Lessons from 2019–2020 Landslide Risk Assessment in an Urban Area of Volcanic Soils in Pereira-Colombia

Abstract

The city of Pereira is in the central-western region of Colombia, in the valley of the Otún River in the Central Cordillera of the Colombian Andes, whose soils are mostly composed of volcanic ash and Lahar flows. On June 11, 2019, after intense rains, a landslide of about 13,000 m³ was triggered. This landslide killed 4 people, interrupted the traffic on an important national road, and endangered the airport runway. Just after the event, it was necessary to evacuate 100 homes, located in the upper part of the slope and 234 in the following days. A second landslide occurred on July 24, 2020, mobilizing 52,000 m³ and affecting 33 of the previously evacuated houses, then there were about 20 more small landslides. A zone of about 2.300 m long and 300 m wide (69 ha) was identified to be susceptible to new landslides that may affect more than 15.000 people. Identification of landslide triggering factors, failure, post-failure behavior, and the retrogressive landslide advance are reviewed because they are key factors for an effective hazard evaluation and risk governance. The landslide event was also used to calibrate a general methodological guide for landslide risk assessment, developed by the Colombian Geological Service and the National University of Colombia (Ávila et al. 2016) that has been extensively applied for the inclusion of risk evaluation in the land use planning, as required by the Colombian legislation.

1 Introduction

Landslide events in volcanic materials triggered by intensive rain periods tend to generate flow-type movements during post failure and as stated by Picarelli et al. (2020) this behavior strongly affects displacements and run out, as it was evident in two landslides that occurred in an urban area in the city of Pereira (Colombia) on June 11, 2019, and on July 24, 2020. The stress reduction caused by the first landslide produced a retrogressive advance, evidenced by the presence of soil cracks in which water accumulated during rain periods, the accumulated water, together with the water absorbed by the volcanic soils, gave the second landslide great mobility and its travel distance was much higher than that estimated with empirical correlations. This article describes the landslide evolution, the significant retrogressive process, the application of a methodological guide developed by the Colombian Geological Service for landslide risk evaluation (Ávila et al. 2016), and the risk management after the emergency caused by the first landslide in a densely populated area.

2 Description and Evolution of the Instability Process

The city of Pereira is located in the Central Cordillera of the Colombian Andes. This mountain range is characterized by having an important volcanic chain, for which many of the cities are founded on lahar deposits and soils derived from volcanic ashes. Figure 1 shows the general site location of the study area and Fig. 2 shows an aerial photograph of the landslide that occurred on June 11, 2019 (the first landslide).

Fig. 1

General location of the study area

Fig. 2

Aerial photograph of the 2019 Pereira landslide

Multitemporal analysis, starting with an aerial photograph of the year 1946 (Fig. 3), which, although it does not have a very good resolution, allows us to see the construction process of the airport runway and the presence of the railway line (built in 1927). In this image also some incipient features of instability can be identified, such as small escarps. A second aerial image from 1969 (Fig. 4) shows geomorphological evidence of ancient scarp surfaces (dashed lines) but no evidence of any significant sign of instability of the slope. A geological fault alignment is also indicated in the image. At that moment the airport and the railroad had normal operation, and some dispersed constructions may be observed. The third image is from 2005 (Fig. 5) and shows the construction process of the Portal de la Villa neighborhood for what was necessary to make significant slope cuts and conformation fills. The National Road 29RSC can be observed in this image. The area remained relatively stable until July 11, 2019, when the first landslide occurred (Fig. 2). This landslide killed 4 people, affected many of the houses located on the top slope (Portal de La Villa and Matecaña, neighborhood) and the 29RSC national road. The instability process continued and on July 24, 2020, a second landslide event, larger than the previous one, was triggered. This new event destroyed 33 houses (previously evacuated), produced partial damming of the Otún river, caused the closure of the national road for almost 20 days, and came close to affecting the airport runway (Fig. 6).

Fig. 3

1946 image: construction of the airport runway. The railroad was constructed in 1927. Some incipient landslide scarps may be identified (Image IGAC, 1946)

Fig. 4

1969 image: No signs of increased instability but some dispersed constructions block up intermittent water courses. Normal operation of the railroad and the airport

Fig. 5

2005 image: explanations for the construction of the Portal de la Villa neighborhood. (Google Earth, 2005)

Fig. 6

Second landslide occurred on July 21, 2020, affecting a greater area in the upper and lower sectors

3 Site Characterization

After the 2019 landslide, a detailed geological and geotechnical site characterization was carried out. This characterization included geological and geomorphological surveys, geophysical exploration, mechanical drilling, sampling, laboratory tests, and a detailed topographic survey. All this information permitted us to obtain a digital elevation model and detailed geological and geotechnical characteristics. A synthetic geological and geotechnical profile is shown in Fig. 7. Table 1 presents the mean water content and Atterberg limits of the surface soils, which are predominantly high-plasticity sandy silts (MH).

