APPENDIX G – LANDSLIDES AND RAINFALL: CASE STUDIES G.1 AUSTRALIA G.2 HONG KONG SAR G.3 ITALY G.4 JAMAICA G.5 NEPAL G.6 NORWAY G.7 SINGAPORE G.8 SLOVENIA G.9 SWITZERLAND G.10 UNITED KINGDOM G.11 UNITED STATES OF AMERICA G.12 OTHER REGIONS AND COUNTRIES
APPENDIX G – LANDSLIDES AND RAINFALL: CASE STUDIES
by M G Winter, I M Nettleton and J A Parsons
Systems to forecast conditions likely to lead to debris flows have been developed for many regions of the world. In this section a selection of case studies is presented. These have been selected to illustrate specific points and on the basis that information on them is relatively easily available.
Flentje and Chowdhury (2006) describe an observational approach to continuous real time monitoring of landslides in the Wollongong city area. Their work encompasses the monitoring of individual slopes, for which the development of pore water pressures and mass movement are related to site-specific measured rainfall. In addition, five stations measuring rainfall, among other parameters, have been established within the Wollongong area (approximately 25km by 15km) to enable alerts to be broadcast in response to rainfall events likely to lead to landslides.
Flentje and Chowdury (2006) represent the intensity, frequency and duration (IFD) of the local rainfall record and they compare this to the threshold proposed by Caine (1980), and reported here as Equation 10.1A, in Figure G.1.
Figure G.1 – Rainfall intensity, frequency and duration analysis of the historical record for a rainfall station in the Wollongong area (from Flentje and Chowdhury, 2006).
Flentje and Chowdhury (2006) have developed both site-specific and regional rainfall triggering thresholds, primarily for deep-seated landslides. The site-specific data is of lesser interest in the current context, but the regional threshold is of considerable interest. Their work involved the spatial and temporal distribution of rainfall that occurred during and prior to an extreme event during August 1998. Data from a total of 147 rainfall stations (including 36 pluviometers) within the region have been analysed and interpolated to give the cumulative rainfall at each landslide location.
The spatial distributions of cumulative rainfall over different antecedent time periods were analysed. The antecedent time periods of six hours and 12 hours prior to 0700 hours on 7 August and 1, 3, 5, 7, 30, 60, 90 and 120 days prior to 0900 hours on the 17, 18 and 19 August were considered in various analyses. Figure G.2 shows the rainfall intensity-durations for each antecedent rainfall period as a series of 142 data points making up each of a series of vertical columns of data points – each vertical column represents one antecedent period and each landslide recorded is represented by one data point in each vertical column.
Figure G.2 – The lower bound intensity-duration ‘regional landslide triggering rainfall threshold’ for the city of Wollongong during the extreme August 1998 event (from Flentje and Chowdhury, 2006).
The red curve extending across the graph near the base of each vertical column of data points represents the lower bound intensity-duration ‘regional rainfall threshold’ for the city of Wollongong for the August 1998 event. The authors emphasise that their threshold is for the Wollongong area and may not be applicable to other areas. In particular their work considers the particular morphology of the flows and slides that are experienced in their areas. They also emphasise that the regional threshold may be at significant variance with that for individual landslide sites.
They state that the shorter duration thresholds (six hour to three days for their study area) are most relevant for shallow debris flow and that this is supported by the work of Wieczorek (1987) amongst others. However, this may be seen as something of a simplification as Wieczorek actually states that 28cm of antecedent rainfall was required before debris flows would be triggered.
Leventhal and Walker (2005) also note that rainfall is a key trigger factor in the Australian Geomechanics Society (AGS, 2000) system for landslide risk assessment and management.
This case study summarises the methods of collating rainfall data used in Hong Kong and how rainfall data has been used to develop a landslide warning system. The Geotechnical Engineering Office (GEO) has been particularly active in investigating the distribution, nature and probable causes of landslide occurrences in natural terrain, and in assessing the hazards of such events. To describe all of their studies and how they interact is however considered outwith the remit of this case study. Instead, only those studies which directly impact on the rainfall correlation with landslides are described. The historic papers collected into a volume to commemorate 30 years of slope safety practice in Hong Kong together provide a comprehensive account of the work undertaken in that part of the world (Anon, 2007b).
While much of this section discussed the rainfall threshold work in Hong Kong for natural terrain landslides, it is important to note that as yet the landslip warning system used in Hong Kong takes into account only the rainfall threshold analyses of man-made slopes.
G.2.1 Key Dates
Early-1970s: Two man-made slopes collapsed on 18 June, killing a total of 138 people: 71 at the Sua Ma Ping Estate in Kowloon (Figure G.3) and 67 at Po Shan Road, in Mid-Levels on Hong Kong Island (Figure G.4). In 1976, another failure at Sau Mau Ping killed a further 18 people.
Figure G.3 – Sua Ma Ping Estate landslide, 18 June 1972. A 40m high road embankment collapsed after 232mm of rain.
1977: The Geotechnical Control Office (now called the Geotechnical Engineering Office, GEO) was formed and development of the Landslip Preventative Measures (LPM) Programme begun. The LPM programme was developed to inspect, and produce an inventory of, all man-made slopes and retaining structures in Hong Kong, although ‘special projects’ involving natural terrain studies were also undertaken. Substandard slopes were, and to some extent are still being, systematically upgraded to progressively reduce the landslide risk from man-made slopes which affect the community, whilst also trying to improve aesthetics of the slope. A major exercise to ‘educate the public’ of the dangers of landslides was also begun.
Mid-1979: GEO undertook a mapping exercise of Hong Kong to identify areas of colluvium, which led to a systematic terrain classification based on i) slope gradient; ii) terrain component (hillcrest, foot slope, side slope, etc); and iii) erosion and instability. This later became known as the Geotechnical Area Studies Programme, GASP.
Figure G.4 – Po Shan Road Landslide, Mid-Levels, Hong Kong Island 18 June 1972. The landslide occurred on a steep hillside above a temporary excavation and demolished a 12-storey building.
Early-1980s: GEO started collecting and reviewing data and producing annual reports (since 1984) of rainfall and landslides in Hong Kong. Rainfall gauge coverage significantly improves from 1985, and in addition rainfall data also began to be collected at five minute intervals (throughout the year).
1994: GEO started using consultants to design and supervise construction of LPM, and to investigate and report on selected landslides.
1995: GEO commenced the Natural Terrain Landslide Study (NTLS) (Evans et al., 1999; Ng et al., 2003). This formed part of a series of integrated studies to investigate the distribution, nature and probable causes of natural terrain landslides and to assess the hazard from such events. Phase 1 produced the Natural Terrain Landslide Inventory (NTLI) (King, 1999) from a review of high level aerial photographs taken between 1945 and 1994. Phase 2 used GIS to examine the spatial distribution of landslides with respect to geology, slope angle, geomorphology, vegetation and slope aspect, etc, to determine causal factors and a preliminary assessment of hazard. Of these, geology and slope angle were found to be most important in determining natural terrain landslide susceptibility at a regional scale. Phase 3 produced regional natural terrain landslide susceptibility and hazard maps, together with detailed studies of some areas with a high incidence of landsliding that are close to existing or proposed developments. Phase 3 also developed procedures for the hazard and risk assessment of natural terrain in Hong Kong, the investigation of hydrological and hydrogeological influences on landslide susceptibility and the continued study into the nature, occurrence and frequency of exceptionally large natural terrain landslides.
