In this report, a study of the exposure to electromagnetic waves at the level of population is proposed. This study is conducted in the case of a plane wave oriented frontally on anatomical models (phantoms) with an incident power of 1W/m2 and at the frequency of 2.1 GHz. For this purpose, the idea is to build a surrogate model of the WBSAR (Whole Body averaged Specific Absorption Rate) as a function of morphology. An identificationof morphological factor is performed using the limited number of phantoms (18phantoms)and different regressions. This analysis shows that to use only external morphological factors (height, weight ...) is not sufficient to build such surrogate model and that internal morphological factors (muscles, fat ...) are also important. The absence of statistical data on internal morphology conducts us to introduce some prior knowledge on these internal factors to study the exposure for a given population. Some parametric laws and Gaussian mixture are used to study their influence on the quantile of the WBSAR at 95 The lack of data on internal morphology led us to consider homogeneous phantoms instead of the heterogeneous one. In homogeneous phantoms, all the internal tissues are substituted by an equivalent tissue. These homogeneous phantoms allow us to overcome the influence of internal morphology for a WBSAR model and to use a morphing technique to increase the database of phantoms. To give an estimation of the WBSAR quantile at 95 %, a sequential experiment design is used. This approach is based on a parametric model and Bayesian inference. This experiment design allows us to estimate the quantile at 95 for a given population and an example of one corresponding phantom. Nevertheless, this approach does not allow one to estimate the whole distribution of the WBSAR for a given population. To give an estimate of this distribution a surface of response is established using polynomial chaos. This approach gives incoherent results with the physical phenomenon. To obtain coherent results, physical knowledge is introduced by variables changing in this responsesurface.
