A modified version of HBV-3 model has been used since 1981 for operational spring-time forecasting in Finland. This model is ready now for eleven river basins ranging from 500 to 30,000 km2. From the experience with these models in forecasting we have learned to know the common difficulties and errors in real-time forecasting and some methods to improve forecasting results.

The main difficulties with forecasting during spring time in Finnish conditions have been: unstable parameters in snowmelt models, heavy unforecasted rainfall, evaluation of areal water equivalent of snow and ice-dams in rivers.

The methods to overcome these difficulties are: the development of more physically based snowmelt models, new methods to determine initial value of areal water equivalent of snow, the use of statistical error models with hydrological model, the automatization of observation and transport of hydrological and meteorological data and the improvement of areal precipitation calculations.