1. Comparing a multi-channel geostationary satellite precipitation estimator with the single channel Hydroestimator over South Africa
Estelle de Coning
South African Weather Service
In collaboration with EUMETSAT (Marianne Koenig)

2. Measurement of precipitation
Rain gauges
Radars
Satellite based estimates of rainfall

3. Satellite QPE in SA
Need is for nowcasting (0-12 hours) thus Geostationary satellite - MSG
The Hydroestimator (HE)
A single channel (IR10.8) of MSG to estimate rainfall based on cloud top temperatures.
Operational over southern Africa since 2007 using the local version of the Unified Model as numerical weather prediction input to the algorithm.
The Convective Rain Rate (CRR)
The EUMETSAT Nowcasting Satellite Application Facilities (SAF) has developed a precipitation estimator called the Convective Rainfall Rate (CRR)
Information on convective and the associated stratiform precipitation.
CRR makes use of either two (IR108 and WV062) or three MSG channels (VIS006 during day light hours).
Not operational, in research mode for case studies

4. Hydroestimator
HE considers some corrections taken from numerical weather prediction models input (Unified Model):
Precipitable water: rain is enhanced/reduced in high/low PW
Relative humidity: reduces rain in dry lower atmosphere
Actual level of neutral buoyancy is accounted for
Interaction of wind with topography

5. CRR CRR uses either 2 or 3 of the MSG SEVIRI channels:
Two dimensional matrices with IR108 and (IR108 – WV062)
Three dimensional matrices with IR108, (IR108 – WV062) and VIS006
IR-WV brightness temperature difference is a useful parameter for extracting deep convective cloud with heavy rainfall.
CRR cases done using UM (instead of ECMWF) as NWP input for “corrections” such as the moisture correction, the cloud top growth/decaying rates or evolution correction, the cloud top temperature gradient correction and the orographic correction.

6. Case study comparison
6 Dec summer convection (widespread heavy falls)
19 Dec – summer convection (wide spread heavy falls)
24 Jan summer convection with tropical showers, (wide spread heavy falls)
26 Feb summer time convection
2 Mar autumn convection with isolated heavy falls
18 April 2010 – autumn convection with isolated heavy falls
19 April autumn convection with isolated heavy falls
1 May 2010 – winter showers (isolated moderate falls)

7. Case 1: 19 Dec 2009

8. Blue – 1 mm
Red – 10 mm
Green 20 mm

10. Case summary Summer time convection with wide spread heavy falls
Intensity - HE overestimates more than CRR
Spatially – all three did well
1, 10 and 20mm: Scores very similar, HE slightly better
10mm threshold had best scores
Rainfall close to coast not captured

11. Case 2: 24 Jan 2010

14. Case summary HE better than CRR for all thresholds
Spatially – handled well
Intensity – overestimated by all
Rainfall close to coastline not captured by any of the algorithms

15. Case 3: 1 May 2010

18. Case summary
HE overestimates area and intensity slightly, CRR underestimates both
1 mm threshold: Very similar, CRR3D better than HE, HE better than CRR2D
10 mm threshold: HE good, CRR no rain more than 10 mm while gauges had 25 to 30mm

21. Summary and conclusion
Initial tests using eight cases with different types of precipitation do not conclusively show that the CRR methodology is better than the HE algorithm.
Reasons:
“calibration matrices” which was set up over Europe and might not be applicable or suitable over southern Africa.
channel difference which is used for estimating cloud depth (IR108 – WV062) is proving to be ambiguous. Setvak (2012) argues that the Brightness Temperature Difference (BTD) between these two channels can indeed reach its maxima approximately above the coldest pixels, but in many other cases the BTD maxima are elsewhere than above the coldest parts of the storm top. “This BTD should thus be used with caution, as it does not have to be related to precipitating clouds only.”
New version of the CRR - in 2013, methodology has been adjusted to exclude the use of calibration matrices and rely more on cloud physical properties – shows improvement over Europe compared to the current method used in CRR.