Fig. 7

General geological profile of the landslide and the influence areas

Table 1 Water content and Atterberg limits

4 Alert Signs and Damage Evolution

One of the most important aspects of landslide risk management is to opportunely identify the instability signs that permit alert people and take mitigation actions. Reports from the local emergency office, show that instability signs were occurring since 2011 (there are reports in 2011, 2015, 2017, and 2019). On tenth July 2019 (one day before the first great landslide) a scarp was observed in the upper part of the slope (Fig. 8) and it was covered by plastic. However, the next day a landslide occurred as shown in Fig. 9, killing 4 people.

Fig. 8

Scarp evidence one day before landslide 2019

Fig. 9

Landslide principal scarp on Jun 11, 2019

Signs of imminent instability are common in most landslides, for example, Hancox (2008), indicated that the Abbotsford Landslide in New Zealand that occurred on 8 August 1979 showed minor cracking and damage on one house located 60 m from the landslide of almost 11 years before the landslide occurrence (1968–1972) due to a graben. Xian et al. (2022) reported crack and small landslide evolution from 2005 to 2019, revealing a significant increase in crack number as in crack length as the instability increased. These warning signs, such as small cracks, small landslides punctual soil deformations, water emanations from the subsoil, etc., should not go unnoticed since, if they are corrected in time, they can allow long-term stability to be achieved and with relatively lower costs. Stabilization measures that are taken very close to the imminent failure (as those shown in Fig. 8) can be dangerous since the speed of the landslide tends to be fast and there is no time to escape. In these cases, the most convenient procedure is to carry out a quick and organized evacuation of the people who are within the possible landslide influence area.

Ávila et al. (2016) published for the Colombian Geological Service a technical guide to performing landslide risk analysis at a detailed scale which has been extensively used for consulting firms to actualize land use planning in several Colombian municipalities. In this guide four main sectors are defined in a landslide profile, in order to define vulnerability and hazard areas to take effective mitigation actions, these sectors are shown in Fig. 10. Zones 2 and 3 are those directly affected by the landslide and all exposed elements located in these areas may result strongly affected or destroyed. The limit between zones 1 and 2 is controlled by geological and geomorphological conditions but not always is easy to define it. Normally it is necessary to estimate how far from the landslide scarp the soil may present cracks or deformations. Numerical models, considering multiple slip surfaces and lateral stress relaxation may be applied for this kind of estimation but, real observations are necessary for calibration.

Fig. 10

Landslide exposure zones (Ávila et al. 2016)

For the Pereira Landslide, the limit between Zone 1 and Zone 2 extended as far as 80 m, from the main scarp, as shown in the aerial photograph of Fig. 11 and in detail, in the photograph of Fig. 12. Another important element in the risk attention process is to have a reasonable estimation of the time evolution of the retrogressive landslide progress, for which, documentation of previous cases, under particular geological and geotechnical conditions, is very useful as a calibration measure. Once the first landslide occurred, a significant crack opening and new cracks were observed after 20 days. Also, the damage in houses was evident, as shown in Fig. 13. for what people had to be evacuated. Figure 14 shows the situation after the second landslide, which occurred on July 21, 2020, one year and 10 days after the first landslide. The previous cracks had converted now.

Fig. 11

Aerial image of the Portal de la Villa on June 12 (one day after the landslide). The distance from the scarp to the initial cracks is about 80 m

Fig. 12

First cracks appeared on June 12, 2019 (One day after the landslide) in block Mz5, 80 m far from the main scarp, as indicated in Fig. 11

Fig. 13

Crack evolution on July 1st (20 days after the landslide

Fig. 14

Total damage observed in block Mz 5 on July 21, the same day of occurrence of the second landslide

The Limit between Zones 3 and 4 of Fig. 10 may be estimated according to different empirical procedures of landslide travel distance as reported by Corominas (1996) or Hungr et al. (2005).

5 Rainfall Behavior Previous to the Landslides

Figure 15 shows the mean monthly rainfall histogram from Matecaña Airport Station (IDEAM 2022). A bimodal regime is observed with a first pick in May (250 mm) and the second pick in October and November (250 mm).

