Lava flow hazard and risk at Mt. Cameroon volcano

Abstract

Mt. Cameroon is one of the most active effusive volcanoes in Africa. About 500,000 people living or working around its fertile flanks are subject to significant threat from lava flow inundation. Lava flow hazard and risk were assessed by simulating probable lava flow paths using the DOWNFLOW code. The vent opening probability density function and lava flow length distribution were determined on the basis of available data from past eruptions at Mt. Cameroon volcano. Code calibration was performed through comparison with real lava flow paths. The topographic basis for simulations was the 90-m resolution SRTM DEM. Simulated lava flows from about 80,000 possible vents were used to produce a detailed lava flow hazard map. The lava flow risk in the area was mapped by combining the hazard map with digitized infrastructures (i.e., human settlements and roads). Results show that the risk of lava flow inundation is greatest in the most inhabited coastal areas comprising the town of Limbe, which constitutes the center of Cameroon’s oil industry and an important commercial port. Buea, the second most important town in the area, has a much lower risk although it is significantly closer to the summit of the volcano. Non-negligible risk characterizes many villages and most roads in the area surrounding the volcano. In addition to the conventional risk mapping described above, we also present (1) two reversed risk maps (one for buildings and one for roads), where each point on the volcano is classified according to the total damage expected as a consequence of vent opening at that point; (2) maps of the lava catchments for the two main towns of Limbe and Buea, illustrating the expected damage upon venting at any point in the catchment basin. The hazard and risk maps provided here represent valuable tools for both medium/long-term land-use planning and real-time volcanic risk management and decision making.

Introduction

Mt. Cameroon, by the Atlantic shore of the Gulf of Guinea in Africa (Fig. 1), is one of the largest active volcanoes on the continent. It has erupted at least 18 times in the last 200 years, most recently in 1999 and 2000 (Suh et al. 2003, 2011; Njome et al. 2008; Bonne et al. 2008). Documented historical eruptions have been characterized by effusive activity accompanied by moderate explosive events from summit or flank vents (Thierry et al. 2008).

Fig. 1

The CVL (dashed yellow line). Mt. Cameroon is the only presently active volcano

The region around Mt. Cameroon is very fertile and populated. There are three major towns (Limbe, Buea, and Muyuka) and minor towns and villages scattered on the flanks of the volcano, with a total of about 500,000 inhabitants.

Given this situation, it is not surprising that lava flows destroyed farmlands and posed a threat to villages in several documented eruptions. The 1922 southwest flank eruption originated from a vent about 1,400 m a.s.l., producing lava flows that reached the Atlantic Ocean near the village of Biboundi (Fig. 2). In 1999, two lava flows formed on the southern flank, the first producing a 3-km-long lava field from a vent located about 2,650 m a.s.l., the second giving rise to a 14-km-long lava field from a vent ∼1,500 m a.s.l. The latter lava flow stopped 200 m from the ocean near the village of Bakingili, cutting off the only coastal road connecting the villages on the eastern flanks of the volcano with those on the western flanks (Déruelle et al. 2000; Suh et al. 2003). The last eruption occurred in 2000, when the lava poured out from a fissure in the summit area above 4,000 m a.s.l. The lava flow spread mostly southeastwards over the topography (Fig. 2), stopping about 4 km from the suburbs of Buea (more than 150,000 inhabitants including the neighboring villages).

Fig. 2

Slope map of Mt. Cameroon volcano. Recent lava flows are reported in red. Towns and villages are reported as black cross-hatched areas. The green line encloses the area where computational vents were placed

Although none of Mt. Cameroon’s historical eruptions have caused loss of human lives, the above events indicate that lava flows pose a significant threat to people and property. Two studies have recently addressed the hazard and risk posed by lava flows at Mt. Cameroon. The basic assumption in these studies has been that past activity at the volcano can be used to predict its behavior in the future. Thierry et al. (2008) approached this issue in a comprehensive study aimed at deriving a global “multi-hazard geological risk” map which combines several different hazard and risk factors, including earthquakes, debris flows, and lava flows. Merging different kinds of risk enables comparison and provides a valuable broad overview. The hazard from lava flows was treated by simply identifying areas at progressively larger distances from the volcano summit and characterized by decreasing probability of inundation.

Bonne et al. (2008) presented a map of the sole risk of lava flow inundation. This map is based on the oversimplified assumption that lava flows behave like water flows and strictly follow the steepest descent path. While this assumption may provide a first general overview of the most hazardous areas, it is inadequate for a more accurate and focused analysis. Lava flows are commonly some meters thick and can thus spread out and override obstacles; therefore, they do not necessarily follow the steepest descent path and can split into different branches, forming complex patterns (Walker 1971).

