9 LANDSLIDES, CLIMATE, RAINFALL AND FORECASTING 9.1 INTRODUCTION 9.2 CLIMATIC INFLUENCES ON LANDSLIDES 9.5 GENERALISED FORECASTING METHODS 9.4 CASE STUDIES 9.5 A TRIGGER THRESHOLD FOR SCOTLAND 9.6 SUMMARY

9 LANDSLIDES, CLIMATE, RAINFALL AND FORECASTING

by M G Winter, A Motion, F Macgregor, J Dent and P Dempsey

9.1 INTRODUCTION

One of the primary factors influencing debris flow occurrence is water. Heavy rainfall and/or snowmelt trigger the majority of flows as the water mobilises the loose sediment and/or infiltrates into the soil (e.g. McMillan et al., 2005).

However, in Scotland the amount of rain that falls during storm events, or in the weeks preceding, and leads to debris flows is currently unknown. Certain of the proposals presented by Winter et al. (2005), in response to the debris flow events of August 2004, included a recommendation to install a system of rain gauges. This was intended initially to gain a better understanding of the amount of rain that has to fall to cause these instabilities but with the intention that, in the longer term, a management strategy would be able to be developed: e.g. a protocol for action, potentially including road closure or increased surveillance when predetermined levels of rainfall are exceeded.

The recommendations for future work included (amongst other things) case studies being ‘assembled from around the world’ in order to capture experience of how rainfall data were collected, analysed and interpreted for the purpose of producing a landslide warning system. These case studies are presented in Appendix G.

Empirical evidence indicates that many Scottish debris flows are triggered by short intense rainfall events preceded by periods of heavy (antecedent) rainfall. However, Crosta (2004) indicates that the ‘meteo-climate’ factors associated with landslide events worldwide are characterised by an extreme variability. They can be sub-divided into short-term (short intense rainfall, snow melting, etc.) and long-term components (antecedent rainfall, snow melting, etc.). Interestingly, Crosta recognises snow melt as both a short and a long-term component depending upon the rate of melting.

Crosta (2004) also indicates that shallow, flow-type landslides are more likely to be triggered by intense rainfall events and that longer duration rainfall is likely to be involved in more deep-seated landslides. However, his comments are inextricably linked to soil type: assuming that flows occur in granular soils and that deep-seated landslides occur in clayey soils. Winter et al. (2005d) noted that debris flows, and in particular their triggering events, in Scotland are by no means confined purely to granular soils and that such generalisations need to be treated with some caution.

Crosta (2004) goes on to state that rainfall analysis is the most frequently adopted approach for forecasting landslides and that worldwide observations have been collected to identify the minimum and maximum rainfalls over various periods of time critical to the triggering of landslides.

In this section of the report observations and approaches to understanding landslides and their rainfall triggers taken from around the world are identified and set in the context of the generalised climate of Scotland. A series of analyses of rainfall events that have led to debris flow has been undertaken in order to develop a tentative rainfall threshold for debris flow formation in Scotland.

9.2 CLIMATIC INFLUENCES ON LANDSLIDES

9.2.1 Rainfall Patterns and Landslides

Landslides are often cited as being caused by storm rainfall and the link between high intensity rainfall and debris flows has been documented in Japan (Fukuoka, 1980), New Zealand (Selby, 1976) and Brazil (Jones, 1973) amongst other places. However, the influence of antecedent rainfall prior to storm events was clear from the events experienced in Scotland in August 2004 (Winter et al., 2007a).

In a study based in the Santa Cruz Mountains of California, Wieczorek (1987) noted that no debris flows were triggered before 28cm of rainfall had accumulated in each season. This clearly acknowledges the importance of pre-storm, or antecedent, rainfall, a factor that has also been recognised in studies in Southern California (Campbell, 1975), New Zealand (Eyles, 1979) and Alaska (Sidle and Swanson, 1982). Wieczorek (1987) also notes that in the case of high permeability soils such as those found in Hong Kong (Brand et al., 1984), the period of antecedent rainfall may be short or even that the amount of necessary antecedent rainfall may be supplied by the early part of the storm event itself.

