EGU Day 2 - morning landslide sessions

For this year's EGU General Assembly I intend to only blog on talks that really catch my eye.  This morning there was a fabulous session on Landslide Forecasting with series of great talks.  Of these,  Samuele Segoni and colleagues presented a very interesting paper on a hugely ambitious project to develop a regional landslide warning system for Tuscany based on rainfall thresholds.  The project appears to be extraordinarily successful – it appears to work with very few false alarms or missed forecasts.  However, to do this the area had to be split into 25 warning zones, using c.330 rain gauges.  As most such systems do, the approach uses an intensity – duration power law relationship. Thus, to make these systems work requires a huge infrastructure.  Interestingly, now that it is clear that such a system can work from a technical perspective, the emphasis needs to shift to the societal problems of trying to disseminate warnings effectively, and getting people to react appropriately to them.  That is a real challenge.

The second paper that caught my attention was by Peter Lehmann and Dani Or, looking at precursor events in the initiation of landslides using concepts of self-organised criticality.  Their starting point was that in the home country, Switzerland, 6% of landscape is prone to instability. In 2005 a rainfall event triggered over 1000 landslides, causing damage estimated at over $3 billion.  Essentially they seek to explain landslide initiation by considering processes that provide a cascade effect, as in the sand pile models of criticality.  Here they model the landslide as being controlled by fibres and fibre bundles, which are analogues of the loss of strength of the landslide material.  They showed that precursor events can be observed as weakening and breakage of the bundles occurs, replicating observed behaviour.  It was very neat and very interesting, and provided potentially important insights into failure initiation.

Next up was Nejan Huvaj-Sarihan from Turkey, presenting her doctoral work undertaken at the University of Illinois.  This was an experimental investigation of failure time prediction in landslides, using creep-rupture as the basic concept.  She built a simple direct shear machine to investigate the creep-rupture process, and showed two key things:
1.    She observed creep at even very low shear stresses (<20% peak strength).  She used this to infer that all slopes creep, which is correct;
2.    She could initiate creep rupture failure when factor of safety was greater than one, but that the time to failure depended upon how close to FoS = 1 the system is. 
Using this data she then explored whether failure prediction can be undertaken using the range of techniques available, concluding that it can in all three cases.  This is a very neat study, although I am surprised to see creep-rupture in pre-sheared materials.

During questions someone made the point that there is a “geotechnical disease” to ignore the time element.  This was a very provocative statement, but is quite correct.

Oded Katz and his colleagues followed this up with a model based study that sought to look  material disintegration in controlling the geometry and size of landslides.  Again, this was a neat bit of work that came to some key conclusions.  For me the most important one was that the power law roll-over in landslides is indicative of the change in material properties, and that the collapsing of the power law relationships onto each other occurs because there is such a narrow range of residual strengths available in natural systems.  However, the talk also demonstrated beautifully that failure is associated with breakage of inter-particle bonds.  Initiation of movement in the models occurred only when shear surface was fully developed through bond breakage, which links with the previous three presentations.

Those three talks (Lehmann, Huvaj-Sarihan and Katz) together present an extraordinary level of insight into landslide processes that on their own justifies my attendance at the meeting.

In the after coffee session I would like to highlight just one talk, that of Monique Fort and her colleagues on debris flow initiation in the Ghatte Khola watershed of Nepal.  This watershed suffers extraordinary pre-monsoon rainfall events – she quoted a storm in 1974 that had over 300 mm of rainfall in an hour – can this really be right?  Anyways this small ( 7.8 square kilometre) catchment generates debris flows that result from shallow slides that block the tributary valley, then collapse, creating flows.  These in turn enter the main valley, which is then blocked in turn, and another flow occurs down the main channel.  There are two interesting things here – first, how a small failure can initiate a bigger flow that in turn blocks the main valley, generating an even larger one – who would try to forecast hazards when this sort of situation occurs?  Second, she highlighted the ignorance of road builders in Nepal to these processes, resulting in inappropriate designs that then fail in the next storm.  This is a hobby-horse of mine; I could not agree with Monique more.

Are satellite-based landslide hazard algorithms useful?

In some parts of the world, such as the Seattle area of the USA, wide area landslide warning systems are operated on the basis of rainfall thresholds. These are comparatively simple in essence - basically the combination of short term and long term rainfall that is needed to trigger landslides is determined, often using historical records of landslide events. A critical threshold is determined for the combination of these two rainfall amounts - so for example, it might require 100 mm of rainfall in hours after a dry spell, but 50 mm after a wet period. These threshold rainfall levels have been determined for many areas; indeed, there is even a website dedicated to the thresholds!

In 1997 NASA and JAXA launched a satellite known as TRMM (Tropical Rainfall Monitoring Mission), which uses a suite of sensors to measure rainfall in the tropical regions. Given that it orbits the Earth 16 times per day most tropical areas get pretty good coverage. A few years ago Bob Adler, Yang Hong and their colleagues started to work on the use of TRMM for landslide warnings using a modified version of rainfall thresholds. Most recently, this work has been developed by Dalia Bach Kirschbaum - and we have all watched the development of this project with great interest. The results have now been published in a paper (Kirschbaum et al. 2009) in the EGU journal Natural Hazards and Earth Systems Science - which is great because NHESS is an open access journal, meaning that you can download it for free from here.