The Slope Safety Technical Review Board was then established. This comprises a panel of three to four renowned experts who interact extensively with GEO, reviewing and advising on various aspects of slope engineering.
1999: The National Landslide Inventory (NTLI) was formed (King, 1999) containing information on more than 26,700 landslides on natural terrain. The associated ‘Landslide Investigation’ methodology was developed (in conjunction with Professor Norbert Morgenstern of the University of Alberta in Canada).
2000: Landslide Investigations became part of the LPM programme.
2004: The Landslide Potential Index was developed – this measures the relative severity of a rainstorm relative to its potential to cause landslides.
G.2.2 Existing Rain Gauge Network in Hong Kong
Rain gauge networks are operated by four separate bodies including the Hong Kong Observatory (HKO), the Water Supplies Department, the GEO of the Civil Engineering Department, and the Drainage Services Department (DSD) of the HKSAR Government. Altogether these four departments are responsible for operating and maintaining more than 200 (as of 2001) of the rainfall, tidal and hydrological gauging stations in the territory.
The rain gauge stations are automatic telemetric stations that transmit data at five minute intervals throughout the year, during both the wet and dry seasons. Telemetric readers in Hong Kong are generally powered from mains electricity as the majority are in built up areas. However, each station has a 72 hour backup battery power in case of a supply failure and a number are now self-powered through solar power and wind power. In Hong Kong it is also necessary to protect the equipment from extremes in temperature (80°C in summer inside equipment cases) and humidity (95% relative humidity).
A typical Hong Kong gauging station may contain the following equipment:
i) Data logger.
ii) Rainfall gauge.
iii) Telemetry connection.
iv) Incoming power supply.
v) Backup power for at least three days.
vi) Lightning protection system (unlikely to be required for the Scottish situation)
vii) Ventilation fan controlled by thermostat (again unlikely to be required for the Scottish situation).
Examining items (i) and (ii) above in more detail:
i) Data Logger – this would tend to be a programmable logic controller (PLC) or a remote terminal unit (RTU). The PLC is easier to install, programme and support but the RTU has superior communication capability, more memory, and is normally designed for extremes in temperature and humidity. The information is transmitted to a central location, generally a PC, via data link/dial up or via wireless transmission. If there is a break in the transmission or an equipment problem, the stored data will be automatically re-transmitted to the office in the next available transmission. Stored data can also be retrieved from site at any time. The data are all in text (ASCII format) for easy transmission and reading. Once the text is received, it is saved in a database such as Oracle or MS SQL. The latter is preferred as it is compatible with Excel™ from which graphical representations (e.g. bar charts) of the rainfall readings can be produced.
ii) Rain Gauge – this would tend to be a Casella tipping bucket, which tips when the rainfall depth reaches 0.5mm. A 0.2mm tipping bucket may be more suitable for a non-tropical (Scottish) situation. In locating a rain gauge the following rules of thumb are observed:
- The rain gauge should be positioned on a reasonably level and flat surface.
- There should be no obstructions in the vicinity. Normally, the height of any object should be less than 1/4 to 1/3 of the horizontal distance from the bucket.
- The rain gauge should be positioned to avoid tall buildings and trees as these can cause eddies which may affect the amount of rain collected.
- Areas that may be susceptible to flooding should be avoided.
- The rain gauge should be positioned in an area where the discharge water from the gauge can drain away quickly.
G.2.3 Determination of a Rainfall Threshold
Initially, correlations of rainfall intensity with landslide activity in Hong Kong concentrated on failures of man-made slopes, as these are incidents that tend to affect developed areas and are therefore reported. There is general agreement that it is possible to define rainfall threshold above which failures of man-made slopes increase in frequency (Lumb, 1975; Brand et al., 1984; Au, 1993; Premchitt et al., 1994).
Thresholds for natural terrain landslides are not so easy to derive, and have not as yet been implemented, as the failure mechanisms may differ and records of events are harder to obtain. However, given that 60% of the land area of Hong Kong is classed as ‘Natural Terrain’ and the ever increasing demand for land pushes new developments and infrastructure closer to the natural terrain, the GEO realised the need to get a better understanding of landslide susceptibility. Hence the Natural Terrain Landslide Studies were set up as a special project, within the LPM programme of works, part of which looked at the correlation between rainfall and natural terrain landsliding.
Evans (1996) was the first to look at the distribution of rainfall over HK and noted that annual rainfall is not uniform, even when expected elevation effects are taken into account. The coastal periphery, outlying islands and the northern New Territories appear to be significantly drier than elsewhere. This led to the suggestion that absolute rainfall thresholds for landslides on natural terrain may also vary across Hong Kong, all other factors being equal. ‘Normalised’ rainfall, in which rainfall at a site is recorded as a proportion of the mean annual rainfall at that site, was considered to be a more appropriate tool for investigating natural terrain landslide susceptibility.
The NTLI allowed Evans (1997) to carry out a semi-quantitative assessment of possible rainfall thresholds (Annex G.1). The method adopted is summarised by Ko (2005) and included as Appendix B for information. Firstly, he looked at aerial photographs for the period between 1985 to 1994 (corresponding to the time when spatial rain gauge coverage was significantly improved) to locate and record natural terrain landslides, from which he produced a series of 1:100,000 plans for each year (1985 to 1994). He then plotted isohyets (lines on a map connecting points that receive equal amounts of rainfall) of the rolling 24 hour rainfall for all significant rainstorms for the same period and superimposed these on the 1:100,000 landslide plans. (Most of this information was obtained from the annual rainfall and landslide reports produced by GEO.)
The plots of rainfall and landslides were examined and for each landslide the maximum rolling 24-hour rainfall in the year of occurrence was recorded. This figure was reduced to a normalised value by dividing it by the approximate mean annual rainfall at the landslide site. A major limitation of this process was obviously that the maximum recorded rainfall may not necessarily have triggered the landslide.
Evans found that there were three points of abrupt change in the gradient (Figure G.5), which were taken as rainfall thresholds where significant increase in the number of natural terrain landslides would occur. Examination of his plots of annual rainfall and landslide distribution showed that for the majority of Hong Kong, where mean average rainfall is in the range 2,000 to 2,400mm, landslide densities of 1 per km2 or more are usually associated with 24 hour rainfall maxima of at least 200mm (0.09 normalised or 9% of annual mean precipitation), while higher densities of over 10 per km2 tend to be associated with 24 hour maxima of at least 400mm (19% of mean annual precipitation). It should be noted that these thresholds were average values, and did not take into account any contributing factors such as geology, slope, etc. He defined approximate landslide densities as the following:
a) Low density – less than 1 landslide per
b) Medium density – 1 to 10 landslides per km2.
c) High density – over 10 landslides per km2.
