This thesis deals with the problem of including the effects of moist convection in the atmosphere in a large scale numerical forecast model and how to interpret and evaluate the predicted convective precipitation.

Two methods for parameterizing convection in a large scale numerical forecast model are tested to find the most appropriate one for precipitation forecasts in the mid-latitudes. The methods are called "penetrative convection" (Ooyama 1964, Kuo 1965) and "moist convective adjustment" (Manabe et. al. 1965). Advantages and disadvantages of the methods are discussed. Forecasts produced by "penetrative convection" show the best agreement with observations.

However, comparisons between observations and predicted precipitation are difficult since the numerical model produces mean values over areas that are large compared to the convective elements. The local variations of the observed precipitation are therefore considerable in these areas. One way to include these variations in the forecasts of precipitation is to formulate the forecasts in terms of probability of precipitation. Therefore, a model to predict the probability of a specified amount of accumulated precipitation in a point at the earth surface in a given period of time is developed. The model utilizes variables calculated by the large scale model and its parameterization schemes. It is based on the assumption that the convective cells have a characteristic precipitation rate, horizontal dimension and life time. If these are not known from the scheme used for parameterizing convection, fixed empirically deduced values are used. Experiments are performed that show that the sensitivity of the model to such prescribed values is moderate.

The probability model has been used in a large scale model and experimental forecasts for two synoptic situations have been carried out. The forecasted probabilities are compared to observations and found to be in good agreement with observed variations in time. The absolute values and the variations in space show less good agreement. Possible reasons for this discrepancy are discussed.