Fig. 15

Mean monthly rainfall histogram from the Matecaña Airport (adapted from IDEAM 2022)

The first landslide (June 11, 2019) occurred just after the first monthly rain pick and it was triggered by an intense rain day (93.7 mm), as shown in Fig. 16.

Fig. 16

Daily rain previous to the first landslide (IDEAM 2022)

The second main landslide (July 21, 2020) did not occur during the strong rainy season, as can be observed in Fig. 17, however, due to the intense soil cracking and the rains of the previous months, part of that water could have accumulated, especially in the ancient natural watercourses that were filled during the construction process and this, in turn, generated the flow-type landslide, of high mobility, capable of reaching a long distance. The situation of general instability of the area between the first and the second landslides made it difficult to build emergency drainage systems, in order to reduce the water accumulation, and the emergency actions concentrated on evacuation and periodical monitoring of land deformations.

Fig. 17

Daily rain previous to the second landslide (IDEAM 2022)

6 Travel Distance, Velocity, and Mobilized Volume

Comparative values of calculated and measured travel distances and mobilized volumes of the 2019 and 2020 events are shown in Table 2. Using the methodology of Hungr et (Hungr et al. 2005), the calculated travel distance for both events was 125 m. For the 2019 event, this result is of the same order as the measured value, however, for the 2020 event, the calculated value of 125 m resulted much lower than the real value of 290 m. The significant difference in travel distance, in this case, may be attributed to the great amount of water present in the soil cracks and internally, in the volcanic soil mass, converting the landslide into a flow-type process.

Table 2 Comparison between calculated and real travel distance and mobilized volume for the 2019 and 2020 events

The calculated volume of the 2019 event was 50.000 m³ but the real volume was only 13.000 m³. Similarly, for the 2020 event, the calculated volume was 71.000 m³ but the measured volume was only 52.000 m³. The difference is due to the presence of a lithological control that did not permit a deep failure surface as initially assumed.

Direct landslide velocity (V) was measured from a video taken by local people. The approximate velocity was 12 m/s. Using the sliding block model, the landslide velocity may be estimated by the fallowing equation:

$$ V=\sqrt{2 gH\left(1-\frac{\tan \varphi }{\tan \beta}\right)} $$

Were:

H:

landslide height

ϕ:

internal friction angle

β:

slope angle

g:

acceleration of gravity

In this case, H = 70 m, β = 30°, and the internal friction angle results to be 27.3°. This value corresponds to the residual strength and is in the range reported for this type of residual volcanic soil in other tropical countries (i.e., Rigo et al. 2006; Wesley 1992).

7 Hazard Assessment

Two elements should be considered in hazard assessment: the possible failure conditions and the reach distance of the potential landslide. Failure conditions were evaluated based on the limit equilibrium method. The modeled circular failure surfaces (Fig. 18) were in general similar to those observed in the field (Fig. 9), while, as previously mentioned, travel distances had good prediction in the 2019 event but large underprediction in the 2020 event.

Fig. 18

Limit equilibrium analysis for hazard modelling

A hazard map was constructed based on the factor of safety (FS) obtained for different terrain profiles and the corresponding computed travel distances. In this case FS < 1.1 was classified as a high hazard level, FS between 1.1 and 1.5 was a medium hazard level and FS > 1.5 was a low hazard level. Although the calculated travel distances represented lower values than those that actually occurred, this possible effect was previously estimated based on the morphological characteristics and on the high saturation observed in the intensely cracked soils, as a result of which a safety corridor was left that finally allowed the construction of a conservative hazard map, as shown in Fig. 19. This map showed good agreement with the 2020 event.

Fig. 19

Hazard map constructed just after the 2019 event

8 Vulnerability Assessment

The previously mentioned methodological guide for landslide risk assessment developed by the Colombian Geological Service (Ávila et al. 2016) presents a detailed procedure for vulnerability evaluation, based on the postulates of Uzielli et al. (2008), Li et al. (2009) and Du et al. (2013). The general steps required for vulnerability assessment are:

• Identification and location of exposed elements.

• Evaluation of structural fragility.

• Calculation of the intensity factor and.

• Evaluation and zoning of vulnerability.

Exposed elements were classified into two main groups: physical elements (buildings, lifelines, and transportation infrastructure) and persons. The typology of buildings was based on 7 possible classifications, according to the structural construction system. Exposed elements were clearly identified on the base map with respect to the landslide exposure zones. Structural fragility was calculated considering the typology of the building, its number of floors, the state of conservation, and the age of the constructions.

