advanced Spatial Technologies

Home » Research » Geoinformatics (GIS/RS) » Remote Sensing » Remote Sensing at Pisco, Peru

Other Remote Sensing areas:

Loading, please wait...

The 2007 M8.0 Pisco, Peru Earthquake

The Mw8.0 Pisco, Peru Earthquake of 15 August 2007 struck the coastal and inland mountainous regions of central Peru (epicenter approximately 150 km south of Lima), causing severe damage to the cities of Pisco, Ica, Cañete, Chincha, and Tambo de Mora, killing more than 500 people (INDECI 2007). INDECI also reports that the earthquake destroyed 55,000 buildings and damaged an additional 21,000. Most of the fatalities occurred in the city of Pisco, where about 80% of the buildings either collapsed or were seriously damaged. The earthquake was a result of the subduction process between the Nazca plate and the South American continental plate, and was recorded at 16 stations within 150 km of the fault. The figure to the right illustrates the approximate area of a massive lateral displacement of the marine terrace in Canchamaná. The boundary of the mobilized area on the north is uncertain and could possibly extend further (background image from Google Earth).

This project is a joint effort between members of the University of Arkansas, Drexel University, Washington State University, and the Center for Advanced Spatial Technologies. Members include Dr. Brady Cox (University of Arkansas College of Engineering), Dr. Joseph Wartman (Drexel University College of Engineering), Dr. Adrian Rodriguez-Marek (Washington State University Department of Civil & Environmental Engineering), Dr. Jack Cothren and Dr. Jason Tullis (University of Arkansas Department of Geosciences and CAST), and Adam Barnes (CAST).

Landslides

Landslides triggered by the earthquake were observed during ground reconnaissance involving vehicular traverses along both the Lima-to-Ica segment of the Pan-American Highway (north-south oriented) and the east-west oriented Highway 024A. The Highway 024A traverse originated near the coast close to Pisco and continued east along the highway roughly 110 km inland to an elevation of approximately 4,500 m. It is estimated that in total the earthquake triggered thousands of landslides, with a significant majority of these being disrupted landslides including rock falls, rock slides, soil falls, soil avalanches, and disrupted soil slides (Keefer 1984). Disrupted slope failures occurred in both natural and altered (e.g. road cuts) terrain. The following paragraphs are descriptions of some of the prominent types of landslides witnessed in different regions, and are presented as examples of typical failures.

The shoulder and pavement from the north-bound lanes of the Pan American highway were pushed into a near vertical face by a 400-m long slope failure induced by liquefaction and lateral spreading at the toe of the slope Juhuay Slope Failure: The Jahuay slope failure (Figure 1) was triggered by liquefaction at the toe of a 30- to 50-m high, steep-faced slope. At this location, the Pan American highway runs right along the interface between the marine terrace and the Cañete Formation. The failed slope material consisted of loose, non-plastic, silty sand eolian deposits that covered the stiffer Cañete. At this location, the swampy marine terrace extends approximately 500 m to the coastline on the south-bound side of the highway. When the slope failed, the shoulder and pavement from the north-bound lane of traffic was lifted into a near-vertical face approximately 3 m high. This likely occurred when the failed slope material plowed under the highway embankment. Massive amounts of failed soil were slumped between the highway and the edge of the slope over the entire 400-m length of the failure. Sand boils were found on both sides of the highway, and a man who lived just south of the slide reported water and sand squirting out of the ground approximately 1-m high during the earthquake. Additional slope failures were documented by the GEER team further to the north along this same marine terrace-Canete Formation interface. It is still unknown whether these failures were also linked to the massive Canchamaná lateral spread.

Rockslide along Highway 024A Highway 024 Massive Rockfall at 500 m: The rockfall shown in Figure 2 occurred roughly 44 km from the fault plane at an elevation of approximately 500 m. Workers were removing the rockslide, which fully closed the highway for 3 days, when the photo was taken. Highway personnel reported that 8,000 m3 of material had already been removed by that time. The total volume of rockfall debris is estimated to be over 20,000 m3. The rockfall involved blocky rock with closely spaced and unfavorably oriented joints. Geologic maps indicate that rocks at the site are Lower Cretaceous monzonites. Most of the slide debris consisted of blocks ranging between 0.2 and 2.0 m in diameter, though several 3.5 m blocks were also observed on the roadway. After filling the roadway, debris spilled over a cliff located to the right (south) of the road, ultimately blocking a water canal. During the visit, crews were blasting debris, including a number of massive rock blocks several meters in diameter, in order to reinstate flow in the canal. It is worth noting that away from the oversteepened rockcuts, the rocks weather to gentle 35 to 50 degree slopes, which showed little if any evidence of instability during the earthquake.