Since the work of Wadge et al. (1994), a number of papers have made use of large sets of lava flow simulations to quantify the susceptibility to lava flow inundation and investigate lava flow hazard and risk to urban areas (Felpeto et al. 2001; Favalli et al. 2006, 2009a, b, c; Crisci et al. 2008; Tarquini and Favalli 2010). In this paper, we adopt a similar approach and assess lava flow hazard and risk at Mt. Cameroon volcano using the probabilistic DEM-driven DOWNFLOW code (for a review of lava flow simulation models, see Costa and Macedonio 2005). DOWNFLOW allows the definition of areas that are susceptible to inundation by lava flows originating from each vent; it has been used extensively to simulate lava flows at Mt. Etna and Nyiragongo volcanoes (Favalli et al. 2005, 2006, 2009a, b, c; Chirico et al. 2009). We derived hazard and risk maps following both (1) a standard approach, which classifies areas according to the probability of inundation and expected damage at each point (e.g., Felpeto et al. 2001; Crisci et al. 2008, 2010; Favalli et al. 2009b) and (2) a novel one introduced by Favalli et al. (2009c), which classifies each point on the volcano in terms of the damage expected from vent opening at that point. Such maps can be readily used to create environmental development plans for the area and to assess the hazard and risk posed by lava flows from any specific vent in real-time.

Geological and volcanological framework

Mt. Cameroon belongs to a 1,600-km-long chain of volcanic centers trending NE–SW known as the Cameroon Volcanic Line (CVL; Fig. 1). The CVL comprises an oceanic sector (Pagalu, Sao Tome, Principe, Bioko) and a continental sector (Mt. Etinde, Mt. Cameroon, Manengouba, Bambouto, Oku, etc.). This structural feature has been interpreted as an intraplate shear zone (Regnoult et al. 1986; Déruelle et al. 1987) that has produced a series of horsts and grabens. Mt. Cameroon has been described as a large horst south of the Tombel Graben (Déruelle et al. 1987; Toteu et al. 2004). The oldest eruptive center is the Mandara volcano in the NE, whereas Mt. Cameroon is the only volcano that is definitely active. The other eruptive centers range in age from less than 1 to 31 Ma, with no evident spatial pattern (Marzoli et al. 2000). For a comprehensive description of CVL volcanism, see Fitton (1987).

Mt. Cameroon has built up since 10 Ma through successive mafic volcanic eruptions (Fitton et al. 1983; Suh et al. 2003); this eruptive activity has been particularly intense since 3 Ma (Marzoli et al. 2000). The Mt. Cameroon stratovolcano has a peculiar shape (Figs. 2 and 3); it is markedly asymmetric and elongated in approximately the same direction as the CVL. This shape is thought to partly reflect the presence of an elongated subsurface dike complex that makes up a substantial part of the edifice volume (Annen et al. 2001). The NW and SE flanks show an average slope exceeding 30°, which is exceptionally steep for a basaltic volcano. The profiles of the SW and NE flanks show more gentle slopes (10° to 15°). The overall shape is seemingly more controlled by tectonics than by volcanic activity (Van Ranst et al. 1989; Suh et al. 2003).

Fig. 3

Example of employed Landsat images. False-color composite images were created by assigning bands 5, 4, and 7 to the red, green, and blue channels, respectively. The persistent partial cloud coverage is due to the very humid equatorial climate of the area. Newly emplaced lava flows are more adsorbing than the surroundings and appear as dark areas

Inward facing topographic steps on the relatively flat summit plateau suggest that the volcano summit has subsided (Mathieu et al. 2011). Parasitic scoria cones (∼400 in number) are scattered across the volcanic edifice, although they are concentrated in an hourglass-shaped zone aligned with the major axis of the volcano (Fig. 4). Eruptive fissures (more than 90 have been mapped; Thierry et al. 2006) mainly occur on the SW and NE flanks and in the summit region, trending mainly in the direction of the CVL (Fig. 4). The main rift zone widens out like a fan towards lower altitudes on both the SW and NE flanks (Fig. 4).

Fig. 4

a Distribution of monogenetic scoria cones (red dots) and eruptive fissures (red lines) derived from the results of the GRINP project (Thierry et al. 2006). b Probability density function for future vent opening (see text). The different colors correspond to different probability densities expressed as percent per square kilometer (contour values are reported in the figure)

Recent lava flows at Mt. Cameroon

Seven documented eruptions have occurred at Mt. Cameroon in the last 100 years. Related lava streams invariably formed a number of flow units generating compound lava flow fields (Walker 1971; Njome et al. 2008). The compound nature of the final flow fields is also evident looking at the shape of the lava coverage (Fig. 2). Compositionally, recent lavas range from basanite to alkali-basalts and hawaiites (Cruz Mermy et al. 2006), with Si0₂ ranging 43–48% (Njome et al. 2008). The flow morphology is of $`{\text{a}}`\overline {\text{a}}$ type with minor incidence of pahoehoe, mostly confined in near-vent settings (Njome et al. 2008; Suh et al. 2003, 2011). The porphyricity index varies in the relatively low range of 10–30% (Suh et al. 2003). Table 1 reports some relevant characteristics of the seven documented eruptions at Mt. Cameroon (estimated volume and areal extension of lava flows, eruption duration, etc.).

