chap3

CHAPTER THREE

HAZARD FORECASTING AND RISK ASSESSMENT

Certainty and probability

Uncertainty surrounds all aspects of natural hazards. Particularly important are the actual uncertainty of the environmental processes and events, and our perception of uncertainty in the absence of perfect knowledge about how the environment operates. The assessment and management of environmental risk are areas of growing importance.

Some events (such as volcanic eruptions and river floods) appear to be largely random in timing and occurrence, whereas others (such as prolonged drought) might appear to be cyclical, although they are not so regular as to be predictable. Yet other events (such as tornadoes and hurricanes) follow seasonal patterns of occurrence, even though their spatial patterns will vary.

However, all natural hazards pose some degree of uncertainty about where and when the next occurrence can be expected, and how large that event might be. We can never be certain (except in the short term for certain types of hazard), so that people are often unprepared when the event does happens. As a result contingency plans (like evacuation), designed to minimise risk to people and property, are often not put into practice.

Uncertainty can be expressed as a probability function - that is, in terms of the probability (or likelihood) that a hazard event of a given size will occur at a given place within a specified time period. Estimates of the likely occurrence of events in the future have to be based on experience of past events; the past is the key to the future, in hazard terms. If there is information available about past events, it can be used to predict the likelihood of hazard occurrences in the future, based on probability estimation.

Detailed forecasting of when the event is likely to occur is much more elusive. It is not unknown for forecasts to turn out wrong; the Meteorological Office issued incorrect short-term forecasts about the severe storm over southern England on 16th October 1987 and people were caught unawares. Nonetheless a number of hazards (like river floods) are understood enough to allow warnings to be given, even if only a day or two before the event. Some hazards (like weather changes and air quality) can be anticipated as they start to occur, and appropriate remedial action taken or started.

Hazard prediction; trend projection

Long periods observations (such as climatological records) can be used to detect periodicities or long-term trends in hazard events. These can then be used to anticipate future events by extrapolation (projecting past trends into the future). For example, the Icelandic volcano Hekla appears to have erupted roughly every seventy years since the sixteenth century. The most recent eruption, in 1991, is entirely consistent with this long-term pattern.

Simple extrapolation is possible only if we can safely assume that no progressive or sudden changes occur through time in the nature, timing and periodicity of the hazard events. This is clearly not always a safe assumption; for example, upstream land-use changes can radically alter the magnitude and frequency of river flooding.

Hazard prediction; magnitude/frequency analysis

The laws of probability tell us that the commonplace event is frequent and moderate in size, whereas extreme events are big but rare. Thus quite high discharges which just overtop river banks occur fairly regularly (about once in 1.5 to 2.3 years on average in many natural rivers), but severe overbank flooding or prolonged drought are less common. Thus the magnitude (size) and frequency (regularity) of events are closely inter-related for any given type of hazard (Figures 4a and 4b).

Figure 4. Magnitude/frequency relationships in natural hazards.

Large events are infrequent but catastrophic, whereas most hazards are of moderate size and common occurrence.
Some natural hazards such as floods and droughts form a continuum of events, with catastrophes occurring with both extremes: normality in this case occurs frequently, and relates to moderate size events.
Hazards vary in the probability distribution of events; hazard B has high frequency of relatively small events (eg river flooding), whereas hazard A has much larger events with similar probabilities.
Probability distributions of many natural hazards can be plotted on special probability graph paper to allow extrapolation or prediction of events beyond the range of observed conditions. The graph relates to river flooding, and the points represent the annual peak flows from the streamflow records. They are plotted after ranking by size, according to the formula R.I.=(n+1)/m, where n is the number of years on record and m is the rank of the flood (flood 1 being the largest flood in that record).

The nature of this relationship becomes clearer if we look at the probability distribution of hazard events, based on past experience or records (Figure 4c). The long-term average time interval between two successive hazard events of a similar size is known as the return period, or the recurrence interval (R.I.), of an event of that specified size. Thus, for example, a particular site on a river may have an estimated R.I. of 10 years, for a discharge of 200 m³s^-1 (cumecs). This means that over the long term the river can be expected to flow at this discharge once every ten years, on average, at this site. The return period is a long-term average based on past events; it is not a definite prediction of when the next event will occur.