Figure G.5 – Cumulative percentage of natural terrain landslides against normalised maximum rolling 24-hour rainfall (1985 to 1994).
G.2.4 Landslide Warning System
The GEO manages and operates the Landslip Warning System with the Hong Kong Observatory (HKO). Landslip warnings are issued by the HKO in consultation with GEO when the recorded and forecast rainfall meets the warning criteria. It is important to note that, as of December 2007, the warnings were based upon man-made slopes and not on natural terrain landslides.
The existing Landslip Warning Criterion (Yu et al., 2003) operates by summing the number of landslide incidents for each of the vulnerable areas, based on the correlation between landslide density (number per km2) and rolling 24-hour rainfall of selected rain gauges. The Landslip Warning level was initially set at 10 landslides, on the basis that on average about 10% of reported landslides were major incidents and that casualties were only caused by major incidents. (This approach is similar to that described above for the unimplemented natural terrain landslide system.)
The landslide warning system was revised in 2001 following a review of landslide statistics. This revealed that, whilst on average major landslides account for about 10% of the total number of reported landslides, the percentage of major landslides was not constant but increased with increasing numbers of landslides (i.e. the percentage of major landslides increased with increasing size of storm event). For smaller rainstorm events, or at the early stages of larger events, the ‘first’ major landslide often occurred after about fifteen landslides were reported to GEO. Therefore, the warning level was increased from 10 to 15 predicted landslides in October 2001.
The action levels for the issuing of Landslip Warnings are as follows:
i) Consultation Level – consultation between HKO and GEO begins when 10 or more rain gauges record rolling rainfall of more than 100mm in 24 hours.
ii) Alert Level – this is a situation wherein continued monitoring of rainfall, and liaison, takes place. This level arises when the average rainfall required to reach ‘warning level’ is less than 100mm in 24 hours.
iii) Warning Level – Landslip Warning issued by HKO after consultation with GEO. The rainfall level has achieved that set for 15 or more predicted landslides.
Following recommendations made by Pun et al. (1999), a performance review of the Landslip Criteria is continuously undertaken. Improvements are made to take into account the experience gained from the operation of the system and correlations between landslide and rainfall are refined.
It is also of interest that the Hong Kong Observatory also operates a Rainstorm Warning to alert the public to heavy rainfall events. It should be noted that the Landslip Warnings are independent from the Rainstorm Warnings, which are set at Amber, Red and Black for 30mm, 50mm and 70mm of rain in 1 hour expected within 24 hours respectively. More emphasis is placed on the rainstorm warnings by the press and TV and during ‘Black Rain’ events, schools and offices are closed, which has led to some complaints about loss of profits from some business sectors. However, on the whole, both types of warnings are well received by the public.
G.2.5 Further Developments and Proposals for Future Studies in Hong Kong
Evans’ (1996; 1997) studies were recently updated in 2005 by Ko (2005), to include landslide data up to the year 2000 (an increase of 75% in the number of landslides), and used geostatistical analyses and GIS to process and analyse data, thus removing human error and improving efficiency and accuracy. Ko concluded that the plots and thresholds produced by Evans had limitations in the establishment of landslide warning criteria because they looked at maximum rolling 24 hour rainfall recorded in a year and not during a storm event. Ko subsequently used statistics to correlate the year-based 24 hour maximum to a storm-based maximum (the reader is referred to Appendix D of Ko, 2005). It is unclear, however, if the landslip warning system has been reviewed in light of her findings and recommendations.
Ko recommended that further refinements were achievable through the use of GIS. These refinements would include the effects of elevation (by locating rain gauges in higher natural terrain), terrain attributes (geology, slope gradient, etc) and terrain susceptibility classification into their rainfall-natural terrain landslide correlation. She also recommended other methods of looking at rainfall data including, the following:
i) Other means of normalisation of rainfall (using rainfall return period instead of the mean annual rainfall at a given site).
ii) Using different durations of rainfall (a maximum three hour rolling with antecedent 30 day rainfall) instead of the 24 hour rolling maximum.
iii) Formulation of a natural terrain landslide warning criterion.
The only ‘measure of success’ that is published relates to man-made slopes (Anon, Undated; Sun and Evans, 1999). Since the adoption of the LPM programme, risk assessment calculations indicate that the overall landslide risk arising from old substandard man-made slopes to the whole community of Hong Kong has been reduced to about 50% of the risk that existed in 1977. The Hong Kong Government’s demanding (but achievable) objective is to further reduce the landslide risk from old man-made slopes to below 25% of the 1977 level by the year 2010.
To put the risk of natural terrain landslides into perspective (Wong et al., 2004), of the 50 fatalities recorded between 1980 and 2003, 16 were as a result of natural terrain landslides and a significant number of these were associated with squatter areas. The historical natural terrain landslide data indicate that the landslide risk from natural hillsides is lower than that from man-made slopes in Hong Kong. However, the data may not fully reflect the inherent landslide risk to the community. Some landslides were ‘near miss’ incidents that could well have resulted in more serious consequences and the situation will only worsen as more new developments take place on, or close to steep natural hillsides.
The Hong Kong Government’s preferred approach is not to carry out stabilisation works to large areas of natural terrain, which would be both impractical and environmentally damaging, but to mitigate the risk through adjustments to the layout of new developments and provision of buffer zones and defence measures (e.g. debris resisting barriers).
A number of case studies have been published describing the effects of rainfall on landslides in Italy, most importantly a national system for the real-time prediction of hydro-geological hazards (floods and landslides). The rainfall detection element of the system is based on a comprehensive radar network (Casagli, 2006)
G.3.1 North Western Tuscany, June 1996
D’Amato Avanzi et al. (2004) report a series of rainfall induced shallow landslides which occurred on 9 June 1996 in the Apuan Alps in north western Tuscany, Italy. The associated rainstorm was concentrated over a 150km2 area and 474mm the rainfall corresponded to 21% of the annual mean.
Some 647 main landslides were recorded and were estimated to have caused damage to the value of hundreds of millions of Euros, in addition to causing the deaths of 14 people. The June 1996 storm occurred after a dry month (17.2mm of rainfall at Pomezzana). Figure G.6 shows the recorded rainfall at two gauges in the affected area. At Pomezzana 474mm of rain was recorded in about 12 hours, with a maximum intensity of 158mm/hour, whilst at Fornovolasco 420mmof rain fell in about 10 hours, before the instrument was destroyed by either a flood or a landslide. At gauges some 7km to 10km away only a few millimetres of rainfall was recorded.
Figure G.6 – Rainfall data from the 9 June 1996 study areas: (a) Pomezzana (597m asl) and Fornovolasco (470m asl) rainfall gauges (from D’Amato Avanzi et al., 2004).