Table 1 Characteristics of recent lava flows

The general features of recent Mt. Cameroon lavas and lava flows (see also Suh et al. 2011 and references therein for further details) are not dissimilar from those of recent Mt. Etna lavas and lava flows, in terms of shape and extent of lava fields, lava morphology, porphyricity index, etc. (e.g., Polacci and Papale 1999; Calvari et al. 2002; Corsaro et al. 2007; Tarquini and Favalli 2011). The emplacement mechanisms also appear to have been similar, with channelized flows evolving into dispersed flows with digitization, lava tunnels, ephemeral vents, and ultimately generating compound lava fields. The estimated time-averaged discharge rates of lava flows—although they can be estimated in a small number of cases at Mt. Cameroon, see Table 1—are also not dissimilar from those observed at Mt. Etna, ranging from a few up to 30–40 m³/s in most cases.

Input data and methods

Lava flow hazard evaluation and risk assessment, as well as the development of safe environmental policies, rely on the analysis of possible future scenarios. This analysis was performed using the probabilistic DOWNFLOW code cited above. The following describes the input data and calibration required for implementation of the code, and the methods used to define lava flow hazard and risk. Input data include the digital elevation model (DEM; representing the topographic basis for the numerical simulation of lava flow paths), maps of past lava flows (required for code calibration), and maps of exposed values such as human settlements and roads for risk assessment. Code calibration requires the definition of the two model parameters described below and of two probability distributions representing lava flow length and the spatial distribution of vents. Lastly, the adopted expressions for lava flow hazard and risk are presented.

Collection of input data

Digital elevation model

As no high-resolution DEM of Mt. Cameroon is available, we made use of DEMs derived from remote sensing datasets freely available on the web. The 90-m-step DEM derived from the Shuttle Radar Topography Mission (SRTM; http://www2.jpl.nasa.gov/srtm) provides a well-known and assessed dataset (e.g., Wright et al. 2006; Bonne et al. 2008; Grosse et al. 2009; Nkono et al. 2009; Karátson et al. 2010). Previous lava flow hazard evaluation at Nyiragongo volcano, Democratic Republic of Congo (Favalli et al. 2006, 2009a; Chirico et al. 2009), has demonstrated the feasibility of using the SRTM DEM in conjunction with the DOWNFLOW code.

Another global dataset has recently become freely available, namely the 30-m-step DEM derived from ASTER data (GDEM; http://www.gdem.aster.ersdac.or.jp). Compared with the SRTM DEM, the ASTER DEM has not been as thoroughly assessed (Hubbard et al. 2007; Huggel et al. 2008). Analysis of ASTER DEM features in the Mt. Cameroon area revealed several artifacts that make it unsuitable as a topographic base for lava flow simulations (Fig. 5). We therefore adopted the SRTM DEM as the basis for our simulations.

Fig. 5

Comparison between SRTM (a) and ASTER (b) slope maps in the western sector of the volcano. The ASTER image is more resolved but includes several topographic artifacts, which are visible in the figure as abrupt changes in slope. Note the arrow indicating a missing tile in the ASTER image

The SRTM DEM at Mt. Cameroon includes all changes in topography due to recent lava flows, except those due to the 2000 eruption (see Fig. 2) which started on May 28th, about 3 months after the end of the SRTM survey. Although the SRTM DEM has a 90-m spatial resolution, we re-sampled it to a 30-m grid for computational purposes. To ensure consistency in spatial resolution, all other information layers (such as human settlements and roads) were also derived at a spatial resolution of 30 m, with a marked gain in resolution with respect to the default 90 m step.

Mapping of recent lava flows

Owing to the local climate, the weather at Mt. Cameroon is typically cloudy (Fig. 3); this limits the possibility of lava flow mapping from any single satellite image. We therefore used multiple satellite images to complete a comprehensive mapping of recent lava flows emplaced on the summit and flanks of the volcano.

Multispectral satellite imagery can be used to map lava flows emplaced in different years because these show different surface properties with respect to light absorption/reflection/diffusion. Differences are due to a number of factors, including different lava compositions and lava flow surface textures, and the different extent and characteristics of surface weathering and vegetation coverage. The multispectral nature of optical satellite imagery allows discrimination between different lava flows since at least one of the recorded bands usually highlights differences with the surroundings (Lu et al. 2004).

In this study, we used two types of multispectral satellite images, namely Landsat and ASTER images. In the case of the Landsat data, we downloaded five items from the website http://glcf.umiacs.umd.edu (see Table 2 and Fig. 3) as multispectral and (for Landsat 7 ETM+ only) panchromatic images. The Landsat images were acquired in the visible, near-infrared and thermal wavelengths, the latter being sensitive to vegetation and land cover. For the ASTER images we downloaded two items (Table 2) from http://ava.jpl.nasa.gov/ as RGB images.

Table 2 Satellite image dataset used to map lava fields and infrastructures/inhabited areas

All the collected images were geo-referenced to an appropriate UTM projection (UTM-WGS84, zone 32 North) and accurately co-registered to guarantee the best possible precision during subsequent processing. Figure 3 shows Landsat false-color composite images created by assigning bands 5, 4, and 7 to the red, green, and blue channels, respectively. The Landsat 7 multispectral images were merged in resolution (Flynn et al. 2001) with the panchromatic bands to obtain 15-m resolution RGB images. Non-vegetated areas such as those covered by the more recent lava flows appear black or brown, whereas vegetated areas appear green. ASTER images were used to fill the residual gaps due to cloud cover in Landsat images. We used this approach to map the 1909, 1922, 1959, 1982, 1999, and 2000 lava flows plus a few still discernable but undated lava fields on the slopes of the volcano (Figs. 2 and 3).

