In the French Mediterranean area where heavy precipitation events can yield devastating consequences, it is essential to obtain reliable estimates of the distribution of extreme precipitation at gauged and ungauged locations. Under mild assumptions, extremes defined as excesses over a high enough threshold can be modeled by the generalized Pareto (GP) distribution. The shape parameter of the GP which characterizes the behavior of extreme events is notoriously difficult to estimate. In regional analysis, the sample variability of the shape parameter estimate can be reduced by increasing the sample size. This is achieved by assuming that sites in a so-called homogeneous region are identically distributed apart from a scaling factor and therefore share the same shape parameter. A major difficulty is the proper definition of homogeneous regions. We build upon a recently proposed approach, based on the probability weighted moment (PWM) for the GP distribution, that can be cast into a regional framework for a single homogeneous region. Our main contribution is to extend its applicability to complex regions by characterizing each site with the second PWM of the scaled excesses. We show on synthetic data that this new characterization is successful at identifying the homogeneous regions of the generative model and leads to accurate GP parameter estimates. The proposed framework is applied to 332 daily precipitation stations in the French Mediterranean area which are splitted into homogeneous regions with shape parameter estimates ranging from 0 to 0.3. The uncertainty of the estimators is evaluated with an easy-to-implement spatial block bootstrap.