Wednesday, 21 May 2014

Glacier retreat and slope instability, an example from Mount Kazbek, Georgia

A large landslide close to the Georgia - Russia border this week killed up to eight people, and disrupted construction of a new hydropower diversion tunnel. The landslide was released from approximately 4,100 m on the north-eastern flank of Mount Kazbek, and is estimated to have involved more than 10 mil. m^3 of rock. The site of the landslide release is likely to have been recently deglaciated, and 2010 aerial images from Google Earth indicate there was ongoing instability in the years prior to this failure.  

Aerial image of the landslide release area at 4100 m elevation

In this case the landslide traveled over 10 km downvalley, and the deposit seems to have formed immediately upstream of outlet tunnels for the hydropower diversion. Although this was clearly an energetic event (in particular in terms of the 400 m run-up apparent on the opposite side of the valley), the total fall was almost 3 km, and the travel distance is not unusual for a landslide of this magnitude.

Map of landslide release, travel, and approximate location of the deposits. 
The profile is derived from the white line indicating the path of the rock avalanche. 

While it's terrible to see loss of life as a result of such events, high alpine landslides such as this can provide critical insight into the possible effect of changing climate and glacier retreat in more heavily populated alpine regions. Although minor depressions or crevasses are evident in the glacier surface in the 2010 aerial image, and smaller events have been noted to be regular in the region, such a large event was not predicted. By studying failures such as this one, we can gain important insight into the failure characteristics, and begin to identify additional information required to predict similar events with enough confidence to evacuate communities at risk. As it happens, the diversion tunnel for the Dariali hydropower scheme appears to be well situated to relieve inflows into the lake impounded behind the landslide deposit, and hopefully prevent any further losses downstream.

Interactive map indicating the estimated source (orange), and mapped deposit (grey), as well the aerial photograph of the release area (double click the camera icon to view). The white line denotes the path of the landslide that appears to have descended as a rock avalanche. A .kmz file (including additional aerial photographs) is available as a Google Earth download here

Sunday, 4 May 2014

The Badakshan landslide: A forseeable tragedy?

A recent rainfall-triggered landslide in Ab Barek (sometimes referred to as Abe Bareek) in the Badakhshan Province of Afghanistan is suspected of taking the lives of up to 2100 villagers. As with so many remote events, details are scarce, although a number of images and videos have appeared on social media websites over the last days. Some images, as that below, indicate the landslide may have crossed a river valley and formed a landslide dam blocking the river running through the town. In a region that has recently experienced heavy rainfall, the formation, and breach of such a dam could have severe consequences to any communities downstream.

Possible landslide dam formation in Ab Barek
Source: Matin Bek/Twitter

As discussed in previous posts (and on a recent EGU presentation), we can use these images in combination with freely available GIS platforms such as Google Earth and QGIS to map the true extents of the landslide source and deposit (a task made easier when the locality is correctly reported). Below is an interactive Google Earth frame with the extents we have been able to map from the imagery to date. The Landslide source is indicated in grey, while the approximate extent of the deposit is in orange. Currently, there are no images of the downstream extent of the deposit, however, the flow of the landslide into a moderately-sized river channel, and upstream is clearly evident both in the photographs and the GIS map.

Interactive map indicating the estimated source (grey), and mapped deposit (orange), as well as four photographs covering the area of the damaged village (double click each image or camera icon to view). The white line denotes the region of the slope that has failed to date. Signs of apparent instability on the slope above this region should be investigated by emergency crews, as further instability behind the steep scarp could be possible. A .kmz file (including additional aerial photographs) is available as a Google Earth download here

The satellite image of the pre-failure hillslope contains a number of arcuate ridges, a clear indication the hillslope has been unstable for a number of years. Some sharper, or more well-defined lineations upslope of the region displaying obvious signs of movement are suggestive of a reactivation of the failure prior to the 2004 DigitalGlobe image. As a natural hazard researcher it's saddening that these signs were not recognised, as such a steep active slope above a populated area should have raised red flags, and at least allowed authorities to educate villagers and encourage the implementation of a monitoring system.