Mapping of exposed values (human settlements and roads)

The above satellite images were also used to detect urban or inhabited areas and the main roads, which represent the two types of exposed values considered in our risk analysis. We derived two maps of local density (Favalli et al. 2009c) representing, for each pixel, (1) the percentage of the pixel surface covered by inhabited areas and (2) the cumulative length of the roads covered by the pixel. These maps are perfectly co-registered with the computational DEM and have the same pixel size. This allows direct overlay of simulated lava flow paths and exposed values, so that quantitative risk analysis is straightforward.

To identify human settlements in our images, we set up a semi-automatic supervised classification procedure based on Landsat multispectral images. Human settlements of any kind (from urban areas to small villages scattered on the flanks of the volcano) were first detected visually and enclosed in hand-traced polygons (Fig. 6a). We ran a supervised classification in each polygon to identify pixels representing human settlements. For each individual polygon, the algorithm was refined to account for the uneven shading of images. To minimize the effects of cloud cover, we systematically analyzed all available images, adjusting the algorithm for each single image. The outputs from all processed images were then combined in a comprehensive preliminary raster layer which was visually checked and manually refined when necessary. The final result was then converted into a grid collimated with the computational DEM and expresses the fraction of area covered by human settlements (Fig. 6b).

Fig. 6

Example of urban area classification along the coast of Mt. Cameroon. a Polygons roughly enclose the inhabited areas in the eastern sector of the town of Limbe. b Refined (supervised) classification of the previously enclosed areas; each pixel is assigned a value representing the percent of surface classified as inhabited area (increasing from light to dark red)

Main roads were derived from satellite images and systematically checked in all available published maps, pictures and Google Map images of the area (e.g., Cruz Mermy et al. 2006; Thierry et al. 2006; Njome et al. 2008). As for inhabited areas, we expressed the final result in a map collimated with the computational DEM. In this case, each pixel expresses the cumulative length (m) of road(s) crossing the pixel area.

Model calibration

Calibration of DOWNFLOW parameters

In order to probabilistically define the area inundated by a lava flow originating from a given vent p _v , DOWNFLOW computes a number N of steepest descent paths from p _v , each referring to a topography the elevation of which is locally and randomly perturbed within a given range ±Δh. Calibration is therefore required to constrain the two model parameters Δh and N. The first parameter (Δh) defines the maximum (in cross-section) basin which can be inundated by lava flows. The second parameter (N) represents the number of steepest descent path computations performed for each given vent and is proportional to the fraction of the maximum basin which is actually inundated (Favalli et al. 2006, 2009a).

Both Δh and N are constrained by finding the best fit between model results and real lava flow paths. An example of the fit is provided in Fig. 7 for the 1959 eruption at Mt. Cameroon. Considering the relatively poor resolution of the initial 90-m pixel DEM, the overall match in Fig. 7 is satisfactory. Tarquini and Favalli (2011) showed that calibration of the DOWNFLOW code at Mt. Etna is robust, since once calibrated on a small number of flows, additional lava flows are well reproduced without the need of any further calibration, irrespective of the variability of conditions and substrate morphology during lava flow emplacement. A similar robustness of calibration is assumed here, although it cannot be rigorously checked at Mt. Cameroon because there are few mapped lava fields, large uncertainties on the vent geometry and extent (single vs. multiple vents or fissures), and lack of adequate pre-eruptive DEMs for the simulations. The similarities between lava flows at Mt. Etna and Mt. Cameroon, discussed above, support such an assumption. It is worth noting that for the purposes of this study, that is, for quantifying lava flow hazard and risk over the entire volcano flanks, small differences between real and simulated flow paths become irrelevant due to the mediating effects of the tens of thousands of simulations from a corresponding number of potential vents, each associated with a given vent opening probability (cfr Favalli et al. 2009a, and below).

Fig. 7

Comparison between real (white lines) and simulated (colored areas) lava flow paths for the 1959 Mt. Cameroon eruption. Lighter and darker colors indicate, respectively, a higher or lower frequency of simulated lava flow passages. Blue hues indicate areas located at a distance from the vent that is greater than the maximum distance attained by the real flows

Best fit values for Mt. Cameroon are Δh = 2.2 m and N = 10,000. Since there is a clear relationship between Δh and the average thickness of lava flows, it is not surprising that the specific best fit values for Mt. Cameroon and Mt. Etna (Δh = 3 m) are quite similar, whereas the best fit value for the Nyiragongo lava flows (well known for their extreme fluidity and very small thicknesses; e.g., Tazieff 1977; Giordano et al. 2007) is much smaller (Δh = 0.4 m). The best fit value of N is the same in all three cases.

Lava flow length

The DOWNFLOW code does not allow computation of lava flow lengths. The “raw” DOWNFLOW output always reaches the boundary of the computational DEM and must be cut to a given length based on additional information. Wright et al. (2008) used cooling-limited lengths computed from a one-dimensional thermo-rheological model for lava flowing in a channel (Harris and Rowland 2001) to cut DOWNFLOW simulation outputs. Here, we use statistical analysis of real lava flow lengths at Mt. Cameroon to define the probabilistic lengths of the simulated lava flows.

