Landslides Triggered by the 2016 Mw 7.8 Pedernales, Ecuador Earthquake: Correlations with ESI-07 Intensity, Lithology, Slope and PGA-h

Abstract

We provide a dataset of the landslides induced by the 2016 Pedernales megathrust earthquake, Ecuador (Mw 7.8, focal depth of 20 km) and compare their spatial distribution with mapped bedrock lithology, horizontal peak ground acceleration (PGA-h) and the macroseismic intensity based on earthquake-induced environmental effects (ESI-07). We studied 192 coseismic landslides (classified as coherent, disrupted and lateral spreads) located in the epicentral area, defined by the VII to IX_ESI-07 isoseismals. Based on our findings, lahar deposits, tuffs and volcanoclastic units are the most susceptible to landslides occurrence. Alluvial plains with fluvial loose fine sand are the most susceptible setting for lateral spreading, with a maximum intensity of IX_ESI-07. The coherent landslides are frequently found in altered shale and siltstone geological units with moderate slopes (8°–16°), with typical intensity ranging between VII and VIII_ESI-07. Our analysis draws a typical framework for slope movements triggered by subduction earthquakes in Ecuador. The most dangerous setting is the coastal region, a relatively highly urbanized area located near the epicenter and where liquefaction can trigger massive lateral spreading events. Coherent and disrupted landslides, dominating the more internal hilly region, can be triggered also in moderate slope settings (i.e., less than 10°). Indeed, the regression analysis between seismic intensity, PGA-h and landslide occurrence shows that most of the events occurred at PGA-h values between 0.4 g and 1.2 g, at a distance of 30 to 50 km from the rupture plane. Our database suggests that lithology and hillslope geometry are the main geological/geomorphological factors controlling coseismic landslides occurrence; while the distance from the rupture plane plays a significant role on determining the landslide size. Finally, we underline that coseismically-triggered landslides are among the most common environmental effects occurring during large subduction events that can be effectively used to properly evaluate the earthquake macroseismic field. The landslide inventory we compiled is suitable for assessing the vulnerability of physical environment from subduction earthquakes in Ecuador, and offers a primary data source for future worldwide analysis.

1. Introduction

Among all earthquake-induced environmental effects (sensu [1,2]), landslides are the most widespread and surveyed along coastal areas, in intermontane basins and in hilly areas [1,3,4,5,6,7,8,9]. Landslide susceptibility is related to lithologic and geomorphic characteristics including highly weathered loose soils, topographic conditions, local hydrogeological and geological setting (e.g., [3,10]). The geological and geotechnical variables play a key role as predisposing factors to earthquake-induced landslides in active tectonic settings such as the Ecuadorian coast region. Large intraslab and megathrust earthquakes due to the convergence between the Nazca, South American and the Caribbean plates (Figure 1) occur in this area. However, there is a lack of information about earthquake-triggered landslides connected to subduction earthquakes in Ecuador (e.g., [11,12,13]). Some earthquake-triggered landslides due to upper-plate events (in the order of Mw 6–7), in the northern Ecuadorian Andes, have been described in detail, whilst less is known about the distribution of earthquake-triggered landslides due to large subduction events [14,15,16,17,18,19].

We present a co-seismic landslide inventory for the 2016 Pedernales earthquake sequence. The earthquake sequence started with a foreshock (ML 5.7, at 18:47), some 10 minutes prior the Mw 7.8 mainshock, that occurred at 18:59 on the April 16, 2016 [20], see Data and Resources; Figure 2a). The extent of physical and environmental damage encompassed an area of thousands of square kilometers [21,22]. Locally, building damage and geological ground effects were enhanced by site amplification due to class E and F water saturated soils [23].

The epicentral area includes the city of Pedernales and surrounding communities (i.e., Coaque, Jama and Canoa), reaching a maximum intensity of IX_ESI-07, assessed from field observation of coastal uplift, soil liquefaction, sinkholes, landslides of natural and stabilized slopes, cracks in natural soils, minor fractures, failure along both the major and minor axis in pavement, particularly asphalt and concrete roads [22]. Coseismic geological effects were observed also across communities near Cojimíes (north of the epicenter, Figure 2b) and near Portoviejo (south of the epicenter, Figure 2b), resulting in a VIII_ESI-07 assessment. The area encompassed by isoseismals VII-IX _ESI07 corresponds to the deformed areas recognized from radar interferometry analysis (e.g., [24]).

