Low visibility events are sometimes associated with delays and accidents related with air and land transportation. An accurate forecast of low visibility events can help to reduce the economical and human life losses associated with this phenomenon. This work contributes to the improvement of visibility forecast proposing a dynamic-statistical model that generates probabilistic visibility forecasts. This model combines retrospective forecast generated with a global model and in-situ observations. The proposed model is used to generate probabilistic visibility forecasts for Ezeiza airport between December 1984 and January 2011. Results show that combining in-situ observations and numerical model outputs increases the skill of the probabilistic forecasts with respect to the probabilistic forecast that are based only on observations or only on numerical model outputs. Considering the dependence of systematic numerical model errors with the time of the year produce an additional increase in the forecast skill.