At other volcanoes where the DOWNFLOW code has been applied extensively, e.g., at Mt. Etna, the length of real lava flows is known in a large number of cases (e.g., Guest 1982). On this basis, several probabilistic laws for the expected length of future lava flows have been proposed (Favalli et al. 2009b, c), all of them showing an inverse relationship between flow length and vent elevation (Lopes and Guest 1982). A similar relationship emerges based on a more limited number of available cases at Mt. Nyiragongo (Favalli et al. 2009a) and Mt. Cameroon (Fig. 8). The statistical sample in the figure is almost two orders of magnitude lower than for the well-known case of Mt. Etna. We therefore accounted for the uncertainty and limited spread of points in Fig. 8 by adding a Gaussian spread with σ = 1 km (red hues in Fig. 8) to the linear relationship between vent elevation and lava flow length. The adopted length distribution assumes that lava flows have a high probability of reaching a maximum length, thus our choice is conservative.

Fig. 8

Lava flow length vs. vent elevation. Black dots refer to well-documented and dated lava flows that did not reach the sea (Fig. 2). Gray dots refer to two undated lava flows, also reported in Fig. 2. The dashed straight line is the best fit for the black dots (R ² = 0.93). A 1-km Gaussian spread (red hues) was superimposed on the best fit line to obtain the lava flow length probability distribution employed in the simulations

Vent opening probability distribution

As explained above and illustrated in Fig. 7, each lava flow path simulation from a single vent provides a probabilistic picture of the areas that are susceptible to inundation. These probabilities must be combined with those associated with the vent opening probability distribution. This is usually expressed as a probability density function (PDF; see also Favalli et al. 2009a, 2009b).

The recent GRINP project (Thierry et al. 2006) provides a valuable collection of volcanological data on Mt. Cameroon and represents the basis for the present analysis. We first made an inventory of all elements labeled as “scoria cone” or “eruptive fissure” (Fig. 4a). We then derived separate probability density functions for the scoria cones and the eruptive fissures and merged them to produce the vent opening PDF in Fig. 4b.

For the scoria cones, assuming that the vent opening location is a random variable, we estimated the spatial probability density function in a non-parametric way using a kernel density estimate (Silverman 1986). If $\overrightarrow {{x_1}}$, $\overrightarrow {{x_2}}$, …, ${\overrightarrow x_N}$ ∼ ƒ is an independent and identically distributed sample of a random variable, then the kernel density approximation of its probability density function is

$$ \widehat{f}\left( {\overrightarrow x } \right) = \frac{1}{{Nh}}\sum\limits_{{i = 1}}^N {K\left( {\frac{{\left| {\overrightarrow x - {{\overrightarrow x }_i}} \right|}}{h}} \right)} $$

(1)

where K is the kernel and h is the bandwidth (smoothing parameter). The kernel is taken as the standard 2D Gaussian function with zero mean and unit variance:

$$ K(x) = c{e^{{ - \frac{1}{2}{x^2}}}} $$

(2)

where c is a normalization constant. In our case, ${\overrightarrow x_1}$, ${\overrightarrow x_2}$, …, ${\overrightarrow x_N}$ represent the locations of scoria cones (Fig. 4a) and $\widehat{f}\left( {\overrightarrow x } \right)$ is the estimate of the vent opening probability density function. The bandwidth h controls the degree of smoothing and can be fixed or adaptive. An adaptive bandwidth allows a balanced result when the local density of objects (scoria cones) varies substantially in the studied area. In our case, an adaptive bandwidth was defined as two thirds of the radius of the circle centered on the vent and enclosing the nearest six neighboring vents. This choice was found to provide a PDF which is adequately smoothed but effectively preserves the local density maxima (Fig. 4b).

Equations 1 and 2 were used again in the case of eruptive fissures; this time, $|\overrightarrow x - {\overrightarrow x_i}|$ represented the distance between $\overrightarrow x$ and the fracture. Each fracture was assigned the same weight regardless of its length.

In merging the two PDFs for scoria cones and eruptive fissures, their relative weight was obviously relevant. We have here adopted the simplest assumption of equal weights, implying that future lava flows may equally stem from monogenic vents that produce scoria cones or from eruptive fissures. This assumption is unlikely to be a critical one because both scoria cones and eruptive fissures display a similar general distribution (Fig. 4a).

Lava flow hazard and risk formulation

After calibration of DOWNFLOW, we created a database of simulations which adequately covers possible future scenarios. A 180-m spaced grid of computational vents was defined for the whole area of the volcano (area enclosed by the green line in Fig. 2) after having verified in preliminary tests that such vent spacing ensures an optimum compromise between accuracy and required computational time. This led to the definition of about 80,000 computational vents and 8 × 10⁸ computed steepest descent paths. The results of simulations were stored in a Geographic Information System which allows the straightforward handling of data.

We compiled hazard and risk maps by combining the database of DOWNFLOW simulations with the input datasets, i.e., the vent opening PDF (Fig. 4b), the estimated lava flow lengths (Fig. 8), and the distribution of inhabited areas and roads (example in Fig. 6b). These maps represent the constrained hazard and risk from lava flow inundation at Mt. Cameroon, i.e., probabilities upon the occurrence of a lava flow eruption.