9.2.2 Scotland’s Rainfall Climate

The climate of Scotland in terms of its rainfall may be very broadly divided into east and west (see Figures 9.1 and 9.2). Data presented by the Meteorological Office (Anon, 1989) indicate that in the east rainfall generally peaks in August while in the west the maximum rainfall levels are reached during the wider period September to January (Figure 9.1). Although rainfall levels in the west are relatively low in August they do increase from a low point in May. Both scenarios indicate that soils may undergo a transition from a dry to a wetter state at or around August, giving rise to an increased potential for debris flow and other forms of landslide activity. The central area, as represented by Pitlochry in Figure 9.1, has a mix between the rainfall characteristics of the ‘east’ and the ‘west’. The rainfall peak is both lower and shorter (December and January) than in the west, but there are also small sub-peaks in August and October. A broadly similar pattern is found for Perth.

Figure 9.1 – Average rainfall patterns for selected locations in Scotland. Edinburgh is in the east of Scotland, Pitlochry in the centre and Tiree in the west.

Clearly, the soil-water conditions necessary for debris flows may be generated either by long periods of rainfall or by shorter intense storms. It is however widely accepted that Scottish debris flow events are usually preceded by extended periods of heavy (antecedent) rainfall in company with intense storms.

Figure 9.2 – Example of Meteorological Office 30-year monthly average rainfall data for October (image courtesy of the Met Office).

9.2.3 Climate Change

The links between greenhouse gas emissions, the rise in the global temperature anomaly and consequent climate change are well-established. Indeed, the Stratigraphy Commission of the Geological Society of London has proposed that the influence of man has supplanted natural forces as the main driver of environmental processes at the Earth’s surface and suggest the formalisation of a new geological epoch, the Anthropocene Epoch, including the last two centuries (Zalasiewicz et al., 2008a; 2008b).

The UKCIP02 (UK Climate Impacts Programme) report considers three periods: the 2020s, the 2050s and the 2080s. In general terms small changes are noted in the predictions for the 2020s. These changes increase slightly for the 2050s and slightly further still for the predictions for the 2080s, reflecting the temporal trends in temperature and precipitation. Whilst climate models generally predict averages and the associated error limits can be substantial, it is also important to note that annual variability is predicted to increase for many climate factors. This means that changes in the averages, as described above for example, may mask more significant variability effects.

Climate change models for Scotland in the 2080s (www.ukcip.org.uk) indicate that there will be a decrease in the precipitation in the summer with an increase in the winter, while overall precipitation levels are expected to decrease (Galbraith et al., 2005a). The climate models, however, are generally considered to be incapable of predicting localised summer storms. Such storms are believed to be at least partially responsible for triggering the events of August 2004, and it must be concluded that climate change data may not give a full picture of the relation between precipitation and landslides⁵.

Taking the analysis of the UKCIP02 data further, predicted changes in the number of ‘intense’ wet days generally indicate a net increase of less than one day per annum by the 2080s, with slightly fewer intense wet days in the summer and more in the winter. However, by the 2080s extreme storm event rainfall depths are predicted to increase by between 10% and 30%, with intense winter rainfall increasing slightly more than this, and spring/autumn rainfall by slightly less. Summer extreme rainfall depths are predicted to increase by between 0% and 10%.

Peak fluvial flows are anticipated to increase progressively during the twenty-first century. Eastern Scotland is expected to experience larger increases than north-west Scotland for example. The occurrence of snow and the associated contribution of snowmelt to both fluvial flow and groundwater are, on the other hand, predicted to decrease. Reductions in snowfall are predicted to be greater for the eastern and southern parts of Scotland and least for the central upland areas.

Changes in the factors discussed above, coupled with increased potential evapotranspiration, particularly in the summer, and a longer growing season – leading to increased root uptake – are expected to have substantial effects on soil moisture. The models predict a 10% to 30% decrease in soil moisture for summer/autumn and an increase of 3% to 5% in the winter. The winter figures reflect the fact that soils can only contain a finite amount of water and most Scottish soils are already close to saturation in the winter.