Realistic probability estimates can only be made within the range of conditions experienced in the past, so the most reliable estimates are derived from long periods of records. But in many situations such records simply do not exist, so probability estimation and rational design of control structures often have to be done with caution and professional judgement.

Probability distributions of natural environmental processes (like river flooding; see Figure 4d) are needed in the design of control structures (like river walls, bridges and dams), which must be large enough and strong enough to protect people and property. Structures are usually designed to contain an event of a stated recurrence interval. For example, major dams in the headwaters of rivers upstream of towns and cities might be designed to withstand a 1 in 500 or 1 in 1,000 year flood. Complete protection against a hazard is unattainable, and control structures must be designed to balance cost against benefit. Even protection against the 20 or 30-year event may be uneconomic. Some rare events are worth massive investment; the Thames flood barrier, designed to protect London against the 1000-year storm surge, cost over £400 million!

The magnitude/frequency approach to prediction highlights two important aspects of the hazard problem. First, all occurrences of a particular hazard at a given location are, in effect, part of a continuum of events which follow a pattern which can be described in probability terms. The corollary is that an event of greater size than all recorded occurrences of a given hazard at a given place is likely to happen sometime in the future (ie "the best (or worst) is yet to come!"). For example, before the 1984 explosion at the Union Carbide plant at Bhopal in India (which killed more than 2,500 people), the largest toxic gas disaster anywhere in the world had killed 60 people.

Hazard forecasting

Ability to forecast the arrival time, location and potential magnitude of hazardous events is clearly necessary if disasters are to be avoided and emergency plans implemented. But such forecasting is generally possible only over short time periods, and for certain types of hazard.

Forecasts of river flooding can be made if information on the upstream generation of floodwaves is available, and it can be communicated downstream efficiently. Telemetric systems, based on river gauging stations which can be interrogated by telephone or satellite from water authority offices, are being installed in many developing countries. Limited prediction and forecasting of earthquake events and volcanic eruptions (such as Mount St Helens in 1980) is also possible, based on monitoring changes in the frequency of small seismic events, tilts and strains in epicentre regions, and changes in the physical properties of rocks near faults as they are strained.

Ideally we would like to be able to predict three things about a likely hazard event - where, when, and how big it might to be. Place is relatively easy to predict for many hazards, both natural and technological. Major earthquake and volcano zones are well known from past experience and geological mapping; floodplain topography and past flood history gives many clues to areas prone to river flooding; industrial accidents are concentrated in major industrial areas. Time is much more difficult to predict, except for events with some detectable periodicity, or short-term predictions when precursors (initial diagnostic stages) are evident. Size is more difficult again to predict, although precursor conditions can throw some light on the probability of events of a given size.

Risk and uncertainty

The UK Health and Safety Executive (1989) make an important distinction between risk and hazard. Risk refers to the likelihood (probability) of a harmful event, such as injury or death from a particular hazard; hazard refers to a situation with a potential to cause harm (regardless of the likelihood of it actually happening).

Considerable uncertainty surrounds any forecasting of rare events, for various reasons. Often reliable data are lacking. Proper evaluations of the health risk to humans is only available for a tenth of the 65,000 chemicals and 18 percent of the drugs in common use today, for example. Uncertainty also arises through the difficulty of identifying all possible causes and consequences of particular hazard events. This is particularly true for technological hazards, where possible failure rate data are limited if not unavailable; human error - which cannot be forecast - might be involved (as in the Chernobyl nuclear disaster in 1986, which was caused by the failure of an unauthorised experiment in the reactor core); and possible injury assessments are difficult (it depends on precisely what happens in the accident).

The Health and Safety Executive also distinguish between individual risk and societal risk -

Individual risk relates to the likelihood that a particular person may be harmed by a particular hazard. It is usually expressed as the probability that a typical person may be harmed in one typical year by that type of hazard. For example, the risk of being killed in a road accident in Great Britain is around 100 per million (or 10,000 to one) per year, on average. This can be compared with some other typical risks shown in Table 4.
Societal risk refers to the likelihood of major events involving mass death and injury. For instance, there is a risk of about 1 in 10 in any one year of a major air travel disaster in Britain. Societal risk is closely related to the prospect of catastrophes.