A Google Earth image of the landslide source and visible deposit. Note the lumpy shapes on the hillslope in the region of the landslide source, tell-tale signs that the slope has been moving in the past. Calculated extents of a future landslide-dammed lake are indicated in dark blue (likely) and light blue (possible).

We can begin making a preliminary analysis of hazards posed from the dam formation and breach by importing the .kml file containing source and deposit extents into QGIS, and using freely available geodata to investigate the catchment characteristics and recent rainfall record. Satellite-based terrain data is notoriously poor near the bottom of valleys, however, the upper limit of the deposit appears to be at an elevation of approximately 1810 m above sea level. This is 10 m above the indicated level of the former river at the same location, although photographs indicate the deposit may be more than 20 m deep in places. Using the 1810 m level as a guide we can plot the approximate extent of a landslide-dammed lake (dark blue in the above image), which could possibly cover an area of 0.04 km^2 and contain 200,000 m^3 of water. An upper estimate of the lake volume can be made by following the 1830 m contour (assuming the dam extents to 30 m elevation at it's crest. In this case the lake could cover 0.26 km^2, and hold 3.9 Mil. m^3 of water. Although these numbers are only rough estimates, the apparently fine-grained nature of the sediments forming the dam may aid both initial water retention, and rapid failure by piping through the dam or scour once the dam is overtopped. The catchment upstream of the dam is approximately 18 km^2, and TRMM data indicates it received ~100mm of rainfall during the 7 days leading up to May 4th. Concentrated, this would total 18 Mil. m^3 of water, more than four times the maximum storage capacity of the potential Ab Barek dam. Although much will have run off during the rainstorm events, it seems likely that the water presently stored in the catchment is more than enough to overtop the dam, and resulting flooding may pose a serious risk, both to residents of Ab Barek, and those in towns such as Balas Shemar and Ahen Jalow downstream of the landslide.

7 days accumulated rainfall in Badakhshan Province

Update 05/05/14: An error in the calculation of surface area from a geographic coordinate system (WGS84) in QGIS mean that initial area estimates were too large. These have now been corrected. The landslide source area has been mapped from video footage.

Wednesday, 27 November 2013

Cartoon contributions to modelling real world physics

Movie physics

Numerical modelling is an important tool for most natural hazard researchers. Excellent codes now allow scientists and engineering practitioners the opportunity to simulate natural processes in both static and dynamic states. These codes typically grow from state-funded natural hazard research initiatives, and are principally aimed at understanding the 'bigger picture' in order to evaluate natural processes at the scale that poses a hazard to people and property.

Every now and then, however, it's worthwhile remembering the hazard researcher's not-so-distant, though often wealthier cousins in animation research. Although often dealing with smaller scale problems, the requirement for animators to come up with realistic-looking dynamic models for movies and computer games is currently driving a large amount of commercial research in scalable physics models. Physics-Based Animation is useful blog detailing recent advances in simulating physics for human visual consumption. This week Gizmodo featured work from the new movie Frozen, including an impressive Material Point Method snow simulation, applying real physical properties (compressibility, tensile strength, strain hardening, density, Young's modulus, and Poisson's ratio) to model snow as a dynamic, compressible granular material.

The models presented in the above video contain between 4x10^6 (snowplow) and 7x10^6 (rolling snowball) particles, sufficient for the evaluation of small to medium-scale rock or soil slope stability hazards, or possibly the design of protection structures which may interact with debris flows or snow avalanches. These simulations commonly use approximate solutions to replicate complex physical behaviors, however, as demonstrated in the above example, 'expert knowledge' (everyone's seen a snow plow in action) can provide good verification for such visual models.

As well as simulating situations with extremely large strains, models developed for animation purposes are also able to reproduce complex fracture behaviours, and examples such as that below may make their way into geohazard studies investigating, for example, particle fragmentation during rock avalanche, or rock slope failure as a result of earthquake shaking.

Blender - the free, open source alternative

Although the commercial nature of these codes often means the software itself is wrapped up in propriety licences, the maths and physics behind the above simulations are published in scientific journals (here and here). Blender is a free and open source 3D 'creation pipeline' (physics-based animation software) designed with fluid, rigid body physics, and particle tracking simulations in mind.