Reduced soil moisture during the summer and autumn months may mean that the short-term stability of some slopes formed from granular materials is enhanced by suction pressures (often described as negative pore water pressures). Soils under high levels of suction are vulnerable to rapid inundation, and a consequent reduction in the stabilising suction pressures, under precisely the conditions that tend to be created by such as short duration, localised summer storms. In addition, non-granular soils may form low-permeability crusts during extended dry periods as a result of desiccation. Providing that these crusts do not crack excessively due to shrinkage, then runoff to areas of vulnerable granular deposits may be increased. Such actions could lead to the rapid development of instabilities in soil deposits, potentially creating conditions conducive to the formation of debris flows. The complicating factors are the potential inability of current climate models to resolve storm events and the precise nature of the localised failure mechanisms that will lead to the initiation of any individual debris flow. It is highly unlikely that the measurement of soil suction could provide a practical and reliable means of debris flow forecast.

Vegetation will also be affected by climate change. Lower overall levels and changed patterns of rainfall might be expected to increase the pressure on vegetation and thus to reduce its beneficial effect upon slope stability. Additionally, extended periods of exceptionally dry weather could potentially lead to wildfires and associated debris flow such as those described by Cannon et al. (2008).

The importance of the potential effects of climate change impacts on slope stability is exemplified by the existence of an Engineering and Physical Sciences Research Council (EPSRC) Network: Climate change impact forecasting for slopes (CLIFFS) (Dixon et al., 2006). This is funded to provide a ‘talking-shop’ for such issues and to develop collaborative working arrangements to study such impacts and to develop coping strategies.

9.2.4 Mechanics of Unsaturated Slope Failure

That rainfall can cause landslides was dramatically demonstrated in February 2005 when catastrophic landslides occurred during intense rainfall in both California in the U.S. and British Columbia in Canada. Property destruction and tragic loss of life were the results of the various landslides. Over approximately a seven-month period, the Malibu area of California received and accumulation of over 585mm (23 inches) of precipitation. Then, in February 2005, the area received an additional 228mm (9 inches) over a period of about four days, at which time the landsides occurred (GeoSlope, 2005).

Analyses by GeoSlope, replicating the rainfall conditions experienced in California and British Columbia in February 2005 yielded some interesting results. The analysis confirmed that a typical model slope remained stable for seven months during which 585mm of cumulative rainfall fell but became unstable after a further 228mm over a period of four days. In general, the failure could not be attributed to increased positive pore water pressures as the failure surface did not penetrate below the water table. GeoSlope attributed the failure to decreases in suction. This type of behaviour corresponds well with that predicted from unsaturated soil mechanics theory (Wheeler et al., 2003) and the broad style of this type of failure mechanism is supported by experiment (Springman et al., 2003).

9.5 GENERALISED FORECASTING METHODS

Caine (1980) and Innes (1983) attempted to empirically quantify the amount of rainfall required to initiate debris flow events. Caine (1980) suggested a threshold for debris flow initiation, based upon worldwide data, albeit predominantly from North and South America, could be expressed in terms of a limiting curve, below which debris flow activity is unlikely to occur:

where I is the rainfall intensity (in mm/hour) and D is the duration of rainfall (in hours).

Caine (1980) suggested that the relation in Equation 9.1A is valid for durations between 10 minutes and 10 days (i.e. across more than three orders of magnitude). It was acknowledged that snowmelt caused by rainfall could significantly increase the apparent rate of rainfall (by up to 4mm/hour) rendering the relation invalid.

A second relation was proposed in the paper, but no description of its use was given. It is essentially an upper bound curve to the lower bound curve of Equation 9.1A. Its potential use is not immediately apparent, but it is reported here for completeness:

Innes (1983) developed a similar (lower bound) curve illustrating the rainfall amount-duration relation that has been reported as triggering a debris flow:

where T is the total rainfall in the period (in mm) and D is the rainfall duration (in hours).

Debris flows in Scotland indicate that anything between 10mm to 75mm of rainfall per hour may be required to initiate these flows, the latter value being significantly in excess of that predicted by the equations developed by Caine (1980). Current annual rainfall in Britain ranges from 1,000mm to 5,000mm (Met Office) and, therefore, these figures represent significant amounts of rain falling in a short time. An early warning system in California suggests that for a rainfall of approximately 15mm per hour, the threshold time for the onset of mud/debris flows varies from 8 to 14 hours depending on slope angles and available material (Bryant, 1991).

Therefore, in the context of antecedent and storm event rainfall triggering landslides, the equations presented above will not provide a complete solution to the identification of likely periods of debris flow activity.