Table 4 Risk assessments for Great Britain

Figures are quoted as the chance in a million that a typical person will die from that cause in any one year, averaged over a whole lifetime; the figures are averaged over the whole population of Great Britain except where there is a specific small group exposed (eg rock climbers).

Cause	Risk (per million per year)

all causes (mainly illnesses from natural causes	11,900
cancer	2,800

(a) all violent causes	396
road accidents	100
accident in private home	93
fire or flame (all causes)	15
drowning	6
excessive cold	8
lightning	0.1

(b) accidents at work (risks to employees)
Deep-sea fishing (UK vessels)	880
coal extraction and manufacture of solid fuels	106
construction	92
all manufacturing industry	23
offices, shops, warehouses inspected by Local Authorities	4.5

(c) leisure (risks to active participants during active years)
rock climbing (assumes 200 hours climbing per year)	8,000
canoeing (assumes 200 hours per year)	2,000
Hang-gliding (average participant)	1,500

SOURCE: Health & Safety Executive (1989)

Catastrophes

The most extreme hazard events create catastrophes, or disasters, which normally arrive without warning. White and Haas (1975) define a catastrophe as any situation in which the damages to people, property or society in general are so severe that recovery and/or rehabilitation after the event is a long and difficult process.

The magnitude/frequency associations of hazard events mean that catastrophes must - by definition - be expected to occur at some time wherever minor events are experienced.

Catastrophic situations are often associated with flooding, hurricanes, tornadoes, tsunamis, volcanoes, earthquakes and large fires (see Table 1). Other processes (like landslides) generally affect smaller areas and have only moderate catastrophe potential. The same is true of drought, which can cover a massive area but generally allows fairly long warning times (although prolonged drought, as in north Africa during the 1980s, can create a major catastrophe). Processes like coastal erosion, frost, lightning strikes and expansive soils have low catastrophe potentials (Table 1).

People often view the sudden, dramatic disaster - with its immediate impact - differently from the ongoing natural hazard which might well kill or injure many more people over a longer period of time. Disasters are usually focussed on one community, which experiences collective shock and sense of loss (such as Lockerbie in Scotland, after the plane crash in 1988 and the 1989 Hillsborough football stadium disaster which affected Liverpool more than anywhere else).

Many disasters (like the 1988 Armenian earthquake) capture news headlines, and thus public interest and concern, around the world. Most disasters are followed by detailed enquiries into causes, and public and official demands that action be taken to reduce such risks. Such collective actions distinguish genuine societal risks from other types of risks (such as road accidents), even if the latter cause more deaths in a typical year (see Table 4).

Some complications

The concept of uncertainty in hazardous events is important, and different types of hazard follow different probability distributions. Most natural hazards (especially the geophysical ones) can be described by probability functions of the type shown in Figure 4d. However, such an approach is not suitable for technological hazards (like an explosion at a chemical plant, or the release of radiation from a nuclear plant), where the continuum concept of occurrence is not appropriate; events like these are discrete in timing and location. The approach is not appropriate either in situations where the environmental systems which give rise to hazard events are themselves changing; examples include the long-term changes associated with global warming, and changes in fault line dynamics and risk distributions associated with major expansions of human populations (eg the San Andreas fault).

It must be stressed that magnitude, threat and impact of individual hazards are not the same thing. A hazard event might be large and cause much environmental change, but cause little damage to people or property. For example, a severe earthquake in a sparsely populated area poses less threat than a small earthquake centred on a populated area (like Tokyo city). Even a relatively small events (like a typical river flood) may have lasting and wide-ranging effects in a heavily populated area.

With advance warning potential it is possible to reduce injury and loss of life by evacuation and contingency planning. The Biblical tale of Noah's Ark and The Flood, for example, illustrates how advance warning allows forward planning and promotes initiatives to minimize hazard threat and impact.