Lithology, slope angle, ESI-07 intensity, and PGA-h (horizontal Peak Ground Acceleration, based on information available by the IGEPN; see Data and Resources) are the seismological and geological factors considered in this study to evaluate the hazard of earthquake-triggered landslides. These factors have been analyzed and spatially correlated with the distribution of the coherent, disrupted and lateral spreading landslides in a GIS environment.

We present the outcome of this study together with landslide hazard maps, describing the characteristics of the surveyed landslides and providing new insights into the controlling factors of coseismic landslides during large subduction earthquakes.

2. Description of the Study Area

The epicentral area of the 2016 Pedernales earthquake sequence is located on the central coast of mainland Ecuador. The mapped area covers about 18,000 km² of narrow valleys in a hilly landscape, uplifted marine terraces and alluvial plains.

Due to its geomorphological setting, many of the hydrographic basins of the province of Manabí are characterized by alternating periods of erosion and enhanced sedimentation and are typically covered by Quaternary loose sediments. In the eastern sector of the study area, large alluvial fans are the main sedimentary source.

The epicentral area is dominated by altitudes lower than 800 m a.s.l. Based on the Koppen climate classification [26], the study region is characterized by different sub-climates ranging from tropical Monsoon to dry and tropical Savanna [27]. Annual rainfall for the dry climate zone in southernmost Manabí region is less than 500 mm, measured between January and April, while the summer is very dry and hot. Annual rainfall ranges from 500 to 1000 mm for the tropical Savanna sub-climate where low rainfall occurs between June and November. The tropical Monsoon sub-climate in northernmost Manabí, is as hot as dry, but includes also a dry season from June to November with an annual total rainfall ranging between 1000 and 2000 mm. The predominant vegetation is a dry to very dry tropical forest, followed by mountains with tropical thorn shrub and premontane thorn shrub [26,28,29]. The most populated urban areas (i.e., the cities of Manta, Portoviejo, Chone, Bahía de Caráquez, Jama and Pedernales) have been settled on floodplains, alluvial terraces and colluvium-alluvium, where low-resistance soft soils prevail as well as soft rocks of claystone and siltstones, and sandstones in a lesser proportion.

3. Seismotectonic Setting and the 2016 Pedernales Earthquake Sequence

The Ecuadorian coastal region is within an active plate margin where the oceanic Nazca plate subducts the South American continent (e.g., [30,31,32,33,34]). The Nazca plate carries the Carnegie Ridge, which is a mountain range of seamounts resulting from the Galápagos hotspot volcanism (Figure 1; [35]). The subduction of this submarine ridge, which has an average elevation rise up to 2.5 km, has influenced the geodynamics of the coastal zone of Ecuador [36,37], as well as in its coastal geomorphology [38,39] and seismogenic characteristics [36,40,41,42,43,44,45,46,47].

Historically, the 1942 Jama (Mw 7.9) and the 2016 Pedernales (Mw 7.8) earthquakes are the largest subduction earthquakes recorded in the central segment of the subduction interface zone (Figure 2a; [48,49]). This tectonic segment registered several megathrust earthquakes, including: i) the May 3, 1896, Mw 7.1; ii) the 1 June 1907, Mw 7.4; iii) the 16 January 1956, Mw 7.4 and iv) the Mw 7.1 on August 4, 1998 earthquakes (Figure 2, [22,50]).

The spatial distribution of the April 16, 2016 Pedernales earthquake sequence indicates a rupture zone ca. 100 to 120 km long [22,24]. The mainshock was located onshore, being less than 10 km northeast from the city of Pedernales, with a hypocentral depth of 21 km (source USGS, see Data and Resources). The focal mechanism obtained by IGEPN indicates a low-angle reverse-fault dipping N119/15. Peak ground acceleration up to 1.4 g was recorded in Pedernales, 0.52 g in Manta, 0.38 g in Portoviejo, 0.23 g in Chone, and 0.19 g in the Esmeraldas (IGEPN, see Data and Resources). The sequence included more than 3500 aftershocks, the most severe ones occurred on May 18, 2016 with Mw 6.7 to 6.9 [22,51] and some 14 months later with a ML 6.3 (IGEPN, see Data and Resources).

