Uncertainty propagation in models driven by remotely sensed data

Often, little importance is given to the problem of how uncertainty propagates in models driven by remotely sensed data, and what the effects of uncertainty might be on the output of these models. In this paper, a general procedure to support a characterisation of uncertainty in the generation of remote sensing (RS) products is proposed. The procedure can be used with models characterised by any degree of complexity and driven by any kind of data. It provides two useful tools to analyse models: uncertainty analysis (UA), which allows the assessment of the uncertainty associated with model output, and sensitivity analysis (SA), which is useful to determine how much each source of uncertainty contributes to model output uncertainty. Uncertainty modelling, i.e. finding suitable tools to represent uncertainty, is a key step in performing UA and SA. A general error model for quantitative raster data is described. Different applications of UA and SA are proposed, and, in the last part of the paper, an example of UA and SA on a model for burned area detection is discussed.