Of course a rainfall threshold on its own doesn't tell you enough about the likelihood of a landslide. For example, it doesn't matter how hard it rains, if the area affected is in a flat, lowland plain then a landslide is not going to occur. To overcome this, the team generated a simple susceptibility index based upon weighted, normalised values of slope, soil type, soil texture, elevation, land cover and drainage density. The resulting susceptibility map is shown below, with landslides that occurred in 2003 and 2007 indicated on the map:


A simple rainfall threshold was then applied as shown below:

Thus, if an area is considered to have high landslide susceptibility and to lie above the threshold line shown above based upon an analysis using 3-hour data from TRMM, then a warning can be issued.

Kirschbaum et al. (2009) have analysed the results of their study using the landslide inventory datasets shown in the map above. Great care is needed in the interpretation of these datasets as they are derived primarily from media reports, which of course are heavily biased in many ways. Examination of the map above does show this - look for example at the number of landslide reports for the UK compared with New Zealand. The apparent number is much higher than in NZ, even though the latter is far more landslide prone. However, in New Zealand the population is small, the news media is lower profile, and landslides are an accepted part of life. However, so long as one is aware of these limitations then this is a reasonable starting point for analysing the effectiveness of the technique.

So, how did the technique do? Well, at a first look not so well:



In many cases the technique failed to forecast many of the landslides that actually occurred, whilst it also over-forecasted (i.e. forecasted landslides in areas in which there were none recorded) dramatically. However, one must bear in mind the limitations of the dataset. It is very possible that landslides occurred but were not recorded, so at least to a degree the real results are probably better than the paper indicates. Otherwise, the authors admit that the susceptibility tool is probably far too crude and the rainfall data to imprecise to get the level of precision that is required. However, against this one should note that the algorithm does very well (as indicated by the green pixels on the map above) in some of the key landslide-prone areas - e.g. along the Himalayan Arc, in Java, in SW India, the Philippines, the Rio de Janeiro area, parts of the Caribbean, and the mountains around the Chengdu basin. In places there is marked under-estimation - e.g. in Pakistan, Parts of Europe and N. America. In other places there was dramatic over-estimation, especially in the Amazon Basin, most of India, Central Africa and China.

All of this suggests that the algorithm is not ready for use as an operational landslide warning system. Against that though the approach does show some real promise. I suspect that an improved algorithm for susceptibility would help a great deal (maybe using the World bank Hotspots approach), perhaps together with a threshold that varies according to area (i.e. it is clear that the threshold rainfall for Taiwan is very different to that of the UK). Kirschbaum et al. (2009) have have produced a really interesting piece of work that represents a substantial step along the way. One can only hope that this is developed further and that, in due course, an improved version of TRMM is launched (preferably using a constellation of satellites to give better temporal and spatial coverage). That would of course be a far better use of resource than spending $4,500 million on the James Webb Space Telescope.

Reference
Kirschbaum, D. B., Adler, R., Hong, Y., and Lerner-Lam, A. 2009. Evaluation of a preliminary satellite-based landslide hazard algorithm using global landslide inventories. Natural Hazards and Earth System Science, 9, 673-686.

Young River Landslide, New Zealand

Interesting news from New Zealand, where for the last few months there has been some concern about a landslide dam on South Island. The landslide itself occurred in a fairly remote area of the Southern Alps at 4:40 am on 29th August 2007. The landslide, which is well-described in a GNS Science poster here (there are some great images and some good data on that poster), blocked the river valley to a depth of about 100 metres. This was a big debris slide, with a volume of about 11 million cubic metres and a runout distance of about 1.8 km. Interestingly it does not appear to have created an air blast as large slides of this type often do. This may mean that the mass moved as a series of events closely spaced in time, rather than one big, instantaneous failure.

Over time a lake built up behind it, finally overtopping on 5th October 2007. Understandably, there was concern about the possibility of an outburst flood. In this case there are few human assets at risk downstream, but this is a popular hiking area, potentially putting people on the trails at risk. As a result, the trails were closed by the Department of Conservation whilst the landslide was monitored in real time by GNS Science and Otago Regional Council. Unfortunately, such a closure was not good for the local community as it reduced the number of tourists.

The images below show the landslide from the downslope side. I have annotated it below to show the key features. Note that the deposit has banked up on the right side (from this view). The natural spillway that has formed is just on the lee of this banked up deposit. This deposit is coarse-grained (bouldery), which may well be the reason why the channel has not eroded downwards to release the water.

The image below is from the landslide dam itself looking across the lake. Note the size of the boulders - these are part of the landslide deposit. This does show that if the landslide dam is stable then the resultant lake can in some circumstances become an asset.

This week it was reported here that a decision has now been taken to reopen the trails downstream of the landslide from 1st November (i.e. just before the start of the summer). Unsurprisingly, the local people are rather pleased: "Makarora Residents Association deputy chairman Devon Miller said the closed valley had affected some of the tourist operators in the township, so the decision to re-open was a welcome one."It's a good positive announcement and the community is happy," he said." (Otago Daily Times).

The decision to reopen the trails but to maintain some restrictions in heavy rain, using a new warning system, appears to be sensible. So often landslide management is about balancing risks - i.e. what is the risk of a collapse of the dam affecting someone on a trail compared to the risk to the local communities associated with the loss of tourism, etc. In this case the stability of the dam suggests that this risk has now dropped to close to the residual level. Note that this does not mean that there is no risk - there are hazards associated with spending time in remote mountains. The risk from the landslide dam is no greater, and may be substantially less, than those other risks.

All-in-all the approach taken by the parties involved in New Zealand has been exemplary, in terms of picking up the event in the first place, in terms of the ways that they have monitored it and in terms of the decision-making process to minimise risk.