In the event of an accidental atmospheric release of radionuclides from a nuclear power plant, accurate real-time forecasting of the activity concentrations of radionuclides is required by the decision makers for the preparation of adequate countermeasures. Yet, the accuracy of the forecast plume is highly dependent on the source term estimation. Inverse modelling and data assimilation techniques should help in that respect. In this presentation, a semi-automatic method is proposed for the sequential reconstruction of the plume, by implementing a sequential data assimilation algorithm based on inverse modelling, with a care to develop realistic methods for operational risk agencies. The performance of the assimilation scheme has been assessed through the intercomparison between French and Finnish frameworks. Three dispersion models have been used: Polair3D, with or without plume-in-grid, both developed at CEREA, and SILAM, developed at FMI. Different release locations, as well as different meteorological situations are tested. The existing and newly planned surveillance networks are used and realistically large observational errors are assumed. Statistical indicators to evaluate the efficiency of the method are presented and the results are discussed. In addition, in the case where the power plant responsible for the accidental release is not known, robust statistical tools aredeveloped and tested to discriminate candidate release sites.