Precipitation forecast has been the most difficult variable to measure and to increase accuracy, and several methods are used for assessing accuracy under different circumstances. We used HSS (Heidke Skill Score) for 3-hourly precipitation events in this study. Individual numerical models (NWP) vary in their accuracy depending on verification parameter, region and season. A combination of several Models, Precipitation Measurements, and Machine learning algorithms is able to increase accuracy by more than 20% compared with individual models.

• Precipitation forecast accuracy can be measured with (Heidke Skill Score).
• The study includes measurements of 3-hourly precipitation events from >30'000 stations worldwide.
• Individual numerical models (NWP) achieve HSS around 0.3.
• An approach combining Multi-Models, Measurements, Statistics and A.I. increases the HSS to 0.4.
• Further increases seem possible with this methodology.