9.4 CASE STUDIES

The review contained in Appendix G highlights a wide range of geographical areas in which landslides are caused by rainfall. Many of these areas are the subject of studies that include the back analysis of rainfall and other records in an attempt to define the levels of rainfall which provide conditions likely to lead to landslides; these are summarised in Table 9.1. While relatively few studies report on the active forecasting of the conditions likely to lead to landslides, many authors of such studies state that the methodologies that they have produced either could be, or will be, used for such purposes. In short the back analysis of such work is widely reported while its use for actual forecasting is less so.

A wide range of methodologies is used in the back analyses, however, these are dominated by intensity-duration analyses which appears to be a viable and well-established way forward. However, an alternative in the form of an analysis of the percentage of mean annual precipitation (MAP) versus duration also shows some promise. This appears to be only a very slight modification of the intensity-duration approach but has the apparent potential advantage of being normalised for local precipitation conditions. A much wider-ranging review of the different approaches to the development of rainfall thresholds is presented by a group of Italian researchers (Anon, 2007a). They identify 16, often subtly, different approaches to the development of rainfall thresholds for landslides in North and South America, Europe, Asia, Africa, and Indonesia and Oceania.

The intensity-duration pattern of each storm during, or immediately after, which landslides have occurred can be analysed. For each storm/landslide event a series of points can be plotted on a graph with intensity on the y-axis and duration on the x-axis. Different durations can be analysed to determine their associated intensities for a given storm/landslide event, giving multiple data points representing multiple durations. Further events can then be analysed in the same fashion. If the same durations are used in the analysis, a series of vertical columns of data points will result, each one representing landslide events corresponding to varying storm intensities at a given duration. The lower boundary of these data points then represents the storm threshold for rainfall-induced landslides.

It is important to note that most analyses consider both storm and antecedent rainfall, the latter usually for periods between five and more than 40 days. Those geographical areas in which antecedent is not considered, or considered over shorter periods, tend to be those in which storm rainfall is particularly intense and/or geological and geomorphological conditions favour the rapid onset of instability. It has proved feasible to plot the intensity-duration relations derived for a number of different areas on the plot presented in Figure 9.3.

Table 9.1 – Summary of landslide forecasting/event causation methodologies.

Country/Region	Data Used in Analysis	Analysis	Rainfall type
Australia	Rainfall records/ rainfall and other site- based monitoring	Intensity-duration: back analysis leading to forecast	Storm and antecedent
Hong Kong SAR	Rainfall monitoring (extensive)	Intensity-duration: forecast	Storm (24 hours)
Italy, NW Tuscany	Rainfall records	Rainfall intensity-time: back analysis	Storm and antecedent
Italy, Sarno	Rainfall records	Cumulative rainfall-time: back analysis	Storm and antecedent
Italy, W Liguria	Rainfall records	Intensity-duration: back analysis	Storm and antecedent
Italy, Piedmont	Rainfall records	Intensity/normalised intensity-duration: back analysis	Mainly storm
Italy, Dolomites	Rainfall records	Intensity/normalised intensity-duration: back analysis	Storm and antecedent
Jamaica	Rainfall records	Intensity/normalised intensity-duration: back analysis	Storm and antecedent (for shallow and deep-seated landslides respectively)
Nepal	None	None	Mainly storm
Norway	Rainfall records	Intensity/normalised intensity-duration: back analysis	Storm and antecedent
Singapore	Rainfall records	1 day versus 5 day rainfall	Storm and antecedent
Slovenia	Rainfall records	Recurrence duration	Storm and antecedent
Switzerland	Rainfall records	Intensity-duration: back analysis	Storm and antecedent
United Kingdom, NW England	Rainfall records/site- based rainfall monitoring	Various intensity-duration figures: back analysis leading to forecast	Storm and antecedent
United Kingdom, SW England	Rainfall records/soil moisture deficit	Percentage of long-term average in a period: back analysis leading to forecast	Storm and antecedent
United Kingdom, Scottish Highlands	Rainfall records/ river gauging	14 day cumulative rainfall	Storm and antecedent
USA, California	Rainfall records	Intensity-duration relation: back analysis	Mainly storm, although antecedent acknowledged as important
USA, Washington State	Rainfall records/site-based rainfall monitoring	Intensity-duration/antecedent water index: back analysis	Storm and antecedent

The back analysis of Scottish events was therefore taken forward on the basis of an analysis of both intensity-duration and of a form of normalised intensity-duration, in this case intensity/MAP-duration.

