In tropical regions (as well as in regions with heavy monsoon rainfalls), precipitation simulation is especially difficult. This is due to the high occurrence of precipitation events, with large volumes of rain in short times. Furthermore, convective precipitation as well as precipitation during thunderstorm events is more common.
Both precipitation forms can be very complex and locally highly inconsistent, and are therefore even more difficult to simulate precisely, compared with other forms of precipitation. The implication of this difficulty in precise simulation is that the precipitation volume in tropic regions generally is clearly underestimated (exceptions apply, which makes a systematic correction very difficult).
In mountainous regions, precipitation can vary on a very small scale. Small differences occur, which even a good calculation cannot track. In our European high-resolution domains, a special effect sometimes occurs in mountainous areas during precipitation: Precipitation that was displayed in the 7-day forecast disappears when the forecast time range is just 3 days.
This is because the weather models for the first 3 days have a grid cell size of 3km. However, the forecast days 4 to 7 are covered by a 12x12 km weather model. As a result, parts of the valley are in the same grid cell as the mountains. This can potentially result in precipitation in the valley, even though precipitation is only expected in the mountains. For the first 3 forecast days, however, the 3x3 km model is used, which is able to distinct between the valley and the mountains. As a result, precipitation forecasts for day 4-7 may disappear on day 3 for locations in the valley. We have not yet found an "easy" way to eliminate this effect, since extending the detailed forecast to 7 days would require too much computational effort, and "reducing" the precipitation for the (more densely populated) valley areas would reduce the precipitation forecast in the less populated, but still important mountain areas.
The dataset CHIRPS2 as well as the CMORPH are produced by geosynchronous satellites, which are placed over the equator, cycling around the Earth with the same speed as the Earth rotation. Therefore, their view at equatorial areas is precise and rectangular, while in higher latitudes, the satellite view must go through more atmospheric layers, at a smaller angle.
Therefore, the quality is best in equatorial areas and decreases steadily towards the poles. CHIRPS2 is only available for regions between 50°S till 50°N, and CMORPH only between 70°S-70°N. However, it is suggested that these datasets are used even more restrictively, from 45°S to 45°N or even between 30°S and 30°N only. From our most recent verification studies, it can be seen that for locations farther away from the equator, other precipitation datasets show the better accuracy.