While D’Amato Avanzi et al. (2004) give few insights into the relations between rainfall and landslides their paper provides some interesting and useful analyses. For example, they show that in this area and on this occasion by far and away the majority of landslides occurred in shallow overburden of between 0.5m and 2m thick.
Frattini et al. (2004) describe a series of more than 400 landslides which occurred in May 1998 near Sarno, to the east of Naples and Vesuvius, in pyroclastic soils. The landslides were triggered by a storm event and destroyed houses and infrastructure in addition to killing a total of 159 people. The events broadly classify as soil slip-debris flows or soil slip-mud flows, with velocities from very to extremely rapid and with high water content (Cruden and Varnes, 1996). According to the Pierson and Costa (1987) classification these would be described as slurry flows evolving into hyperconcentrated flows, with estimated velocities of 9.3m/s to 10m/s (see Figure 2.3 of Winter et al., 2005a).
Detailed rainfall gauge information was not available from within the authors’ study area, making rainfall analysis very difficult due to both the high areal variability of intense rainfall and orographic effects. However, data from five gauges was reported and Figure G.7 illustrates this data along with the locations of the rainfall gauges relative to the study area.
Figure G.7 – Cumulative rainfall for 4 to 5 May 1998 recorded by rainfall gauges at Lauro (4.5km north of the study area; 192m above sea level, asl); S. Pietro (12km east; 215m asl); Ponte Camerelle (12.5km south; 97m asl); S. Mauro (10.5km south; 31m asl); and Sarno (5.5km south-east; 36m asl) (from Frattini et al., 2004).
The data from the Lauro gauge was considered to be most relevant to the events due both to its distance from the hillside initiation areas and also its position with respect to the path of the storm. The cumulative rainfall recorded by the Lauro gauge during the 48 hour event was 173mm. The first low intensity fall occurred between 0000 and 0500 hours on 4 May and after a break of 11 hours it rained continuously until the early morning of 6 May. A maximum rainfall intensity of 15mm/hour was recorded at 1500 on 5 May and the mean intensity over the 48 hour period was 3.6mm/hour (Frattini et al., 2004).
Antecedent rainfall between 28 April and 3 May contributed a further 61.4mm and the rainfall return period was relatively short, with a maximum return period of 33 years for the 24 hour rainfall recorded on 5 May at Lauro. However, this must be set against the events occurring at the end of the rainy season and if this period is considered then the return period rises to greater than 100 years (Figure G.8).
The authors maintain that antecedent rainfall played a significant part in the triggering of this series of landslides, not least because of the high water retention (up to 100% of dry weight) of the volcaniclastic deposits. In such case rainfall infiltration over a prolonged period of time can cause significant increases in the unit weight making such an effect potentially more significant than in some other materials.
The rainfall and other data acquired by Frattini et al. (2004) were used to drive a hydrological model and there is no evidence that this has been used in any way to attempt to forecast future events. Indeed, Frattini et al. stated that they believe that such hydrological models were impractical for reliable physically-based distributed modelling, largely due to their complexity, associated data requirement and the difficulties associated with calibration.
Figure G.8 – Antecedent and event rainfall at the S. Pietro gauge, 215m asl and 12km to the east of the study area. The inset upper left shows the daily rainfall for late-April and early-May (from Frattini et al., 2004).
Sirangelo and Braca (2004) studied the same area as Frattini et al. (2004), but from a substantially different viewpoint. Their work involved the creation of a hydrological model, based upon a back analysis of the May 1998 events. The model produced is highly complex and comprises two parts:
- ‘Rainfall-Landslide’ for correlating precipitation and landslide occurrence, intended for model calibration through the reproduction of historic events.
- ‘Stochastic Rainfall’ for real-time forecasting of landslide events.
The model has been operated using data from the Sarno events and predictions performed over a period of approximately four years. The model enables three levels of elevated landslide potential status to be implemented, as follows:
- Attention status: with real time monitoring of instruments (when the mobility function, dependent upon the antecedent rainfall, reaches 40% of its critical value).
- Alert status: involving civil protection agencies (when the mobility function reaches 60% its critical value).
- Alarm status: involving the evacuation of the local population (when the mobility function reaches 80% its critical value).
During the period October 1999 to May 2002, 21, five and one respectively of each of the above status levels were implemented.
The ‘Rainfall-Landslide’ model is currently being used as a warning system for the Sarno area by the local authorities. However, it would appear that no events have as yet been successfully forecast using the system.
G.3.3 Imperia Province, Western Liguria, November 2000
From mid-October to 22 November 2000, the Western Liguria Region (Figure G.9) experienced prolonged and intense rainfall, with cumulative values exceeding 1,000mm in 45 days. This was followed on 23 November by a high intensity storm of 180mm of rain in 24 hours.
Figure G.9 – Cumulative rainfall distribution for 23 November in Imperia Province. The grey lines show the extent the post-event aerial photography. Black dots show the locations of rainfall gauges. Irregular black lines show the locations of landslides, which have been exaggerated for illustration purposes (from Guzzetti et al., 2004).
More than 1,000 landslides, including debris flows and a few large complex slides, were triggered causing severe damage to roads, private homes and agriculture as well as leading to three deaths. The landslides commenced between eight and 10 hours after the start of the storm and the most intense areal landslide activity occurred as a consequence of rainfall intensities of 8mm/hour to 10mm/hour (Guzzettti et al., 2004). Mean annual precipitation ranges from between 750mm and 1,250mm in the west to between 1,350mm and 1,850 in the central and eastern parts of the region.
Figure G.9 relates the spatial distribution of cumulative rainfall in Imperia Province to landslide activity. Although this Province has experienced less rainfall and fewer landslides than others within Liguria Region. The map shows that the highest intensity rainfall coincides with the area in which landslides were most abundant.
Figure G.10 shows patterns of rainfall intensity versus duration for a gauge at Imperia (Figure G.10a) and the synthetised rainfall pattern constructed for San Romolo (Figure G.10b), the latter based on a cumulative rainfall fof 241.2mm (i.e. at the San Romolo gross measurement gauge) and the same intensity as recorded at the Imperia gauge. Each graph begins at 15 minutes (0.25 hours) at the left hand side of the graph and ends at 28 hours on the right hand side. The times of landslide occurrence as observed at nearby Ceriana are over-plotted. Figure G.10c corrects the timings of landslides for a two-hour apparent lag time observed between the highest intensity rainfall at Imperia and Ceriana.
G.3.4 Piedmont Region
In dealing with debris flows and soil slips triggered by short intensity storms in the Piedmont Regionof NW Italy, Aleotti (2004) usefully defines some of the key rainfall parameters relating to the potential to trigger landslides (Figure G.11).
Aleotti (2004) proposes an equation similar in form to equation (10.1A) as follows:
This equation is claimed to account for 90% of the available data for which rainfall is believed to have led to landslides in the Region. It has been refined by normalising the intensity of the rainfall (NI) with respect to the mean annual precipitation (MAP) such that two equations collectively describe the triggering threshold, as follows:
where the normalised intensity (NI) is expressed as a percentage by I/MAP ¥ 100.
