Page 3 - Volume 4, Issue 1, January 2011

INRA, UMR 211 INRA AgroParisTech 78850 Thiverval-

Grignon, France

e-mail:

david.makowski@grignon.inra.fr

Hellenic Plant Protection Journal

4:

1-11, 2011

REVIEW ARTICLE

Uncertainty and sensitivity analysis for models used in pest

risk analysis

D. Makowski

Summary

Quantitative models have several advantages compared to qualitative methods for pest

risk analysis; quantitative models do not require the definition of categorical ratings and can be used to

compute numerical probabilities of entry and establishment, and to quantify spread and impact. How-

ever, quantitative models include several sources of uncertainty that need to be taken into account by

risk assessors. In this paper, we review the four main sources of uncertainty in models used for pest risk

analysis, namely input variables, parameter values estimated from expert knowledge, parameter val-

ues estimated from data and equations. We discuss the practical interest of uncertainty and sensitivity

analysis for pest risk assessors. Uncertainty analysis consists in describing the different uncertain ele-

ments of a model, and deducing an uncertainty distribution for each output variable rather than a sin-

gle value. The aim of sensitivity analysis is to determine how sensitive the output of a model is with re-

spect to elements of the model which are subject to uncertainty. Uncertainty analysis typically com-

prises three main steps: i) definition of uncertainty ranges and/or of probability distributions for uncer-

tain model elements, ii) generation of values of the uncertain model elements, iii) model output com-

putation and description of model output distribution. Sensitivity analysis includes another step to

compute sensitivity indices (step iv). When several model equations are available for predicting a given

quantity of interest, a further step is to analyse uncertainty about model equations using specific tech-

niques. Several methods were illustrated in a case study on

Sclerotinia sclerotiorum

. Results showed

that a moderate uncertainty on parameter values can induce a large uncertainty on model output.

Additional keywords

: biological invasion, model prediction, model selection

Introduction

Risk analysis includes a series of steps from

initiation, throughqualitative or quantitative

assessments of risk, to the resultant man-

agement decisions. It also includes com-

munications with stakeholders throughout

the process. In plant health, Pest Risk Anal-

ysis (PRA) consists of the assessment of the

probabilities of entry and establishment of

an invasive species, the magnitude of the

impact resulting from an invasive species,

and of management options. Both quanti-

tative and qualitative methods have been

used for PRA. Qualitative methods based

on scoring systems are a primary choice for

assessing risk in plant health but, in several

cases, quantitative models have been devel-

oped and used in PRA (e.g. Stansbury

et al

.,

2002; Peterson

et al

., 2009).

Qualitative methods for PRA are based

on categorical ratings (e.g. low, moderate,

high) and the use of such ratings may lead to

problems of consistency due to inaccurate

definitions of ratings. Qualitative methods

also make the computation of an overall risk

level difficult because categorical ratings

can be combined using many different tech-

niques which may lead to different conclu-

sions (Holt, 2006). In addition, the perform-

ance of qualitative PRA methods depends

on the technique chosen for combining cat-