9.5 A TRIGGER THRESHOLD FOR SCOTLAND

9.5.1 Introduction

In terms of forecasting conditions potentially leading to debris flow, the current rainfall gauge network in Scotland is sparse in most of the areas of interest. In addition, the rainfall radar system covers some of the areas of interest at a resolution of 2km, but most are at a resolution of just 5km. Accordingly, data are not available on a routine basis for the key areas of interest.

Figure 9.3 – Comparison of landslide trigger thresholds based on rainfall intensity-duration for different parts of the world.

At the time of writing, two rainfall gauges are to be installed, as part of a trial, on land adjacent to the A83 between Ardgarten and Cairndow. Experience indicates that this area is probably one of the most active debris flow areas in Scotland; certainly it is one of the areas of the major road network most frequently affected by such events. Whilst a requirement for planning permission has created significant delay, it is hoped that the installation of these rain gauges will be completed in time to be functional during the landslide seasons of 2008: July to August and November to January (Winter et al., 2005a).

A back analysis, using analytical techniques to retrospectively examine historical radar data obtained in the lead up to known landslide events, has been undertaken and is reported in Section 9.5.2 to 9.5.4. The events to be studied encompass a wide geographical area and a diverse range of geological settings⁶. The data are used to develop a preliminary threshold based upon rainfall in terms of rainfall intensity-duration. The data have also been analysed with the intensity normalised for mean annual precipitation (MAP); this is intended to allow possible further comparison with threshold data produced for other regions of the world.

An additional element to this approach can be to undertake the same type of analysis for storm events that do not trigger landslides (Winter et al., 2007c). This allows the threshold to be defined from below as well as from above, lending an additional degree of surety to the process. Figure 9.4 illustrates the development of a purely hypothetical threshold in this manner. It should, however, be noted that, while the approach is sound, it is difficult to justify the expenditure of resources to analyse ‘non-events’. Accordingly, no such analysis has been undertaken for this purpose.

Figure 9.4 – Development of a purely hypothetical rainfall intensity-duration threshold for landslides.

The hypothetical rainfall data of Figure 9.4 have been utilised to develop three (also hypothetical) threshold levels:

• A threshold level above which landslides might be expected to occur.
• A lower threshold level at which a warning could be issued and action taken. This is set at this lower level so as to give adequate lead-in time for notifications and actions to be effective.
• A still lower threshold level is set at which instruments are checked and key personnel alerted to the possibility of the development of conditions likely to lead to landslides. This would be a precursor to the issue of a landslide warning.

It is important to recognise that threshold levels developed in this way are in no way absolute. They may simply represent the transition between the landslide density and/or short-term frequency in a given area reaching a limit that is significant in the context of infrastructure operation, for example. This transition is likely to be more complex in larger areas of varied and complex geology such as Scotland.

Observation of debris flow events in the trial area will allow further development and/or validation of the preliminary threshold (as developed in Section 9.5.4), by analysing the rainfall data collected in the lead-up to the events. Furthermore, it may also be feasible to analyse the rainfall leading up to storms that do not lead to debris flow events. Thus, the threshold above which landslides may be expected to occur can be defined using data points lying both above and below it which improves confidence in its accuracy.

Once sufficient confidence in the threshold has been established, the objective is to introduce the system as routine in forecasting debris flow events in this area. Essentially, the forecast will be used as the detection element of the DNA (detection-notification-action) sequence in the management procedures described in Section 8 of this report.

Provided that the system proves successful, it is envisaged that the system be introduced to other areas prone to rainfall-induced debris flows.