The intensity factor is related to the potential energy of the landslide, therefore, it is necessary to identify, firstly, to which vulnerability scenario is referred (rock fall, rapid landslides, or slow landslides) and secondly, what is the position of the exposed elements in relation to the landslide. In this case, a rapid landslide is the adopted scenario (according to Cruden and Varnes (1996) when velocity is higher than 5 m/s, as in this case, the landslide is classified as very rapid), and the exposition is based on the location of the elements, as shown in Fig. 10.

According to Du et al. (2013), the vulnerability of the exposed elements (V) may be expressed by:

$$ V=\frac{1}{2}{\left(\frac{I}{1-S}\right)}^2\ if\ I\le 1-S\ \mathrm{or} $$

$$ V=1-\frac{1}{2}{\left(\frac{1-I}{S}\right)}^2\ if\ I>1-S $$

Where:

I:

Intensity factor

S:

Structural fragility

For structures, a vulnerability value of 1 means the complete destruction of the property, and a vulnerability value of 0 indicates no damage. Values between 1 and 0 represent a proportion of expected structural damages. In people, a vulnerability value equal to 1 means loss of life, and values less than 1 are considered as the corresponding probability of loss of life. A vulnerability value equal to zero means no loss of life is expected. Table 3, presents the vulnerability parameters for each type of structure that could potentially be involved in the landslide.

Table 3 Vulnerability parameters

Based on this procedure, in the Matecaña neighborhood, 49 houses were classified as vulnerable equal to 1. For the Portal de la Villa neighborhood two main susceptibility areas were identified: one with S = 0.69 (187 houses) and the other with S = 1 (63 houses), however, due to the high-intensity factor, in both cases, vulnerability is equal to 1 and therefore consequences will be catastrophic for both, structures and people. For lifelines and transportation fragility the intensity factors were also evaluated and, as in buildings, their vulnerability factors were equal to 1.

Once the previously described factors have been analyzed, a zoning map was obtained (Fig. 20). The Portal de la Villa neighborhood, the Matecaña neighborhood and the 29RSC road were classified as highly vulnerable (red), while the Matecaña International Airport was classified as medium vulnerability (yellow). The rest of the studied area was classified as low vulnerability.

Fig. 20

Vulnerability map prepared just after the 2019 event

9 Risk Assessment

Vulnerability maps and hazard maps may be used to take decisions on-site about risk management; however, a risk map is better because it takes into account both, hazard and vulnerability. Classical risk calculations are based on the product of the probability of occurrence of event times the expected consequences, expressed in cost or adversity derived from that event (Beacher and Christian 2003). However, when circumstances require rapid and safe decisions, related to evacuation processes or suspension of a high-traffic road, a simplified method of risk assessment may be a good alternative. In this case, a risk matrix method was used, based on the combination of the hazard map and the vulnerability map as shown in Fig. 21.

Fig. 21

Risk matrix based on hazard and vulnerability maps

The risk map constructed after the 2019 event is shown in Fig. 22. This map was very effective at that moment for the risk management and the evacuation process because it was clear and easy to explain to local authorities and to the affected community. In the 2020 event, people had already been evacuated from the impact area and for this reason, there were no deaths or injuries, showing the effectiveness of the procedure. A formal risk analysis and a clear risk map undoubtedly represent a fundamental tool for risk governance.

Fig. 22

Risk map constructed after the 2019 landslide

10 Conclusions

When landslides occur in highly populated urban areas with critical public services and transportation infrastructure, efficient risk management is required in order to save people’s lives and keep local governance. The methodological landslide risk assessment guide, developed by the Colombian Geological Service (Ávila et al. 2016) showed to be a practical and effective tool to apply in these cases.

The application of this systematic methodology requires detailed geological and geotechnical data and a precise digital elevation model. Gathering this information may require some time (time is a critical factor in these conditions) but it is essential for safe decision-making, based on objective parameters.

Landslide travel distances based on empirical correlations resulted in good agreement with the first landslide (2019 landside), however, a significant underestimation of the travel distance was observed during the second landslide (2020 landslide). This can be explained by the presence of a large amount of water in the cracks that were formed in the upper part of the landslide and the high moisture content of the shallow volcanic soils. These two factors reduced soil resistance and produced a high-mobility flow-type landslide.

In many tropical regions, urban developments are carried out in hillside areas of volcanic soils, for which land cuts and fillings are required. This procedure often obstructs the natural water channels and if efficient drainage is not implemented, long-term instability processes may develop.

When initial instability signals are observed (surface cracks land deformations or water flows) it is necessary to act immediately, to take appropriate corrective measures. If not, the problem can become uncontrollable, as in the case of the Pereira landslide.