Absolute probabilities, i.e., probabilities within a reference time frame, are not considered here. These could be taken into account by introducing a multiplication factor depending on the lava flow recurrence time (Kauahikaua et al. 1995; Favalli et al. 2009c) in the expressions for hazard and risk (Eqs. 3 to 5 below). In the present case, there is an exceedingly large uncertainty in the recurrence time of lava flows due to the small number of dated flows. We therefore limited our investigation to constrained hazard and risk, as defined above.

Hazard map

The hazard at any pixel i in the computational DEM is here defined as the probability H _i that the pixel is inundated by a future lava flow:

$$ {H_i} = \sum\nolimits_j {{\rho_{{Vj}}}\Delta x\Delta y \cdot {P_{{ij}}} \cdot {P_{{Lij}}}} $$

(3)

where the sum extends over all possible vent locations j with coordinates x _j , y _j , Δx, and Δy define pixel size, ρ _Vj is the probability density function for vent opening at j as defined in Eq. 1 (including both scoria cones and fissures), P _ij takes the value of 1 or 0 according to whether the pixel i is or is not, respectively, inundated by a lava flow originating from pixel j (irrespective of the distance between i and j), and P _Lij is the probability that a lava flow originating from j reaches i along the calculated flow path.

Risk map

The risk from lava flow inundation is computed by assuming that the vulnerability of infrastructures (either buildings or roads) is 1, i.e., assuming total loss of infrastructures upon impact by lava flows. The risk R _i at each pixel i is given in dimensionless units by:

$$ {R_i} = {\nu_i} \cdot {H_i} $$

(4)

where ν _i is the fraction of pixel i that is occupied by infrastructures, and H _i is the hazard at pixel i defined in Eq. 3.

Vent location impact

Standard hazard and risk maps (Eqs. 3 and 4) rank areas according to the probability of inundation by lava flows and expected damage at each location. While this information is very useful for long-term planning, it is almost useless when dealing with an ongoing eruption. At the onset of an eruption the vent opening PDF no longer provides a good representation of the expected scenario; this is because the vent opening PDF (ρ _Vj in Eq. 3) becomes zero everywhere except at the known vent location. In this case the relevant information is the expected damage from that specific vent.

Once a vent opens at a generic position j it is possible to quantify the damage D _j at all pixels i using the following expression:

$$ {D_j} = \sum\nolimits_i {{\nu_i}\Delta x\Delta y \cdot {P_{{ij}}} \cdot {P_{{Lij}}}} $$

(5)

where the sum is now over all the pixels i of the computational domain, and the symbols are defined as in Eqs. 3 and 4. The product ${P_{{ij}}} \cdot {P_{{Lij}}}$ ensures that the sum involves values different from zero only for pixels i that can be inundated by a lava flow originating from a vent in j. The impact D _j can be assessed for different kinds of exposed values ν _i . Below are maps of the impact on inhabited areas (in which D _j is expressed as square meters of inhabited land invaded by the lava flow) and on roads (in which D _j is expressed as linear meters of road reached by the lava). Note that, according to Eq. 5, a similar value of D may involve little damage with high probability or great damage with low probability.

Results

Hazard map

The resulting lava flow hazard map (Fig. 9) shows two fan-shaped areas of highest probability on the SW and NE flanks of the volcano. The axis of these downslope-spreading high hazard areas coincides with the direction of the CVL and with the elongation of the volcano edifice. These two high hazard areas are well separated by a central zone with low probability of inundation by lava flows extending outward from the summit of the volcano along the shorter (NW–SE) axis of the volcano edifice. Two major shield effects are noticeable. The most relevant one is due to the presence of Mt. Etinde (see Figs. 2 and 6), which protects part of the inhabited area along the coast from lava flows. The second one is linked to the presence of the Elephant Opening Valley on the NW flank (see Fig. 2), which reduces hazard in the areas downhill by locally diverting lava flows.

Fig. 9

Lava flow hazard map of Mt. Cameroon volcano. Cross-hatched areas indicate towns and villages

In general, the probability of lava flow inundation is higher in the inhabited centers along the coast than in the villages around the volcano and far from the coast. This is most clearly seen when comparing the situation in the two largest towns in the region: Limbe, along the coast, is characterized by generally high hazard, whereas Buea, just southeast of the summit, is characterized by low hazard despite its vicinity to the summit. A wide strip of high and locally very high lava flow hazard extends towards the coast parallel to the lava flows of 1982 and 1999; maximum hazard values are attained along the coastline, just west of Mt. Etinde between the villages of Bakingili and Batoke. High to intermediate-high hazard also characterizes the northeastern flanks of the volcano, decreasing to intermediate-low, low, or very low at greater distances, where most of the villages in the area are located.

Risk map

The lava flow risk map (Fig. 10) reflects the specific distribution of hazard in Fig. 9, with striking differences in risk between the two largest towns of Buea on the eastern flank (very low to negligible risk) and Limbe on the Atlantic coast (high to intermediate-high risk). It is noteworthy that the oil refinery of Sonata in the western suburbs of Limbe lies in an area of significant risk, and may therefore represent an additional concern in the case of lava flow inundation of the area.

