12 December 2009
On the flight to San Francisco yesterday I spent some time preparing for one of my AGU presentations – the one on Friday on the impact of landslides on society. I have been looking at my landslide fatality database, which now stretches back for over seven years. There is still some way to go with this to really understand long term trends in fatal landslides, but the dataset is now big enough to do some quite interesting things.
I have divided each year into five-day blocks (we often call one of these blocks a bin), and then taken the average number of recorded fatal landslides within that bin over the seven year period. So, bin number one is 1st to 5th January, the second is 6th to 10th, etc. I have then looked at the cycle through time by plotting a graph in which I have smoothed the data using a 25 day filter – this is a noisy dataset, so this is needed given the comparatively short window:
The resulting graph is pretty interesting I think. First, the peak in aggregate recorded fatal landslides (the black line) clearly occurs in the northern hemisphere summer, the minimum is around about now. The peak is actually on about 25th July. This is coincident with the peak of the SW Monsoon over the Indian subcontinent. Notice though that the graph is asymmetric – i.e. it rises to a peak more quickly (about 100 days) than it then declines (about 150 days). I assume that this is because in this post-SW monsoon peak the influence of the monsoon in East Asia and of tropical cyclone landfalls becomes significant.
The minimum period coincides I think with the onset of winter in the northern hemisphere (which is a dry period for many of the most landslide prone areas) but is before the rainy season really gets going in SE Asia. By early January the rains in for example Indonesia are really under way, and the occurrence of landslides increases.
The standard deviation is a measure of variability between years. So, if for a specific bin the number of fatal landslides was always three then the standard deviation would be low. If however one year there were none, the next six, the next ten and the next two then the standard deviation would be much higher. It is interesting that as the average number of fatal landslides increases in the N. Hemisphere summer so does the standard deviation – this is to be expected. However, in the post-peak period the standard deviation remains high for a while before declining. I think that this probably reflects the influence of tropical cyclones in this period, which tend to landfall rather sporadically but then to cause many landslides over a small area. Over the seven year period many of the bins in this period have been affected by a tropical cyclone.
I hope to see you at the session!