An interesting statistical paper on UAP was published in 2015 on the arXiv.org  University site.  Michaël Vaillant, a CNES/GEIPAN collaborator and two other persons from the Toulouse School of Economics, issued a serious analysis on the French UAP cases from the GEIPAN database. It should be recalled that GEIPAN is still currently the only official organism in the world in charge of collecting and assessing the UAP testimonies, and in a spirit of transparency makes available on line its information and updated statistics. http://www.cnes-geipan.fr/index.php?id=196

In that respect, the latest available yearly statistics show a reduction of the number of testimonies reaching their center during the last years: http://www.cnes-geipan.fr/?id=198.

 
2014= 368 ; 2015= 394; 2016= 126; 2017 (beg. November)= 42 (unless delay in processing files).

It is interesting to read in this paper about a strong relationship with the sites of nuclear activities and contaminated land. This analysis covered a total 1969 UAP cases, covering the period 1951-2013, and at that time in 2015 some 381 observations were still classified non-identifiable, hence falling under the category D. Already at the time it was stressed that the level of UAP classified as UAP D had substantially decreased during the previous 10 years.

Link to the online paper: https://arxiv.org/abs/1509.00571

“…We model the unidentified aerial phenomena observed in France during the last 60 years as a spatial point pattern. We use some public information such as population density, rate of moisture or presence of airports to model the intensity of the unidentified aerial phenomena. Spatial exploratory data analysis is a first approach to appreciate the link between the intensity of the unidentified aerial phenomena and the covariates. We then fit an inhomogeneous spatial Poisson process model with covariates. We find that the significant variables are the population density, the presence of the factories with a nuclear risk and contaminated land, and the rate of moisture. The analysis of the residuals shows that some parts of France (the Belgian border, the tip of Britany, some parts in the SouthEast , the Picardie and Haute-Normandie regions, the Loiret and Correze departments) present a high value of local intensity which are not explained by our model.”