Comparing the Multi-sensor Precipitation Estimates products and ground observation data in the western part of Georgia during a strong storm in June 2017.
11 October 2022
14 June 2017
By Elena Nikolaeva (Georgian National Environmental Agency Hydrometeorological Department) and Jochen Kerkmann (EUMETSAT)
Satellite Multi-sensor Precipitation Estimates (MPE) products are widely used in climate studies, numerical weather prediction, nowcasting and other applications. Therefore, it is important to estimate accuracy and limitations of satellite precipitation estimates.
The MPE algorithm assumption is that colder clouds are more likely to produce precipitation than warmer clouds. It uses available Meteosat Second Generation satellite images with 15 min temporal resolution.
On 15 June 2017 there were several heavy rainfall events in western parts of Georgia. The first started around 3am local time (Wed 14 June 23:00 UTC) and, according to ground-based observations data from the Zugdidi meteorological station there was 5mm of precipitation. The MPE product coincides with ground-based data for that time (Figure 1, see arrows).
The heavy rains started again at around 9am local (05:00 UTC) for around two hours, and bringing approximately 8mm. However, MPE products do not show precipitation for that time (Figure 1), indicating the heavier precipitation was actually more to the north of the observation site. According to the Zugdidi data the rain was heavier between noon and 3pm local time (11:00 UTC).
Peak rainfall values of more than 8mm were recorded around 5pm local time (13:00 UTC) (Figure 1) and there was local flooding north of the Zugdidi station.
However, there is a very weak signal and an extremely small area of precipitation in the MPE product for that time (Figure 2, a). At the same time, the Meteosat-8 High Resolution Visible RGB image shows most of the area was covered by thick clouds (Figure 2, b).
So, why are MPE products not always accurate about rain events? Probably the answer is not trivial and depends on several factors. One of the limitations of the MPE is that this product does not well describe mid-latitude frontal rain due to the satellite observation angle. Also, in these cases the calculation/estimation of the hourly accumulations of precipitation in the MPE and ground observation are not same.
When comparing MPE products with the products using the Cloud Physical Properties (CPP) algorithm developed at KNMI, it is obvious that CPP algorithm is better at estimating the precipitation from low and middle level warmer clouds, showing a fairly realistic precipitation rate in Georgia. The CPP precipitation rate from KNMI on 15 June 2017 at 13:00 UTC is shown in Figure 3.