Finally, Aleotti (2004) expresses the critical normalised intensity in terms of the normalised critical rainfall (NCR) to encompass 90% of events studied, as follows:
where the NCR = R/MAP ¥ 100.
Aleotti (2004) used hourly rainfall in the study, but appears to have analysed only the storm events taking no account of longer-term antecedent rainfall perhaps accounting for some of the poor correlations reported.
Figure G.10 – Landslide timings at Ceriana relative to intensity-duration plots: (a) rain gauge at Imperia; (b) synthetically derived rainfall at San Romolo; (c) synthetic San Remolo data corrected for a two hour time lag (from Guzzetti et al., 2004).
Figure G.11 – Definition of rainfall parameters (from Aleotti, 2004).
In recent years there have been a number of debris flow events that have exposed the population of the Cancia area of the Dolomites to significant risk. In response, an alarm and monitoring system was set up with data from three rain gauges being monitored during debris flow events.
Data from the rain gauges was analysed, taking into account the elevation of the gauges, to determine debris flow initiation and rainfall relations. The findings were then compared with results from geologically similar areas in the Eastern Alps.
The geology of the area is typically Triassic to Jurassic of the Dolomitic stratigraphic sequence. The deposits that have proved susceptible to debris flows are gravels with a low content of sand and fine particles.
The climatic zone is a cold Alpine Climate (Köppen Class D) with an annual rainfall of 1,000mm, which falls mainly in spring and summer.
The drainage basin for the Cancia debris flow area covers a surface area of approximately 1.8km2, and the profile of the debris flow channel is shown in Figure G.12. Debris flows are recorded from 1868 (100,000m3) to 1996 (40,000 m3 to 45,000m3), with activity over period 1986 to 1996 being one event every 1 to 2 years.
Figure G.12 – Longitudinal profile of a flow channel, the upper part of the source area and mean slope angles in the different sectors (from Bacchini and Zannoni, 2002).
Thresholds based on Ceriani et al. (1994) were found to be too high, with most of the observed events falling in the stable zone (Figure G.13). Thresholds were developed for debris flows in terms of mean intensity (I), duration (D) and mean annual precipitation (MAP). These utilised normalised rainfall and normalised intensity expressed as a percentage of the MAP (Figure G.13).
Figure G.13 – Normalised rainfall intensity (intensity/MAP) versus duration and debris flow correlation. The dashed line shows the debris flow threshold proposed for the study area (from Bacchini and Zannoni, 2002).
Thresholds for debris flows, written in terms of the normalised rainfall (Rn = R/MAP) were as follows:
where I > 2 mm/hour.
Normalised rainfall and normalised rainfall intensity should only be used in limited areas where the annual frequency of rain storms is fairly constant (Wilson, 2000).
Typically, triggering rainfall events were found to be 20mm to 30mm in 1 or 2 hours (i.e. not particularly high rainfall levels) but due to the short duration relative to the data reading frequency they may be of intermediate intensity. The role of storm cells in defining rainfall intensities leading to potential debris flow conditions is thus clear.
Rainfall thresholds were found to be an unsuitable medium for the purposes of debris flow prediction but useful in determining a suitable level at which actions by management and monitoring personnel might be undertaken as part of an overall management strategy.
Landslides are a common occurrence and a recurring problem on the mountainous island of Jamaica (R Ahmad, 2006; Personal Communication, 2006). These are usually associated with tropical storms, including hurricanes, the paths of which often pass close to the island. Typically, disruption and damage takes a number of forms, including:
- Severance of transport routes leading to stranded communities.
- Loss of income through economic activity, including loss of productive agricultural areas, especially coffee farms and farm-to-market access roads.
- Closed schools.
- Damage to property and community facilities.
- Interruption to domestic water supplies.
- Addition of sediment to river profiles raising channel levels and thus increasing future flood hazard.
In particular the social fabric of communities may be severely disrupted by many of these consequences and, in addition, individuals are exposed to the trauma of evacuation and the loss of their homes. Much of the impact of such landslides is due to transported landslide debris, especially along debris chutes and deposition areas, which may often be far removed from the landslide source.
Ahmad (2003) reports the development of two thresholds:
- For debris flows that commonly develop from shallow landslides during intense rainfall.
- For deep-seated landslides that are usually triggered by prolonged rainfall.
Also noted is the fact that rainfall amounts for storms that did not trigger landslides are equally important in that they allow the population of the threshold graph from both directions. The threshold established by Ahmad (2003) is presented in Figure G.14.
Figure G.14 – Rainfall intensity-duration threshold for shallow landslides in eastern Jamaica, using data from 19 storms between 1951 and 2002 (from Ahmad, 2003).
Ahmad (2003) notes that the rainfall threshold relation is defined for storm durations between 1 and 168 hours and average rainfall intensities between 2 and 93mm/hour. The threshold relation indicates that, for rainfall of short duration (about 1 hour), intensities greater than 36mm/hour are required to trigger landslides.
There is a relation between landslide characteristics and the position of the landslide-triggering storm on the threshold line. Storms near the short-duration/high-intensity end of the threshold line trigger mostly shallow landslides (e.g. Figure G.15) by causing an excess pore pressure in shallow colluvial zones.
In contrast, storms near long-duration/low-intensity end of the threshold have triggered the largest, deepest landslides in eastern Jamaica (e.g. Figures G.16 and G.17).
Figure G.15 – Shallow landslide induced by rainfall between Ramble and Somerset on the Yallahs River in St Thomas Parish, eastern Jamaica. The road followed the shoulder of the hill to either side of the landslide.
Figure G.16 – Deep rainfall induced landslide on the A2 road between Whitehall and Martins in St Mary Parish, eastern Jamaica.
Figure G.17 – Deep rainfall-induced landslide on the B1 road at Section in Portland Parish, eastern Jamaica.
Landslides in Nepal are often associated with high intensity rainfall in combination with the highly active slope processes that, in such an active mountain environment, are driven by gravity. Monsoon rainfall patterns mean that more than 80% of the annual rainfall occurs within a four month period between June and September, with the 50-year average for Kathmandu in July being around 375mm. At the Arughat Bazar rainfall gauge (near the Privthi Highway, H04: Figure G.18) in excess of 550mm of rain fell in August 2000; while the highest recorded rainfall in a 24 hour period was at Kulekhani, where 540mm of rain fell on the 19 and 20 July 1993, an average of 22.5mm/hour. Sunuwar et al. (2005) compare this to figures reported by Wieczorek (1996) of 6.3mm/hour for the triggering of landslides in California.
Rainfall-induced landslides are thus frequent and often block the major roads of Nepal, causing particular problems of the effects of severance of access for rural populations. There appears to be no effort to forecast landslides using rainfall data in Nepal; there remains a suspicion that conditions are sufficiently extreme that such an exercise might be unproductive in that the entire monsoon season would be seen as high risk period.