Fig. 10

Risk map for urban areas at Mt. Cameroon volcano. The risk is expressed in dimensionless units as the probability of inundation multiplied by the fraction of surface which is inhabited (see Eq. 4). Colored areas define non-zero risk. Dotted lines are contour lines of the vent opening probability distribution in Fig. 4b (numbers indicate the percentage per square km)

The belt of villages surrounding the volcano edifice shows generally intermediate risk on the northern and western flanks (e.g., Ikata, Bafia, Munyenge, and Bova) and low risk on the eastern flanks (Tiko, Mutengene, Muea, and Muyuka), except at Ekona where the risk is locally intermediate.

Impact of vent location

Figures 11 and 12 show the impact of vent location on human settlements and roads, respectively. The colored areas in Fig. 11 (damage to human settlements) comprise all the locations where vent opening corresponds to different amounts of expected damage, as indicated in the legend. Vent opening in the gray (not colored) areas is not expected to cause any damage to infrastructures, whereas vent opening in, e.g., blue areas is expected to result in substantial damage (corresponding to the destruction of 2–5 million m² of human settlements). When a new vent originates, the map in Fig. 11 can be consulted to quickly assess the possible threat to towns and villages from the associated lava flows. Similarly, the map in Fig. 12 can provide an overview of the expected damage to roads in terms of km of roads destroyed. Venting in regions uphill of the dashed yellow line in this figure have a less than 5% probability of impacting (and interrupting) a road, whereas venting downhill of the solid yellow line has more than 95% probability of disrupting a road.

Fig. 11

Classification of areas at Mt. Cameroon in terms of the expected impact on towns and villages associated with vent opening in the area. Note that this classification does not depend on the local vent opening probability distribution (dotted contours). The impact is expressed as the square meters of human settlements that are expected to be inundated by the lava. Colored areas define non-zero expected impact. Dotted lines are contour lines of the vent opening probability distribution in Fig. 4b (numbers indicate the percentage per square km)

Fig. 12

Classification of areas at Mt. Cameroon in terms of the expected impact on roads associated with vent opening in the area. Note that this classification does not depend on the local vent opening probability distribution (solid black contours). The impact is assessed as the linear meters of roads that are expected to be inundated by the lava. Colored areas define non-zero expected impact. The yellow dotted line identifies the area on the summit of the volcano where vent opening results in a less than 5% probability of reaching a road. Venting from areas at elevations below the solid yellow line results in a more than 95% probability of reaching a road

Equation 5 can also be used to define areas where venting would cause damage to a specified target (a village or town, a strategic infrastructure, etc.). In this case, the quantity ν _i represents only the specified target. As an example, Figure 13 shows the areas where venting would pose a threat to the towns of Buea and Limbe. In this case the colored areas identify regions where venting would pose a direct threat to the two towns, with the different colors indicating the amount of damage expected from any single vent. The white contours in the figure represent the simple hydrological catchment for each of the two towns; clearly, water catchments defined through steepest descent path calculations cannot adequately represent lava flow catchment areas (Favalli et al. 2009c) because lava can fill in topographic depressions and overcome obstacles (accounted for in the DOWNFLOW code).

Fig. 13

Classification of areas in which venting would pose a threat to the towns of Limbe and Buea (cross-hatched areas). All possible vents are ranked according to the expected impact of the lava flow, expressed as the square meters of human settlements that are expected to be inundated by the lava. The white contours enclose the hydrological catchments of the two towns. Colored areas define non-zero expected impact

Discussion

The present study provides a global high-resolution hazard and risk analysis of lava flow inundation at Mt. Cameroon volcano based on the DOWNFLOW probabilistic lava flow simulation code.

The hazard map is critically dependent on the input probability distribution of vent openings, the lava flow length distribution and the characteristics of the topography. Of these factors, the lava flow length distribution is the major source of uncertainty, since few of the lava flows available from the literature or satellite images are useful for length statistics (8 in total, see Fig. 8). Bonne et al. (2008) reported a statistical analysis of lava flow lengths comprising 42 real cases, most of which were not available to us. This analysis, reported in Fig. 14, shows a significant number of flows at <6 km long, which are instead poorly represented in Fig. 8. Note that the two flow length statistics in Fig. 8 and in Bonne et al. (2008) are not conflicting. When the probability distribution for lava flow lengths as a function of vent altitude (Fig. 8) is combined with the vent opening probability distribution as a function of altitude (obtained from the distribution in Fig. 4b), a distribution similar to that of Bonne et al. (2008) emerges (Fig. 14). This means that the probability distributions in Figs. 4b and 8 are fully consistent with the observations summarized in Fig. 14.