This sector of the subduction zone, including the entire province of Manabí, has a short seismic record covering the last 120 years that suggests a recurrence interval of approximately 20 years for Mw >7.0 events, and between 70 and 80 years for Mw >7.5 [22,52]. Such subduction-related earthquakes have a recurrence relatively short in comparison with the subduction zones north and south of it [22].

4. Methods and Materials

Shortly after the mainshock, we mapped in the field all the observable earthquake-induced environmental effects (sensu, [1]), accessible through the main access roads and towns near Pedernales, which is the location of the instrumental epicenter where the peak ground acceleration of 1.4 g was recorded (IGEPN Accelerograph Ecuadorian Network; see Data and Resources). Government institutions such as MTOP (i.e., Ministerio de Transporte y Obras Pública) and MIDUVI (i.e., Ministerio de Desarrollo Urbano y Vivienda) provided logistics and transportation management due to partial destruction of the main access roads to towns, bridges and buildings by creating fieldwork camp sites for geotechnical engineers and geologists reaching out to the entire Manabí province and the southern part of Esmeraldas province.

Preliminary field mapping evaluated the macroseismic intensities of the primary coseismic geological effects (i.e., coastline uplifting along beaches and cliffs; e.g., [22,53] and secondary effects (i.e., soil liquefaction causing lateral spreading, subsidence, sinkhole and sand boils, as well as cracks in the soil, pavement failure in asphalt roads as well as landslides in stable and unstable slopes), using the ESI-07 intensity scale, with values ranging from VII to IX_ESI-07 in a coseismic affected area of about 18,000 km² [21,22]. Along with this database, an inventory of 192 seismically induced landslides (in attached pertinent Ⓔ electronic supplement to this article) by the 2016 Mw 7.8 event is herein provided (Figure 2), grouping them as follows:

(a) lateral spreads, landslides which moved in a translational mode in areas of liquefied gravel, sand, or silt, or rarely in sensitive clay transformed into fluid by disturbance [3,54], out of which 58 coseismic landslides were surveyed (Figure 3);

(b) coherent landslides, generally deep-seated, consisting of one or multiple blocks sliding on planar or smoothly curved basal shear surfaces, out of which 78 landslides in the epicentral area were surveyed (Figure 4);

(c) disrupted landslides associated with rock falls, rock slides, soil falls and disrupted soil slide, adding up to 56 surveyed coseismic effects, between earthquake-induced landslides and reactivated older landslides (Figure 5).

In this study, we also manually digitized some landslides on satellite imagery (SPOT-CNES imagery, courtesy of Google Earth) to i) check the extent of landslides at some of the surveyed events, ii) assure that no pre-earthquake landslides were present and iii) check field-observed mapping, in the closeness of some surveyed sites, especially in remote areas where only limited access was possible (Figure 4). Moreover, we checked the coseismic landslide inventory provided by the British Geological Survey BGS and GEER- ATC (see Data and Resources), referred only to the areas of Portoviejo, Bahía de Caráquez, Chone, Muisne and Crucita. This resulted in a statistically significant database of earthquake-induced landslides, though not comprehensive over the entire area interested by coseismic slope movements.

We compared the landslide occurrence with a set of selected variables that can possibly be considered as triggering or predisposing factors: lithology type, horizontal peak ground acceleration (PGA-h), topographic slope (derived from the 1 arc second SRTM digital elevation model) and distance from both the epicenter and the subduction rupture plane.

We derived the statistical correlation of landslide occurrence for each class of variables (i.e., PGA-h and lithology) using a statistical approach, modified after the Information Value Method (InfoVAL). The method allows the quantified prediction of susceptibility by means of a score (Wi), calculated according to landslide occurrence on each class, weighed according to the class distribution over the entire study area:

$$Wi =\ln \frac{Densclas{s}_i}{Densma{p}_i}$$

where Wi is the score for the i^th class, Densclass is the landslide occurrence for the i^th class and Desnmap is the i^th class occurrence on the whole area. This formula normalizes the event occurrences over the spatial distribution of each considered class: positive values indicate positive statistical correlation, negative values a negative one, while values close to zero indicate a random distribution of the data.