Experience in Norway has indicated that 8% to 10% annual precipitation in one day (24hrs) is likely to lead to debris flows in ‘exposed’ (or susceptible) locations (U Domass, Personal Communication, 2006). If there is significant antecedent rainfall (several days) then this threshold may be lower.
Figure G.18 – Privthi Highway, H04, Nepal.
An investigation of 30 debris flows in Norway was undertaken by Sanderson et al. (2005). The work indicates that steep Norwegian slopes are often partially covered with glacial till, which in many places is itself covered with colluvium. The silt and clay content of these is typically in the range of 10% to 30% (Jorgensen, 1978). The upper 0.5m to 1.0m of soil has high permeability due roots and organisms, and this enables frost to influence the structure of the soil profile. The permeability of the lower soil is much lower.
Norway comprises two climatic areas:
- Marine west coast climate (western Norway), typically 1,000 to 3,000mm annual rainfall falling in predominantly south-westerly winds during the passage of warm fronts. Daily rainfall can exceed 200mm.
- Continental sub-arctic climate (eastern Norway), typically 300 to 1,000mm annual rainfall falling predominantly during convective summer storms.
Slope aspect plays an important role with the greatest rainfall on windward slopes (south-west facing slopes). The high relief on the west coast also leads to large differences in precipitation even over small distances. South-west facing slopes are also most prone to intense meltwater production due to the exposure to wind and solar radiation.
Field measurements indicate the presence of slip surfaces along a relatively impermeable layer at 0.5m to 1.0m depth. This surface is a boundary between relatively high permeability material and underlying lower permeability material, leading to increased pore pressures.
Climatic monitoring stations in the areas of the 30 debris flows investigated record the following information three times a day (at 0700, 1300 and 1900):
2. Snow depth.
3. Air temperature/humidity.
4. Wind speed/direction.
Records of precipitation and calculated snowmelt over the 12 hour, 24 hour, 7 day, 15 day and monthly time periods were assessed. For the continental climatic areas debris flows activity was found to be most frequent in April and May, whilst for the marine climate August to December were the most active months. For the marine west coast climate areas the weather patterns triggering the majority of events were:
1. Heavy rainfall of one day duration with a concentrated period
of 1 hour to 4 hours.
2. Rainfall in combination with snowmelt over 3 days to 7 days.
Two examples of this are:
- In this example event the 24-hour precipitation in excess of 64mm, with the 24-hour rainfall return period being >150 years. The period prior to this had been relatively dry, with only 29.5mm of precipitation over 14 days.
- In the second example event the probable cause was rainfall and snow melt, yielding 190mm in a week (211% of monthly average) – a figure corresponding to a return period of several decades.
The resulting intensity-duration relations for the sites studies were compromised by a high degree of uncertainty, mainly due to following factors:
1. The widespread rain gauge network does not cover all local regions where heavy precipitation is experienced.
2. The frequency of recordings was too low to reflect variations in precipitation with time – Sanderson et al. (2005) found that climatic stations recording at 6 and 12 hour frequencies could not be used for generation of water supply/debris flow relations.
3. The rate of snowmelt depends largely on wind speed.
Caine (1980) plotted rainfall intensity against duration for worldwide debris flows and found a lower bound as given in Equation (9.1A)
Sanderson et al. (2005) discuss the fact that time is a very significant factor, with rainfall over as little as one hour being potentially critical in the generation of debris flow (Figure G.19). Also identified was a lower intensity-duration threshold (Figure G.20), derived from the 30 debris flows, and this is expressed as:
where P is the ‘critical water supply’ expressed as a percentage of mean annual precipitation and D is duration (hours).
For example, the 12-hour critical water supply expressed as a percentage of mean annual precipitation is given by:
If the mean annual precipitation is then 2,000mm then the ‘critical rainfall level’, R12hour Crit, is (2,000 ¥ 5.33)/100 = 106.6mm.
Sanderson et al. (2005) conclude that debris flows exhibit the following characteristics:
1. They are triggered by rare climatic events with return
periods of 50 years or more.
2. They show short response times to climatic events (e.g. 4 to10 hours).
3. Many recent cases are apparently due to human activity affecting slope hydrological regime: e.g. forest roads, forest harvesting.
Figure G.19 - Debris flow trigger due to intense rainfall within the west-coast climatic region (after Sanderson et al., 2005).
Figure G.20 - Critical water supply for debris flow initiation. Data points indicate water supply in debris flow events (after Sanderson et al., 2005).
Toll (2001; 2006) reports on rainfall leading to landslides in Singapore and presents a graph of the rainfall occurring on the day of the landslide against that in the five days preceding the landslide (Figure G.21).
Figure G.21 – Rainfall events leading to landslides in Singapore (from Toll, 2006).
While a few minor landslides have occurred after intense one-day rainfalls with little antecedent rainfall others have occurred with low one-day rainfall and higher antecedent rainfalls. Toll (2006) concludes that this indicates that total rainfall, over an extended period, is more important that either daily or antecedent rainfall.
The solid diagonal line in Figure G.21 represents a total rainfall of 100mm in a six-day period appears to define the minimum rainfall conditions that can lead to minor landslides in Singapore.
Mikos et al. (2004) report on a study of two debris flows that occurred near Stoze in NW Slovenia on 15 and 16 November 2000.
A rain gauge at the nearby village of Log pod Mangartom recorded 1,638mm (more than 60% of the average annual precipitation) in the 48 days leading up to the events (average rainfall intensity 1.42mm/hour), corresponding to a return period of more than 100 years. Other rainfall depths for shorter durations within the same time window (481.6mm in 7 days, 174.0mm in 24 hours, 70mm in 1 hour) had recurrence intervals of much less than 100 years (Table G.1).
Several short periods of intense rainfall events were recorded in Log pod Mangartom during 2000, as follows:
- 407.4mm (11 to 13 October).
- 380.2mm (14 to 16 November).
- Daily rainfall of 174mm (12 October).
- Daily rainfall of 165.3mm (14 November, a day before the first landslide).
These levels of rainfall are not extreme for the area. In contrast, the precipitation depths for one and two months measured at the gauge were extreme, with return periods of around 100 years. Only the measured rainfall intensity of 1.42mm/hour in the last 1,152 hours (48 days) lies outside the collected historical data for critical rainfall intensity and duration (Crosta, 2004); all others of shorter duration lie within these
Table G.1 – Measured rainfall depths at rainfall gauging station in Log pod Mangartom compared with statistical values given for different recurrences intervals for that station (reference period 1961 to 1990) (from Mikos et al., 2004).
The comparison with empirical (Caine, 1980) rainfall-intensity relations shows that all measured data in Log pod Mangartom in late-Autumn 2000 lie above but close to the lower bound threshold for shallow landslides worldwide (Equation 10.1A). Only the rainfall intensity of 70mm/hour measured in a one hour period on the evening of 16 November 2000 came close to Caine’s upper bound threshold (Equation 10.1B).