Fig. 14

Comparison between the expected frequency of lava flow lengths (line derived by combining the vent opening probability density function in Fig. 4 and the lava flow length vs. vent elevation relationship in Fig. 8) and the observed lava flow length frequency from Bonne et al. (2008) (histogram)

The vent opening probability distribution is assessed with greater confidence because a large number of structural features (scoria cones and eruptive fissures) define a clear trend on the volcano (Fig. 4). Lastly, the SRTM DEM with 90-m pixel resolution over which the DOWNFLOW simulations were performed has been found to provide sufficient topographic detail for lava flow path reproduction and prediction, as in previous applications at other volcanoes (Favalli et al. 2009a). These three elements—the lava flow length distribution, vent opening probability distribution and DEM—are the only inputs required by DOWNFLOW to construct the lava flow hazard map in Fig. 9. The probabilistic character of each single lava flow path predicted by DOWNFLOW, the superposition and statistical processing of a very large number of path predictions, and the use of a probabilistic description of lava flow lengths help decrease uncertainties and improve the reliability of the lava flow hazard map in Fig. 9 and therefore of the risk maps in Figs. 10, 11, and 12.

The relatively low hazard in the summit area of Mt. Cameroon (Fig. 9) despite the high probability of vent opening in the same area (Fig. 4) may seem peculiar. Fig. 4b reveals that the probability that a vent opens in the summit area is significant. However, the local convex cone-like topography tends to regularly distribute lava flows into many equally probable paths rather than into preferential channels with high probability as for lower elevations. In other words, although the probability that a summit lava flow occurs is substantial, there is no way of knowing where exactly on the summit the lava will flow: the higher the number of possible lava paths, the lower the probability of each path.

Figure 15 compares the present hazard map with those proposed previously in the literature. The map from Thierry et al. (2008) (Fig. 15a) represents only a very rough approximation, reflecting the more general aim of multi-hazard mapping. The map from Bonne et al. (2008) (Fig. 15b) is instead more detailed, although it does not consider the wide area above an altitude of 1,000 m. The approach used to compile this map was similar to the one illustrated in this paper, but lava flow paths were assumed to coincide with hydrological paths. The authors thus overlooked the fact that lava flows can surmount obstacles and spread over the topography, a factor which was taken into account in the present simulations. The difference is better illustrated in Fig. 13. The solid white lines indicate the hydrological catchment basins of the towns of Buea and Limbe, calculated according to Bonne et al. (2008). The DOWNFLOW simulations employed here show that the true catchment basin for lava flows can be much wider, depending on the details of the topography and the general characteristics of lava flows. The case of Limbe in Fig. 13 clearly illustrates this point.

Fig. 15

Comparison among lava flow hazard maps of Mt. Cameroon. a Thierry et al. (2008). b Bonne et al. (2008); c this work. The dashed white line in (c) corresponds to the boundary of the masked summit area (roughly above an elevation of 1,000 m) in (b)

Comparison between Fig. 15b, c reveals that the general features of lava flow hazard are similar in the areas of overlap between the two maps. Differences (apart from high elevation areas not included in Bonne et al. 2008) are mostly related to an apparent extension of high hazard to lower altitudes southeast and north of the volcano in the map of Bonne et al. (2008). However, as the scale provided in Bonne et al. (2008) is only qualitative, it is impossible to make an exact comparison with our results. In general, the hazard map from this work provides greater detail, discerning differences at smaller spatial scales.

The lava flow hazard and risk maps presented here are valuable tools for medium/long-term land-use planning as well as decision making during emergencies. The standard hazard map in Fig. 9, which classifies areas according to the lava flow inundation probability, is a fundamental medium/long- term planning tool which can be used to make decisions on where to safely develop future infrastructures. The risk map in Fig. 10 classifies the presently inhabited areas according to the probability of damage, indicating where more urgent measures should be taken in order to reduce risk.

In the case of a new or impending eruption, the constraint on the vent opening location makes the above standard hazard and risk maps no longer useful for emergency management. Valuable information can instead be derived from the maps in Figs. 11, 12, and 13. The “reversed” risk maps in Figs. 11 and 12 are useful not only for quickly assessing the possible threat posed by vent opening anywhere on the volcano but also for identifying the “safer” corridors towards which lava flows should be diverted in order to minimize their impact (e.g., through the construction of diversion walls, as illustrated in Chirico et al. 2009). Many of these corridors can be readily identified in the map in Fig. 11, and to a lesser extent, in that in Fig. 12. Details on the areas where damage is expected as a result of venting at each given location can be obtained through the approach illustrated in Fig. 13, where vent positions are classified in terms of damage to specific targets (e.g., a given village, as in the figure, or important infrastructures such as oil refineries, power plants and power lines, airports, major communication lines, etc.).

Conclusions

This study shows that hazard along the coast south of the volcano is high and locally very high. This translates into a comparably high risk of lava flow inundation in the town of Limbe (85,000 inhabitants) and in the many coastal villages (only Batoke is partly shielded by Mt. Etinde). All the villages lying further inland (including Buea, the second largest city) are generally located in zones of low probability of inundation by lava flows. Among these, a higher risk characterizes Bafia and the nearby villages on the lower northeastern slopes of the volcano.

The results of this investigation include “reverse” risk maps, which are particularly useful tools for rapid response during volcano emergencies. Based on such maps, the expected damage can be assessed as soon as a new vent forms, or even in advance if observations constrain the area of possible venting. Reverse risk maps therefore provide operative tools that can be readily used to mitigate hazard and risk long before, just before, and during effusive events.