The objective of this research is to evaluate and compare the impact of the 3DVAR and 3DEnVAR methods on short-term and very short-term fog forecasting, applied to Short-range Forecast System (SisPI). The experiments were developed using the initialization of 00:00 UTC on December 30^th and 31^th, 2019, where there was a continuous event of dense fog recorded in the region of Havana, Artemisa and Mayabeque. The research combines data related to conventional observations contained in prepbufr format, radiance information from microwave sensors available in bufr format, and observations from the KBYX and KBMA radars that include measurements of radial velocities and reflectivity, something that constitutes a novelty at the national level in relation to fog investigations in the country. As a whole, domain-dependent covariance matrices (BECs) are used, generated with the inclusion of hydrometeors as additional control variables. Given the high degree of subjectivity inherent in the registration of fog and haze in the conventional stations of the study area, the binary analysis uses the data of the present time code with the visibility predicted by the model, which is obtained through an empirical algorithm. The results suggest that the 3DVAR method leads to improve the CSI values by only 1% and the correct detections by 2%. These discrete values respond to a limited modification of the background field as a consequence of an inadequate dispersion of the impact of the observations on the domain. 3DEnVAR generates a more realistic analysis field compared to 3DVAR and achieves a more efficient spread of the impact of observations over the domain. The CSI values manage to be up to 17% higher than the runs without assimilation and 11% at 3DVAR.