Disrupted soil/rock slide along Highway 024A. Disrupted soil/rock slide at 2020 m: The landslide shown in Figure 3 occurred at an elevation of 2020 m along Highway 024A. The closest distance to the fault plane is estimated at 65 km. A significant quantity of slide debris completely covered the road surface after the earthquake, closing off the highway at this location. Debris was removed to allow passage of one lane of traffic prior to the GEER team visit to the site. The slope consisted of sedimentary deposits of large (up to 1 m), subrounded igneous rocks in a silty sandy soil matrix. Other nearby stable road cuts in this material were inclined at 60 to 65 degrees. In contrast to many of the rockslides seen in the region, this disrupted soil/rock slide had significantly more run out, and ultimately spilled over the road surface and into the canyon to the right (north) of the road. It is worth noting that a similar failure occurred in this material across the canyon in natural terrain.

Rockslides in natural terrain. Natural Terrain Rockslides at 2250 m: Figure 4 shows the upper portion of one of two adjacent rockslides occurring in natural terrain along Highway 024A about 65 km from the fault plane. The landslide debris cone was 40 m wide at its base and extended to an estimated height of 150 m. The rock slope consisted of moderately fractured and jointed igneous rock. Rock slide debris, consisting of blocks as large as 3 to 4 m in diameter, spilled into a major river below restricting - but not blocking - the flow of water.

Satellite Image Processing: Landslide Identification

Remote sensing (with combined field methods) will be used to identify and inventory landslides over an approximately 1,800 km² portion of the mesoseismal area. This study area spans a roughly 40 km stretch of the coast centered near the Canchamaná lateral spread and extends inland approximately 40 km along a variable-width zone located along and north of a corridor containing Highway 024A. This study area was defined to span a variety of terrain, but also by the availability of cloud-free pre- and post-earthquake satellite images from the NGA archive. Images in this archive scenes commissioned 2 days after the earthquake by USAID.

Several researchers have applied standard supervised and unsupervised maximum likelihood classification methods to identify landslides in remotely sensed images (Danneels et al. 2007, Metternich et al. 2007) with varying degrees of success. Using the large store of pre- and post- NGA archive imagery available to us (note that this is not necessarily the same imagery used in the high-accuracy photogrammetric analysis) the research team will study whether newly developed methods based on object oriented image analysis offers additional robustness in automatically detecting and counting landslides. Traditional methods rely on a pixel by pixel spectral classification to identify landcover but neglect other identifying features such as texture, shape and orientation of near-homogenous image regions. Classification approaches that take these features into account have shown promise in urban and environmental change detection (Walter 2004, Im et al. 2008). Definiens ‘eCognition' software exposes a variety of these segmentation tools in a flexible and easy to use application programming interface (API) that can be used to develop application specific object oriented image analysis procedures in which the image is segmented into distinct objects based on a user defined set of characteristics. These characteristics are then used to classify each object into similar groups. This study will attempt to develop a set of distinguishing characteristics for landslides in the Canchamaná area. Objects can be defined and described by additional geospatial datasets including elevation models, soil maps and landcover types. These datasets will be included in the analysis.

A selection of the landslides identified by automated change detection algorithms will be manually viewed to confirm automated analysis. Based on this, refinements will be made as needed to the algorithm and the confirmation process will be repeated until the algorithm is judged to produce accurate results. A large selection (hundreds) of significant landslides will receive additional manual visual image investigation on a site-by-site basis so that the slides can be classified base on the criteria of Keefer (1984), and so that zones of source, run out, and deposition can be delineated. Additionally, information on the consequences of these landslides will be noted, as appropriate (e.g. lane closures, damage to structures, delivery of sediment to channel, etc.). This information will be tagged and linked to the final project database.