User-developed plugins and tutorials provide the opportunity for enthusiasts (or practitioners from generally unrelated fields) the opportunity to access these advanced simulation techniques, and drawing inspiration from commercial software, are often not far behind the state-of-the-art. Although limitations currently exist in terms of particle numbers and physical calibration, mean that animation software is currently no match for specifically designed and calibrated geohazard simulation software such as RAMMS, these are not insurmountable obstacles. As demonstrated by the simulation below, projects such as Blender may soon provide some exciting opportunities for research and mitigation of dynamic gravitational hazards, and at least so far, provides useful insight which may improve snowman mortality in the upcoming winter.  

Friday, 20 September 2013

Location and volume of the La Pintada landslide

Heavy rainfall from hurricane Manuel initiated a large landslide which struck the village of La Pintada (Mexico) this Monday. 58 people have been reported missing in the small village, and at least 20 buildings have been destroyed. Although helicopters have rescued most of the villagers, 45 people remain, and authorities are concerned that "The rest of the hill could fall.". It is almost inevitable that such a large landslide in the narrow valley will cause major difficulties for rescue crews in the coming days, and access into much of the remote coffee growing region is likely to be challenging in the wake of the storm.

An aerial view of the La Pintada landslide
Source: Amador Narcia/Twitter

The above image is one of the first released of the region, and clearly conveys the devastating effect of the landslide. Getting information out to concerned parties can be critically important in such situations, although an accurate sense of scale and location data for early social media images is rarely available. Using the 'Add Photo' dialogue in Google Earth, however, we can use media images to locate, map, and make early decisions regarding ongoing risks and emergency response very soon after disastrous events. Using the Google Earth API we are able to approximately locate the position from which the above photograph was taken, as well as the position of an image which seems to provide an overview of the deposit area prior to the failure. The interactive map below represents an initial interpretation of the spatial extent of the landslide deposit (outlined in brown), as well as an initial guess as to what may be the source of the failure (outlined in grey).

Interactive map indicating the estimated source (grey), and mapped deposit (brown), as well as two photographs covering the area of the damaged village (double click to view). An updated .kmz file (including additional aerial photographs and and revised map of the scarp from 21/09/13) is available as a Google Earth download here

As is evident above, the landslide initiated on a partially vegetated slope to the north of the village, and as reported, appears to have buried much of the eastern side of the community. Survivors report a loud rumbling, and very rapid failure, which is evident from the map as at least one building seems to have been pushed around 30 m from it's foundations (white arrow).  A depression formed at the back of the deposit (adjacent to the hillslope) indicates the energetic landslide mass probably scoured sediment out from the toe of the slope, and did not stop moving until the whole mass reached the valley floor. Calculation of the landslide volume using a combination of the measurement tools in Google Earth and an estimation of the deposit thickness from the aerial photograph suggests approximately 100,000 m^3 of debris remains in the village, although the failure may have been larger as there has been some erosion of the deposit adjacent to the river channel. 

An image taken looking east, across the village prior to the landslide. The mass travelled from left to right, through the buildings in the farground.
Source: gatopelu008\Panoramio

While it is not immediately apparent why this particular slope failed during the storm, there appears to be some evidence for planar failures on similarly oriented slopes elsewhere in the valley. Judging from the weathered appearance of the landslide deposit, it seems likely that bedrock degradation has been ongoing as part of the landscape's natural weathering and erosion cycle. As in many sub-tropical regions, it is possible that either historical, or recent, removal of vegetation led to a local acceleration of these natural processes, decreasing the stability of slope materials, and may have increased the chance of failure during the extreme environmental event.

Map of La Pintada landslide deposit (brown) and assumed source area (grey) derived from the above Twitter photograph

Although there is definitely room to streamline the process of mapping from media photographs, combining Google Earth with social media in this case provides important quantitative insights into the landslide occurrence in a relatively short period of time (~1 hr). Given the limited resources available in many remote or undeveloped regions, these technical aspects are often not addressed, and even more rarely published. Expanding on methods such as those presented here could therefore be a useful means for both local authorities, and the wider research community to investigate and document natural disasters in the future.