Since we did not map the areal extent of landslides and considering that the survey coverage was limited by site accessibility, we adopted simple univariate statistics for the description of topographic slope intervals typically associated with each landslide category.

We calculated the distance of each event from both the mainshock location (i.e., epicentral distance) and from the seismogenetic rupture plane. The latter has been calculated considering a rupture plane orientation on N119/15, as derived from seismological data and the distance have been calculated as the closest distance to the plane (i.e., Rrup distance).

5. Results and Discussion

Most landslides have shallow shear planes, between 2 and 4 m depth. On the contrary, few landslides have deep rupture planes (up to 15–30 m below ground level), as documented in the Briceño and San Isidro sites. A large number of landslides fall in the range of 200 to 3200 m³ (VIII_ESI-07)., while only few of them fall in the range of 5000 to 8000 m³ (marked as intensity IX_ESI-07). Isolated rotational landslides and rockfalls smaller than 200 m³ are regarded as intensity VII_ESI-07 (southern part of Esmeraldas province). Landslides located in plains and alluvial terraces are constrained to low resistance cohesive soils and are mainly lateral propagations caused by liquefaction. The earthquake-induced landslides extend as far as to the southern Manabí province, 134 km away from the instrumental epicenter, where lateral-spreading displacements was recorded near the city of Portoviejo in a site within the VIII_ESI-07 and where 0.35 g of PGA-h was recorded.

According to the geological map at a scale of 1: 500,000 of the coasts of Ecuador [55], we divided the study area into 5 lithological units (Figure 6): loose fluvial deposits and/or fine to coarse grained coastal units (AL); mainly sandstones and conglomerates (SS); shales, siltstones and mudstones (SS); tuffs and volcanoclastic deposits (VV); mainly basalts, hyaloclastite and associated greywackes (BB).

Our analysis shows that tuffs and volcanoclastic units recorded the largest number of coseismic landslides. Shales and siltstones host most coherent landslides while sandstones are mostly associated with disrupted landslides. Fluvial loose deposits and alluvial plains constitute the most susceptible setting for lateral spreading (Figure 7).

Lateral-spreading landslides are located in plains and alluvial terraces as well as few inclined slopes (Figure 3). Coherent landslides are located on moderate slopes with inclination between 8° and 16° and are mostly developed in Pliocene and Pleistocene geological units, which are composed of volcanic units and highly weathered clastic rocks. Disrupted landslides are developed on escarpments with inclination ranging from 16° to 55°, many of which are formed in high relief within the epicentral area as highlighted in Figure 5, where Miocene sandstone and greywacke lithological units (SS) predominate. These results coincide well with those obtained for the coseismic landslides caused by the 2011 Tohoku (Mw 9.0) and the Maule (Mw 8.8) thrust earthquakes, where most landslides, in the epicentral area, occurred in Neogene geological units [56,57,58].

The seismic intensity and the PGA-h characterize the degree of destruction and the seismic energy of earthquakes [8]. The ESI-07 intensity map is spatially well-correlated with the PGA-h isolines of the Pedernales earthquake (Figure 8) and shows a distribution of the intensity VII to IX_ESI07 contours elongated along the with the rupture strike and encompassing the area affected by earthquake-triggered landslides. The isoseismal zone of VI_ESI-07 has much fewer landslides, with volumes lower than ca. 10³ m³, especially where the slope is unstable and particularly steep.

For the area encompassing the ruptured sector we can estimate a relationship between ESI-07 degree and the PGA-h: IX_ESI-07 ranges between 0.73 g and 1.35 g; VIII_ESI-07 from 0.55 g to 0.73 g and VII_ESI-07 from 0.30 g to 0.55 g.

