Forecasting hazards, averting disasters

Table 1    Methods for integrating hazard, vulnerability and exposure information to predict impact

---

14

Box 2    Selecting hazard triggers

Hazard triggers are established and agreed in 

advance, based on forecasts or measurements. 

For flood-related disasters, for example, 20mm of 

rainfall within a specific time period could be a 

trigger to initiate a set of action(s). The threshold 

is referred to as the ‘danger level’ of interest. Often 

multiple increasing thresholds are selected to trigger 

different levels of action (e.g. a forecast of 10mm 

or 20mm or rain will trigger amber or red alerts). 

However, this simple approach is complicated 

because forecasts are inherently uncertain, and so 

are often expressed in a probabilistic form (for 

example, there is a 30% probability of exceeding 

the threshold of 20mm of rainfall; see Section 2.1). 

In these systems, triggering actions requires defining 

both the danger level threshold (e.g. 20mm of rain) 

and the probability of an occurrence of that danger 

level in the forecast (e.g. a 30% probability). Both 

values have to be carefully selected so that actions 

are triggered with an acceptable level of frequency.

We can distinguish two types of trigger systems:

1.  Deterministic systems involving a single trigger 

(i.e. the danger level of some parameter), which 

can be applied to either a deterministic forecast 

(which provides a single predicted outcome) 

or, more usually, real-time monitoring of some 

precursor to disaster combined with biophysical 

information, e.g. upstream river flow or 

vegetation condition.

2. Probabilistic forecast systems, which require 

both a danger level and probability thresholds. 

For climate extremes, this would be an ensemble 

forecast system.

Many methods for integrating impact-relevant 

information overlap (for example, statistical modelling is 

just a more complicated version of the threshold method, 

and the qualitative method still requires some sort of 

threshold in order to start a discussion). The method 

selected is likely to be a function of several factors, 

including data availability, how well we understand the 

hazard–impact relationship (and if it is too complicated 

to model), whether unexpected events can sway the result 

and the scale of the hazard itself. The lead time of the 

hazard is also a factor: it might not be possible to use a 

complex qualitative method for a rapid-onset event like a 

flash flood. Finally, the characteristics of the infrastructure 

or the population at risk will also determine the kind of 

assessment method that can be used: the threshold method 

might be best for a specific situation, such as whether a 

particular wind speed will cause a bridge to collapse or a 

particular water level will cause a dam to burst.

The field is rapidly evolving due to advances in 

computing power and data availability. While all the 

methods reviewed here are being used more frequently 

than in the past, quantitative modelling is growing 

particularly rapidly. It is important to note, however, 

that the examples reviewed in this paper are all led or 

supported by international agencies. Low-income countries 

face significant cost and capacity limitations in developing 

impact models.