Conducting Environmental Impact Assessment in Developing Countries (UNU, 1999, 375 pages)
 5. EIA tools
 5.1 Impact prediction
 (introduction...) 5.1.1 Application of methods to different levels of prediction 5.1.2 Informal modelling 5.1.3 Physical models 5.1.4 Mathematical models 5.1.5 Modelling procedure 5.1.6 Sensitivity analysis 5.1.7 Probabilistic modelling 5.1.8 Points to be considered when selecting a prediction model 5.1.9 Difficulties in prediction 5.1.10 Auditing of EIAs 5.1.11 Precision in prediction and decision resolution

### 5.1.8 Points to be considered when selecting a prediction model

Many models used at varying levels of sophistication depend on the quality of information required to select suitable methods. With an increase in the sophistication, the complexity of model also increases. This leads to an increasing number of uncertain parameters. The error in estimating parameters is carried over through the model, thus producing a less accurate model.

Box 5.2 Modified mathematical model

 Modified mathematical model If the model calculations are linear, with the inputs only being multiplied by constants and being added or subtracted, it is possible to calculate the probability distribution of the outputs from those of the inputs. For example, suppose that a process has inputs a, b, and c, all normally distributed with means Ma, Mb, and Mc, and standard deviations Sa, Sb, and Sc. If these are combined to form: g = Pa · a + Pb · b + Pc · c, with Pa, Pb, and Pc being factors or numbers, the output g will also be normally distributed with mean: Mg = Pa · Ma + Pb · Mb, + Pc · Mc and standard deviation:

Table 5.1 Generation of random numbers for calculating means and standard deviations

 Set No. Input A Input B Output D Output E 1 82 79 26.1 12.0 2 62 107 26.1 12.1 3 74 93 26.4 12.1 4 70 95 26.1 12.0 5 79 101 27.4 12.2 6 67 121 27.6 12.1 7 65 117 27.2 12.2 8 70 92 25.9 12.0 9 88 105 28.4 12.4 10 73 85 25 12.0 Mean 26.7 12.1 Standard deviation 0.90 0.13

Where limited resources are available, decisions have to be made about information needs for different effects and therefore about the allocation of these resources between the different effects. If specially designed methods are already available this will help to reduce the use of resources.