The number of landslide events and their size can be scaled with the epicentral distance up to VIII_ESI-07 (Figure 9 and Figure 10, Supplementary Materials). Several landslides ascribable to IX_ESI-07 have been mapped in the coastal region but these are mainly lateral spreads and are more influenced by the local geomorphological and geological setting (i.e., coverage lithology and thickness, depth of the aquifer etc.) than by distance or other factors (Figure 11). Collectively, the coseismic landslides caused by the Pedernales earthquake (Mw 7.8) are strongly dependent from the distance to the rupture plane and PGA-h. Most of the landslides are triggered for values that are bounded between 0.4 g and 1.2 g and within a Rrup distance of less than 50 km (Figure 9, Supplementary Materials).

6. Conclusions

With this first database, we produced an inventory of 192 seismically-induced landslides related to the 2016 Pedernales earthquake (Mw 7.8). Landslides were analyzed through a bivariate approach against: slope, lithological setting, PGA-h and macroseismic intensity data.

The spatial distribution of PGA-h, ranging between 0.4 g and 1.2 g, is consistent with the maximum VII–IX_ESI-07, recorded over an area of about 18,000 km². The disrupted landslide type is preferably located on fractured greywacke/sandstone units and in high relief areas, with slopes ranging between 16° and 55°. Our database suggests that lithology and hillslope geometry are the main geological/geomorphological factors controlling coseismic landslides occurrence; while the distance from the rupture plane plays a significant role on determining the landslide size. Indeed, the regression analysis between seismic intensity, PGA-h and landslide occurrence shows that most of the events occurred at PGA-h values between 0.4 g and 1.2 g, at a distance of 30 to 50 km from the rupture plane.

Our analysis draws a typical framework for slope movements triggered by subduction earthquakes in Ecuador. The most dangerous setting is the coastal region, a relatively highly urbanized area located near the epicenter and where liquefaction can trigger massive lateral spreading events. Coherent and disrupted landslides, dominating the more internal hilly region, can be triggered also in moderate slope settings (i.e., less than 10°).

We highlight that both the dimensions and frequency of landslides scale with distance in accordance with PGA-h and earthquake intensity (ESI-07) and through macroseismic intensity scenarios or strong motion modeling (e.g., USGS SHAKE maps) this information can be effectively used for the prediction of landslides distribution following strong subduction events. Calculated regressions can be considered valid for the South American subduction setting only, still the proposed approach can also be exported in different seismotectonic settings worldwide.

We note that timely field mapping after the mainshock allows to calibrate with unprecedented accuracy the geographic distribution of earthquake environmental effects, specifically landslides; and to understand the factors which control their distribution. We remark that coseismically-triggered landslides are among the most common environmental effects occurring during large subduction events. This effectively augment the relevance of coseismic landslide investigations for properly evaluating the earthquake macroseismic field, as initially suggested by the pioneers in the study of seismic intensity (e.g., [59,60,61,62]. We argue that our results represent a clear illustration of this point.

Data and Resources: Seismicity and macroseismic data were collected from IGEPN (http://www.igepn.edu.ec/servicios/eq20160416, last accessed June 2019);

USGS (http://earthquake.usgs.gov/earthquakes/eventpage/us20005j32#dyfi, last accessed May 2019) and the Geological Survey of Colombia (http://studylib.es/doc/6527439/informe-del-sismo-del-16-de-abril-de-2016-en, last accessed June 2019).

The isoseismal map for EMS-98 intensity realized by IGEPN is available at http://www.igepn.edu.ec/servicios/noticias/1324-informe-sismico-especial-n-18-2016 (last accessed May 2019).

Additional information on the Pedernales earthquake made available by BGS (British Geological Survey) and GEER-ATC was obtained at https://www.bgs.ac.uk/research/earthHazards/epom/ecuadorEarthquake.html (last accessed June 2019) and http://www.geerassociation.org/administrator/components/com_geer_reports/geerfiles/ECUADOR_Report_GEER-049-v1b.pdf (last accessed May 2019), respectively.

The DTM used for analysis is the NASA SRTM 3.0 global model, 1 arc second resolution: https://lpdaac.usgs.gov/news/nasa-shuttle-radar-topography-mission-srtm-version-30-srtm-plus-product-release/ (last accessed 6 August, 2019).

Some of the figures were realized using Qgis and Arcmap® software.