9.5.2 Threshold Development

The work to develop a rainfall threshold for debris flow potential was assisted by a sub-contract to the UK’s Meteorological Office to examine rainfall conditions at specific locations where landslides disruptive to the road system have occurred. A set of 16 events of known location and for which an apparently robust estimate of timing was available were selected for the study, as follows:

1. October 2001: Stromeferry Bypass, Loch Carron.
2. January 2003: Rest and be Thankful, Glen Croe.
3. November 2003: Rest and be Thankful, Glen Croe.
4. January 2004: Rest and be Thankful, Glen Croe.
5. February 2004: Laide, Wester Ross.
6. August 2004: Glen Kinglas
7. August 2004: Cairndow.
8. August 2004: Glen Ogle, Lochearnhead.
9. August 2004: Dunkeld, Perthshire.
10. August 2004: Pitcalnie, Nigg, Easter Ross.
11. August 2004: Eathie, Black Isle.
12. October 2004: Avoch-Fortrose, Black Isle.
13. December 2004: Cnoc Fhionn, Shiel Bridge-Glenelg.
14. January 2005: Letterfinlay, Loch Lochy.
15. September 2005: A87 Junction, Inverinate-Morvich.
16. September 2005: Kylerhea Glen, Skye.

The locations of the analysed landslide events and the rain gauges used in the analyses are illustrated in Figure 9.5, along with the coverage from Met Office weather radar installations.

9.5.3 Objectives

The overall objective of the work described in this section and Section 9.5.4 was to investigate the pattern of rainfall events associated with landslide occurrences and to analyse the data for both short duration and extended antecedent periods, in order to test analytical methods that could have an application to forecasting similar events in the future. The following objectives were thus set, as follows:

• To extract comprehensive data sets of rainfall from rain gauge and radar sources for each of the 16 events.
• To analyse the data in order to make four graphical representations for each of the 16 events:

i) Cumulative rainfall over an extended antecedent period (up to 150 days).

ii) Storm rainfall, presented as accumulation and intensity for a period of 18-24 hours leading up to the time (if known) of the landslide occurrence.

iii) The relation between intensity (mm/hr) and duration for the combined storm and antecedent periods.

iv) The relation between rainfall intensity as a function of mean annual precipitation (Intensity/MAP) and duration of the storm and antecedent period.

• To compare all of the intensity-duration relations for individual events and also on the basis of temporal and geographical spread.
• To prepare a spreadsheet for analysis of future events (‘Future Back Analysis’), based on the methods for data manipulation and analysis above.

Figure 9.5 – Location of landslides, rain gauges and radar (prepared by P Dempsey and J Dent of the Met Office).

The form of the analyses, the results for individual events and associated discussion are given in Appendix H. The following section details combinations of intensity-duration data from, the 16 analysed events, which may be used to develop event thresholds.

9.5.4 Results

Intensity-duration plots for different combinations of the 16 events analysed are presented in Appendix H. These include plots for Events 1 to 8 and Events 9 to 16 in order that the data for each event to be seen more clearly. The results are discussed in more detail in Appendix H.

In terms of understanding the important issue of how rainfall may cause debris flow, groups of plots for intensity-duration are presented for summer and winter events and also for events that occurred in the eastern and western parts of Scotland. While the details of the data are discussed in Appendix H, perhaps the most important observation is that all of the data sets broadly occupy the same space on the intensity-duration diagrams and that there is thus no compelling case for different thresholds for summer and winter events or for events that occur in the east and west of Scotland.

It is thus appropriate to combine all of the intensity-duration data for the 16 events onto a single diagram (Figure 9.6).

Figure 9.6 – Combined plot of intensity versus duration data for the 16 analysed debris flow events.

As might be expected there is a considerable amount of scatter in the data. However, the key point is that once certain ‘outlying’ data points are removed from consideration (Appendix H), then a reasonably clear tentative trigger threshold can be drawn (Figure 9.7). The blue crosses on Figure 9.7 represent data that are considered to be ‘outliers’ and as a result were not considered in the formulating the tentative threshold (illustrated in Figure 9.7 by red dots connected by a red line).

Clearly there is an issue as to how such a threshold may be used in so far as that illustrated in Figure 9.7 is tentative, requiring validation from future events. Indeed such work is ongoing, concentrating upon the Rest and be Thankful area as previously described.

Figure 9.7 – Tentative trigger threshold for Scottish debris flows in terms of intensity-duration.

Once validated, however, and with any suitable adjustments made it should be possible to set both ‘Wake-Up’ and ‘Warning’ thresholds, as described in Section 9.5.1 (Figure 9.4). There does, of course, remain the question as to how these thresholds should be operated.