Debris flows are a geomorphological process common in the Swiss Alps, and in 2000 four significant flows (between 5,000m3 and 35,000m3) occurred which were monitored by debris flow observation stations. These comprised video cameras, ultrasonic devices, radar, geophones and rain gauges (Hurlimann et al., 2003).
The debris flows occurred in the Illbach and Schipfenbach catchments, both of which appear to be characterised by channelised debris flow activity.
The Schipfenbach monitoring system incorporated a rainfall gauge recording every 10 minutes. This indicated a rather dry June period (Figure G.22a) followed by a high July rainfall of 189mm. During a 3 hour period before the debris flow the maximum intensity was 11mm/hr, yielding a total rainfall for 6 August of 106mm. Comparing this event to Zimmermann et al. (1997) the authors proposed a relation between intensity and duration as follows:
Hurlimann et al. (2003) concluded that the threshold was most likely too high. However, they did establish that the critical rainfall fell in a period of 4 to 24 hours before the event.
Rainfall gauges were not installed in the Illbach catchment until after the 2000 events. However, the authors indicate that the 100 year return rainfall intensity is between 35mm/hour and 57mm/hr for 0.5 hour and 1.0 hour rainfall durations.
Figure G.22 – Precipitation analysis of the Schipfenbach debris flow. (a) Cumulative rainfall during the 24 hour prior to the debris flow event (the arrow indicates the time of initiation). Inset shows the daily precipitation during the month prior to the debris flow. (b) Comparison between the climatic threshold for debris flow initiation in the outer parts of the Swiss Alps and the data for the Schipfenbach event (after Hurlimann et al., 2003). (Note: Equation 2 referred to in Figure G.22a is reproduced as Equation G.9 in this report.)
The superficial deposits in the Illbach catchment typically comprise 35% to 40% sand with less than 5% clay. The Schipfenbach catchment superficial deposits typically comprise 45% to 70% gravel with a clay fraction of less than 5%.
The authors concluded that:
- The debris flows were triggered by intense rainfall leading to in-channel mobilisation.
- Large landslides in both catchments provided debris for flows.
- Ultrasonic and radar measurements were practicable for defining debris flow hydrographs (channelised debris flows).
- Monitoring indicated a wide spectrum of flow behaviour even within the same channel.
- A critical factor was the rainfall in a period of 4 hours to 24 hours before the debris flow.
G.10.1 North-West England
A rainfall and early warning system was set up to monitor the condition of earthworks on the Settle to Carlisle line following a landslide which caused a train derailment at Ais Gill, Cumbria on 31 January 1995
Rainfall gauges were installed at several locations where earthworks were classed as ‘Poor’. Hourly, daily, weekly and 28 day rainfall levels were recorded and trigger levels set. These trigger levels were based on a study by Lancaster University of rainfall levels that had caused landslides in Cumbria.
The levels set were as follows:
1. 24-hour total threshold set at 80mm.
2. Antecedent Precipitation Index (API) threshold set at 130mm.
3. 30-day total threshold set at 300mm.
The system was used to put in place train speed restrictions when trigger levels were exceeded. The system was removed two years later when remedial measures had been undertaken on the railway earthworks.
G.10.2 South-West England
Network Rail (Personal Communication, 2006) report on a system on trial in southern England incorporating three levels of alert status, as follows:
1. Earthwork Failures Likely.
2. Earthwork Failures Possible.
3. Earthwork Failures Unlikely.
4. Embankment Desiccation Possible.
The alert levels are based on Soil Moisture Deficit (SMD) (Figure G.23) and rainfall as a percentage of the Long Term Average (LTA). The threshold rainfall is defined as 175% of the LTA.
G.10.3 Scottish Highlands
A series of debris flows occurred in the Scottish Highlands between 1999 and 2001 adjacent to the A890 Stromeferry Bypass road and the railway which runs on a close by. As the debris flows had been triggered by rainfall events, a review of existing rainfall data was undertaken (Nettleton et al., 2005a).
The nearest automated rainfall gauge was at Plockton 10km to the west, on a low relief peninsula, and was not initially considered to be representative of the rainfall at Stromeferry. However, assessment of the 1999 to 2001 daily rainfall data from this gauge indicated a good correlation of peak rainfall events with debris flow activity. In particular, the 14-day cumulative rainfall indicated clear peaks that correspond well with the January 1999 and October 2001 debris flow events and the smaller event of October 2000, thus indicating the importance of antecedent as well as high intensity rainfall.
Figure G.23 – Soil Moisture Deficit (SMD) and rainfall graph (from Network Rail, Personal Communication, 2006).
Figure G.24 shows a graph of the normalized rainfall from the gauge at Plockton and a Scottish Environmental Protection Agency (SEPA) river flow-gauging station some 5km north-east of Stromeferry at the head of Loch Carron. There are good correlations between both sets of data and debris flow occurrence, probably as the principal weather fronts track in from the west. This indicates that the Plockton rainfall is, in fact, representative of the Stromeferry/River Carron catchments in terms of peak events. The magnitude of rainfall is however likely to be lower at Plockton due to its lower relief. There is a rainfall and river flow peak in November/December 1999 which has no corresponding debris flow event, but this may be because the main gully in question had a major clear out in January 1999.
For an early warning system at Stromeferry an automated local rain gauge, appropriate trigger levels and some form of automated barrier or signs would be required (Nettleton et al., 2005a). Figure G.24 suggests that a trigger level for the 14 day antecedent rainfall could be developed based on the Plockton rainfall. Similar trigger levels would have to be developed for daily rainfall and a range of other antecedent rainfall periods.
The current rainfall readings are only daily and the response of the system to high intensity rainfall events correspondingly would be limited. Hence, a system recording hourly rainfall would be required to provide greater response sensitivity to high intensity events which follow a moderate antecedent build-up.
Between 1986 and 1995 the United States Geological Survey (USGS) and the National Oceanic and Atmospheric Administration (NOAA) undertook an exploratory program for predicting debris flows in the San Francisco Bay area. Circular 1283 (Anon, 2005) presents the findings and recommendations of a joint USGS/NOAA task force tasked with developing a plan for the implementation and operation of a NOAA/USGS system to issue joint Outlooks, Watches and Warnings for areas deemed to be at risk from debris flows as a result of current or forecast precipitation.
Figure G.24 – Normalised graph of 14 day cumulative rainfall at Plockton and River Carron flow for 1999 to 2001 showing the major debris flow events (from Nettleton et al., 2005a).
The task force reviewed several operational rainfall intensity-duration landslide warning systems from around the world, including:
- Hong Kong (Chan et al., 2003)
- San Francisco Bay 1986-1995 (Wilson 1997)
- Rio de Janeiro (1998-2003, 42 warnings) (D’Orsi et al., 2004)
- The State of Oregon (Mills 2002)
- Lyme Regis, UK (Cole and Davis 2002)
- Seattle, Washington (Godt et al., 2005)
The task force identified that an antecedent rainfall threshold and an intensity duration threshold would be required for a warning system. To achieve this, methods for quantitative precipitation estimation (QPE) and quantitative precipitation forecasting (QPF) were reviewed.
The report provides elements of a worked up proposal for the research and development of a full debris flow warning system.
Wieczorek (1987) studied debris flows in the Santa Cruz Mountains of California over a 10-year period, including 110 debris flows triggered during 10 storms. Analysis of the rainfall records indicated that two conditions had to be met for debris flows to be initiated: antecedent rainfall had to exceed a minimum threshold, and the storm rainfall had to exceed certain a level of intensity for a specified duration.
In the low permeability clay, silt and clayey silt soils of the study area, antecedent rainfall was found to be important over periods from seven days to two months. Seasonal rainfall of at least 28cm was observed prior to any debris flows being triggered. It was also found that rainfall values during the preceding seven to 30 days accounted for about 80% of the antecedent seasonal value and that the seven to 30 day antecedent rainfall values for storms that triggered debris flow was about twice that of storms that did not trigger debris flows.
Wieczorek (1987) derived the expression defining the storm events capable of triggering debris flows, provided that sufficient antecedent rainfall had fallen, as follows:
where I is the rainfall intensity (in cm/hour) and D is the duration of rainfall (in hours).
The equation is best defined within the range of intensities 0.5cm/hour to 1.0cm/hour and the relation is assumed to be asymptotic at its extremes.
Figure G.25a shows a plot of duration for different levels of intensity for a number of storms and the threshold (Equation G.10) separates those that did and those that did not trigger debris flows. Each of these storms followed antecedent rainfall of at least 28cm. Each storm is represented by a family of data points, each point corresponding to a duration of particular intensity. In contrast, Figure G.25b illustrates storms that were not associated with at least 28cm of antecedent rainfall. While in Figure G.25a the intensity-duration data sited to the right of the curve defined by Equation (G.10) generally triggered debris flows and those sited to the left of the curve did not, in Figure G.25b none of the data are associated with debris flow activity.
The data presented by Wieczorek (1987) presents a very simple, threshold-based approach to coping with the effects of antecedent rainfall. While the intensity-duration approach then used to deal with the subsequent storm rainfall is potentially difficult to achieve in real-time this is broadly true for all related approaches.
G.11.2 Washington State (Seattle)
The Seattle area experiences shallow landslides in the colluvium deposits triggered during or immediately following heavy rainfall or snowmelt. Previous studies in the Seattle area have indicated that both antecedent and storm rainfall have significant effect. Seattle has a dense network of rain gauges with hourly recordings dating back over 25 years. This coupled with records of landslides (Laprade et al., 2000) has enabled development of empirical rainfall / slope stability models (Godt, 2004).
Recent analysis of data between 1933 and 1997 showed a combination of three day triggering rainfall and 15-day antecedent precipitation can be used to forecast when three or more landslides can be expected during a three day period (Chleborad, 2003).
The Seattle rain gauge network comprises 17 tipping bucket gauges providing a dense coverage (2km to 5km between gauges). Mean rainfall intensity, Imean, and duration, D, were compiled from rainfall gauge data. A rainstorm was defined as a period of rain bounded by at least 3 hours of no rainfall. Analysis of six rainstorms, which triggered shallow landslides, between 1978 and 1997 yielded a rainfall intensity-duration graph with a threshold defined by:
Figure G.25 – Intensity-duration data for storms in the Santa Cruz Mountains in California: (a) data for storms following 28cm of antecedent rainfall; (b) data for storms that did not follow 28cm of antecedent rainfall (from Wieczorek, 1987).
The authors employed the Antecedent Water Index (AWI), calibrated with measurements of soil-water content and rainfall to provide a general assessment of the soil-moisture conditions (Figure G.26).
A decision tree for assigning warnings was developed based on the AWI and the rainfall threshold, as shown in Figure G.27.
The authors concluded that, based on landslide events during the previous 25 year period, the rainfall intensity-duration and the water balance model would have flagged some 56 rainstorms that exceeded the intensity-duration threshold, with three rainstorms below the intensity-duration threshold (‘Null’) which were associated with evidence of shallow landsliding.
Some 28 rainstorms were assigned a ‘Watch’ status and evidence of shallow landsliding was noted in 42.9% of these. A further 13 rainstorms were assigned a ‘Warning’ status and shallow landsliding occurred in 61.5% of these.
Figure G.26 – Rainfall, volumetric water content, and the Antecedent Water Index (AWI) for the Edmonds field site for the period 17 October 2003 to 14 February 2004 (from Godt et al., 2005).
This research was also applied specifically for rail transportation (Baum et al., 2005). For this application rain gauges were normally set to record hourly but this increased to every 15 minutes during times of high precipitation (>2.54mm/hour). The data were transmitted by radio telemetry system and graphs were produced on a web server in near to real time.
For this application the alerts were as follows:
1. Advisory – Days in advance.
2. Watch – 3 hours to 72 hours in advance.
3. Warning – Near real time.
Other regional studies of landslide risk assessment that have been studied in order to obtain information useful to this work include:
Albania: Bozo et al. (2005) report on landslide risk assessment for roads and include rainfall events as one of the seven most important factors in their triggering. Around half of all landslides in Albania are thought to occur during or just after ‘rainy weather’. It is not, however, entirely clear how this is translated into an assessment mechanism although it seems likely that seismic activity is more of a potential trigger than in Scotland.
Figure G.27 – Decision tree for assigning warnings (from Godt et al., 2005).
Brazil: Ortigao et al. (2001) report on a system based upon intensity and accumulated 96-hour rainfall. This system appears to be adapted more to slower moving landslides that may be triggered by relatively short periods of rainfall with little or no influence from longer term antecedent rainfall.
Mainland China: Zhou and Chan (2005) note that the understanding of debris flow mechanics is at a relatively immature level of understanding and that qualitative evaluation parameters currently predominate over quantitative ones. Other recent work on regional landslide management in China has been conducted by Wen et al. (2005) and Yin and Wang (2005).
Columbia: Montero Olarte and Ojeda Moncayo (2005) report that 70% of the Columbian national road network suffers the consequences of frequent obstruction or destruction due to the actions of rainfall-triggered landslides and that landslides in Columbia are mainly triggered by rainfall.
Cuba: Castellanos Abella and van Westen (2005) report on a proposed landslide risk assessment method for Cuba. Rainfall is equally ranked with seismic activity as a triggering factor.
Ethiopia: Woldearegay et al. (2005) report on landslide hazard mitigation strategies for the northern highlands of Ethiopia. While the authors implicitly acknowledge the role of rainfall in this region (where the bimodal annual average can vary between 500mm and 2,000mm their paper pays relatively little attention to this issue).
United States of America (Alaska): Sidle and Swanston (1976) estimated a return period of less than two years for a storm that caused a small debris flow in Alaska. They also noted that around 54% of the rain fell in the final three hours of the storm (total duration 10 hours). This early work perhaps points to the importance of the relation between intensity and duration in understanding how debris flows are triggered.