In viewing the data there is a tendency to view the data from a position equivalent to time, t=0 (although as the scale is logarithmic this is simply a very low number). In real terms this, of course, corresponds to the time of the actual event. Therefore, the data must be viewed as if from a point in time in advance of the event.

It is suggested that the ‘Wake-Up’ be viewed from a point of view of three days (t=36 hours) in advance of any potential event, with the ‘Warning’ being viewed from the point of view of half of one-day (t=12 hours) and actual event threshold being observed from the point of view of a very short time in advance of any actual event. These viewpoints are illustrated in Figure 9.8 (which is based upon Figure 9.4) with the threshold observation point being set at time, t=6 hours. These timings have been assigned very much on an initial basis and require further work prior to the finalisation of a fully-developed threshold suitable for implementation.

Early testing of the threshold has been undertaken using the results of an analysis of the storm that led to the debris flow event at the A83 Rest and be Thankful on 28 October 2007 (see Section 2.3). The analysis was performed in precisely the same manner as those analyses reported earlier and which facilitated the development of the tentative rainfall intensity-duration threshold (Figure 9.7). The important difference is, however, that the October 2007 analysis was carried out after the tentative threshold had been determined, thus providing some degree of validation to the threshold. The threshold and the new data are both illustrated in Figure 9.9. It can be clearly seen that for the major part of the precursor period the data plot well above the tentative threshold; only during the last two hours before the landslide event do the data plot below the threshold. This may mean that by a point in time two hours before the event the rainfall had been sufficient to cause the debris flow to be inevitable. The break in the data at 288 hours (12 days) is also interesting. This coincides with a change in the slope of the threshold and may imply that this approximates to the longest period before the event that is significant; it may thus be reflective of the true limit of the antecedent period. This is, however, a very tentative conclusion and needs to be verified or otherwise by further data sets (Winter et al., 2008).

Figure 9.8 – Hypothetical rainfall intensity-duration threshold for landslides, illustrating observation points for the different thresholds, described in Figure 9.4 may be used.

Figure 9.9 – Tentative trigger threshold for Scottish debris flows in terms of intensity-duration showing the back analysis of the October 2007 event at the A83 Rest and be Thankful.

9.6 SUMMARY

Clearly weather and climate are key influences upon the triggering of debris flows in Scotland. In addition climate change models generally indicate a potential for such events to become more frequent and/or more intense in the future.

The ability to forecast conditions during which such events may occur is important now, but is likely to become still more important in the future. Such forecasts potentially enable the Detection and Notification of such events prior to their occurrence. In some circumstances it may even be appropriate to take action in advance of such events.

A wide variety of international approaches to the back analysis and forecast of landslide events resulting from rainfall has been studied. It was found that, back analyses to determine the relations between rainfall and debris flow events are relatively common. However, the implementation of practical systems to forecast the likelihood of debris flow events occurring seems to be relatively rare – albeit with notable exceptions.

A tentative debris flow trigger threshold, in terms of rainfall intensity-duration, has now been developed for Scotland. This threshold needs to be tested against observations in the future to validate it use prior to its implementation as a management tool. Notwithstanding this, the first test of the threshold (in the form of the October 2007 event at the A83 Rest and be Thankful) indicates that it has the potential to be successful. Work is ongoing to capture and analyses further such data for the purposes of validation.

A series of high quality data sets from a variety of geographical locations will be needed in order to validate and/or modify the threshold prior to its introduction to the management of the road network in any formal sense. Further data will also be required to enable the limit of the antecedent period of rainfall that influences the formation of debris flows. Given the frequency of such major events in Scotland it is estimated that this process may take of the order of approximately five years.

During this five year period there is a need to develop a system to allow the ‘real-time’ capture and analysis of appropriate rainfall data, including forecast rainfall data, to enable the forecast of potential debris flow events. This work could be most effectively taken forward in collaboration with the Met Office as both expertise in meteorology and landslides is required.

Once confidence in the threshold has been established working simulations and trials of its use should be conducted. This would enable lower thresholds for ‘Wake-Up’ and ‘Warning’ thresholds, as described in Figure 9.4 for Figure 9.8, to be set. It would also enable firm rules for the use and operation of the threshold to be set, again as described in Figure 9.8.